Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
HLA-mediated control and CD8+ T
cell response mechanisms in persistent
viral infections
A thesis submitted to
Imperial College London
for the degree of Doctor of Philosophy
by
Nafisa-Katrin Seich al Basatena
Department of Medicine
Imperial College London
St Mary’s Campus
September 2012
2
Acknowledgements
I would like to most cordially thank my supervisor Becca Asquith for her
constant help, guidance and support. She truly created a great environment for
advancing during the course of my PhD research. I have certainly learned a lot
from her that I will always value. I would also like to express my gratitude to
the Wellcome Trust for funding this work. Special thanks should also go to my
second supervisor Charles R. B. Bangham for many useful discussions and vital
input to my research. I am deeply grateful to all my colleagues in the Department
of Immunology and most of all to U. Kadolsky and M. Elemans for encouraging
me and sharing their insightful advice on many aspects of my work.
This project would not have been feasible without the data provided by our
collaborators Alison M. Vine, Koichiro Usuku, Mitsuhiro Osame, Chloe L. Thio,,
Jacquie Astemborski, Gregory D. Kirk, Sharyne M. Donfield, James J. Goedert,
Salim Khakoo, Mary Carrington, Nicole Klatt, Guido Silvestri and Emma Thom-
son; for this I am very grateful. Additionally, I would like to especially extend
my appreciation to S. Khakoo, M. Carrington, J. Trowsdale, J. Traherne, F.
Graw, R. Regoes, G. O’Connor and D. McVicar for very useful comments and
productive collaborations.
Finally and very importantly, I want to express my whole-hearted appreci-
ation for the genuine, very altruistic and all-around support of my partner, K.
Chatzimichalis and warmly thank my lovely parents and all my great friends for
always being there for me during these demanding three years.
... To my family and friends ...
... for always being there for me ...
... και πανω απ′ oλα στην αξǫχαστη γιαγια µoυ ...
Declaration of Originality
All the work presented in this thesis is the author’s own unless clearly stated in
the statement of collaboration and the relevant sections of the thesis.
Signature ..................
Date ..................
5
Statement of Collaboration
The work presented in Chapters 2 and 3 would not have been possible without
the experimental work and the datasets collected by our collaborators; HTLV-1
cohort: Alison M. Vine, Koichiro Usuku, Mitsuhiro Osame and Charles R. M.
Bangham and HCV cohort: Chloe L. Thio, Jacquie Astemborski, Gregory D.
Kirk, Sharyne M. Donfield, James J. Goedert, Mary Carrington and Salim I.
Khakoo. The cohorts were HLA and KIR genotyped for other studies and were
generously provided for further analyses. Additionally, Dr Aidan Macnamara
contributed a part of the epitope prediction results presented in Chapter 3. The
findings of this project were published in [1].
For the part of Chapter 3 that refers to the role of KIR3DS1 in HTLV-1
infection, the study was lead by Dr Geraldine O’Connor and Dr Daniel McVicar
and I contributed the statistical analysis for the Japanese HTLV-1 cohort. The
latter study was published in [2].
The study-design presented in Chapter 4 involves the analysis of experimental
data (a longitudinal HCV cohort) gathered by our collaborator Dr Emma Thom-
son. The KIR-typing was performed by Dr Susanne Knapp and the preparation
of the samples for next-generation sequencing was done by Dr Heather Niederer
and Aviva Witkover. The sequencing was performed by Niall Gormley and his
team at Illumina. The study design was part of a grant proposal submitted by
my supervisor Dr Becca Asquith to the MRC. The grant has now received full
funding in order to be implemented by our group.
The work presented in Chapter 5 was a collaborative effort with Dr Marjet
Elemans. The R code for fitting the models presented in this chapter was written
by myself and implemented by both myself and my collaborator. Other parts of
6
this study were done only by Dr Marjet Elemans and Dr Becca Asquith and are
not presented in this thesis. The experimental data obtained from SIV-infected
macaques were provided by Dr Nicole Klatt and Dr Guido Silvestri. All the
study results (including the part presented here) were published in [3].
Konstantinos Chatzimichalis provided valuable advice on coding in the C++
programming language for the 3D Cellular Automaton computational model pre-
sented in Chapters 6 and 7. Additionally, the rules for the motility of T cells
were provided in text format by Dr Frederik Graw and I implemented them from
scratch in C++. This work is currently submitted for publication.
7
Dissemination
Publications
1. Seich al Basatena N.K. and Asquith B., Is the probability of target recogni-
tion by CD8+ T cells low? (In preparation).
2. Seich al Basatena N.K., Chatzimichalis K., Graw F., Frost S. D. W., Regoes
R. R. and Asquith B., Can non-lytic CD8+ T cell responses drive viral escape?
(Submitted).
3. O’Connor G. M., Seich al Basatena N.K., Olavarria V., MacNamara A., Vine
A., Qi Y., Hisada M., Goedert J., Carrington M., Galvo-Castro B., Asquith B.
and McVicar B. W., In Contrast to HIV, KIR3DS1 Does Not Influence Outcome
In HTLV-1 Retroviral Infection, Hum. Immunol., 2012.
4. Elemans M., Seich al Basatena N.K. and Asquith B., Review: The efficiency
of the human CD8+ T cell response: How should we quantify it, what determines
it and does it matter?, PLoS Comput. Biol., 8(2), 2012.
5. Elemans M., Seich al Basatena N.K., Klatt. N., Gkekas C., Silvestri G. and
Asquith B., Why don’t CD8+ T cells reduce the lifespan of SIV-infected cells in
vivo?, PLoS Comput. Biol., 7(9), 2011.
6. Seich al Basatena N.K., MacNamara A., Vine A.M., Thio C. L., Astem-
borski J., Usuku K., Osame M., Kirk G.D., Donfield S.M., Goedert J.J., Bang-
ham C.R.M., Carrington M., Khakoo S.I. and Asquith B., KIR2DL2 enhances
protective and detrimental HLA-class I mediated immunity in chronic viral in-
fection, PLoS Pathogens 7(10), 2011.
7. Seich al Basatena N.K., Hoggart C., Coin L. and O’Reilly P.F., The effect of
inversions on population genetics methods of inference (In press, Genetics).
8
8. Valcarcel B., Wurtz P., Seich al Basatena N.K., Tukiainen T., Soininen P.,
Kangas A.J., Jarvelin M.R., Ala-Korpel M., Ebbels T.M. and de Iorio M., A
differential network approach to exploring differences between biological states:
an application to pre-diabetes, PLoS One 6(9), 2011.
9. Abebe T., Hailu A., Woldeyes M., Bilcha K.D., Cloke T., Fry L., Seich al
Basatena N.K., Corware K., Modolell M., Munder M., Tacchini-Cottier F.,
Mller I., Kropf P., Investigation of arginase activity and T cell function in pa-
tients with cutaneous leishmaniasis in Ethiopia, PLoS Negl. Trop. Dis. 6(6),
2012.
Presentations
1. Poster Presentation in Viruses, Genes and Cancer Workshop in Venice (Travel
award) - September 2010
2. Poster Presentation in British Society of Immunonology Congress in Liverpool
- December 2010
3. Oral Presentation in International Workshop: T lymphocyte dynamics in acute
and chronic viral infection in London - January 2011
4. Oral Presentation at Division of Infectious Diseases ’Away-Day’ in Imperial Col-
lege London - January 2011
5. Poster Presentation at Keystone meeting on HIV Evolution, Genomics and
Pathogenesis in Vancouver - March 2011
6. Poster Presentation at the European Mathematical Genetics Meeting at Kings
College London - April 2011
7. Poster Presentation at the 19th HIV Dynamics & Evolution International Con-
ference at Ashville, North Carolina - April 2012
9
8. Poster Presentation at Mathematical Immunology Affinity Group Meeting, Cam-
bridge - June 2012
Outreach activities
1. Project management team for the ‘Beautiful Science’ discussion and exhibition
events and leader of the outreach event, Imperial College and Wellcome Trust
People’s Award - Jul. 2011 - Jul. 2012
2. Project management team for the British Society of Immunology stand at the
Big Bang Science Fair, London Excel - Nov. 2010 - Mar. 2011
3. Co-author of the article ‘Immunology... Do the Maths?’ article featured at the
British Society of Immunology Newsletter - Jan. 2012
4. Volunteer at the Royal Society Summer Science Exhibition, Imperial College
and Exscitec - Jul. 2011
5. Soapbox Scientist at the Imperial College Festival discussing the topic: ‘Is science
too advanced to explain to the public?’ - June 2012
6. Volunteer at the ‘Greying Matters’ event at the Dana Center discussing with the
public what happens to our brains as we grow older - June 2011
7. Member of the ‘Reaching Further’ scheme that communicates science to stu-
dents. Developed together with other colleagues an activity that introduces
PCR to A-level students
10
Abstract
Background There are many viruses that result in persistent infections affecting
millions of people worldwide. Although our immune system deploys different strategies
to eliminate them, many times they prove unsuccessful and call for a better under-
standing of the host-virus interplay. One important weapon of the immune system
is CD8+ T cells which identify infected cells and limit the spread of infection using
different effector mechanisms.
Aim The aim of this study is two-fold; the investigation of 1) the impact of im-
munogenetic factors such as HLA class I molecules and Killer cell immunoglobulin-like
receptors (KIRs) on CD8+ T cell responses and 2) the efficiency of lytic and non-lytic
CD8+ T cell responses and how they shape viral escape dynamics.
Methods The methods used to address the aims include statistical models, high-
throughput sequence analysis, ordinary differential equation models and agent-based
models.
Results We find that HLA class I molecules explain a small percentage of the het-
erogeneity observed in the outcome of HCV, HTLV-1 and HIV infections. However,
we show that an inhibitory KIR, namely KIR2DL2, can enhance both protective and
detrimental HLA class I-restricted anti-viral immunity, for both HCV and HTLV-1 in-
fections and in a manner compatible with the modulation of CD8+ T cell downstream
responses. Furthermore, for HIV/SIV infection, we show that the CD8+ T cell control
of the infection can be consistent with a non-lytic mechanism. Additionally, we find
that lytic CD8+ T cell responses are more efficient than non-lytic responses which can
lead to slower and less frequent viral escape explained by spatial factors.
Conclusions We conclude that KIRs can play an important role in shaping HLA
class-I mediated immunity and suggest that this occurs in synergy with CD8+ T cells
whose lytic and non-lytic effector functions can differ in efficiency and lead to variable
viral escape rates.
11
Contents
Acknowledgements 1
Declaration of Originality 5
Statement of Collaboration 6
Dissemination 8
Abstract 11
List of Abbreviations 18
List of Figures 21
List of Tables 25
1 Introduction 27
1.1 HLA class I molecules . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.1.1 Antigen presentation . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.1.2 Polygenism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.1.3 Polymorphism . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.1.4 Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.1.5 Associations with viral infection . . . . . . . . . . . . . . . . . . 32
1.2 CD8+ T cell function in viral infection . . . . . . . . . . . . . . . . . . . 33
1.2.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
1.2.2 CD8+ T cell effector mechanisms . . . . . . . . . . . . . . . . . . 34
1.2.3 Evading CD8+ T cell responses . . . . . . . . . . . . . . . . . . . 36
1.3 Killer Cell Immunoglobulin-like receptors . . . . . . . . . . . . . . . . . 37
1.3.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
12
1.3.2 Associations with disease . . . . . . . . . . . . . . . . . . . . . . 38
1.4 Persistent viral infections . . . . . . . . . . . . . . . . . . . . . . . . . . 39
1.4.1 Hepatitis C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
1.4.2 Human T-Lymphotropic Virus 1 . . . . . . . . . . . . . . . . . . 42
1.4.3 Human Immunodeficiency Virus 1 . . . . . . . . . . . . . . . . . 45
1.5 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2 HLA class I impact on HCV and other viral infections 52
2.1 Aim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.3.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.3.2 Statistical analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.3.3 Epitope prediction . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.4 HLA class I and HCV infection . . . . . . . . . . . . . . . . . . . . . . . 66
2.4.1 Heterozygotes’ advantage . . . . . . . . . . . . . . . . . . . . . . 66
2.4.2 Rare allele advantage . . . . . . . . . . . . . . . . . . . . . . . . 73
2.4.3 Individual alleles . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
2.4.4 HLA class I breadth . . . . . . . . . . . . . . . . . . . . . . . . . 77
2.4.5 Protein specificity . . . . . . . . . . . . . . . . . . . . . . . . . . 79
2.4.6 Epitope specificity . . . . . . . . . . . . . . . . . . . . . . . . . . 83
2.5 Quantification of HLA class I impact on viral infection . . . . . . . . . . 85
2.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3 Impact of KIRs on HLA class I-mediated immunity 93
3.1 Aim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
3.3.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
13
3.3.2 Statistical analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.3.3 Epitope prediction . . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.4 KIR2DL2 in HTLV-1 infection . . . . . . . . . . . . . . . . . . . . . . . 99
3.4.1 Disease status . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.4.2 Proviral load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
3.4.3 HLA class I specificity . . . . . . . . . . . . . . . . . . . . . . . . 102
3.5 KIR2DL2 in HCV infection . . . . . . . . . . . . . . . . . . . . . . . . . 103
3.5.1 Spontaneous viral clearance . . . . . . . . . . . . . . . . . . . . . 103
3.5.2 Viral load in chronic infection . . . . . . . . . . . . . . . . . . . . 103
3.6 No KIR2DL2 main effect on outcome . . . . . . . . . . . . . . . . . . . 107
3.7 Linkage disequilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.7.1 Linkage between HLA class I alleles . . . . . . . . . . . . . . . . 111
3.7.2 Linkage between KIR genes . . . . . . . . . . . . . . . . . . . . . 113
3.7.3 Conclusions on LD . . . . . . . . . . . . . . . . . . . . . . . . . . 116
3.8 Canonical KIR-HLA binding . . . . . . . . . . . . . . . . . . . . . . . . 116
3.9 The role of KIR2DL2 ligands . . . . . . . . . . . . . . . . . . . . . . . . 118
3.10 The role of other KIRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
3.10.1 Other Inhibitory KIRs . . . . . . . . . . . . . . . . . . . . . . . . 118
3.10.2 Activatory KIRs . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
3.10.3 KIR haplotypes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
3.11 Potential mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
3.12 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
4 Do KIRs affect CD8+ T cell selection pressure? - Study design 131
4.1 Aim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.3 Description of study design . . . . . . . . . . . . . . . . . . . . . . . . . 133
4.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
14
4.4.1 Longitudinal HCV cohort . . . . . . . . . . . . . . . . . . . . . . 136
4.4.2 Predicted viral peptides . . . . . . . . . . . . . . . . . . . . . . . 136
4.5 NGS data post-processing . . . . . . . . . . . . . . . . . . . . . . . . . . 139
4.5.1 Error sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
4.5.2 Quality control . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
4.5.3 Sequence alignment and coverage . . . . . . . . . . . . . . . . . . 145
4.6 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
5 Lytic and non-lytic CD8+ T cell responses 153
5.1 Aim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
5.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
5.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
5.3.1 Experimental data . . . . . . . . . . . . . . . . . . . . . . . . . . 155
5.3.2 Lytic models of infection . . . . . . . . . . . . . . . . . . . . . . . 156
5.3.3 Non-lytic models of infection . . . . . . . . . . . . . . . . . . . . 158
5.3.4 Model fitting and selection . . . . . . . . . . . . . . . . . . . . . 159
5.4 Model comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
5.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
6 A Cellular Automaton model of CD8+ T cell responses 167
6.1 Aim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
6.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
6.3 Model description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
6.4 Model assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
6.5 T cell motility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
6.6 CD4+ T cell influx . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
6.7 CD4+ T cell reproduction and death . . . . . . . . . . . . . . . . . . . . 177
6.7.1 Reproductive rate . . . . . . . . . . . . . . . . . . . . . . . . . . 177
6.7.2 Death rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
15
6.8 CD8+ T cell effector function . . . . . . . . . . . . . . . . . . . . . . . . 180
6.8.1 Conjugate formation . . . . . . . . . . . . . . . . . . . . . . . . . 180
6.8.2 Target recognition by CD8+ T cells . . . . . . . . . . . . . . . . 181
6.8.3 Lytic and non-lytic CD8+ T cell viral suppression . . . . . . . . 181
6.9 Quantifying CD8+ T cell killing and viral escape . . . . . . . . . . . . . 186
6.9.1 Simulated killing rate . . . . . . . . . . . . . . . . . . . . . . . . 186
6.9.2 Mass-action killing model . . . . . . . . . . . . . . . . . . . . . . 186
6.9.3 Saturated killing model . . . . . . . . . . . . . . . . . . . . . . . 187
6.10 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
7 CD8+ T cell effector function and viral escape 191
7.1 Aim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
7.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
7.3 Lytic CD8+ response and viral escape . . . . . . . . . . . . . . . . . . . 193
7.3.1 The CA model captures HIV/SIV dynamics . . . . . . . . . . . . 193
7.3.2 Probability of recognition by CD8+ T cells and killing rate . . . 193
7.3.3 Higher CD8+ T cell killing leads to faster viral escape . . . . . . 196
7.3.4 CD8+ T cell killing rate is higher during ‘escape phase’ . . . . . 203
7.3.5 Saturation term better estimates killing rate . . . . . . . . . . . 206
7.4 Non-lytic CD8+ response and viral escape . . . . . . . . . . . . . . . . . 211
7.4.1 Non-lytic suppression leads to slower viral escape dynamics . . . 211
7.4.2 Cluster formation of infected cells . . . . . . . . . . . . . . . . . 218
7.4.3 The efficiency of the non-lytic control mechanism . . . . . . . . . 219
7.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
8 General Discussion 228
Appendices 236
A Appendix A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
16
A.1 The HCV genome and gene products . . . . . . . . . . . . . . . . 236
A.2 The HCV-1a protein sequence (isolate H77) . . . . . . . . . . . . 237
A.3 The HTLV-1 protein sequence . . . . . . . . . . . . . . . . . . . . 239
A.4 HLA class I associations with HCV infection outcome . . . . . . 242
A.5 Confounding factors in HCV infection . . . . . . . . . . . . . . . 244
A.6 The Explained Fraction depends on factor frequency . . . . . . . 244
B Appendix B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
B.1 False Discovery Rate of cohort stratifications . . . . . . . . . . . 246
B.2 Canonical KIR-HLA binding not supported by linkage effects . . 246
B.3 A*02 binds peptides strongly . . . . . . . . . . . . . . . . . . . . 251
C Appendix C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
C.1 Alignment to ambiguous reference sequence . . . . . . . . . . . . 254
D Appendix D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
D.1 AICc differences of the fitted models . . . . . . . . . . . . . . . . 257
E Appendix E . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
E.1 3D Cellular Automaton - Cell motility rules . . . . . . . . . . . . 258
E.2 Model quantities: look-up tables . . . . . . . . . . . . . . . . . . 260
F Appendix F . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
F.1 No significant effect of the grid size on the estimated escape rates 263
F.2 Killing rate increases with effector cells size . . . . . . . . . . . . 264
F.3 The percentage of simulations for each model that do not result
in viral escape . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
F.4 Quantifying the immune control of a non-lytic response that
blocks viral production . . . . . . . . . . . . . . . . . . . . . . . . 267
F.5 Equivalence of non-lytic models in chronic infection . . . . . . . 270
17
List of Abbreviations
ABMs Agent-based Models
ACs Asymptomatic Carriers
AIC Akaike Information Criterion
AICc bias-adjusted Akaike Information Criterion
AICD Activation-Induced Cell Death
ARF Alternative Reading Frame
ARVs Antiretroviral Drugs
ATL Adult-T cell Leukemia
CA Cellular Automata
DNA Deoxyribonucleic Acid
dpi days post infection
ELISPOT Enzyme-linked immunosorbent spot
ER Endoplasmic Reticulum
FDR False Discovery Rate
GWAS Genome-Wide Association Studies
HAART Highly Active Antiretroviral Therapy
HAM/TSP HTLV-I associated myelopathy/tropical spastic paraparesis
HCV Hepatitis C Virus
HIV-1 Human Immunodeficiency Virus 1
18
HLA Human Leukocyte Antigen
HTLV-1 Human T-Lymphotropic Virus 1
IFN Interferon
IL Interleukin
IS Immunological Synapse
KIRs Killer Cell Immunoglobulin-like Receptors
LIF Leukemia Inhibitory Factor
MΦs Macrophages
MDC Macrophage Derived Chemokine
MHC Major Histocompatibility Complex
mRNA messenger RNA
MTOC Microtubule-organising Center
NKs Natural Killer Cells
ODE Ordinary Differential Equation
OR Odds Ratio
PBMC Peripheral Blood Mononuclear Cell
pMHC peptide MHC complex
PVL Proviral Load
R0 Basic Reproductive Ratio
RBV Ribavirin
19
RN Reticular Network
RNA Ribonucleic Acid
SDF-1 Stromal Cell-Derived Factor-1
SIV-1 Simian Immunodeficiency Virus 1
SNP Single Nucleotide Polymorphism
STLV-1 Simian T-Lymphotropic Virus 1
SVR Sustained Virological Response
TAP Transporter associated with Antigen Processing
TCR T Cell Receptor
TNF Tumour Necrosis Factor
VL Viral Load
VPA Valporic Acid
20
List of Figures
1.1 Antigen presentation by HLA class I molecules to CD8+ cytotoxic T
lymphocytes (CTLs). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.2 Polymorphism of HLA class I genes. . . . . . . . . . . . . . . . . . . . . 31
1.3 Schematic outline of the thesis. . . . . . . . . . . . . . . . . . . . . . . . 51
2.1 Consistency of predictions between Metaserver and NetMHCPan. . . . . 65
2.2 Predicted HLA class I breadth in HCV (genotype 1a) infection. . . . . . 71
2.3 Predicted HLA class I breadth in HTLV-1 infection. . . . . . . . . . . . 72
2.4 HLA class I associations with HCV infection outcome in UK, USA and
pooled cohorts. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
2.5 Binding preference of HCV proteins for protective and detrimental HLA
class I molecules. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
2.6 Quantification of the HLA class I impact on persistent viral infections. . 88
3.1 The protective effect of binding HBZ in HTLV-1 infection is enhanced
by KIR2DL2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.2 The impact of HLA-B*57 and KIR2DL2 on HCV viral load . . . . . . . 108
3.3 KIR2DL2 enhances HLA class I-mediated antiviral immunity . . . . . . 109
3.4 Potential mechanisms of KIR2DL2 impact on HLA class I-mediated
immunity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
4.1 Experimental setup for obtaining the NGS HCV data . . . . . . . . . . 134
4.2 The HCV amplicons sequenced in this study. . . . . . . . . . . . . . . . 135
4.3 Pipeline for post-processing the HCV NGS data in silico . . . . . . . . . 135
4.4 A longitudinal cohort of HCV/HIV co-infected individuals . . . . . . . . 137
4.5 Predicted HCV epitopes for viral proteins NS3, NS5A and NS5B. . . . . 138
4.6 Errors arising during the NGS data generation. . . . . . . . . . . . . . . 140
21
4.7 Base calling quality control (Filter a) . . . . . . . . . . . . . . . . . . . . 142
4.8 Base calling quality control (Filter b) . . . . . . . . . . . . . . . . . . . . 143
4.9 Optimal base calling quality filter . . . . . . . . . . . . . . . . . . . . . . 144
4.10 Adapter sequence bias . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
4.11 Sequence variation of 78 HCV genotype 1a strains . . . . . . . . . . . . 147
4.12 Coverage of HCV amplicons . . . . . . . . . . . . . . . . . . . . . . . . . 149
4.13 Coverage per HCV amplicon . . . . . . . . . . . . . . . . . . . . . . . . 150
5.1 Best fitting lytic and non-lytic models to experimental data. . . . . . . . 163
5.2 Model comparison using AICc. . . . . . . . . . . . . . . . . . . . . . . . 164
6.1 A snapshot of the 3D cellular automaton model . . . . . . . . . . . . . . 169
6.2 Cell populations considered in the 3D cellular automaton model . . . . . 169
6.3 The speed of CD8+ T cells in the 3D cellular automaton model. . . . . 172
6.4 The mean displacement of CD8+ T cells in the 3D cellular automaton
model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
6.5 The ‘probability of successful scan’ and CD8+ T cell speed . . . . . . . 174
6.6 The CD4+ T cell influx model . . . . . . . . . . . . . . . . . . . . . . . 176
6.7 The lifespan of wild-type infected cells . . . . . . . . . . . . . . . . . . . 179
6.8 Possible CD8+ T cell responses after the scanning time of the infected
target is completed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
6.9 Non-lytic models simulated with the 3D cellular automaton model . . . 184
6.10 Simulated secretion patterns of soluble factors . . . . . . . . . . . . . . . 185
6.11 Schematic of a model which includes saturated epitope-specific CD8+
T cell killing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
7.1 Dynamics of wild-type infected cells over a course of 150 days. . . . . . 194
7.2 Probability of successful scan by specific CD8+ T cells and killing rate . 197
7.3 Duration of CD8+ T cell lysis and killing rate . . . . . . . . . . . . . . . 198
22
7.4 Dynamics of wild-type and variant infected cells over a course of 150 days.200
7.5 Estimation of the variant escape rate. . . . . . . . . . . . . . . . . . . . 201
7.6 Higher CD8+ T cell killing leads to faster viral escape . . . . . . . . . . 202
7.7 CD8+ T cell killing rate is higher during ‘escape phase’ . . . . . . . . . 204
7.8 The absolute number of wild-type infected cells killed per day is constant205
7.9 Estimation of the CD8+ T cell killing rate using a saturated killing term.208
7.10 The AICc of the fitted models over different simulated datasets . . . . . 209
7.11 Correlation of simulated and estimated killing rate using a mass-action
or a saturated killing term. . . . . . . . . . . . . . . . . . . . . . . . . . 210
7.12 Non-lytic suppression leads to slower escape dynamics . . . . . . . . . . 214
7.13 Number of uninfected cells protected’ from infection under a non-lytic
CD8+ T cell response . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
7.14 New infections prevented under a non-lytic CD8+ T cell response that
blocks infection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
7.15 Set-point of productively infected cells . . . . . . . . . . . . . . . . . . . 217
7.16 Formation of clusters of infected cells during infection spread . . . . . . 219
7.17 New infections prevented under a non-lytic CD8+ T cell response that
blocks infection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
7.18 Number of uninfected cells protected’ from infection under a non-lytic
CD8+ T cell response . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
7.19 New infections prevented under a non-lytic CD8+ T cell response that
blocks infection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
A.1 The HCV genome and gene products. . . . . . . . . . . . . . . . . . . . 236
A.2 Dependence of Explained Fraction on factor frequency. . . . . . . . . . . 245
B.1 A*02 is a strong epitope binder - Experimental data. . . . . . . . . . . . 252
B.2 A*02 is a strong epitope binder - Predicted data. . . . . . . . . . . . . . 253
C.1 Sequence alignment using ambiguous reference strain . . . . . . . . . . . 254
23
C.2 Coverage of HCV amplicons - ambiguous reference . . . . . . . . . . . . 255
C.3 Coverage per HCV amplicon - ambiguous reference . . . . . . . . . . . . 256
D.1 AICc differences between fitted models. . . . . . . . . . . . . . . . . . . 257
F.1 No significant effect of the CA grid size on escape rates . . . . . . . . . 263
F.2 Killing rate increases with effector population size . . . . . . . . . . . . 265
F.3 Number of infected cells ‘blocked’ from viral production under a non-
lytic CD8+ T cell response . . . . . . . . . . . . . . . . . . . . . . . . . 267
F.4 New infections prevented under a non-lytic CD8+ T cell response that
blocks production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
F.5 Set-point of productively infected cells under a non-lytic CD8+ T cell
response that blocks production . . . . . . . . . . . . . . . . . . . . . . . 269
24
List of Tables
1.1 Groups of HLA class I molecules which are known ligands for KIRs. . . 38
2.1 Evaluation of epitope predictors. . . . . . . . . . . . . . . . . . . . . . . 63
2.2 Impact of zygosity on HCV infection outcome. . . . . . . . . . . . . . . 69
2.3 Impact of supertype zygosity on HCV infection outcome. . . . . . . . . 70
2.4 Allelic distribution per HLA class I locus. . . . . . . . . . . . . . . . . . 70
2.5 Impact of HLA-A and B supertype frequency on HCV infection outcome. 74
2.6 The HLA alleles that are consistently associated with the outcome of
the HCV infection. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
2.7 Overall breadth of HLA-A, B, C molecules in HCV infection. . . . . . . 78
2.8 Breadth per protein of HLA-A, B, C molecules in HCV infection. . . . . 80
2.9 No predicted epitopes were consistently associated with the outcome of
HCV infection. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
2.10 Factors explaining part of the heterogeneity in HCV infection outcome. 86
2.11 Factors explaining part of the heterogeneity in HTLV-1 infection outcome. 87
2.12 HLA class I alleles that explain part of the heterogeneity in HIV-1 in-
fection outcome. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.1 KIR2DL2 enhances the protective effect of C*08 and the detrimental
effect of B*54 on HAM/TSP risk and, independently, on proviral load . 101
3.2 KIR2DL2 enhances the protective effect of HLA-B*57 on HCV status
and, independently, on HCV viral load. . . . . . . . . . . . . . . . . . . 105
3.3 KIR2DL2 enhancement of the B*57 protective effect in HCV infection
stratified by race . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
3.4 No KIR2DL2 main effect on status or viral burden . . . . . . . . . . . . 110
3.5 Linkage between the KIRs in HTLV-1 and HCV cohorts. . . . . . . . . . 114
25
3.6 No role for KIR3DS1 in HTLV-1 disease outcome . . . . . . . . . . . . . 121
3.7 No role for KIR3DS1 in HTLV-1 proviral load . . . . . . . . . . . . . . 121
3.8 The role of KIR haplotypes . . . . . . . . . . . . . . . . . . . . . . . . . 123
3.9 KIR-B haplotype enhancement is not seen in the absence of KIR2DL2 . 123
5.1 List of ODE models fitted to the experimental data. . . . . . . . . . . . 161
5.2 Comparison of ODE models of lytic and non-lytic control . . . . . . . . 161
A.1 List of HlA class I associations with HCV outcome reported in the
literature. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
A.2 Confounding factors in the HCV cohort. . . . . . . . . . . . . . . . . . . 244
B.1 False Discovery Rate of cohort stratifications. . . . . . . . . . . . . . . . 246
B.2 Impact of HLA-C1 alleles on HTLV-1 disease status and proviral load. . 248
B.3 Only the effect of B*57 and not of alleles with similar binding is en-
hanced by KIR2DL2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
B.4 KIRs which are known to bind B*57 do not enhance the B*57 protective
effect as much as KIR2DL2 does. . . . . . . . . . . . . . . . . . . . . . . 250
E.1 CA model initialisation values and motility parameters. . . . . . . . . . 261
E.2 CA model parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
F.1 Percentage of simulations not resulting in viral escape. . . . . . . . . . . 266
26
Chapter 1
Introduction
The present thesis studies the CD8+ T cell response to persistent viral infections.
We investigate the immunogenetics of persistent viral infections, focusing on the effect
of Human Leukocyte Antigen (HLA) molecules and Killer-cell Immunoglobulin-like
Receptors (KIRs) on the disease outcome. We also study the CD8+ T cell effector
mechanisms, focusing on the dynamics of lytic and non-lytic CD8+ T cell immune
pressure and how these relate to ‘viral escape’ dynamics. The persistent viral infec-
tions discussed in the thesis are HTLV-1, HCV and HIV. The methods used include
statistical models, high-throughput sequence analysis, ordinary differential equation
models and agent-based models.
1.1 HLA class I molecules
Human Leukocyte Antigen complex (HLA) is the name of the major histocompati-
bility complex (MHC) in humans1. It is a set of genes located at chromosome 6 and
constitutes the most gene dense region of the human genome. They are expressed on
all the cells of the body that have a nucleus.
HLA class I molecules are important in shaping the response of two different types
of immune cells: CD8+ T cells and Natural Killer (NK) cells. Specifically, they are
responsible for presenting antigen (e.g. viral peptides) mainly from the inside of the
cell to CD8+ T cells as opposed to MHC class II molecules which present extracellular
antigens via Antigen Presenting Cells (APCs) to helper CD4+ T cells. Additionally,
1In this thesis, since we predominantly focus on human disease, we use the terms HLA and
MHC interchangeably.
27
they tune Natural Killer cell responses by controlling their cytotoxic activity. NK cells
have both activating and inhibitory receptors on their surface and it is the balance
between activating and inhibiting signals that determines the response of NK cells to
infected cell targets. Many of these receptors are ligated by HLA class I molecules.
In this study, we primarily focus on the effect of HLA class I on CD8+ T cell
responses.
1.1.1 Antigen presentation
MHC class I molecules present endogenous antigens such as viral peptides (9-12 amino
acids long) on the surface of infected cell via an extensive process. During the synthesis
of the viral material, some of the peptides: 1) undergo cleavage by the proteasome in
the cytoplasm, 2) are transported by the Transporter associated with Antigen Process-
ing (TAP) heterodimer to the endoplasmic reticulum (ER), 3) are loaded on partially
folded MHC class I molecules and 4) are exported in a peptide:MHC (pMHC) com-
plex to the cell surface via the Golgi apparatus. The binding strength of the pMHC
complex is defined by the binding affinity (measured in the IC50 scale).
The viral peptides presented by the MHC class I molecules act as a signal to inform
the immune system that the cell has become infected (see Figure 1.1). CD8+ cytotoxic
T lymphocytes (CTLs) can receive this signal and eliminate the pathogen infected cell.
CTLs recognise both a part of the MHC class I molecule and the viral peptide bound
on it; a peptide that is recognised by the immune system is also called an epitope. If
the self-MHC is not recognised then the viral peptide is also not recognised and this
characteristic of the antigen presentation process is referred to as an MHC-restricted
immune response. Furthermore, an affinity threshold of approximately 500nM (prefer-
ably 50nM or less) has been shown to determine the capacity of the pMHC complex to
elicit a CD8+ T cell response [4]. Interestingly, significant functional differences have
been reported between CD8+ T cells recognising identical peptides but restricted by
28
different, albeit closely related MHC class I molecules [5], suggesting that the response
to antigen presentation is a complicated process controlled by multiple factors.
New virion
production
Virus specific
CD8+ T cell
TCR
viral peptide
Infected cell
HLA
Figure 1.1: Presentation of a viral peptide from HLA class I molecules to virus-specific
CD8+ T cells. The image is adapted from [6].
Epitope binding prediction
Peptide binding to MHC class I molecules is a key element in CD8+ T cell mediated
immunity. There are many available in silico models that attempt to predict which
viral peptides can bind specific HLA class I molecules and they are referred to as epi-
tope prediction algorithms. These algorithms exploit properties of the pMHC binding
complexes in order to obtain reliable predictions. The first epitope prediction models
were based on sequence motifs shared by experimentally defined complexes [7, 8]. As
the availability of data grew and quantitative measurements such as the affinity of the
pMHC complex became possible more complex algorithms were developed. These in-
clude position-specific matrices which assign different probabilities that specific amino
acids are found in different peptide positions [9], Hidden Markov Models [10] and ma-
chine learning approaches such as Artificial Neural Networks (ANNs) [11] and Support
Vector Machines (SVMs) [12]. The latter algorithms can capture the influence of the
sequence context on the binding contribution of a given amino acid in the binding pep-
tide [13]. More elaborate algorithms integrate several steps of the antigen presentation
29
process in their models such as proteasomal cleavage and TAP-transportation [14,15];
generated peptides need to be properly ‘chopped’ at the proteasome, loaded and car-
ried by the TAP complex before they can be displayed by the MHC class I molecules on
the surface. This process may result in a limited number of peptides that can actually
trigger a downstream response.
1.1.2 Polygenism
The MHC class I genetic locus consists of several different genes. Specifically, its
human version, the HLA class I gene family includes, amongst others, three highly
polymorphic bi-allelic loci: HLA-A, HLA-B and HLA-C2. The expression of these genes
is co-dominant. Every individual possesses a set of different HLA class I molecules
(minimum 3, maximum 6) with different ranges of peptide binding specificities.
1.1.3 Polymorphism
HLA genes are the most polymorphic genes in humans. For each of the different HLA
class I molecules there are multiple variants of each gene within the whole population
(see Figure 1.2). The fact that different people have different shapes (alleles) of the
HLA class I molecules and thus present different parts of the pathogen (peptides) to
the immune cells impacts on the effectiveness of an individual’s immune response.
Three models have been proposed to explain maintenance of HLA polymorphism in
the population [16]: (1) balancing selection, where alleles considered protective against
one disease can confer susceptibility to another, (2) heterozygote advantage, where a
higher HLA repertoire increases the breadth of peptide recognition and immune de-
fense against pathogens and (3) frequency-dependent selection, where a pathogen has
evolved to escape an efficient immune response driven by alleles commonly found in
the population but remains prone to responses mediated by rare, i.e. low-frequency,
2In this study when we refer to HLA class I genes we only consider the A, B and C loci.
30
HLA−A HLA−B HLA−C
HLA class I gene
Num
ber o
f alle
les
050
010
0015
0020
0025
00
1381
1927
960
Figure 1.2: Polymorphism of HLA class I genes (numbers based on hla.alleles.org)
alleles. Linkage association of neutrally selected elements to positively or negatively
selected ones also plays a role [16]. However, establishing the relative importance of
the three proposed mechanisms has been a challenging task. Interestingly, using a
mathematical model, it has been shown that the heterozygote advantage on its own is
insufficient to explain the high population diversity of the MHC molecules, even in a
very large host population, and a high degree of polymorphism is only achieved under
unrealistically similar allelic fitness contribution [17]. Using again a theoretical ap-
proach, another analysis has shown that provided a sufficiently large host population,
selection for rare MHC alleles driven by host-pathogen co-evolution can account for re-
alistic MHC polymorphism [18]. In a very recent study, the latter model was supported
by experimental data demonstrating that antagonistic co-evolution is a viable mecha-
nism explaining the evolution and maintenance of MHC polymorphism in vertebrate
populations given fitness trade-offs associated with pathogen adaptation [19]. This
mechanism predicts that MHC alleles that confer resistance to a subset of diseases can
lead to susceptibility to another subset as a natural consequence. Nevertheless, this
explanation is not excluding an additive effect from a heterozygote advantage-driven
31
selection pressure.
1.1.4 Nomenclature
The nomenclature of HLA alleles has recently been revised [20,21]. Currently, an allele
name may be composed of four, six or eight digits depending on its sequence. The
different pairs of digits are separated by colons. The convention is to use a four-digit
code to distinguish HLA alleles that differ in the proteins they encode. Alleles whose
numbers differ in the first four digits must differ in one or more nucleotide substitutions
that change the amino-acid sequence of the encoded protein. Based on hla.alleles.org
the following rules are used for naming HLA molecules:
• The 1st and 2nd digits describe the allele group.
• The 3rd and 4th are assigned to specific HLA proteins.
• The 5th and 6th digits distinguish alleles that differ only by synonymous
nucleotide substitutions within the coding sequence.
• The 7th and 8th digits discriminate between alleles that only differ by sequence
polymorphisms in non-coding regions.
1.1.5 Associations with viral infection
The important role of HLA class I molecules in viral infection has made them the
focus of many studies. Different HLA alleles have been significantly associated with
different disease outcomes in a range of infections including HIV- 1, HTLV-1 and HCV
(the ones studied in this project). Importantly, many of the reported associations
between HLA class I genes and the outcome of HCV, HIV-1 and HTLV-1 infection
have been observed in independent cohorts. Both the impact of individual alleles and
their homozygosity/heterozygosity status has been investigated. Several alleles have
been suggested by these studies to have a protective (e.g. A*02 in HTLV-1, B*57 in
32
HIV and HCV) or detrimental (e.g. B*35 in HIV, B*54 in HTLV, C*04 in HCV) role
in the progress of the infection [22–28]. HLA class I associations with disease outcome
have also been shown in other viral infections including malaria, HBV and dengue virus
infection [29–31] as well as autoimmune diseases such as ankylosing spondylitis [32].
1.2 CD8+ T cell function in viral infection
1.2.1 Background
CD8+ T cells can recognise the presence of antigen, in the context of MHC class I
molecules, via their T Cell Receptors (TCRs). Antigen presentation can trigger a cas-
cade of coordinated molecular interactions that may result in the limiting the infection.
CD8+ T cells primarily recognise endogenous antigen which is directly presented by
infected cells but they can also recognise exogenous antigen which is ‘cross-presented’
by other cells such as dendritic cells (DCs). Specifically, cross-presentation of antigen
by DCs can stimulate naive CD8+ T cells and induce their proliferation; however,
under normal circumstances it is probably less efficient than direct presentation, since
it requires the additional step of transfer from one cell to another [33].
There are many lines of evidence, whilst not all equally strong, demonstrating
the importance of CD8+ T cells in antiviral immunity. First, there are statistically
significant associations between specific MHC class I molecules and the outcome of
viral infection [22–28, 30, 31, 34]. Second, the existence of escape mutants associated
with CD8+ T cell pressure strongly implies their role in viral control [35, 36]. Third,
there are in vitro experiments indicating that CD8+ T cells can inhibit replication
[37,38]. Fourth, there is a temporal correlation between CD8+ T cell appearance and
vireamic control [39, 40]. Fifth, and perhaps most convincingly, the in vivo depletion
of CD8+ T cells, mainly in animal models, leads to higher viral burden in many viral
infections [40–44]. Sixth, the infusion of infected subjects with CD8+ T cells leads to
33
the dramatic, but transient, reduction of virus-positive cells [45].
1.2.2 CD8+ T cell effector mechanisms
CD8+ T cells can elicit two main effector mechanisms upon specific antigen recognition.
Although the manifestation of the two effector functions is different, they both can lead
to the limitation of infection.
Lytic function
The lytic function is the one that gives CD8+ T cells the name cytotoxic T cells. Under
a lytic mechanism the extracellular secretion of perforin [46], a pore-forming protein,
allows the breach of the membrane of the infected cell followed by the entrance of
granzyme A and B proteins [47] in the target cell cytoplasm which eventually prompt
the degradation of the cells DNA. Cytotoxic CD8+ T cells can also kill through the
Fas/CD95 pathway which requires neither calcium nor perforin [48]. Binding Fas, on
the infected cell, with its ligand (FasL), on the T-cell, activates the caspase cascade.
In culture, the granule pathway is the one that mainly destroys target cells and only
when this pathway is compromised there are significant levels of Fas-mediated killing;
however, the situation is more complicated in vivo [49]. Under both pathways, once
the infected cells undergo apoptosis, phagocytic cells can identify and ingest them.
Non-lytic function
The non-lytic CD8+ T cell function involves the secretion of non-cytolytic soluble
suppressor factors such as cytokines and chemokines. Several studies on different viral
infections have provided evidence supporting this effect. It has been shown that certain
immunoregulatory cytokines, including type I and II interferons (IFN-α,β and IFN-γ
respectively) and tumour necrosis factor (TNF) can activate a number of intracellular
pathways that directly suppress viral replication without killing the host cell [50–53].
34
Interferons have been shown to suppress hepatitis B virus (HBV), hepatitis C virus
(HCV), herpes simplex virus, vesicular stomatitis virus, vaccinia virus, picornaviruses,
retroviruses, influenza viruses and other types of viruses in vitro in a non-cytolytic
manner [50,54]. Interestingly, clinical studies have shown that HBV replication can be
suppressed by acute hepatitis A virus-induced production of soluble factors including
IFN-γ [55]. This suggests a non-specific ‘bystander’ suppression effect which can occur
in certain tissues and viral systems [50] and might act on a short but not necessarily
strictly adjacent distance.
In this thesis we particularly study the non-lytic CD8+ T cell response in the
context of HIV/SIV infection. It has been shown that in HIV infection, suppressor
activity can be mediated by diverse soluble factors [50]. Many studies have explored
the role of these soluble factors. 1) The ‘mystery’ suppressor named CD8 Antiviral
Factor (CAF) [56] was the first reported to be released by primary CD8+ T cells
upon activation in vitro. Studies indicated that an unknown factor is responsible for
CAF activity, but did not eliminate the possibility that it is a collection of known
antiviral cytokines with redundant functions [50]. 2) Secretion of IFN-γ by activated
T cells has been shown to inhibit HIV in macrophages and was associated with viral
suppression and a lack of disease progression [57]. 3) Activated CD8+ T cells or
PBMC obtained from HIV-infected individuals produced increased levels of MIP-1α,
MIP-1β and RANTES [58, 59]. These are CC-chemokines that bind to and activate
the chemokine receptor CCR5 blocking the entry of HIV-1 strains that use it as a co-
receptor (R5). 4) Other agents that have been studied as potential viral suppressors
are: IFN-α, Macrophage Derived Chemokine (MDC) , IL-13 and Leukemia Inhibitory
Factor (LIF) [60]. Depending on the secreted factor, CD8+ T cells can either block
viral production from infected cells or ‘protect’ uninfected cells from viral entry.
35
1.2.3 Evading CD8+ T cell responses
CD8+ T cells target epitopes derived from viral proteins and also cryptic epitopes
encoded by viral alternative reading frames (ARF). On the other hand, viruses ‘ex-
ploit’ opportunities and ‘formulate’ mechanisms in an attempt to avoid recognition
and elimination.
The main evasion mechanism employed to thwart CD8+ T cell activity is muta-
tional escape from antigen-specific responses [61, 62]. If a genetic mutation in a viral
peptide occurs during the antigen presentation process and abrogates its recognition by
a specific CD8+ T cell response then the variant viral strain might acquire a survival
advantage over the wild-type strain. However, the same mutation might also involve a
fitness cost for the virus. This cost usually occurs because of structural constraints in-
fluencing the kinetics of viral replication; specifically, conserved viral regions are often
considered vital to viral replication. The balance between the survival advantage and
the fitness cost will define the outgrowth rate of the variant strain over the wild-type.
A mutation that does not offer a survival benefit is expected to eventually die out
unless it is neutral and in linkage disequilibrium to one that does.
Interestingly, the loss of recognition of a specific epitope due to a viral mutation
can result in a new response against subdominant epitopes [63]. On the other side
however, if a mutation bares a fitness cost for the viral replication, compensatory viral
mutations that restore the viral replicative capacity can arise [64, 65].
Importantly, the mutations of the viral peptides are frequently associated with
the presence of specific MHC class I molecules. Identical MHC class I genotypes are
associated with very similar escape variants suggesting viral constraints [66]. The im-
portance of MHC-restriction in viral escape evolution is also supported by the reversion
to the wild-type strain in the absence of the restricting MHC class I molecule [67, 68]
albeit not always [69].
Apart from mutations that affect epitope presentation, there are other strategies
36
that viruses employ in order to avoid immune control. For example, in the case of
HIV-1, the viral accessory protein Nef has been shown to downregulate MHC class
I expression. Nef can downmodulate all HLA-A and HLA-B but not HLA-C and
HLA-E allotypes [70]. This selective downregulation may lead to decreased antigen
presentation to CD8+ T cells, whilst simultaneously avoiding NK-mediated killing by
retaining HLA-C and HLA-E expression [71].
1.3 Killer Cell Immunoglobulin-like receptors
1.3.1 Background
Killer cell immunoglobulin-like receptors (KIRs) are a family of transmembrane pro-
teins that are expressed on natural killer (NK) cells and subsets of T cells [72–74].
They are both polymorphic and polygenic and are found at chromosome 19. KIRs
bind HLA class I molecules and have both activatory (DS) and inhibitory (DL) iso-
forms [75]. For example, KIR2DL2 binds group C1 HLA-C molecules which have
asparagine at residue 80, and with a weaker affinity, group C2 alleles which have a
lysine at position 80 [76] while KIR3DL1 binds Bw4 alleles which are distinguished
from Bw6 alleles by the motif spanning amino acid positions 77-83 [77]. Although for
most of the inhibitory KIRs their ligands are known, this is not true for activatory
KIRs [78] (Table 1.1).
KIRs contribute both directly and indirectly to antiviral immunity. Directly, KIRs
on NK cells sense the loss of HLA class I molecules from the cell surface and trigger NK-
mediated cytolysis. Additionally, inhibitory KIRs expressed directly on T cells have
been suggested to increase cell survival by reducing activation-induced cell death [79,
80]. Indirectly, NK cells can regulate adaptive immunity via crosstalk with dendritic
cells and by the production of chemokines and cytokines [81, 82].
37
KIRs Known ligands
KIR2DL1 HLA-C1
KIR2DL2 HLA-C1/C2
KIR2DL3 HLA-C1
KIR3DL1 HLA-Bw4
KIR3DL2 HLA-A3
KIR2DS1 HLA-C1
KIR2DS2 HLA-C2
KIR2DS3 ?
KIR3DS1 ?
Table 1.1: Groups of HLA class I molecules which are known ligands for KIRs. In
this study we denote HLA-CAsp80 as HLA-C1 and HLA-CLys80 as HLA-C2.
1.3.2 Associations with disease
Early research on KIRs investigated protection and/or susceptibility for disease by
studying associations with KIRs in the context of their HLA class I ligands. In HCV
infection, homozygosity of KIR2DL3 and its HLA-C1 ligand has been associated with
viral clearance [83] while the HLA-Bw4I80/KIR3DS1 compound has been shown to
have a protective effect against the development of HCV-associated hepatocellular
carcinoma [84]. In HIV, the epistatic interaction of KIR3DS1 and HLA-B delays the
progression to AIDS [85] while various distinct allelic combinations of the KIR3DL1
and HLA-B loci have significant and variably strong influence both on protection
from progression to AIDS and plasma HIV RNA abundance [86]. In agreement with
the latter result, the presence of the inhibitory allele KIR3DL1 in combination with
the HLA-B*57 allele had a highly protective effect against progression to AIDS in
Zambian patients [84]. In chronic myeloid leukemia, KIR2DL2 and/or KIR2DS2 in the
presence of its ligand was found to be protective [87]. In [88], susceptibility to Crohn’s
disease is shown to be mediated by KIR2DL2/KIR2DL3 heterozygosity and their HLA-
C ligand. Also in psoriatic arthritis, individuals with activating KIR2DS1 and/or
38
KIR2DS2 genes were found to be susceptible to developing disease, but only when
HLA ligands for their homologous inhibitory receptors, KIR2DL1 and KIR2DL2/3,
were missing [89]. An extended review about the role of KIRs in disease is given
in [90].
Although the associations of KIR genes (with or without their ligands) with disease
outcome are many, they are not always confirmed across different cohorts and they can
be puzzling and pointing to contradictory suggestions. In the case of HIV for example,
both the inhibitory receptor KIR3DL1 and its activatory counterpart, KIR3DS1 are
associated with protection from disease progression making the underlying mechanism
hard to decipher.
1.4 Persistent viral infections
1.4.1 Hepatitis C
Epidemiology
Hepatitis C is among the most frequent viral infections in humans with 170 million
infected people worldwide. It is caused by the Hepatitis C virus (HCV) which is a mem-
ber of the Flaviviridae family. Infected individuals show considerable heterogeneity in
the outcome of infection. HCV persists in approximately 70% of infected individuals
while the rest spontaneously clear the infection (usually within the first 6 months).
Chronic HCV infection can cause serious liver damage such as cirrhosis, liver failure
and hepatocellular carcinoma [91].
Virology
The hepatitis C virus (HCV) is a small positive-stranded RNA virus (≈ 9.6Kb). It
encodes a single large polyprotein which is 3000 amino acids long (see Appendix Figure
39
A.1). During the post-translational stage the polyprotein is chopped in 10 distinct ma-
ture viral proteins unless there is a frame-shift and then 11 proteins are produced (with
the inclusion of F). There are 3 structural proteins (Core, E1, E2) and 7 non-structural
proteins (P7, NS3, NS4A, NS4B, NS5A, NS5B). A description of the structure and
function of the HCV proteins is given in [92, 93]. HCV, like other hepatitis viruses, is
predominantly a hepatotropic virus.
Treatment
The available treatment is based on pegylated-interferon-α (PEG-IFN-α) plus ribavirin
(RBV) for patients with chronic hepatitis C infection and achieves sustained virologic
response (SVR) in 40%-52% of treated patients infected with HCV genotype 1 [94].
The SVR can be higher for genotype 2 and 3 [95]. However, side effects are common and
sometimes serious, leading to premature termination of treatment in many patients.
Vaccine
Currently, there is not an available preventative or therapeutic HCV vaccine although
promising attempts are being made [96, 97].
Immunogenetics
In HCV infection, similar to other viral -or even non-viral infections- the origins of
the observed heterogeneity in disease outcome are not yet understood. Several host
genetic factors are suggested as key players in HCV spontaneous clearance. Some of
the most important factors are: 1) several HLA class I (see Appendix Table A.1) and II
alleles; B*57 and C*01 which can have a protective effect and most importantly have
been replicated in several independent cohorts [24–26, 98, 99], 2) compounds of KIRs
with their relevant HLA class I ligands which are associated with disease outcome; the
HLA-C1/KIR2DL3+KIR2DL3+ compound has been linked to HCV clearance [83,100]
40
while the HLA-Bw4I80/KIR3DS1 compound has been shown to have a protective effect
against the development of HCV-associated hepatocellular carcinoma [84], 3) the single
nucleotide polymorphism (SNP) rs12979860 which is found 3kbs upstream of the IL28B
gene (encodes the type III interferon, IFN-λ3) has also been linked with both natural
and treatment-associated control of HCV in multiple studies [101, 102]. Other SNPs
that influence hepatitis C outcome [103–105] include polymorphisms in genes such as
interleukins, chemokines and interferon-stimulated genes and have been linked to both
spontaneous and treatment-controlled infection [106–109].
CD8+ T cell responses
Strong and maintained virus-specific CD4+ and CD8+ T cell responses are thought
to be required for spontaneous viral clearance and can be detected in resolved patients
for more than 20 years after successful elimination of HCV [110]. There are several
lines of evidence supporting a strong role for specific CD8+ T cell in HCV infection
including immunogenetic data of HLA class I associations with disease outcome. Many
studies report enriched mutational changes in experimentally described or predicted
epitopes in patients with the restricting-HLA genotype [62, 111–115] suggesting the
exertion of a strong CD8+ T cell response. Additionally, clearers have been shown to
mount significantly broader CD8+ T cell responses of higher functional avidity and
with wider variant cross-recognition capacity than non-clearers [116]. In chimpanzees,
antibody-mediated depletion of CD8+ T cells before re-infection with HCV led to
prolonged virus replication until HCV-specific CD8+ T cells recovered in the liver
[40]. Furthermore, suppression of acute viremia in vaccinated chimpanzees occurred
as a result of massive expansion of peripheral and intrahepatic HCV-specific CD8+ T
lymphocytes that cross-reacted with vaccine and virus epitopes [117].
41
Animal models
The only existing HCV animal model is the chimpanzee model which involves a lot
of limitations since the experimentally infected chimpanzees may clear HCV infection
more readily than humans, and those that develop persistent infection typically show
mild disease characteristics [91]. Of course, the ethics of animal experimentation limit
the use of animal models in scientific research.
Challenges
There are many challenges posed in the study of the hepatitis C virus such as the
high mutation rate of the virus - 1.5− 2× 10−3 base substitutions per genome site per
day [118], the high replication rate - 1012 virions per day [119] and the high diversity
of HCV sequence - 7 major genotypes and more than 80 subtypes [110]. Additionally,
the existence of quasispecies - different but closely related viral genomes in the same
host [120] and the lack of a small animal model burden the discovery of a potential
vaccine.
1.4.2 Human T-Lymphotropic Virus 1
Epidemiology
Human T-Lymphotropic Virus 1 (HTLV-1) is a persistent retrovirus that infects 10-
20 million people worldwide. Most infected individuals remain lifelong asymptomatic
carriers (ACs). However, approximately 1% of infected individuals develop virus-
associated diseases including HTLV-I associated myelopathy/tropical spastic parapare-
sis (HAM/TSP), an inflammatory disease of the central nervous system that results in
progressive paralysis. In addition, another 2-3% develop Adult-T cell Leukemia (ATL)
and a small number of other less well-defined inflammatory disorders. It is poorly un-
derstood why some individuals remain asymptomatic whereas others develop disease,
42
but one strong correlate of disease is the proviral load, which is significantly higher in
HAM/TSP patients than in ACs [121].
Virology
HTLV-1 is a human retrovirus. Once the diploid genome of HTLV-1 is copied into a
double stranded-DNA (≈ 8.5kb) form, it is integrated in the genome of the host. Then
the virus is referred to as provirus and the viral load of the host is usually quantified as
proviral load (PVL). The HTLV-1 proteome consists of two structural proteins (Gag,
Env) and 10 non-structural proteins (Pro, Pol, Rof, P12, Tof, P13, Rex, P21, Tax,
HBZ). About 90-95% of the HTLV-1 proviral load is carried by CD4+ T cells and only
5-10% is carried by CD8+ T cells [122].
Treatment
Different therapies, mainly anti-inflammatory agents, have been considered for the
treatment of HAM/TSP . These include corticosteroids [123], nucleoside analogues such
as zidovudine and lamivudine [124] and valporic acid (VPA) [125]. Corticosteroids
may help ease the disease symptoms -although not rigorously tested- but treatments
with nucleoside analogues and VPA have shown no significant decrease of proviral
load or improvement of HAM-TSP related symptoms. However, in [126] the combined
treatment with valproate and azidothymidine is shown to be a safe and effective means
to decrease PVL in vivo in Simian T-Lymphotropic Virus 1 (STLV-1) infection.
Vaccine
There is no available vaccine for HTLV-1 related diseases.
43
Immunogenetics
HLA class I molecules have been reported to influence both disease status and vi-
ral load. Specifically, HLA-A*02 and C*08 are associated with a reduced risk of
HAM/TSP and a reduced proviral load in ACs, HLA-B*54 is associated with an
increased prevalence of HAM/TSP and an increased proviral load in HAM/TSP pa-
tients [23, 127]. Other genetic factors that influence the risk of HAM/TSP include a
polymorphism in the TNF-α promoter, the cytokine gene IL-15 and the chemokine
gene SDF-1α [128].
CD8+ T cell responses
Apart from the association of specific HLA class I molecules with disease outcome,
additional evidence for the ability of CD8+ T cells to control the infection include the
association of high mRNA levels of cytolytic genes in CD8+ T cells with low HTLV-1
proviral load [129], the spontaneous killing of HTLV-1 expressing cells by autologous
CD8+ T cells in vitro [130] and ex vivo [131] and the higher variation of Tax coding
sequences in ACs compared to HAMs that suggests a higher CD8+ T cell selection
pressure exerted in ACs [132].
Animal models
The are no animal models that can mimic human HTLV-1 infection reliably. How-
ever, the experimental infection of rabbits, rats and non-human primates has been
been reported in the literature [133–135] but none of these animal models develops
inflammatory disease of the central nervous system that is similar to HAM/TSP in
humans.
44
Challenges
The lack of a reliable small animal model and the absence of samples derived during
acute infection are key challenges in the understanding of the reasons differentiating
ACs from HAM/TSP and ATL patients.
1.4.3 Human Immunodeficiency Virus 1
Epidemiology
Based on the 2010 Joint United Nations Programme on HIV-1/AIDS (UNAIDS) there
are worldwide approximately 33 million individuals infected with Human Immunod-
eficiency Virus 1 (HIV-1). The number of new incidents is declining, mainly due to
prevention efforts, and there is a higher survival period for HIV-1-infected individuals
thanks to effective treatments.
Virology
HIV-1 is a retrovirus composed of two copies of single-stranded RNA (≈ 9.8kb) enclosed
by a capsid. The HIV-1 genome consists of Gag, Pol, Env, the transactivators Tat, Rev,
Vpr, other regulators, Vif, Nef, Vpu and rarely Tev. Pol codes for the viral enzymes
reverse transcriptase, integrase, and HIV protease which are vital for the infection of
news targets. The tropism of the virus depends on the Env protein defining the viral
envelope which enables the virus to attach to and infect target cells. Viruses that bind
the CCR5 chemokine receptor are macrophage tropic (M-tropic or R5) and can infect
CD4+ T cells, macrophages and dendritic cells. Conversely, the lymphotropic strains
(T-tropic or X4) can enter only CD4+ T cells and use a different chemokine receptor
known as CXCR4, which is not bound by CCR5 ligands.
45
Treatment
The treatment of HIV-1 mainly consists of a variety of Antiretroviral Drugs (ARVs)
which target different viral proteins and therefore different phases of the viral cycle.
The drugs include: 1) entry inhibitors, 2) nucleoside/nucleotide reverse transcriptase
inhibitors, 3) non-nucleoside reverse transcriptase inhibitors, 4) integrase inhibitors and
5) protease inhibitors. When taken in combination, the treatment is known as Highly
Active Antiretroviral Therapy (HAART) and although effective can have serious side-
effects. The ARV therapy can fail, amongst other reasons, because of low adherence
to the treatment regimen or the emergence of viral strains resistant to the drugs.
Vaccine
Unfortunately, there is no available preventative or therapeutic vaccine although vac-
cines based on both humoral and cellular responses are being considered. T-cell based
vaccines might only be able to control the infection by lowering the viral load but not
provide sterilising immunity [136]. However, containment of viral load would not only
enhance survival but also limit the risk of transmission [137]. Many vaccines are in
pre-clinical or clinical trials but the high viral diversity and the challenges associated
with generating broadly reactive neutralizing antibodies and cellular immune responses
are key obstacles to be overcome [138]. For both nucleic acid- or protein-based vac-
cines, another difficulty will be to achieve a high effector/target ratio during the viral
expansion phase [139]. Two big T cell based vaccine studies, the STEP [140] and the
RV144 [141], showed none to limited protective effect. Many researchers now believe
that vaccine candidates need to induce both sustained broadly neutralizing antibodies
and a strong cell-mediated response [142].
46
Immunogenetics
In many [143–145] -but not all [146]- genome-wide association studies (GWAS) in multi-
ethnic cohorts of HIV-1 infected individuals, HLA class I molecules are found to be the
major genetic determinants of HIV-1 control. The HLA class I allele B*57 is perhaps
the one most consistently associated with both slower disease progression and better
control of viral replication [147–152]. HLA-B*27 has been shown to have a protective
effect on progression while HLA-B*35 (specifically, a subset of HLA-B*35 molecules
[153]) has a detrimental effect on outcome [16,22,154–156]. Furthermore, the haplotype
B*35/Cw*04 is associated with rapid disease development in Caucasians [22]. In
macaques, MHC class I molecules also influence the outcome of SIV infection [157,158].
Additionally, more rapid disease progression is observed in HLA class I homozygote
individuals [159] and in line with this, maximum HLA class I heterozygosity (A, B,
and C) delays disease progression [22]. In a very well-defined study in SIV-infected
macaques [160], a clear protective effect of a MHC class I heterozygote advantage is
demonstrated; the animals are all infected by the same strain and they share a low
number of MHC alleles forming as such the best system to study this effect.
Apart from HLA class I genotype, KIR/HLA compounds are also associated with
disease outcome. In a sample size of over 1,500 HIV-infected individuals, distinct allelic
combinations of the KIR3DL1 receptor and the HLA-B loci (including B*57 ) were
associated with slow progression to AIDS and decreased plasma HIV RNA abundance
[86]. In another study, an epistatic interaction between KIR3DS1 and HLA-B alleles
delays the progression to AIDS [85]. However, the ligand for the KIR3DS1 receptor
is not well-defined and HLA-B Bw4-80I alleles that encode molecules with isoleucine
at position 80 are thus far only a putative binder; only a rare KIR3DS1 allotype
(KIR3DS1*014) has been experimentally observed to bind HLA-Bw4 molecules [161].
These findings have been suggestive of a critical role for NK cells in the natural history
of HIV infection and although there are some functional data [162] consistent with
47
this hypothesis, the underlying mechanism of the KIR/HLA effect on disease outcome
remains unclear.
CD8+ T cell responses
Indirect evidence for the role of CD8+ T cell responses in controlling HIV infection
include specific HLA class I genes that are consistently associated with disease outcome
[163] and selection of viral escape mutants which avoid CD8+ T cell recognition and
lead to the loss of immune control [164,165]. Direct evidence come from human studies,
were infusion of HIV-1 infected subjects with CD8+ T cells led to the dramatic, but
transient, reduction of HIV RNA-positive cells [45] and animal studies where SIV-
macaques depleted of their CD8+ T cells exhibited a high rise in plasma viremia
[41, 42, 166, 167]. Furthermore, in a cohort of 578 untreated HIV-infected individuals
from KwaZulu-Natal, increasing breadth of Gag-specific responses has been associated
with decreasing viremia while increasing Env breadth with increasing viremia [168].
However, it is important to note that the impact of T-cell responses on the control of
viral replication cannot be explained by the mere quantification of the magnitude and
breadth of the CD8+ T-cell response and the quality of the response (polyfunctional,
avid etc.) can also be crucial for viral control [169–171]. Finally, a temporal correlation
between CD8+ T cell appearance and vireamic control has also been reported [39].
Animal models
There are many animal counterparts of HIV with the main one being the Simian
Immunodeficiency Virus 1 (SIV-1) which infects many African non-human primate
species. Chimpanzees were initially studied but then there was a switch to study-
ing monkeys, in which AIDS develops sooner, in order to expedite the testing of new
hypothesis. In contrast to HIV-infected humans, the natural SIV hosts (for exam-
ple, sooty mangabeys, African green monkeys and mandrills) typically do not develop
48
AIDS despite chronic infection with a highly replicating virus [172]. Limited immune
activation and preserved mucosal immunity are the two main mechanisms which may
explain why SIV infection of natural hosts remains non-pathogenic [173]. However,
in many species such as rhesus macaques, SIV infection resembles many features of
HIV infection [173] and can ultimately generate new approaches that can lead to the
development of a successful HIV vaccine [172, 173]. Humanised mouse models [174]
and feline models [175] are also being considered.
Challenges
There are many obstacles to surpass before obtaining a successful HIV vaccine. Many
of these challenges are similar to the ones involved in the study of HCV infection.
The virus has extensive clade and sequence diversity, it can evade both humoral and
cellular immune responses and it can establish latent viral reservoirs quite early in
infection [176]. Additionally, no small animal model exists.
1.5 Thesis Outline
The present thesis focuses on the study of the CD8+ T cell response to persistent
viral infections (see Figure 1.3). In Chapters 2-4 we study the immunogenetics of
persistent viral infections, focusing on the effect of Human Leukocyte Antigen (HLA)
molecules and Killer-cell Immunoglobulin-like Receptors (KIRs) on disease outcome
and viral burden. In Chapters 5-7, we investigate CD8+ T cell effector mechanisms
and particularly the dynamics of lytic and non-lytic CD8+ T cell responses and how
these relate to ’viral escape’. The persistent viral infections considered in this thesis
are HTLV-1, HCV and HIV. The methods applied include statistical models, high-
throughput sequence analysis, ordinary differential equation models and agent-based
models. Specifically,
In chapter 2, HLA class I-mediated determinants of outcome are investigated
49
in the context of HCV infection. These are: 1) the heterozygote advantage, 2) the
rare allele advantage, 3) individual alleles, 4) allelic breadth, 5) protein specificity and
6) epitope specificity. Furthermore, the impact of HLA class I on disease outcome
is quantified for several molecules and for three different viral infections using the
Explained Fraction.
In chapter 3, known HLA class I associations with disease outcome are studied
in different KIR genetic backgrounds for both HCV and HTLV-1 infections. The effect
of KIRs on HLA class I-mediated immunity is measured in terms of both viral burden
and disease status. Furthermore, potential effector mechanisms that can explain the
findings are discussed.
In chapter 4, a preliminary study design for testing the hypothesis that ’KIRs
can influence the immune pressure exerted by CD8+ T cells during HCV infection’ is
presented and discussed.
In chapter 5, an extensive set of differential equation models that describe lytic
and non-lytic CD8+ T cell responses to SIV/HIV infection are fitted to experimen-
tal viral load, CD4+ and CD8+ T cell data of SIV-infected macaques in order to
investigate the different CD8+ T cell response mechanisms.
In chapter 6, the development and implementation of a 3D Cellular Automa-
ton (CA) model of acute and chronic HIV/SIV infection is described. A CA is an
individual-based computer simulation of a system that evolves in time on a multidi-
mensional (usually 2D or 3D) lattice of nodes and simulates both spatial and temporal
characteristics of a system.
In chapter 7, the CA model is used in order to study the efficiency of lytic and
non-lytic CD8+ T cell effector mechanisms and how they shape the dynamics of viral
escape. Using an extensive set of simulations, the hypothesis that non-lytic CD8+ T
cell effector function can lead to the outgrowth of the variant strain over the wild-type
is tested and compared to the outgrowth observed under a lytic effect. In addition, the
quantification of CD8+ T cell lytic killing rate is discussed and two different models:
50
mass action killing and saturated killing are explored.
Finally, in chapter 8, the main conclusions of the thesis are discussed along with
suggestions for future avenues of research originating from our findings.
Figure 1.3: Schematic outline of the thesis.
Many of the results presented in the following chapters have been published in
peer-reviewed journal articles listed in the ’Dissemination’ section.
51
Chapter 2
HLA class I impact on HCV and
other viral infections
Part of the analysis and findings presented in section 2.5 have been published in
Elemans M., Seich al Basatena N.-K. and Asquith B., PLoS Comput. Biol. 8(2),
2012 [177].
2.1 Aim
Our aim was to investigate the impact of HLA class I-related factors on the outcome
of HCV infection and quantify the influence of HLA class I molecules on the outcome
of HCV, HTLV-1 and HIV-1 infections.
2.2 Introduction
An effective CD8+ T cell response is considered crucial in controlling many viral in-
fections. Polyfunctionality, frequency, breadth, avidity, specificity as well as location
and persistence of the response are likely to be important factors in shaping its ef-
fectiveness. However, measurements of CD8+ T cells such as frequency, phenotype
and function are informative but readily influenced by antigen load; hence, it can be
difficult to ascertain if a particular observation is the cause or the effect of the immune
control. An alternative approach is to examine CD8+ T cell mediating host genotype
factors such as HLA class I genes, where the direction of causality is unequivocal. HLA
class I alleles and other related factors such as zygosity, frequency, allelic breadth and
52
specificity can potentially explain part of the heterogeneity observed in the outcome of
viral infection. In this study, we focus on the impact of individual alleles, heterozygote
advantage, rare allele advantage and binding peptide:HLA properties such as allelic
breadth, protein specificity and epitope specificity on disease outcome, mainly in HCV
infection and partly in HIV and HTLV-1 infection.
In HCV infection, many HLA class I molecules (A, B and C), have been linked
to spontaneous viral clearance but also to viral persistence that can lead to cirrhosis,
liver failure or even hepatocellular carcinoma. Although many HLA class I alleles have
been reported in the literature as significant determinants of outcome (see Appendix
Table A.1 for an extensive list), very few have been replicated in independent studies
(e.g. B*57 [24, 25, 98, 178]). This can be the result of a high false positive discovery
rate but it could also be explained by ethnic and geographical differences between the
cohorts studied. Additionally, the outcomes of infection explored are not always the
same; some studies investigate healthy individuals versus infected individuals while
others compare cases of different disease progression (e.g. spontaneous clearance v
viral persistence). Furthermore, since HCV is characterised by the existence of many
quasispecies, different cohorts might be infected with various viral strains leading to
differences in HLA class I-mediated immune control; a clear example is the study of a
large cohort of HCV genotype 1b-infected Irish women with a single strain [26] which
failed to detect a protective effect of HLA-B*57. The virus in the Irish outbreak
is suggested to contain polymorphism that are different from the general genotype
1b consensus sequence and would not allow recognition by the HLA-B*57 -restricted
responses characterized in the other reports [178]. However, when specific HLA class
I alleles are repetitively associated with outcome across independent studies they are
highly likely genuine determinants of disease course.
Apart from individual HLA class I alleles, HLA class I zygosity can also influence
disease status. In the context of viral infections, under the ‘heterozygote advantage’
hypothesis (also referred to as overdominant selection) MHC heterozygous individuals
53
can present a more diverse repertoire of viral peptides to the immune system than
homozygous individuals resulting in a potentially beneficial effect that may even delay
the emergence of viral escape mutants. In HIV infection, maximum heterozygosity of
all HLA class I loci delayed progression to AIDS among HIV-1 infected individuals
whereas individuals who were homozygous for one or more loci progressed rapidly to
AIDS and death [22]. These findings were confirmed by another study relating HLA
class I homozygosity and rapid progression to late-stage HIV-1-related conditions [159].
An MHC heterozygote advantage has also been shown in SIV-infected macaques [160].
In HTLV-1 infection, HLA class I heterozygosity has an effect on lowering HTLV-1
proviral load which is a known important factor in the risk of developing HTLV-1
associated disease. In HCV infection, evidence for heterozygote advantage has been
reported for HLA class II alleles of the HLA-DRB1 supertype [179] and similarly,
heterozygotes for HLA class II alleles have exhibited protection against hepatitis B
virus infection [180].
Another HLA-mediated factor that can potentially influence the epidemic of infec-
tious diseases is the ‘rare-allele advantage’, also termed as frequency-dependent selec-
tion. Individuals who express rare MHC molecules might have a selective advantage
over those that express common alleles to which the pathogen has adapted. However,
as the selected allele increases in frequency, the virus will adapt to it and the advantage
will be lost. In HIV infection, there is a positive correlation of the frequency of HLA
supertypes with viral load suggesting a selective advantage for individuals with a rare
supertype [181]; HLA class I supertypes can be considered as a plausible classification
scheme for alleles with similar peptide binding specificities. In the case of malaria in-
fection in the Gambian population, a frequency-dependent selection may have caused
the increase of the rare in other racial groups and protective allele, HLA-B*53 [29,182].
Although specific HLA class I alleles are associated with a protective or detrimental
effect on outcome, the reason underlying this effect remains in most cases unknown.
Importantly, it is both the HLA class I molecule and the viral peptide bound by it that
54
trigger CD8+ T cell recognition and activation. Therefore, the study of pMHC binding
properties such as breadth (number of binding peptides), protein and epitope specificity
(preference for binding a specific viral protein or a specific viral peptide) can help
in that direction. A positive correlation between pMHC binding strength, although
for a limited amount of complexes, and eliciting of CD8+ T cell responses has been
shown in [4, 183], indicating an impact of MHC binding affinity on immunogenicity.
Crucially, the constantly improving epitope prediction software that can identify in
silico which viral peptides can form complexes with specific MHC molecules facilitates
such an analysis. Using epitope prediction software and focusing on pMHC binding
properties, a study showed that in HIV-1 infection, HLA class I alleles associated
with slow progression to AIDS prefer to present the p24 Gag viral protein while non-
protective HLA alleles preferentially target the viral protein Nef [184]. In HTLV-1, the
same approach revealed that HLA Class I binding of the viral protein HBZ determines
the outcome of the infection [185]. Here, using epitope prediction software, we also
explore whether pMHC binding properties can influence disease outcome in the context
of HCV infection.
In the first part of this Chapter, using two large HCV cohorts, we study three
HLA class I-mediated factors which have been associated with the outcome of viral
infection: 1) the heterozygote advantage, 2) the rare allele advantage and 3) individual
alleles. In the second part we use epitope prediction software to obtain a list of binding
peptides for each HLA class I molecule present in the cohorts and study whether 1)
allelic breadth, 2) protein specificity or 3) epitope specificity can explain the variability
in the outcome of HCV infection (viral clearance v persistence). We, also quantify the
role of individual HLA class I alleles in explaining heterogeneity of disease outcome in
three different viral infections, namely HCV, HTLV-1 and HIV.
55
2.3 Methods
2.3.1 Data
Cohorts The HCV cohort consists of three sub-cohorts from the US: AIDS Link
to Intravenous Experience (ALIVE, N=262) [186], Multicenter Hemophilia Cohort
Study (MHCS, N=320) [187], Hemophilia Growth and Development Study (HGDS,
N=110) [188] and a UK cohort (N=341) [83]. 251 individuals were excluded due to
incomplete information. The cohort had 257 resolved and 525 chronic patients. The
HTLV-1 cohort (N=431) [23] consists of individuals recruited in Japan. All individuals
were of Japanese ethnic origin and resided in Kagoshima Prefecture, Japan. The cohort
includes 229 HAM/TSP patients and 202 asymptomatic carriers (ACs). The HLA and
KIR genotyping, as well as the viral load measurements were obtained by collaborators.
HLA genotyping For the HCV cohort, genomic DNA was amplified using locus
specific primers as described in [83]. The resulting PCR products were blotted onto
nitrocellulose membranes and hybridized with sequence specific oligonucleotide probes
(SSO). US: alleles were assigned according to the reaction patterns of the SSO probes,
ambiguities were resolved by sequencing analysis. UK: PCR products were typed by
direct sequencing. HLA types that were not resolved by sequencing or which gave
unusual results were also tested by SSO typing using commercial kits (Dynal, RELI
SSO.,Wirral, UK). For the HTLV-I cohort, 96 PCR-sequence-specific primer reactions
were performed to detect all known HLA-A, B, and C specificities in an allele or group
specific manner [189].
56
2.3.2 Statistical analysis
Multiple logistic regression
For studying disease status the multiple logistic regression model is used (see Equa-
tion 2.1). The outcome of the infection (response) which is a binary variable (HCV:1-
Cleared and 0-Persisted and HTLV-1:1-ACs and 0-HAM/TSPs) is studied based on
potential explanatory variables, xi, such as HLA class I mediated factors (e.g. zygos-
ity, breadth etc.) and potential confounders (see below). Some of the explanatory
variables are categorical (e.g. HBV status, mode of infection etc.) and some others are
continuous (e.g. breadth and specificity). The available cohorts (in HCV) are studied
separately and also pooled in order to gain more statistical power. The Odds Ratio
(OR) is given by the exponent of the coefficient, bi, of the relevant covariate, xi. The
OR is a descriptive statistic that is a measure of effect size. It describes the strength of
association between two binary data values (e.g. having a specific HLA class I molecule
and the outcome of infection). Based on the logistic regression model we also obtain
the two-tailed p-values for the statistical significance of the explanatory variables. The
model is fitted using the statistical software R v2.9.2.
logit(p) = ln
(
p
1− p
)
= b0 + b1 ~x1 + ...+ bk ~xk (2.1)
where p is the probability of the outcome of interest, e.g. spontaneous clearance, b0
is the intercept and bi is the change in logit(p) for a one unit change in the predictor
xi.
Multiple linear regression
For studying viral burden the multiple linear regression model is used (see Equation
2.2). The logarithms of the viral load (HCV) or the proviral load (HTLV-1) and not the
raw measurements are considered in the model so that the data are normalised. The
57
viral burden (response), y, which is a continuous variable is studied based on potential
explanatory variables, xi, such as HLA class I mediated factors (e.g. zygosity, breadth
etc.) and potential confounders (see below). Some of the explanatory variables are
categorical (e.g. HBV status, mode of infection etc.) and some others are continuous
(e.g. breadth and specificity). The coefficients of the continuous explanatory variables,
bi, indicate the increase or decrease in log-(pro)viral load when this variable, xi, changes
by 1 unit. If the explanatory variable is binary then the coefficient indicates the
difference in mean log-viral load between the categories. Based on the linear regression
model, we also obtain the two-tailed p-values for the statistical significance of the
covariates. The available cohorts (in HCV) are studied separately and also pooled in
order to gain more statistical power. The model is fitted using the statistical software
R v2.9.2.
y = b0 + b1 ~x1 + ...+ bk ~xk (2.2)
Confounding factors
In the literature, there are examples which suggest that HLA class I associations studies
an be confounded by several other factors. To account for potential confounding in
our models we first examine which variables have an impact on disease status or viral
burden and then we include them in the model used to explain the variation in the
outcome of infection or in viral burden. For HCV infection the confounding factors
included in the model are (Appendix Table A.2): mode of infection, HBV status,
the HLA-C1C1/KIR2DL3++ compound, the SNP rs12979860 and the cohort. For
the HTLV-1 infection the confounding factors are age and gender [23]. For the HCV
cohort, we first consider the UK and USA cohorts separately to investigate whether
the results can be replicated and then we pool the cohorts to gain more statistical
power.
58
Linkage Disequilibrium
Linkage disequilibrium (LD) occurs when genotypes at two loci are not independent
of each other. Similarly, we say that two alleles are in positive LD when their oc-
currence together happens significantly more often than expected by chance while we
refer to negative LD when they segregate significantly more often than expected by
chance. The LD for HLA class I alleles is calculated using the relevant tool available
at www.hiv.lanl.gov/content/immunology. The latter tool calculates a two-sided exact
Fisher’s p-value and corrects the results for false discovery rate using Storey’s q-value
calculation.
Wilcoxon rank-sum test
The Wilcoxon rank-sum test or MannWhitney U test is a non-parametric test for
assessing whether two independent observation samples have similar distributions. The
Wilcoxon rank-sum test is similar to the t-test when no underlying distribution is
assumed for the data. Where applicable, independent p-values are combined using
Fisher’s combined test. The statistical software R v2.9.2 was used for applying the
test.
Explained Fraction
The Explained Fraction (EF) suggested in [190] can be used to estimate how much
of heterogeneity in disease outcome is explained by known genetic or environmental
factors, and hence how much is yet to be explained by unknown or non-included
factors. In [190], the EF was used to estimate the amount of variation in progression
to AIDS that can explained by 13 different genetic factors. The EF can provide
insight into how much a disease status can be explained by a factor but also how
much the absence of the factor explains the absence of disease [190]. This statistic
is based on mutual information. Assuming only cases were both causal factors and
59
disease outcomes are described by categorical variables, we construct a contingency
table of frequencies, denoted [aij ], with rows i indicating causal factors and columns
j representing disease outcome. For multiple causal factors, its combination of factors
is denoted by a row. The row marginals, ai. correspond to the frequency of the factor
and the column marginals, a.j to the frequency of the disease. Now, the EF can be
calculated using the values in the contingency table, the marginal values and Equations
2.3 and 2.4. In our analysis, we calculate both the combined EF that takes into account
the combinations of causal factors and the additive EF which is the summation of the
EFs of the individual factors. For independent factors, the combined EF is expected to
be larger that the additive EF due to the convexity of the information measure [191].
The EF takes values between 0 and 100%. To estimate the 95% confidence intervals
(CI) for the EF we used bootstrapping of the available data.
EF =
∑
i,j aijlog(aij
ai.a.j)
−Imax(1, 2)(2.3)
where the nominator is the mutual information I(1, 2) and the denominator, Imax(1, 2)
is calculated by Equation 2.4 for two disease states and can be extended to include n
disease states. Imax(1, 2) is the maximal value of I(1, 2) achieved when knowledge of
the causal factors completely predicts an individual’s disease status.
Imax(1, 2) =∑
i
ai1log(ai.
ai.a.1) +
∑
i
ai2log(ai.
ai.a.2) (2.4)
Multiple testing correction
In many cases we explore a lot of different hypotheses and multiple testing correction is
required. However, a multiple testing correction approach might be too strict because
a lot of the tests are not independent. In these cases, we only consider a result valid
when it can be replicated in an independent cohort. Where applicable we also perform
Monte Carlo simulations to estimate the False Discovery Rate (FDR) for our findings
60
(see Chapter 3).
2.3.3 Epitope prediction
Software
Epitope prediction is a very useful tool in immunoinformatics. Prediction of T cell
class I epitopes is now highly accurate (algorithms can achieve prediction accuracy
(i.e. correct identification of true positives) of up to 94%) and has a wide allelic cover-
age [192]. This is particularly useful since many known HLA class I molecules do not
have their binding specificity characterised. Some examples of epitope prediction soft-
ware include Epipred [193], NetMHCPan [194], Stabilized Matrix Method (SMM) [195],
Kernel-based Inter-allele peptide binding prediction SyStem (KISS) [12] and Adaptive
Double Threading (ADT) [196]. These algorithms have been used in several studies in
order to identify potential epitopes associated with differential outcome of viral infec-
tions and hence guide vaccine design [184, 197]. Many epitope prediction algorithms
(including the ones used in this study) estimate the binding affinity of the pMHC com-
plex (IC50). An affinity threshold of approximately 500nM (preferably 50nM or less)
has been shown to determine the capacity of a peptide epitope to elicit a CD8+ T
cell response [4]. It is worth noting that overall, prediction is based on the presence of
specific motifs which can lead to a skewed sampling of peptides towards high-affinity
binders and therefore potentially miss intermediate or weak binders.
We use the following epitope predictors:
1. Metaserver v1.0 [198] is a combination of two web-based prediction meth-
ods, NetCTL v1.2 [14] and NetMHC v3.0 [199, 200]. NetCTL is an integrated
method that predicts TAP transport, proteasomal cleavage and HLA binding.
It has allelic predictors for 12 different class I alleles that are chosen to be repre-
sentative of each of 12 supertypes. NetMHC v3.0 predicts pMHC binding, using
artificial neural networks to predict binding affinities for 43 HLA molecules.
61
Metaserver combines the two methods (TAP transport and proteasomal cleav-
age from NetCTL and pMHC binding from NetMHC) and removes a normalising
assumption (which held that all alleles bind the same number of peptides) to
produce an estimate that shows improved accuracy in epitope prediction [198]
and predicts epitopes for 43 HLA molecules. Binding predictions for HLA-C
molecules are not available in Metaserver due to scarcity of HLA-C binding data
necessary for training the neural networks).
2. NetMHCPan v2.2 and v2.4 [194] is a pan-predictor which uses an artificial
neural network trained on the hitherto largest set of quantitative MHC binding
data available, covering HLA-A, HLA-B, HLA-C as well as chimpanzee, rhesus
macaque, gorilla, and mouse MHC class I molecules. Because it does not use a
different artificial neural network for each allele NetMHCPan can be used (with
caution) to predict peptide binding strength for any HLA allele. The model is
built such that data from alleles that are similar to the allele under study are
also taken into account.
We predict epitopes for both HCV and HTLV-1 strains. The specific protein se-
quences used for the epitope prediction can be found in Appendix A.
Evaluation
In order to investigate the accuracy of the epitope predictors we compared the pre-
dicted affinities with the experimentally quantified affinity for a set of 4 HLA molecules
(A-02:01, A-24:02, B-07:02, B-35:01 ) for the HTLV-1 proteome (Table 2.1). This
‘evaluation’ set of experimental values has not been used in the training of the predic-
tors, i.e. overfitting is avoided, and therefore constitutes an independent measure of
benchmarking the performance of the predictors. Although, many different predictors
were investigated, Metaserver, NetMHCPan and SMM were overall outperforming the
other predictors based on the evaluation dataset used. Here, we use Metaserver and
62
NetMHCPan for our analysis.
Spearman correlation (p-value)
Predictor Allele(number of peptides)
A-02:01 (50) A-24:02 (49) B-07:02 (44) B-35:01 (49)
Metaserver [198] 0.76 (1.53 × 10−10) 0.68 (6.97 × 10−8) 0.67 (7.07 × 10−7) 0.66 (2.60 × 10−7)
NetMHCPan [194] −0.76 (1.23 × 10−10) −0.72 (3.96 × 10−9) −0.78 (3.94 × 10−10) −0.77 (6.84 × 10−10)
SMM [195] −0.80 (2.02 × 10−12) −0.62 (1.59 × 10−6) −0.70 (1.40 × 10−7) −0.78 (5.44 × 10−11)
Epipred-Web [193] 0.48 (4.27 × 10−4) 0.68 (7.65 × 10−8) 0.65 (1.95 × 10−6) 0.47 (6.44 × 10−4)
Epipred-Local 0.56 (2.20 × 10−5) 0.70 (2.76 × 10−8) 0.60 (1.66 × 10−5) 0.32 (2.35 × 10−2)
Epipred-New 0.54 (5.37 × 10−5) 0.68 (6.69 × 10−8) 0.67 (6.94 × 10−7) 0.40 (4.18 × 10−3)
KISS [12] 0.67 (1.36 × 10−7) 0.65 (1.98 × 10−6) 0.52 (3.34 × 10−4) 0.54 (8.30 × 10−5)
ADT [196] −0.68 (6.32 × 10−8) −0.65 (3.66 × 10−7) −0.51 (4.31 × 10−4) −0.80 (6.29 × 10−12)
Table 2.1: Evaluation of epitope predictors. The measure used to evaluate the pre-
dictor’s is the non-parametric Spearman correlation. Depending on the binding score
returned by the predictor (affinity is not always the default measure, other measures
are used, such as binding energy) the correlation with experimental affinities might be
positive (e.g. for affinity) or negative (e.g. for binding energy). The local version of
the Epipred software was slightly different from the web version and was downloaded
from the Epipred server. The new version of the software was provided after personal
communication with the authors.
Consistency
The allelic coverage of Metaserver includes 43 HLA class I molecules (only HLA-A and
B molecules are considered). In order to broaden the coverage and be able to study
all the HLA alleles present in the HCV cohort we chose to use the combination of two
epitope predictors: Metaserver and NetMHCPan. To verify that their predictions can
be combined we explored the consistency of the predicted scores using 1) Spearman’s
rank correlation and 2) a simple linear regression model that used the Metaserver
predicted values as predictor and the NetMHCPan predicted values as response and
vice versa. Since the Spearman correlation was very significant and the slope of the
regression line was converging to the unit (Figure 2.1) we considered that the two
63
predictors can be combined.
Rank measure
Not all epitope prediction software provide the same measures of peptide binding
strength. Some are only focused on binary classification (epitope or non-epitope)
and some have a continuous measure that is usually given in units of binding affinity
(IC50). The predictors used here return a continuous peptide binding score. In theory,
the predicted binding score for each pMHC complex could allow the comparison of
predicted binding affinities between alleles. However, between allele comparisons can
be problematic. Firstly, within-allele comparisons (i.e. predictions for different peptides
to the same allele) are thought to be more comparable than predictions between alleles.
Secondly, whether or not a normalisation procedure should be applied for between-
allele comparisons is still being debated in the community [198]. To avoid the potential
problem of between-allele comparisons we used the rank measure technique introduced
by [184]. For the rank measure, the strength of an alleles preference for a particular
protein is quantified by ranking the strength of binding of the top binding peptide
from the protein of interest amongst the strength of binding of peptides from the
entire proteome to that allele. Specifically, we split the relevant proteome (HCV or
HTLV-1) into overlapping nonamers offset by a single amino acid and predicted a
binding affinity score for each nonamer. For each allele we rank all nonamers from
the proteome from the weakest to strongest predicted binding scores. This produces a
list of rank values for each protein to that particular allele that quantifies the binding
relationship between that allele and the protein.
64
0 50 100 150
05
01
50
25
0
Strong Binders
NetMHCPan breadth
Me
tase
rve
r b
rea
dth
slope=1.1y=xr=0.8
0 50 100 150
05
01
50
25
0
Weak Binders
NetMHCPan breadth
slope=1y=xr=0.8
0 100 200 300
01
00
20
03
00
40
0
All Binders
NetMHCPan breadth
slope=1y=xr=0.8
0 50 100 150
05
01
50
25
0
Metaserver breadth
Ne
tMH
CP
an
bre
ad
th
slope=0.8y=xr=0.8
0 50 100 150
05
01
50
25
0
Metaserver breadth
slope=0.9y=xr=0.8
0 100 200 300
01
00
20
03
00
40
0
Metaserver breadth
slope=0.9y=xr=0.8
Figure 2.1: Comparison between the breadth calculated based on Metaserver and
NetMHCPan respectively for the same alleles. Since Metaserver has a smaller cover-
age we were able to compare the breadth only for the 36 alleles covered by Metaserver.
We examined the consistency of strong binders (<50nM), weak binders (>50nM and
<500nM) and all binders (strong and weak) between the two predictors. The agree-
ment is very good, as indicated by the Spearman correlation (r) and the slope of the
regression line, and this allows us to combine the results of the two software with
confidence in order to study the impact of allelic breadth on the outcome of the HCV
infection. The line of equality (y = x) is also displayed to ease the comparison with
the regression line.
65
2.4 HLA class I and HCV infection
In this part of the thesis, we tested a number of hypothesis to investigate the impact of
HLA class I molecules on HCV infection outcome. The HLA class I related factors that
we examine, have been significant in determining the outcome of other viral infections
such as HIV and HTLV-1 but have not been explored extensively in the context of
HCV infection. Interestingly, we find a weak overall impact of HLA class I-related
factors on HCV outcome, suggesting a less efficacious role for HLA class I in HCV
infection.
2.4.1 Heterozygotes’ advantage
First, we test the hypothesis that the patients who are heterozygotes in terms of their
HLA class I alleles can present a broader spectrum of viral peptides to the immune sys-
tem and obtain an advantage in clearing HCV infection. To investigate this hypothesis
we examine the following possibilities:
1. Homozygotes at HLA-A locus have disadvantage over heterozygotes at HLA-A
locus.
2. Homozygotes at HLA-B locus have disadvantage over heterozygotes at HLA-B
locus.
3. Homozygotes at HLA-C locus have disadvantage over heterozygotes at HLA-C
locus.
4. Homozygotes at exactly one locus have disadvantage over heterozygotes at all
loci.
5. Homozygotes at exactly two loci have disadvantage over heterozygotes at all
loci.
66
6. Homozygotes at all three loci have disadvantage over heterozygotes at all loci.
7. Homozygotes at any one or more loci have disadvantage over heterozygotes
at all loci.
The analysis is performed on 4-digits (HLA-YXX:XX, Y=A,B,C), 2-digits (HLA-
YXX) allele resolution and supertypes, i.e. the broader HLA grouping suggested in
[201] (for HLA-C alleles there are no defined supertypes). HLA class I supertypes are a
plausible classification scheme for alleles with similar peptide binding specificities [181].
It is necessary to perform all three resolutions because in the literature results are
given for different resolutions and we need to obtain comparable results. Additionally,
because of data aggregation for coarser resolution such as 2-digit and supertypes our
analysis obtains more statistical power. Finally, for this part of the analysis we only
consider individuals with full known HLA background (N=782).
The results for the 4- and 2- digits resolution are presented in Table 2.2 and for su-
pertypes in Table 2.3 for each cohort, respectively. Although the direction of the Odds
Ratio indicates a heterozygote advantage in almost all the hypothesis tested, the only
statistically significant indication for a heterozygote advantage is observed for HLA-B
alleles in the UK cohort (see Table 2.3). Interestingly, in [150] a dominant involvement
of HLA-B in influencing HIV disease outcome is also reported. However, although the
OR is in the same direction, the same result is not statistically significant in the larger
USA cohort. As a consequence, the heterozygote advantage at the supertype level
for the HLA-B locus is only a trend in the pooled cohort (see Table 2.3). The main
explanation that we can provide for the lack of replication, other than the case of a
false positive result, is the significantly different underlying allelic distribution between
the two cohorts. To examine this hypothesis we compared the allelic distributions of
the two cohorts for each locus (A, B and C) using a paired Wilcoxon sum-rank test.
For a more detailed comparison we calculated the allele frequencies for heterozygotes
and homozygotes in each case. We found that the distribution of the alleles between
67
the two cohorts when taking only the homozygous individuals into account are not
significantly different, i.e. the alleles which are frequent among homozygotes have sim-
ilar prevalences in the two cohorts. However, the distribution of the alleles at B locus
for the individuals that are heterozygotes at the specific locus is significantly different
between the two cohorts (Table 2.4). This implies that the HLA-B alleles that are
driving the heterozygote advantage in the UK Cohort might not be as prevalent in the
USA cohort.
Additionally, when we predict the breadth (i.e. number of binding peptides) of
the HLA class I molecules for all individuals in the pooled cohort and categorise the
individuals based on their zygosity we can readily observe that the HLA breadth dis-
tribution of heterozygotes is not significantly different from the one of homozygotes at
any locus (Figure 2.2); nevertheless, consistently with a stronger heterozygote advan-
tage in the UK cohort there was a trend for higher allelic breadth in heterozygotes in
the UK cohort (p=0.09) but not in the USA cohort (p=0.8). When we pool the two
cohorts we find that at the supertype level, there is a statistically significant advantage
of heterozygotes (OR=0.7, p=0.03) over individuals that are homozygotes at one or
two loci (Table 2.2(c)). In general, the overall lack of a strong heterozygote advantage
in HCV is supported by findings in others studies [99,202,203]. To examine the valid-
ity of our approach we also applied this analysis in the context of HTLV-1 infection
where there is a heterozygote advantage associated with proviral load [127]. We hy-
pothesised that this can be partially explained by the breadth of HLA binding and we
indeed found that a higher breadth of HLA-A alleles is significantly associated with
lower log-proviral load in ACs and a higher breadth of HLA-B alleles is significantly
associated with a lower proviral load in HAM/TSPs; overall, in the HTLV-1 cohort
heterozygotes have a higher allelic breadth compared to homozygotes at one or two
loci (Figure 2.3). This indicates that breadth can be linked to heterozygote advantage
but it is not necessarily the only driving factor. We further examine the role of HLA
class I allelic breadth in HCV infection in section 2.4.4.
68
(a) UK cohort
Homozygosity Odds Ratio p-value
HLA-A alleles 0.86(0.86) 0.78(0.77)
HLA-B alleles 0.23(0.31) 0.09(0.09)
HLA-C alleles 0.54(0.58) 0.40(0.35)
One allele 1.69(1.23) 0.32(0.63)
Two alleles 0.23(0.23) 0.19(0.19)
Three alleles 0.34(0.34) 0.42(0.42)
One or more alleles 0.89(0.81) 0.80(0.61)
(b) USA cohort
Homozygosity Odds Ratio p-value
HLA-A alleles 0.55(0.83) 0.07(0.51)
HLA-B alleles 1.15(0.93) 0.75(0.85)
HLA-C alleles 0.87(0.82) 0.69(0.46)
One allele 0.70(0.90) 0.20(0.64)
Two alleles 0.88(0.87) 0.80(0.74)
Three alleles 0.80(0.47) 0.86(0.51)
One or more alleles 0.72(0.86) 0.18(0.46)
(c) Pooled cohort
Homozygosity Odds Ratio p-value
HLA-A alleles 0.68(0.90) 0.18(0.69)
HLA-B alleles 0.72(0.68) 0.38(0.23)
HLA-C alleles 0.78(0.75) 0.41(0.25)
One allele 0.86(0.97) 0.53(0.86)
Two alleles 0.64(0.68) 0.32(0.34)
Three alleles 0.47(0.38) 0.43(0.28)
One or more alleles 0.76(0.84) 0.21(0.36)
Table 2.2: Impact of zygosity on HCV infection outcome for UK, USA and pooled
cohort for 4- and 2- digits allelic resolution. OR<1 implies a heterozygote advantage.
The results in the brackets are for 2-digits resolution.
69
(a) UK cohort
Homozygosity Odds Ratio p-value
HLA-A alleles 0.70 0.45
HLA-B alleles 0.33 0.02
One allele 0.52 0.12
Two alleles 0.41 0.26
One or two alleles 0.43 0.04
(b) USA cohort
Homozygosity Odds Ratio p-value
HLA-A alleles 0.81 0.34
HLA-B alleles 0.84 0.45
One allele 0.79 0.22
Two alleles 0.89 0.79
One or two alleles 0.78 0.19
(c) Pooled cohort
Homozygosity Odds Ratio p-value
HLA-A alleles 0.80 0.25
HLA-B alleles 0.68 0.07
One allele 0.72 0.07
Two alleles 0.76 0.45
One or two alleles 0.70 0.03
Table 2.3: Impact of supertype zygosity on HCV infection outcome for UK, USA and
pooled cohort for HLA-A and B supertypes. OR<1 implies a heterozygote advantage.
There are no defined supertypes for HLA-C molecules.
Zygosity Locus p-value
Homozygotes
HLA-A 0.44
HLA-B 0.79
HLA-C 0.92
Heterozygotes
HLA-A 0.04
HLA-B 0.01
HLA-C 0.49
Table 2.4: Comparison of the allelic distribution for each HLA class I locus between
the UK and USA cohorts stratified for zygosity. The results are based on paired
Wilcoxon sum-rank test.
70
A B C One Two Three Any None
050
100
150
200
HLA class I homozygosity
HLA
cla
ss I
brea
dth
(SB)
p=0.4
Figure 2.2: Breadth of HLA class I molecules based on strong predicted binders versus
zygosity, in the pooled CV cohort. The x-axis indicates the homozygosity status of
the individuals included in each boxplot. A, B, C=Homozygotes only at HLA-A, B,
C locus, respectively, One, Two, Three=Homozygotes only at one, two or three loci,
Any=Homozygotes at any number of loci, None=Heterozygotes at all loci.
71
A B C One Two Three Any None
050
100
150
200
HLA class I homozygosity
HLA
Bre
adth
(SB)
p=0.001
Figure 2.3: Breadth of HLA class I molecules based on strong predicted binders versus
zygosity, in the HTLV-1 cohort. The x-axis indicates the homozygosity status of the
individuals included in each boxplot. A, B, C=Homozygotes only at HLA-A, B, C
locus, respectively, One, Two, Three=Homozygotes only at one, two or three loci,
Any=Homozygotes at any number of loci, None=Heterozygotes at all loci.
72
2.4.2 Rare allele advantage
The rare allele advantage hypothesis is based on the underlying assumption that the
patients who are expressing rare HLA alleles can have an advantage because the virus
adapts to the most commonly encountered HLA-restricted immune responses but not
to the rare ones. The frequency of the supertypes is calculated for 1) the USA Cohort,
based on the data from the Allele Frequency Net Database [204] while for 2) the UK
Cohort, based on the present data. The supertype grouping defined in [201] is used.
We find a statistically significant rare allele advantage for the HLA-B locus at
the supertype level (see Table 2.5) for the UK but not the USA cohort. However,
because the direction of the OR is the same, when combining the cohorts the rare allele
advantage is also present. The rare allele advantage hypothesis cannot be considered
independently from the heterozygote advantage hypothesis. An individual with rare
HLA alleles is more likely to be heterozygote for these alleles. Therefore, it might
not surprising that the weak heterozygote advantage that we have already found is
followed by the rare allele advantage. This is further supported by the fact that the
rare allele advantage for the HLA-B supertypes is no longer significant if we correct
for the heterozygote advantage. In summary, there is a weak indication of a rare
allele advantage for HLA-B alleles at the supertype level but this cannot be decoupled
from the heterozygote advantage studied above. However, the heterozygote advantage
constitutes a broader hypothesis and it does not follow immediately from the rare allele
advantage. Of note, we found no evidence suggesting a role for HLA-A alleles in HCV
infection.
2.4.3 Individual alleles
We also investigated whether individual HLA class I alleles have an important impact
on the outcome of HCV infection. Many HLA class I associations with HCV clearance
and persistence have been found in different cohorts (Appendix Table A.1). However,
73
(a) UK cohort
Alleles Odds Ratio p-value
HLA-A 2.26 0.84
HLA-B 0.004 0.03
(b) USA cohort
Alleles Odds Ratio p-value
HLA-A 20.50 0.09
HLA-B 0.10 0.23
(c) Pooled cohort
Alleles Odds Ratio p-value
HLA-A 16.28 0.08
HLA-B 0.06 0.03
Table 2.5: Impact of HLA-A and B supertype frequency on HCV infectionoutcome
for pooled cohort. OR<1 implies a rare allele advantage.
many of the associations have not been replicated and some others give contradictory
results in different studies. Here, we use the UK and USA cohorts as well as known
associations reported in the literature in order to obtain a consistent and accurate set
of HLA class I alleles which can explain part of the heterogeneity of the HCV infection
outcome. In the multiple logistic regression model used, all identified confounding
variables are included as covariates (Appendix Table A.2). We perform the analysis
for both 4-digits (data not shown) and 2-digits allelic resolution. The 2-digit resolution
analysis allows for more statistical power. Our results for all the cohorts are shown
in Figure 2.4. In summary, there are three HLA molecules (see Table 2.6) which have
a statistically significant effect on HCV outcome :1) B*57 and C*01 are associated
with viral clearance (protective effect) and 2) C*04 is associated with viral persistence
(detrimental effect). Since this is a retrospective study and the UK and USA cohorts
have been studied before, these associations have been reported in the literature [24].
Another important factor that might confound these results is linkage disequilib-
rium (LD). Therefore, we investigated which HLA alleles are in LD with B*57, C*01
and C*04. These are:
74
1. B*57 : A*01, C*06, C*18
2. C*01 : B*27, B*56
3. C*04 : A*36, B*07, B*35, B*53
Only C*06 which is in LD with B*57 should be further explored since C*06 is statis-
tically significant associated with viral clearance in the pooled cohort while the other
linked alleles do not have an effect on outcome. Two observations can rule out the
possibility that C*06 is driving the protective effect of B*57 : 1) When both molecules
are included as covariates in the multiple regression model only the B*57 effect is still
significant and 2) the B*57 effect is still significant among C*06 -negative individuals.
Highly likely-Protective
B*57 (01,02,03)
P in Pooled Cohort (OR=1.99, p=0.005, nc=86)
P in Ghana in [25]
P in USA in [98]
C*01 (02)
P in Pooled Cohort (OR=2.24, p=0.02, nc=40)
P in Japan in [99]
P in Ireland in [26]
P in USA in [98]
Highly likely-Detrimental
C*04 (01)
D in Pooled Cohort (OR=0.65, p=0.03, nc=187)
D in Ireland in [27]
Note: In two studies reported P [99,205]
Table 2.6: The HLA alleles that are associated with the outcome of the HCV infection
and are replicated in independent cohorts. P=Protective, D=Detrimental, OR=Odds
Ratio, nc=Number of allele carriers. The bold numbers in the parentheses refer to the
4-digit alleles considered in each case.
75
0 2 4 6 8 10 12OR
A68
C08
B14
B15
C06
A24
B57
C03
A02
A32
A01
C05
B18
B44
B07
C12
C07
A03
B40
B35
B08
C04
C15
B51
C16
A11
B27
A29
C02
(a) UK cohort
0 1 2 3 4 5OR
C01C18B45A66A25B50B57A11C16B13C06A26B81B27B41B52B38C14A01B58C08C03C17A02B35A03B44C05B15A68B08A30B51B55A32A74C12A23C02A33C15B40B18A31C07B07B42B39B14A29A36C04B53B49A24
(b) USA cohort
0 1 2 3 4 5OR
C01C18B57B45B50A66B13A25C06B52B81C08B41C03B15C16A01A02B39A11B58B55A68A26A32C14C17A03C05B44B27B35B14A24B08C12B18B51B40B37A31C07A30B07C15B38A74C02B42A33A23C04B53A36B49A29
(c) Pooled cohort
Figure 2.4: HLA class I associations with HCV infection outcome in UK, USA and pooled cohorts. Only alleles with 10
or more observations are shown. The OR is calculated based on a multiple logistic regression model which includes all potential
confounding variables as covariates. The bars represent 95% confidence intervals (CI). The dashed vertical line indicates OR=1.
Here, OR>1 indicates a protective effect (significant associations in green) while OR<1 indicates a detrimental effect (significant
associations in orange). The asterisks (*) are dropped from the HLA names for a clear display.
76
2.4.4 HLA class I breadth
Using predicted binding peptides for each HLA class I molecule present in the UK and
USA cohort we test the hypothesis that the HLA molecules which have a larger breadth
i.e. can present a larger number of HCV peptides to the immune system (in particular
to CD8+ T cells) and give an advantage to the host in clearing HCV infection. Based
on the values of affinity measures used by the predictors we can divide the predicted
epitopes in 1) strong (>50nM) and 2) weak binders (>50nM and <500nM). Hence, in
the analysis we consider three categories of pMHC binding: a) strong, b) weak and
c) strong and weak, i.e. all binding peptides. The total or per protein breadth (see
definitions below) are considered for each HLA class I locus separately.
Total breadth For each HLA type (A, B and C) we count the distinct number of
HCV epitopes (de) which we call total breadth of each HLA locus (A, B or C). The
number of epitopes is defined based on the default threshold used by the epitope pre-
diction software. For example, if an individual recognises x HCV epitopes based on
his HLA-A1 molecule and y epitopes based on his HLA-A2 molecule and z of these
epitopes bind both A1 and A2 then de = x+ y − z distinct epitopes can bind in total
the HLA-A molecules. We found that total breadth of HLA-A, B or C molecules did
not have a statistically significant impact on the outcome of infection (Table 2.7). In-
terestingly, as already mentioned homozygosity was not associated with a significantly
narrower predicted breadth (Figure 2.2).
Breadth per protein Next, we investigated whether the breadth of an HLA
molecule for peptides of a specific HCV protein has an effect on the outcome of HCV
infection. Although we found that the total breadth of HLA class I molecules does not
have an impact on outcome this can be masking an effect which might be attributed
to the breadth for peptides belonging to only one or few HCV proteins. We call this
quantity breadth per protein. To calculate it, we count the distinct epitopes per HLA
77
(a) UK cohort
Binding HLA-A HLA-B HLA-C
OR p-value OR p-value OR p-value
Strong 1.01 0.48 0.99 0.69 1.22 0.20
Weak 1.00 0.30 1.00 0.91 1.03 0.32
All 1.00 0.32 1.00 0.96 1.03 0.22
(b) USA cohort
Binding HLA-A HLA-B HLA-C
OR p-value OR p-value OR p-value
Strong 1.00 0.82 1.00 0.45 0.95 0.51
Weak 1 0.98 1.00 0.58 0.99 0.28
All 1 0.95 1.00 0.55 0.99 0.34
Table 2.7: Overall breadth of HLA-A, B, C molecules in HCV infection. The variable
expressing the total breadth of each HLA molecule is fitted simultaneously in the model
for both UK and USA cohorts. All confounding variables are included as covariates.
Here, OR>1 indicates a protective effect while OR<1 indicates a detrimental effect.
78
locus as in the case of overall breadth but we split the overall breadth in breadth per
HCV protein. We did not find any statistically significant impact of the HLA breadth
per HCV protein on the outcome of infection (Table 2.8). Hence, we can conclude
that based on this approach, allelic breadth does not influence the outcome of HCV
infection. This is consistent with the lack of a strong heterozygote advantage that we
reported above (see section 2.4.1).
2.4.5 Protein specificity
Following the analysis performed in [184, 185] we hypothesised that HLA molecules
which can present peptides from specific HCV proteins to CD8+ T cells, can be as-
sociated with viral clearance. Hence, we explored the protein specificities of the HLA
molecules in the context of HCV infection. The binding strength measure which we
used is rank (see Methods). First, we looked at the protein specificities of protective
and detrimental HLA molecules (Table 2.6). The ranks of the strongest 5 binding pep-
tides from each protein to the alleles B*57 and C*01 (10 rank values) were compared
against the ranks of the strongest 5 binding peptides to the allele C*04 (5 rank values)
(C*01:02, C*04:01 and B*57:01 are the most frequent alleles among the C*01, C*04
and B*57 allelic groups, respectively). A Wilcoxon-Mann-Whitney test was performed
for each protein to test for differences between the two sets of rank values. We could not
establish a statistically significant difference in preference of binding for HCV proteins
between protective and detrimental HLA molecules (Figure 2.5). The only significant
difference observed was for protein NS3 (protective alleles have a stronger binding pref-
erence for NS3 compared to the detrimental allele) but it was lost after FDR correction
(q-value>0.05). The results were robust when the number of peptides was changed to
8 or 10. Interestingly, although no significant difference was detected between protec-
tive and detrimental molecules, we did observe that all three HLA molecules had an
overall stronger binding preference for proteins NS3 and NS5B which are shown to be
79
(a) UK cohort
Binding HLA-A HLA-B HLA-C
Protein OR p-value OR p-value OR p-value
Strong
F6 1.02 0.87 0.99 0.97 - -
core 1.036 0.73 0.99 0.84 - -
E1 1.18 0.20 0.95 0.92 - -
E2 1.19 0.09 0.93 0.79 - -
P7 1.53 0.02 0.43 1 1.11 0.80
NS2 1.05 0.39 1.56 0.34 1.54 0.38
NS3 1.06 0.44 0.95 0.83 - -
NS4A 1.08 0.83 - - - -
NS4B 1.01 0.88 0.57 0.24 1.78 0.17
NS5A 1.07 0.35 0.97 0.75 - -
NS5B 1.03 0.50 0.92 0.70 - -
Weak
F 1.03 0.70 0.97 0.46 1.35 0.16
core 1 1.00 0.97 0.77 2.28 0.05
E1 1.12 0.15 1.16 0.07 - -
E2 1.04 0.31 1.13 0.12 1.03 0.96
P7 0.92 0.52 0.98 0.88 0.91 0.65
NS2 1.07 0.11 0.99 0.91 1.02 0.95
NS3 1.02 0.67 1.00 0.96 1.15 0.11
NS4A 1.41 0.08 1.16 0.62 - -
NS4B 1.09 0.14 1.03 0.65 1.23 0.29
NS5A 1.05 0.55 0.95 0.40 1.06 0.80
NS5B 1.02 0.45 1.03 0.43 1.03 0.70
(b) USA cohort
Protein OR p-value OR p-value OR p-value
Strong
F 0.99 0.78 0.81 0.02 - -
core 0.99 0.82 0.91 0.18 - -
E1 1.02 0.77 1.03 0.63 - -
E2 0.98 0.63 0.97 0.37 - -
P7 0.97 0.71 0.93 0.42 0.75 0.13
NS2 1.01 0.67 1.00 0.99 1.00 1.00
NS3 1.02 0.56 0.99 0.57 - -
NS4A 1.10 0.57 1.46 0.32 - -
NS4B 1.04 0.29 0.98 0.73 0.99 0.95
NS5A 1.02 0.63 0.97 0.41 - -
NS5B 0.99 0.61 0.99 0.59 - -
Weak
F 0.99 0.68 0.98 0.26 0.91 0.29
core 1 1.00 0.97 0.41 0.93 0.74
E1 1.01 0.86 1 0.98 - -
E2 0.99 0.56 1.00 0.79 0.96 0.83
P7 0.93 0.18 1.03 0.40 0.87 0.15
NS2 1.00 0.97 0.99 0.54 1.09 0.61
NS3 1.01 0.41 0.99 0.36 0.98 0.67
NS4A 1.01 0.82 0.97 0.75 - -
NS4B 1.00 0.89 0.99 0.49 0.88 0.18
NS5A 1.00 0.99 0.99 0.37 0.92 0.44
NS5B 1.00 0.92 1.00 0.88 0.96 0.25
Table 2.8: Breadth per protein of HLA-A, B, C molecules in HCV infection. The variable expressing the breadth per
protein of each HLA molecule was fitted in the same model for UK and USA cohorts. When no predicted epitopes were identified,
a model could not be fitted (dash). All confounding variables are included as covariates. Here, OR>1 indicates a protective
effect while OR<1 indicates a detrimental effect. 80
more immunogenic compared to the other HCV proteins [115, 206, 207] (Figure 2.5)
while there was no binding preference for NS4A.
F core E1 E2 P7 NS2 NS3 NS4A NS4B NS5A NS5B
010
020
030
040
050
060
00
100
200
300
400
500
600
HCV proteins
Ran
k (p
refe
renc
e of
bin
ding
)
ProtectiveDetrimental
p=n.s.
Figure 2.5: Binding preference of HCV proteins for protective and detrimental HLA
class I molecules. Protective molecules: B*57:01 and C*01:02 and detrimental
molecules: C*04:01 (Table 2.6). To establish the binding preference of the HLA
molecules we used the top 5 ranking peptides per molecule. The results are unchanged
if other thresholds are used instead. The p-values were corrected for multiple compar-
isons using the Benjamin-Hochberg method (q-values).
Secondly, we explored whether the protein specificity was different between resolved
and chronic individuals. We found similar results when we looked at two different
measures: 1) overall specificity and 2) specificity per HLA locus.
For overall protein specificity, for each HLA molecule (A1, A2, B1, B2, C1, C2),
we took only the top ranking epitope out of the whole HCV proteome and recorded
81
the protein to which it belonged. Then we counted how many out of the 6 HLA
molecules of each individual preferred binding to each HCV protein. For example, if
all the HLA molecules of an individual preferred binding to epitopes from the same
HCV protein then this protein has value 6 while all the others have value 0. This gave
us a specificity measurement to which we fitted a multiple logistic regression model
for each HCV protein including all the identified confounding factors (see Methods).
No statistical significant impact of protein specificity on the outcome of infection was
identified (data not shown). We also looked at more than just the top ranking epitope
and followed the same process in order to avoid being biased by the immunodominant
proteins. The results were not altered. Additionally, for each HCV protein we set
the top 40% of the HLA molecules which bind to these protein to be strong binders
for it. Then we counted how many of the 6 alleles (A1, A2, B1, B2, C1, C2) that
an individual has, binds to each HCV protein. We also performed our analysis for
different thresholds for defining the strong binding alleles for each protein (10-40%).
The results were the same as in the previous approach.
To examine protein specificity per HLA type, for each HLA molecule, we took only
the top ranking epitope for each HCV protein. Then we calculated the mean rank
(mr) of each HLA molecule per protein. For example, if the top ranking epitope of
an HCV protein for HLA-A1 has rank x and respectively for HLA-A2 has rank y then
the binding preference of the individual’s HLA-A molecules for that HCV protein can
be represented by mr =x+y2 . No statistical significant impact of protein specificity on
the outcome of infection was observed (data not shown). For completeness, for each
HLA molecule, we also calculated the mean of the top 3 ranks for each HCV protein
(less biased towards the top rank) and calculated the mean rank of this quantity per
HLA type but the results were not altered.
In summary, we investigated 1) whether protective and detrimental HLA class I
alleles have different protein specificity in HCV infection and 2) if the protein specificity
differentiates between resolved and chronic individuals but no statistical significant
82
differences in protein specificity were identified.
2.4.6 Epitope specificity
The last hypothesis that we tested is whether the HLA molecules which can present
specific peptides (epitope specificity) to CD8+ T cells give an advantage to the host in
clearing HCV infection. For each individual, we looked at the top 10 binding peptides
for each HLA molecule (maximum 60 peptides if there are no overlapping ones). We
considered this list to be a list of epitopes for each individual. Then, for each peptide
in turn (N=3084) we attributed a score of 1 if an individual had this peptide in
the list and 0 otherwise. This provided a binary variable for each peptide which we
fitted in a logistic regression model to predict disease outcome. All the confounding
variables were included as covariates. This approach, because of the very large number
of peptides examined, requires a correction for multiple testing. Hence, applying the
Bonferroni correction would require a statistical significance level of p<0.00005. We
found no such statistical significance for any of the peptides neither in the UK nor
in the USA cohorts. The peptides with statistical significance p<0.05 are given in
Tables 2.9. Importantly, none of these peptides were included in the prediction of
both the cohorts so no result could be replicated. The same results were reached when
we created the list of epitopes for each individual considering only the strong binding
or the strong and weak binding peptides. Therefore, because of lack of high statistical
significance and lack of replication our findings suggest that epitope specificity cannot
explain the heterogeneity in the outcome of HCV infection and the identified peptides
with p <0.05 are most likely spurious associations.
83
(a) UK cohort
Peptide OR p-value Protein
TYRSSAPLL 2.66 0.04 F
MNWSPTAAL 5.06 0.007 E1
RMYVGGVEH 2.70 0.04 E2
NIVDVQYLY 0.10 0.03 E2
FYGMWPLLL 2.66 0.04 P7
MWPLLLLLL 2.66 0.04 P7
LLALPQRAY 2.70 0.04 P7
MAIIKLGAL 1.94 0.03 NS2
RQAEVITPA 2.70 0.04 NS4B
NMWSGTFPI 1.81 0.03 NS5A
LLAAGVGIY 2.70 0.04 NS5B
(b) USA cohort
Peptide OR p-value Protein
RSGAPTYSW 1.74 0.02 E2
LPALSTGLI 0.73 0.006 E2
FSLDPTFTI 1.39 0.02 NS3
RAYMNTPGL 2.67 0.01 NS3
SWLGNIIMF 0.69 0.01 NS5B
Table 2.9: No predicted epitopes were consistently associated with the outcome
of HCV infection. These predicted epitopes are associated with the outcome of
HCV infection at p<0.05 statistical significance but not corrected for multiple testing
(p<0.00005) for both UK and USA cohorts. None of these peptides were observed in
both cohorts and are most likely spurious associations. All confounding variables are
included as covariates. Here, OR>1 indicates a protective effect while OR<1 indicates
a detrimental effect.
84
2.5 Quantification of HLA class I impact on vi-
ral infection
Having examined the impact of six different HLA class I-related factors on the outcome
of HCV viral infection (see section 2.3), we find that apart from a weak heterozygote
advantage and a significant effect of three individual HLA molecules, the other HLA
class I-mediated factors do not have a substantial influence on hepatitis C infection
outcome. Therefore, in this part of the project we aim at quantifying the impact of
HLA class I molecules on HCV viral clearance and persistence. Since HLA class I
associations with infection outcome are reported in other viral infections, we estimate
the impact of HLA class I alleles in three different viruses: 1) HCV, 2) HTLV-1 and 3)
HIV in order to have a more complete assessment. The quantification is based on the
Explained Fraction (EF) which estimates how much of the variation in disease outcome
is explained by a given factor (see Methods). We calculate both the combined EF and
the additive EF. Additionally, because the EF depends on the frequency of the disease
outcome and the disease factor (Appendix Figure A.2) we also normalise it; we set the
frequency of disease outcome to known population frequencies (HCV: Resolved=30%
and Persistent=70% [110]; HTLV-1:ACs=99% and HAM/TSPs=1% [127]; HIV: Elite
controllers=0.3%, Vireamic Controllers=10% and Chronic progressors=89.7% [208])
and the frequency of the disease factor to 50%. The normalisation of the EF allows
us to compare between alleles. The HCV and HTLV-1 cohorts are described here (see
Methods) while the HIV cohorts are reported in [155] and [156].
The proportion of variation in disease outcome explained overall by the known
HLA class I associations is nearly 3% and 6.6% in HCV and HTLV-1 (Tables 2.10 and
2.11), respectively. Additionally, the amount of variation in HCV outcome explained
by the combination of all the protective factors available in the pooled HCV cohort
(B*57, C*01, the SNP rs12979860, the KIR2DL3/2DL3+HLA-C1C1 and the HBV
85
infection) is 9.5%. Over all the infections studied here, we find that individual HLA
class I molecules can explain 1.5% (median) of the heterogeneity observed in the disease
outcome and no more than 6.5% (using the EF normalised for population frequency).
Interestingly, the EF varies substantially among the viral infections studied. In sum-
mary, the impact of the individual HLA class I alleles is greater in HIV infection, less
pronounced in HTLV-1 infection and very limited in HCV infection (Table 2.12 and
Figure 2.6).
Factors Comb EF(%) LCI(%) UCI(%) Add EF(%) Pop Freq Freq=0.5
B*57 0.74 0.05 2.25 - 0.73 1.79
C*01 0.66 0.03 2.12 - 0.66 3.23
C*04 0.98 0.15 2.53 - 0.96 1.43
B*57C*01 1.57 0.49 3.7 1.4 - -
B*57C*04 1.93 0.8 4.27 1.72 - -
C*01C*04 1.85 0.68 4.02 1.64 - -
B*57C*01C*04 2.92 1.57 5.71 2.38 - -
Homoz 1 or 2 0.54 0.02 1.85 - - -
rs12979860 3.56 1.57 6.27 - - -
HBV 1.68 0.42 3.64 - - -
HIV 0 0 0.52 - - -
2DL3/2DL3+C1C1 0.95 0.11 2.63 - - -
Protective 9.52 7.64 14.8 7.59 - -
All 13.57 12.83 21.35 9.11 - -
Table 2.10: The proportion of heterogeneity in outcome of HCV infection which is
explained by HLA class I alleles and other known factors. In the protective factors we
include: B*57, C*01, the SNP rs12979860, the KIR2DL3/2DL3+HLA-C1C1 and the
HBV infection. The confidence intervals are obtained using the bootstrap method. Ab-
breviations, Homoz: Homozygosity, LCI: Lower 95% Confidence Interval, UCI: Upper
95% Confidence Interval, Comb EF: Combined EF, Add EF: Additive EF, Pop Freq:
EF normalised for disease outcome at population frequency, Freq=0.5: EF normalised
for 50% factor frequency.
86
Factors Comb EF(%) LCI(%) UCI(%) Add EF(%) Pop Freq Freq=0.5
A*02 3.13 0.98 6.57 - 1.61 3.27
B*54 2.55 0.54 5.26 - 1.51 4.02
C*08 1.61 0.23 4.26 - 0.78 3.09
A*02B*54 5.06 2.43 9.6 5.43 - -
A*02C*08 4.36 1.96 8.58 4.74 - -
B*54C*08 3.67 1.56 7.58 3.91 - -
A*02B*54C*08 6.6 4.08 12.28 7.04 - -
Table 2.11: The proportion of heterogeneity in outcome of HTLV-1 infection which
is explained by HLA class I alleles. The confidence intervals are obtained using the
bootstrap method. Abbreviations, LCI: Lower 95% Confidence Interval, UCI: Upper
95% Confidence Interval, Comb EF: Combined EF, Add EF: Additive EF, Pop Freq:
EF normalised for disease outcome at population frequency, Freq=0.5: EF normalised
for 50% factor frequency.
Factors EF(%) LCI(%) UCI(%) Add EF(%) Pop Freq Freq=0.5
B*57 -Emu et al. 5.40 1.86 11.33 - 3.33 7.69
B*57 -Pereyra et al. 4.56 1.68 9.75 - 6.45 5.69
B*27 -Emu et al. 0.14 0.02 2.37 - 0.02 0.38
B*27 -Pereyra et al. 1.69 0.31 5.14 - 4.97 3.96
B*35 -Pereyra et al. 1.87 0.49 5.13 - 2.60 5.14
Table 2.12: The proportion of heterogeneity in outcome of HIV-1 infection which is
explained by specific HLA class I alleles. The confidence intervals are obtained using
the bootstrap method. The data for the HIV calculations are obtained from [155]
and [156]. Abbreviations, LCI: Lower 95% Confidence Interval, UCI: Upper 95%
Confidence Interval, Comb EF: Combined EF, Add EF: Additive EF, Pop Freq: EF
normalised for disease outcome at population frequency, Freq=0.5: EF normalised for
50% factor frequency.
87
Figure 2.6: The proportion of variation in viral infection outcome explained by known
HLA class I associations (the non-normalise EF is used here). The data for the HIV
calculations are obtained from [155] and [156]. 95% confidence intervals were esti-
mated by bootstrapping the data 5,000 times, trimming the 5% extremes, and then
calculating the range in which 95% of the remaining data lay. Due to linkage between
the HLA alleles, the EF is not additive; so for instance, in HTLV-1 infection, the three
alleles HLA-A*02, C*08, and B*54 together only explain 6.6% of the outcome. The
outcomes explained are HCV: spontaneous clearance v persistence; HTLV-1: asymp-
tomatic carriage v HAM/TSP; HIV-1: elite control v viraemic control v progression.
The asterisks (*) are dropped from the HLA names for a clear display.
88
2.6 Discussion
HLA class I molecules are responsible for presenting viral peptides to CD8+ T cells and
also for modulating NK cell activation. They are therefore considered a very important
factor in determining the response of the immune system to viral infection. The fact
that HLA class I genes are highly polymorphic suggests that they are maintained
through selective forces, such as infectious disease morbidity [209].
In this study, we investigated six HLA class I-related factors that can potentially
shape a protective or detrimental CD8+ T cell response in the context of hepatitis C
infection. Furthermore, we quantified the impact of individual HLA class I alleles on
HCV, HTLV-1 and HIV infection.
In HCV infection for HLA class I supertypes, we found a weak heterozygote ad-
vantage for HLA-B alleles as well as a weak overall protective effect for heterozygote
individuals over individuals who are homozygotes at one or two loci. A rare allele
advantage for HLA-B supertypes was also observed. However, an individual with rare
HLA alleles is more likely to be heterozygote for these alleles. Therefore, it is not
surprising that the weak heterozygote advantage that we found is followed by the rare
allele advantage. Nevertheless, the heterozygote advantage constitutes a broader hy-
pothesis and it does not follow immediately from the rare allele advantage. The lack of
a strong heterozygote advantage is consistent with the fact that the predicted HLA al-
lelic breadth distribution of heterozygotes is not significantly different from that of the
homozygotes and is also further supported by findings in others studies [99, 202, 203].
However, it remains interesting that in HCV infection, HLA-B, as in the case of HIV-
1 [150], is the locus that presents the strongest protective effect suggesting a special
role for HLA-B alleles in controlling viral infections.
We also re-identified three HLA class I alleles which are associated with disease
outcome: 1) the protective molecules: B*57 and C*01 and 2) the detrimental molecule:
89
C*04. All associations were corrected here for known confounding factors such as
linkage disequilibrium and mode of infection. These HLA class I associations have
been reported before and confirmed in multiple independent cohorts [25–27, 98, 99].
However, we should note here that the evidence for C*04 -a detrimental allele in the
context of HCV infection- remain controversial. C*04 has also been found to convey
protection from HCV infection [99, 205] and no C*04 effect was identified in a large
HCV-infected cohort which was recently analysed in [98].
Interestingly, the B*57 protective effect has also been reported in HIV infection.
In [210,211] a plausible explanation of this effect is the broad cross-reactivity of B*57
against variants of a dominant Gag-epitope as well as its plasticity in peptide binding
which was also shown for B*27 in [212]. In the latter study, the ability of both bulk
CD8+ T cells as well as epitope-specific TCR clonotypes to inhibit viral replication
in vitro and upregulate perforin and granzyme B were also linked to the protective
effect. Our analysis cannot rule out or verify that B*57 acts in a similar manner in
HCV infection. Furthermore, 1) the frequency of B*57 in the studied cohorts would
not allow for enough statistical power so that B*57 could drive a significant difference
in epitope specificity between resolved and persistent individuals and 2) if the effect is
moderate, the lack of a strong T-cell related detrimental HLA class I molecule hinders
the possibility of identifying a difference in protein or epitope specificity. However, the
overall lack of breadth and protein or epitope specificity impact on disease outcome
suggests that an alternative mechanism to cross-reactivity might explain the protective
effect of B*57 in HCV infection; such us driving the selection of costly viral escape
mutations [213].
Next, we used epitope prediction software to examine whether there are differences
between resolved and persistent individuals in the 1) number of peptides that their
HLA class I molecules are predicted to bind, 2) the HCV proteins that their HLA class
I molecules prefer to bind or 3) the specific predicted epitopes that their HLA class
I molecules prefer to recognise. We found that none of these factors were significant
90
determinants of the outcome. In particular, for protein specificity we found no signifi-
cant differences in binding preference, either between protective and detrimental HLA
class I alleles, or between persistent and resolved individuals. All this evidence, taken
together suggests that the impact of HLA class I related factors, as described here,
on the outcome of HCV infection is rather limited. The lack of a significant differ-
ence of breadth, protein specificity and epitope specificity is perhaps not surprising.
The C*01 -protective effect in HCV infection has been shown to be NK-mediated [83].
The only detrimental allele in our study is an HLA-C allele, C*04 for which epitope
prediction performs poorly (similarly for C*01 ) compared to HLA-A and HLA-B and
which can also be NK-mediated. If the effect is NK-mediated, this approach would
not be expected to relate pMHC properties to disease outcome because of 1) the dif-
ferent way that HLA class I molecules tune NK cell responses compared to CD8+ T
cell responses and 2) the less pronounced effect of peptides in NK cell inhibition or
activation by HLA class I molecules.
Motivated by the rather small role of HLA class I-mediated factors in HCV out-
come, we quantified the impact of HLA class I alleles in HCV infection and compared
it with known associations reported for HTLV-1 and HIV infection. We calculated
that the proportion of heterogeneity in the outcome of these viral infections that can
be explained by individual HLA alleles is on average 1.5% (median), with an observed
maximum of 6.5%. Interestingly, that proportion varies substantially among the vi-
ral infections studied. The impact of the individual HLA class I alleles, in terms of
the normalised explained fraction, is found to be greater in HIV infection, less pro-
nounced in HTLV-1 infection and very limited in HCV infection. This is in line but
not necessarily linked with functional and sequence-based studies on CD8+ T cell re-
sponses which suggest that a strong and effective CD8+ T cell response is detected in
HIV [214,215] and HTLV-1 [216] but not consistently in HCV infection [217]. Notably,
HLA class I associations with disease outcome inform us about differences in the effect
of the HLA molecules and a small explained fraction does not imply that CD8+ T
91
cell responses are dismal; it could be that all CD8+ T cell responses are rather similar
regardless of their restriction so no significantly different ’best’ molecules exist. On the
other hand, we should not assume that the CD8+ T cell response is highly important
just because HLA class I associations are highly statistically significant. Furthermore,
the small explained fraction implies that HLA class I genotype is unlikely to be the
major determinant of between-individual variation in clinical outcome. Nevertheless,
the normalised comparison of the ’most protective’ and ’most detrimental’ molecules
in each viral infection, as performed here, can provide a potential ranking of HLA class
I-mediated impact between the different infections.
92
Chapter 3
Impact of KIRs on HLA class I-
mediated immunity
Most of the analysis and findings presented in this chapter have been published in
Seich al Basatena N.-K. et al., PLoS Pathogens 7(10), 2011 [1] and in O’Connor G M.,
Seich al Basatena N.-K. et al., Hum. Immunol., 2012 [2].
3.1 Aim
Our aim was to investigate known HLA class I associations with disease outcome in
different KIR genetic backgrounds for both HCV and HTLV-1 infection.
3.2 Introduction
Killer cell immunoglobulin-like receptors (KIRs) are a family of transmembrane pro-
teins that are expressed on natural killer (NK) cells and subsets of T cells [72–74]. They
bind HLA class I molecules and have activatory and inhibitory isoforms [75]. KIRs
contribute directly and indirectly to antiviral immunity. Directly, KIRs on NK cells
sense the loss of HLA class I molecules from the cell surface and trigger NK-mediated
cytolysis. Indirectly, NK cells regulate adaptive immunity via crosstalk with dendritic
cells and by the production of chemokines and cytokines [81, 82, 218].
HLA class I molecules can be grouped into allotypes with similar KIR binding
properties [219]. For example, KIR2DL2 binds group C1 HLA-C molecules which
93
have asparagine at residue 80, and, with a weaker affinity, group C2 molecules which
have a lysine at position 80 [76].
Early research on KIRs investigated NK-mediated protection by studying disease
associations with KIRs in the context of their HLA class I ligands [83, 86]. There is
now compelling evidence that KIRs also regulate adaptive immunity [81,82,218], but it
is not known whether this has a significant impact on the response to infection in vivo.
Differences between human KIRs and their mouse functional homologues (the Ly49
receptors) and the paucity of KIR allele-specific antibodies have hindered work on the
role of KIRs in controlling adaptive immune responses. Here we used immunogenetics
to investigate whether KIR genotype modulates HLA-mediated anti-viral protection
in vivo. We focused on HLA class I alleles which have previously been associated
with disease outcome and investigated whether these effects were altered by the KIR
background. We studied 4 well-documented HLA class I allele-disease associations in
two viral infections: human T lymphotropic virus type 1 (HTLV-1) and hepatitis C
virus (HCV).
HTLV-1 is a persistent retrovirus that infects 10-20 million people worldwide. Most
infected individuals remain lifelong asymptomatic carriers (ACs). However, approx-
imately 10% of infected individuals develop associated diseases including HTLV-1-
associated myelopathy/ tropical spastic paraparesis (HAM/TSP), an inflammatory
disease of the central nervous system that results in progressive paralysis. It is poorly
understood why some individuals remain asymptomatic whereas others develop dis-
ease, but one strong correlate of disease is the proviral load, which is significantly higher
in HAM/TSP patients than in ACs [121]. it has previously been shown that HLA-A*02
and C*08 are each associated with both a reduced risk of HAM/TSP and a reduced
proviral load in ACs and that HLA-B*54 is associated with an increased prevalence
of HAM/TSP and an increased proviral load in HAM/TSP patients [23, 127].
HCV is among the most widespread viral infections, with 170 million infected
people worldwide. As in HTLV-1 infection, the outcome of HCV infection is hetero-
94
geneous: the virus persists in approximately 70% of infected individuals while the
remainders clear the infection spontaneously, usually within 6 months. Chronic HCV
infection can cause serious liver damage including cirrhosis and hepatocellular carci-
noma [91]. The origins of this heterogeneity are not completely understood but several
genetic determinants have been identified, including HLA-B*57 which is associated
with spontaneous clearance in several cohorts [24, 25, 98, 178].
The aim of this study was to test the hypothesis that KIR genotype determines
the efficiency of HLA class I-mediated anti-viral immunity. We tested this hypothesis
for 4 HLA class I associations: HLA-C*08, A*02 and B*54 in HTLV-1 infection and
B*57 in HCV infection. We show, using multiple independent measures, that for both
HCV and HTLV-1, possession of the KIR2DL2 gene enhanced HLA class I-restricted
immunity.
3.3 Methods
Many parts of the methodology applied in this Chapter are very similar to the one
presented in Methods of Chapter 2, so we only describe below the parts of the analysis
that have not already been presented.
3.3.1 Data
The HCV and HTLV-1 cohorts and the HLA genotyping are described in the Methods
of Chapter 2 and were collected by our collaborators.
KIR genotyping For the HCV cohort, the presence or absence of ten KIR genes
(KIR2DL1, KIR2DL2, KIR2DL3, KIR3DL1, KIR2DS1, KIR2DS2, KIR2DS3, KIR2DS4,
KIR2DS5 and KIR3DS1) was determined by PCR using sequence specific primers
as described in [85]. Additionally, KIR2DL4, KIR3DL2, KIR3DL3, KIR2DL5 and
KIR2DP1 were typed in the USA samples. For the HTLV-1 cohort, the presence or
95
absence of ten KIR genes (KIR2DL1, KIR2DL2, KIR2DL3, KIR2DL4, KIR3DL1,
KIR2DS2, KIR2DS3, KIR2DS4, KIR2DS5 and KIR3DS1 ) was also determined by
PCR using sequence specific primers.
Viral load HCV RNA was assessed by a branched DNA (bDNA) assay (Quantiplex
HCV RNA 2.0 assay; Chiron Corporation) [ALIVE] or the HCV COBAS AMPLICOR
system (COBAS AMPLICOR HCV; Roche Diagnostics) [MHCS]. The HTLV-1 provirus
load in peripheral blood mononuclear cells (PBMC) was measured as described in [121].
Quantitative PCR was performed using an ABI 7700 sequence detector (Perkin-Elmer
Applied Biosystems). All DNA standards and samples were amplified in triplicate. A
standard curve was generated by using the β-actin gene from HTLV1 -negative PBMC
and the Tax gene from TARL-2, a cell line containing a single copy of HTLV-1 proviral
DNA.
3.3.2 Statistical analysis
Overview
Two independent cohorts were studied: HCV-infected individuals (N=782) and HTLV-
1-infected individuals (N=402).
Ten KIR genes were typed; of these, 4 were present at an informative frequency
(see definition below) in the HTLV-1 cohort and 9 in the HCV cohort. The modulation
of HLA class I impact on infection outcome by the informative KIRs was analysed; the
models included all other known determinants of outcome in the cohorts as covariates.
All HLA class I alleles which were significantly associated with outcome in our
cohorts and that were independently verified (either in a large independent cohort or
on an independent outcome) were studied. For HTLV-1 there are three such alleles:
HLA-A*02 and C*08 which are associated with reduced proviral load in ACs and
reduced risk of HAM/TSP and HLA-B*54 which is associated with increased proviral
96
load in HAM/TSP patients and increased risk of HAM/TSP [23,127]. For HCV there
are two such associations: HLA-B*57 and C*01 which have both been associated
with increased odds of viral clearance in two independent cohorts (see Chapter 1 and
[98]). The C*01 -protective effect in HCV infection has been shown to be NK-mediated
[83, 220] and therefore we did not consider it here.
We investigated all KIR genes which were present at an informative frequency. We
define an informative frequency as sufficient to detect a ‘moderate’ effect size (∆=0.5);
this is equivalent to at least 32 people carrying the gene and 32 not carrying the
gene [221]. Of the KIR genes typed, 4 were present at an informative frequency in the
HTLV-I cohort (KIR2DL2, KIR2DS2, KIR2DS3 and KIR3DS1) and 9 in the HCV
cohort (KIR2DL2, KIR2DL3, KIR3DL1, KIR2DS1, KIR2DS2, KIR2DS3, KIR2DS4,
KIR2DS5 and KIR3DS1 ).
The effect of individual HLA class I alleles in different KIR genetic background was
investigated by stratifying the cohorts for the absence and presence of the informative
KIRs and the effect of the already described HLA class I associations were re-evaluated
in each stratum. The impact of the HLA alleles on two response variables was stud-
ied: 1) status (AC vs. HAM/TSP for HTLV-1 infection and resolved vs. chronic for
HCV infection) and 2) viral burden (log10[proviral load] for HTLV-1 infection and
log10[viral load] for HCV infection). Multiple logistic regression was used to study
the variation in status and multiple linear regression was used for the variation in
viral burden. All statistically significant and potentially confounding variables were
included in the models (HTLV-1: age and gender; HCV: HBV status, mode of infec-
tion, SNP rs12979860 and cohort). We focused on the difference between KIR+ and
KIR- individuals in the size of the protective/detrimental effects associated with the
individual HLA class I molecules (i.e. odds ratios and differences in viral load) rather
than p values as the latter comparison is confounded by differences in strata sizes. As
opposed to Chapter 2, here OR<1 implies a protective effect and OR>1 implies
a detrimental effect so that the comparison with the previous studies is immediate.
97
HCV viral load
The HCV cohorts were analysed separately as viral load was measured using different
assays. Two cohorts (MHCS and ALIVE) had sufficient numbers of individuals with
measurements of all variables for this analysis. For MHCS we used the median of
multiple measurements of viral load; for ALIVE, a single measurement was available
for each subject. We only analysed viral load in patients with chronic infection, so any
observed impact on viral load is independent of the impact on viral clearance.
Linkage Disequilibrium
The LD for HLA class I alleles is calculated as described in Chapter 1. The LD between
KIRs is calculated based on the Chi-squared test on a 2x2 contingency table.
False Discovery Rate
We quantified the false discovery rate for our analysis using Monte Carlo methods. For
each of the HLA associations studied, we performed 10,000 random stratifications of
the relevant (HCV or HTLV-1) cohort with the size of the two strata being the number
of KIR2DL2+ and KIR2DL2- individuals in that cohort and asked how many times
we would see odds ratios equal to or more extreme than we observed in the actual
cohorts. The probability of making our observations by chance is less than p=2x10−11
(Appendix Table B.1).
3.3.3 Epitope prediction
We produced a list of rank values (see Methods of Chapter 2) for each protein to each
allele that quantifies the binding relationship between that allele and the protein. We
then invert the rank so the bigger 1/rank, the stronger the preference of the allele for
the protein; the logarithm of this measure is plotted on the y axis of Figure 3.1 as ‘HBZ
binding score’. Each individual therefore contributes up to 4 values (alleles for which no
98
predictive algorithms were available were excluded from the analysis). Binding scores
were compared between ACs and HAM/TSP patients using the Wilcoxon rank sum
test and reported both as separate p values for HLA-A and B molecules and combined
(since we found no evidence to reject the null hypothesis that the HBZ binding score
of an individual’s A and B molecules was independent, spearman correlation=0.05
p=0.46). The median difference in binding score is the median of the difference of
average HBZ binding between ACs and HAM/TSP patients expressed as a percent of
the AC binding score for HLA-A and -B molecules.
3.4 KIR2DL2 in HTLV-1 infection
3.4.1 Disease status
In the cohort from Southern Japan, HLA-C*08 was associated with a significantly re-
duced odds of developing HAM/TSP (OR=0.47, p=0.03, OR<1 indicates a protective
effect while OR>1 indicates a detrimental effect) [23]. We investigated the impact
of KIRs on this protective effect by stratifying the cohort by KIR genotype. Of the
KIR studied, one particular KIR, KIR2DL2, had a noticeable interaction with C*08
(Table 3.1 and Figure 3.3). We found that the C*08 protective effect was weakened
and no longer statistically significant in the subset of individuals who were KIR2DL2-
(OR=0.67, p=0.4) but enhanced in KIR2DL2+ individuals (OR=0.16, p=0.02). There
were more KIR2DL2- individuals (N=300) than KIR2DL2+ individuals (N=102) so
the absence of significance in the KIR2DL2- individuals was not simply due to reduced
cohort size. Similarly, HLA-B*54, which is associated with a significantly increased
risk of HAM/TSP (OR= 3.11, p=0.0009), had a weakened impact on disease risk in
the absence of KIR2DL2 (OR=1.70, p=0.2) but an enhanced impact in the presence
of KIR2DL2 (OR=12.05, p=0.004). Again, the absence of a significant effect of B*54
in KIR2DL2- individuals was not attributable to a loss of power. In contrast, although
99
HLA-A*02 was associated with a reduced risk of HAM/TSP, there was no significant
additional impact of KIR2DL2 genotype which could not be attributed to power.
3.4.2 Proviral load
As an independent test of the observation that KIR2DL2 enhanced the effect of both
protective and detrimental HLA class I alleles in HTLV-1 infection, we investigated the
interaction between HLA class I alleles, KIR2DL2 and HTLV-1 proviral load (pvl). We
investigated pvl in ACs and HAM/TSP patients separately, so any observed impact on
pvl is independent of the impact on disease status. C*08 has previously been associated
with a low pvl in ACs (difference in log10 pvl between C*08+ and C*08- ∆=-0.33,
p=0.05); again, this effect was weakened in KIR2DL2- individuals (∆=-0.29 p=0.18)
but enhanced in KIR2DL2+ individuals (∆=-0.66 p=0.07); Table 3.1. Similarly, HLA-
B*54, which is associated with a high pvl in HAM/TSP patients (∆=+0.24 p=0.01)
showed a weakened effect in the absence of KIR2DL2 (∆=+0.22, p=0.05) but an
enhanced effect in the presence of KIR2DL2 (∆=+0.42, p=0.01).
Two previous observations on HTLV-1 immunogenetics have, until now, remained
unexplained. Firstly, although C*08 has been associated with a low pvl in ACs it
has no detectable impact on pvl in HAM/TSP patients; similarly, B*54, which was
associated with a high pvl in HAM/TSP patients, had no impact on pvl in ACs [23,127].
Why some HLA class I alleles apparently ‘cease working’ in some populations was
unknown. We hypothesised that the lack of the expected C*08 and B*54 effects in
HAM/TSP patients and ACs respectively was due to a low frequency of KIR2DL2 in
these groups and that the decrease or increase in pvl due to C*08 or B*54 respectively
would be manifest only in KIR2DL2+ individuals. Consistent with this hypothesis we
found that the frequency of KIR2DL2 carriage in the groups that did not show the
expected effect of HLA genotype on pvl was approximately half that of the groups
in which HLA-associated effects were observed (prevalence of KIR2DL2+ amongst
100
HTLV-1 disease status (HAM/TSP v AC)
OR KIR2DL2 OR HLA Allele Cohort size
in whole cohort Genotype in stratified cohort Carriers
(p value) (p value)
+ 6.249 14 102
C*08 2.126 (p=0.02)
(p=0.032) - 1.504 44 300
(p=0.364)
+ 0.083 21 102
B*54 0.322 (p=0.004)
(p=0.0009) - 0.588 64 300
(p=0.173)
HTLV-1 proviral load
Difference in KIR2DL2 Difference in HLA Allele Cohort size
logPVL in Genotype logPVL in Carriers
whole cohort stratified cohort
+ -0.66 10 48
-0.33 (p=0.066)
ACs (p=0.047) - -0.286 26 132
(p=0.181)
C*08 + -0.856 4 54
-0.173 (p=0.005)
HAM/TSP (p=0.208) - -0.087 18 168
(p=0.578)
+ -0.218 3 48
0.057 (p=0.709)
ACs (p=0.778) - 0.095 21 132
(p=0.676)
B*54 + 0.417 18 54
0.24 (p=0.014)
HAM/TSP (p=0.012) - 0.224 43 168
(p=0.049)
Table 3.1: KIR2DL2 in HTLV-1 infection: KIR2DL2 enhances the protective effect of C*08 and the detrimental effect
of B*54 on HAM/TSP risk and, independently, on proviral load. In the disease status models an odds ratio (OR)<1 indicates
a protective effect (decreased risk of HAM/TSP), an OR>1 indicates a detrimental effect (increased risk of HAM/TSP). In
the proviral load models the dependent variable was log10[proviral load], a difference in PVL<0 indicates a protective effect
(decreased pvl with the HLA allele), a difference in PVL>0 indicates a detrimental effect (increased pvl with the HLA allele).
All models also included the two variables which can act as confounding variables in this cohort: age and gender. The impact
on proviral load was considered separately in ACs and HAM/TSP and so the observation of an impact on proviral load is
independent of the observation of an impact on status. If variables which were not significant predictors (p<0.05) were removed
by backwards stepwise exclusion the conclusions were unchanged.
101
B*54+ individuals: 12.5% in ACs vs 29.5% in HAM/TSPs; prevalence of KIR2DL2+
amongst C*08+ individuals: 18.2% in HAM/TSPs vs 27.8% in ACs; Table 3.1). The
small numbers of individuals in the stratified cohorts (HAM/TSP KIR2DL2+: C*08+
N=4, C*08 - N=50. ACs KIR2DL2+: B*54+ N=3, B*54- N=45) precluded a reliable
test for an impact of HLA on pvl in KIR2DL2+ individuals. However, in the larger of
these groups there was a significant impact; i.e C*08 was associated with a significant
reduction in pvl in KIR2DL2+ individuals (∆=-0.86, p=0.005). This provides, for
the first time, a plausible explanation for the reported observation [23] that the B*54
effect on pvl was not manifest in ACs and the C*08 effect on pvl was not manifest in
HAM/TSP patients.
3.4.3 HLA class I specificity
It was recently reported that in HTLV-1 infection, HLA class I molecules that bind
peptides from the virus protein HBZ are associated with a reduced risk of HAM/TSP
and, independently, a reduced pvl [185]. In the same study it was shown, using IFN-
γ ELISpot, chromium release and CD107 staining, that HBZ-specific CD8+ T cells
were present and functional in fresh PBMC from infected individuals. We therefore
investigated the interaction between KIR2DL2 and the protective effect of binding
HBZ. We used epitope prediction software [198] to predict the strength of binding of
HBZ peptides to an individual’s HLA-A and B molecules. It was already reported
that ACs carry HLA-A and -B molecules that are predicted to bind HBZ signifi-
cantly more strongly than those carried by HAM/TSP patients (median difference
12%, p=0.00005). We found that this effect was stronger in KIR2DL2+ individuals
(median difference 25%, p=0.00006) than in KIR2DL2- individuals (median difference
7%, p=0.06); Figure 3.1. We reasoned that this difference in HBZ binding between ACs
and HAM/TSP patients was due to HLA-A*02 and B*54, which differ in their HBZ
peptide-binding affinities [185] and are associated with different outcomes in HTLV-1
102
infection. We therefore removed all individuals with A*02 or B*54 from the cohort
and repeated the analysis. Surprisingly, we still found the same pattern: possession
of HLA molecules that bind HBZ strongly was significantly associated with remaining
asymptomatic (median difference 10%, p=0.038) and this effect was strengthened in
KIR2DL2+ individuals (median difference 23%, p=0.022) but not in KIR2DL2- in-
dividuals (median difference 3%, p=0.177); Figure 3.1. This demonstrates that the
protective effect of binding HBZ peptides by multiple HLA class I molecules, both A
and B, is enhanced by KIR2DL2.
3.5 KIR2DL2 in HCV infection
3.5.1 Spontaneous viral clearance
As previously reported, HLA-B*57 was associated with significantly decreased odds
of chronic infection (OR=0.571, p=0.023). This protective effect was enhanced in the
presence of KIR2DL2 (OR=0.40, p=0.007) but weakened in the absence of KIR2DL2
(OR=0.83, p=0.63) (Table 3.2). Furthermore, the impact of B*57 was strongest in
KIR2DL2 homozygote individuals (OR=0.21, p=0.024), weaker in KIR2DL2 het-
erozygote individuals (OR=0.48, p=0.07) and absent in KIR2DL2 -negative individuals
(OR=0.83, p=0.63). KIR2DL2 enhanced the association between B*57 and sponta-
neous clearance independently and with similar strength in both African-Americans
and Caucasians (Table 3.3).
3.5.2 Viral load in chronic infection
Next we investigated the impact of B*57 and KIR2DL2 on HCV viral load. We only
considered patients with chronic infection, so any observed impact on viral load is
independent of the impact on viral clearance. This analysis was possible in two co-
horts: MHCS and ALIVE. We found that B*57 was associated with reduced chronic
103
Figure 3.1: In the whole cohort (left column) individuals with HLA-A or B molecules
which were predicted to bind peptides from HBZ strongly were significantly more
likely to be asymptomatic (top row); this effect was stronger in the individuals with
KIR2DL2 (middle row) than without (bottom row). When individuals with A*02 or
B*54 were removed (right column) the pattern remained the same. This indicates
that the protective effect of binding HBZ peptides by multiple, different HLA class
I molecules, both A and B, is enhanced by KIR2DL2. We did not predict HLA-C
binding as the relevant algorithms are not available in Metaserver (due to scarcity of
HLA-C binding data necessary for training the neural networks).
104
HCV status (spontaneous clearance v chronic infection)
OR in whole cohort KIR2DL2 OR in stratified cohort HLA Allele Cohort size
(p value) Genotype (p value) Carriers
+ 2.482 49 408
(p=0.007)
- 1.202 35 374
1.75 (p=0.631)
(p=0.023)
+/+ 4.796 16 84
(p=0.024)
-/+ 2.104 33 324
(p=0.072)
-/- 1.202 35 374
(p=0.631)
HCV Viral load
Difference in logVL KIR2DL2 Difference in logVL HLA Allele Cohort size
in whole cohort Genotype in stratified cohort Carriers
in whole cohort (p value)
(p value)
MHCS ALIVE MHCS ALIVE MHCS ALIVE MHCS ALIVE
-4.485 -0.432 5 13 46 76
+ (p<0.0001) (p=0.064)
-3.106 -0.180 (pcomb = 0.00003)
-1.64 0.116 5 5 25 69
- (p=0.243) (p=0.741)
(pcomb = 0.5)
Table 3.2: KIR2DL2 in HCV infection: KIR2DL2 enhances the protective effect of HLA-B*57 on HCV
status (spontaneous clearance v chronic infection) and, independently, on HCV viral load in chronic patients. In
the HCV status models an odds ratio (OR)<1 indicates a protective effect (decreased odds of chronic infection),
an OR>1 indicates a detrimental effect (increased odds of chronic infection). In the viral load models the
dependent variable was log10[viral load]. Cohorts were analysed separately as viral load was measured using
different assays. Two cohorts (MHCS and ALIVE) had sufficient numbers of individuals with measurements
of all variables for this analysis. A difference in VL<0 indicates a protective effect (decreased vl with the
HLA allele), a difference in VL>0 indicates a detrimental effect (increased vl with the HLA allele). All models
included all variables which can act as confounders (see Methods). The impact on viral burden was considered
within chronic carriers and so the observation of an impact on viral burden is independent of the observation
of an impact on status.
105
Unstratified cohort
Cohort OR LCI UCI p-value Allele carriers Cohort size
All 0.569 0.351 0.923 0.023 84 782
African-Americans 0.482 0.222 1.046 0.065 34 240
Caucasians 0.556 0.283 1.092 0.087 43 483
KIR2DL2+
All 0.403 0.208 0.781 0.007 49 408
African-Americans 0.354 0.130 0.960 0.041 25 125
Caucasians 0.363 0.134 0.985 0.047 21 252
KIR2DL2-
All 0.832 0.392 1.764 0.632 35 374
African-Americans 0.649 0.150 2.801 0.563 9 115
Caucasians 0.755 0.286 3.759 0.57 22 231
Table 3.3: There was a trend for B*57 to be associated with an increased odds of HCV
clearance in both African-Americans and in Caucasians. On stratifying the cohorts by
KIR2DL2 genotype this trend became significant for KIR2DL2+ individuals but was
lost for KIR2DL2- individuals. The same pattern and a similar strength of effect were
seen in both African-Americans and Caucasians.
106
HCV viral load, particularly in the MHCS cohort (MHCS: difference in log10 VL
∆=-3.1, p=0.0003; ALIVE: ∆=-0.18, p=0.336. Combined p=0.0006). Consistent
with our observations in HTLV-1 infection, this reduction was enhanced in the pres-
ence of KIR2DL2 (MHCS: ∆=-4.5 p<0.0001, ALIVE: ∆=-0.46 p=0.046. Combined
p=0.00003) but weakened in the absence of KIR2DL2 (MHCS ∆=-1.64, p=0.24,
ALIVE ∆=+0.32, p=0.35. Combined p=0.66); Table 3.2 and Figure 3.2. In MHCS
(but not ALIVE) we also observed a progressive effect with KIR2DL2 copy number
(2 copies: ∆=-6.5, p=0.0005. 1 copy: ∆=-4.1, p=0.001. 0 copies ∆=-1.6, p=0.24);
however, in this case, the number of homozygous individuals is too small and does not
allow to draw conclusions.
3.6 No KIR2DL2 main effect on outcome
These data show that KIR2DL2 enhances both protective and detrimental HLA class I
associations (Figure 3.3) with disease outcome and viral burden. Therefore, KIR2DL2
would not be predicted to have a significant net impact across all HLA class I molecules.
That is, possession of KIR2DL2 (alone or with its C1 ligand) without a particular pro-
tective or detrimental HLA allele, would not be expected to be significantly protective
or detrimental. This prediction was verified (Table 3.4).
3.7 Linkage disequilibrium
There is strong linkage disequilibrium between the KIR genes and between the HLA
class I alleles (see below). Here, we thoroughly analyse linkage between HLA class I
alleles and amongst KIR genes in order to identify which molecules are driving the
effect on infection outcome.
107
Figure 3.2: In HCV infection, B*57 is associated with a reduced HCV viral load.
This protective effect is enhanced in KIR2DL2+ individuals and reduced or absent
in KIR2DL2- individuals. The cohorts were analysed separately as viral load was
measured using different assays. Two cohorts, MHCS and ALIVE, had enough in-
dividuals with measurements for all factors to perform this analysis. HIV-1 status
was a significant determinant of viral load in the ALIVE cohort, including HIV sta-
tus in the model as a covariate did not change any of our conclusions (ALL p=0.336;
KIR2DL2+ p=0.046; KIR2DL2- p=0.350). The p values were obtained based on the
linear regression model. Summary data from these plots is provided in Table 3.2.
108
Figure 3.3: The effect of both protective (A) and detrimental (B) HLA class I alleles on both
status (first column) and viral load (second column) is stronger in KIR2DL2+ individuals than
in KIR2DL2- individuals. This effect was seen in both HTLV-1 and HCV infection. We also
saw a progressive effect of KIR2DL2 copy number (C). We could not test for a progressive
effect in the HTLV-1 cohort because there are no KIR2DL2 homozygotes. In all cases, the
impact on viral burden was considered within (not between) disease status categories and so
the observation of an impact on viral burden is independent of the observation of an impact
on status. Effect sizes, p values and cohort sizes are provided in Tables 3.1 and 3.2.
109
KIR2DL2 KIR2DL2:C1
OR
(p-value)
HTLV-1 status 1.11 1.13
(p=0.71) (p=0.67)
HCV status 0.86 0.9
(p=0.34) (p=0.51)
Difference in log(p)VL
(p-value)
HTLV-1 pvl AC 0.05 0.05
(p=0.77) (p=0.77)
HTLV-1 pvl HAM/TSP -0.02 -0.03
(p=0.85) (p=0.78)
HCV vl MHCS 0.09 1.11
(p=0.89) (p=0.07)
HCV vl ALIVE 0.22 0.09
(p=0.08) (p=0.47)
Table 3.4: KIR2DL2 does not have an effect on status or viral burden, with or without
its C1 ligand.
110
3.7.1 Linkage between HLA class I alleles
HTLV-1: C*08 rather than linked HLA alleles appears to be the pri-
mary allele driving protection which is enhanced by KIR2DL2
In the HTLV-I cohort, C*08 was in linkage disequilibrium with B*40 and B*48 ; we
therefore sought to establish which was the primary HLA class I gene driving the
protective association which was enhanced by KIR2DL2.
C*08 & B*40 In a logistic regression model to predict disease status (HAM/TSP
v AC) when both C*08 and B*40 were included as factors C*08 retained a protective
trend (OR=0.52 p=0.07) whereas B*40 lost significance (OR=0.88 p=0.3); stratifying
on KIR2DL2 we again found that in KIR2DL2+ individuals, C*08 retained a pro-
tective trend (OR=0.22 p=0.07) and B*40 lost significance (OR=0.49, p=0.20) and
that, as expected, neither was significant in the absence of KIR2DL2. Similarly, in a
linear regression model to predict log10(proviral load) in ACs when both C*08 and
B*40 were included as factors C*08 retained significance (difference in logVL=-0.37
p=0.03) whereas B*40 lost significance (indeed went in the other direction, difference
in logVL=+0.14 p=0.3); stratifying on KIR2DL2 we again found that in KIR2DL2+
individuals the C*08 effect was strengthened (difference in logVL=-0.64 p=0.07) and
B*40 lost significance (difference in logVL=+0.3, p=0.20) and that, as expected, nei-
ther was significant in the absence of KIR2DL2. We therefore concluded that C*08
rather than B*40 was the HLA gene most likely to be associated with protection whose
effect was enhanced by KIR2DL2.
C*08 & B*48 In a logistic regression model to predict disease status (HAM/TSP
v AC) when both C*08 and B*48 were included as factors both factors lost significance.
However, stratifying on KIR2DL2 we found that in KIR2DL2+ individuals C*08 re-
tained a protective trend (OR=0.21, p=0.07) and B*48 lost significance (OR=0.33,
p=0.34) and that, as expected, neither was significant in the absence of KIR2DL2.
111
In a linear regression model to predict log10(proviral load) in ACs when both C*08
and B*48 were included as factors C*08 retained a trend (difference in logVL=-0.36
p=0.08) whereas B*48 lost significance (indeed went in the other direction, difference
in logVL=+0.08 p=0.78); stratifying on KIR2DL2 we again found that in KIR2DL2+
individuals the C*08 effect was strengthened (difference in logVL=-1.1 p=0.03) and
B*48 lost significance (again went in the opposite direction, difference in logVL=+0.7,
p=0.24) and that, as expected, neither was significant in the absence of KIR2DL2.
Taken together, these data suggest that C*08 rather than B*48 was the gene most
likely to be associated with protection whose effect was enhanced by KIR2DL2.
HTLV-1: B*54 rather than linked HLA alleles appears to be the pri-
mary allele driving susceptibility which is enhanced by KIR2DL2
In the HTLV-I cohort, B*54 was in linkage disequilibrium with C*01. We therefore
investigated i) whether B*54 or C*01 was the primary gene associated with increased
susceptibility to HAM/TSP and ii) whether B*54 or C*01 -associated susceptibility
was enhanced by KIR2DL2. In a logistic regression model when both B*54 and C*01
were included as factors B*54 retained significance (OR=3.84 p=0.0009) whereas C*01
lost significance (indeed went in the opposite direction OR=0.73 p=0.3); stratifying on
KIR2DL2 we again found that in KIR2DL2+ individuals B*54 retained significance
and C*01 lost significance and that, as expected, neither was significant in the absence
of KIR2DL2. If all B*54+ individuals were removed from the cohort then C*01 was
no longer detrimental (OR=0.88, p=0.4, C*01+=96). Unfortunately, due to the large
number of C*01+ individuals in the cohort (N=184) it was not possible to reverse this
analysis and investigate the impact of B*54 in the absence of C*01.
112
HCV: B*57 rather than linked HLA genes appears to be the primary
gene driving protection which is enhanced by KIR2DL2
In the HCV cohort, B*57 is in linkage disequilibrium with A*01, C*06 and C*18.
Hence, we investigated whether the observed protective effect of B*57 can be at-
tributed to the other linked alleles. We found that A*01, C*06 and C*18 do not have
a significant impact on disease status neither overall nor in the context of KIR2DL2.
We therefore conclude that B*57 is the HLA allele associated with HCV clearance and
also the one enhanced by KIR2DL2.
3.7.2 Linkage between KIR genes
The KIR genes are in tight linkage disequilibrium (Table 3.5), making it hard to
definitively ascertain which KIR enhances the HLA-associated effects (i.e. the associa-
tion between C*08 and asymptomatic status in HTLV-1 infection, between B*54 and
HAM/TSP in HTLV-1 infection and between B*57 and spontaneous viral clearance
in HCV infection). To try to determine which was the primary KIR driving the en-
hancement of the HLA-associated effects we constructed a logistic regression model to
predict status (HAM/TSP v AC for HTLV-1 infection, spontaneous clearance v per-
sistence for HCV infection) in which the HLA molecule with the presence or absence
of each KIR, depending on the stratum in which the HLA had the more significant
effect, was included along with the known confounding factors. Then, we remove the
HLA:KIR factors by stepwise backwards exclusion (i.e. by the highest p-value one at
a time, refit the model and repeat). In all 3 cases (C*08, B*54 and B*57 ) the only
HLA:KIR compound that remains in the model is the HLA:KIR2DL2+. This analysis
suggests that KIR2DL2 is most likely to be the primary gene driving the observed
effect.
KIR2DL2 is in particularly tight LD with KIR2DS2, an activating receptor. It
could be argued that an activatory receptor is more likely to modulate CD8+ T cells
113
(a) HTLV-1 Cohort
<0.001 2DL2
0.04 <0.001 3DL1
<0.001 0.76 <0.001 2DS2
<0.001 <0.001 <0.001 <0.001 2DS3
0.1 0.02 0.69 0.8 <0.001 2DS4
<0.001 <0.001 <0.001 <0.001 0.27 <0.001 3DS1
2DL2 3DL1 2DS2 2DS3 2DS4 3DS1
(b) HCV Cohort
<0.001 2DL1
0.001 <0.001 2DL2
<0.001 <0.001 <0.001 2DL3
0.78 0.04 0.06 <0.001 3DL1
0.91 0.002 <0.001 <0.001 <0.001 2DS1
0.001 <0.001 <0.001 0.02 <0.001 <0.001 2DS2
0.22 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 2DS3
0.69 0.09 0.02 <0.001 <0.001 0.04 0.001 <0.001 2DS4
0.73 <0.001 0.002 <0.001 <0.001 0.006 0.42 <0.001 <0.001 2DS5
0.35 0.12 <0.001 <0.001 <0.001 0.008 <0.001 <0.001 <0.001 <0.001 3DS1
2DL1 2DL2 2DL3 3DL1 2DS1 2DS2 2DS3 2DS4 2DS5 3DS1
Table 3.5: Linkage between the KIRs in the HTLV-1 (top) and HCV (bottom) cohorts.
Positive linkage disequilibrium (LD) is shown in gray and negative in black. The
statistical significance (p-values) of the LD is displayed the cells of the tables.
114
and so we specifically investigated whether the observed effect was more likely to
be driven by KIR2DL2 or KIR2DS2. In all cases logistic regression models to pre-
dict status where HLA allele:KIR2DL2 and HLA allele:KIR2DS2 (plus confounders)
were included as simultaneous covariates then the covariate HLA:KIR2DL2 was more
significant than HLA:KIR2DS2 in every case (i.e. for HLA-B*57 in HCV, B*54 in
HTLV-1 and C*08 in HTLV). However, in a model where HLA:KIR2DS2 but not
HLA:KIR2DL2 was a covariate then HLA:KIR2DS2 was significant for B*54 and B*57
(but not C*08 ). For B*54 and B*57 we therefore also investigated whether the HLA
allele was significant in the cohort that was KIR2DL2+/KIR2DS2-. Although the
cohorts were small to draw concrete conclusions, both alleles had a significant effect
in KIR2DL2+/KIR2DS2- (B*54 : OR=18.5, p=0.016, n=10, N=48; B*57 : OR=0.47,
p=0.02, n=2, N=13). The size of theKIR2DL2-/KIR2DS2+ did not allow further
investigation (B*54 : n=2, N=6; B*57 : n=1, N=8). The small cohort sizes makes it
impossible to conclude that KIR2DS2 does not also have an effect but it is clear that
even in the absence of KIR2DS2, KIR2DL2 does have an effect. Together these three
observations indicate that KIR2DL2 rather than KIR2DS2 is more likely to be the
primary gene driving the enhancement of HLA class I-mediated antiviral immunity.
Additionally, HLA-KIR factors which may be particularly relevant because they
are known receptor-ligand pairs (e.g. HLA-B*57 with KIR3DL1 or KIR3DS1) were
examined in more detail (Appendix Table B.3 and B.4) but all evidence indicated that
the effect is driven by KIR2DL2.
In summary, the evidence suggests that KIR2DL2 is most likely to be the KIR
which is enhancing immunity. The one KIR for which it is impossible to assess whether
it has a stronger effect than KIR2DL2 is KIR2DL3 as KIR2DL2 and KIR2DL3 segre-
gate as alleles of the same locus. Both are inhibitory but KIR2DL2 provides stronger
inhibitory signals than KIR2DL3 [76]. The 2DL2/L3 locus is present in one copy in
the majority of haplotypes so the observation that KIR2DL2 is present in an individ-
ual (1 or 2 copies) implies that there are 0 or 1 copies of KIR2DL3. So the statement
115
that the presence of KIR2DL2 enhances class I mediated immunity can be restated in
the reciprocal as lack of KIR2DL3 homozygosity enhances class I mediated immunity.
However, functionally it is difficult to understand how the lack of homozygosity for
a weaker receptor should enhance immunity more effectively than homozygosity for a
stronger receptor.
3.7.3 Conclusions on LD
Analysis of the linked genes/alleles indicates that the primary molecules driving the
observed associations are most likely to be KIR2DL2 in combination with HLA-B*54,
C*08 and B*57 rather than individual linked KIR, multiple stimulatory linked KIRs or
linked HLA class I alleles. We cannot rule out an effect of linkage between KIR2DL2
and neighbouring loci outside the KIR genes. However, there is little evidence of
significant linkage between KIRs and even the next closest gene cluster, the LILR [222].
Furthermore, we observed the same effect of KIR2DL2 in three different populations
(Japanese, African-American and Caucasian) so a putative linked locus driving the
effect would have to be linked to KIR2DL2 in all three populations.
3.8 Canonical KIR-HLA binding
We next explore whether direct binding of KIR2DL2 with any of the HLA class I
molecules studied (namely C*08, B*54 and B*57) can explain the enhancement that
we observe.
Although HLA-C*08, as a group C1 molecule, is expected to bind KIR2DL2, the
most frequent subtype in our cohort (C*08:01, 88%) binds KIR2DL2 very weakly (com-
parable to background [223]), furthermore HLA-B*54 and HLA-B*57 are not expected
to bind KIR2DL2 and the most frequent subtypes in our cohorts (B*54:01 and B*57:01)
have been shown not to bind KIR2DL2 [76,223]. Finally, KIR2DL2 enhanced the pro-
116
tective effect of binding HBZ peptides by multiple HLA-A and -B molecules. With the
exception of B*46:01 and B*73:01 (which were not responsible for the enhancement,
data not shown) KIR2DL2 is not thought to bind HLA-A and B molecules and has
been shown not to bind 29/29 HLA-A and 54/56 HLA-B allotypes tested in [76]. We
therefore hypothesised that the effect of KIR2DL2 on HLA class I-mediated immunity
we have observed is not attributable to KIR2DL2 directly binding the HLA molecule
whose effect is enhanced. To test this hypothesis we first investigated whether the
other group C1 alleles had the same effect as C*08 in HTLV-1 infection. Grouping
all the C1 alleles we found no significant association between C1 and decreased risk of
HAM/TSP either in the whole cohort or in KIR2DL2+ individuals. Similarly, there
was no relation between pvl and C1 in either ACs or HAM/TSP patients. Analysis of
the individual C1 alleles confirmed the hypothesis that the C*08 effect we observed
was not exhibited by other group C1 alleles (Appendix Table B.2).
HLA-B*54, a group Bw6 HLA allele, is not known to bind any KIR molecule. We
therefore tested whether the observed B*54 effect was attributable to C*01, which is
in linkage disequilibrium with B*54 and which encodes molecules that bind KIR2DL2.
This analysis suggested that B*54, not C*01, was the gene driving the observed detri-
mental effect on HTLV-1 outcome (see Appendix B). This result, and the observation
that no other C1 allele shows ‘B*54 -like’ behaviour, indicate that, as postulated, the
interaction between B*54 and KIR2DL2 cannot be explained by direct KIR-HLA
binding.
The most frequent B*57 allele in our cohort is B*57:01, which does not bind
KIR2DL2 [76]. There are therefore two ways in which the observed interaction between
KIR2DL2 and HLA-B*57 could be attributed to ‘classical’ KIR-HLA binding: either
KIR2DL2 might bind a class I HLA molecule whose encoding gene is linked to HLA-
B*57, or the effect might be due to KIR3DL1/S1, which does bind B*57. Analysis of
both these possibilities indicated that they did not explain the KIR2DL2-B*57 effect
(see Appendix B).
117
As shown so far, the enhancement of C*08, B*54 and B*57 -restricted immunity by
KIR2DL2 is not explained by direct binding between the respective HLA molecules and
KIR2DL2. Instead, we suggest that KIR2DL2 binds its HLA-C ligands and indirectly
modulates C*08, B*54 and B*57-restricted T cells. Consistent with this, we found
some weak evidence that KIR2DL2 enhanced HLA class I effects more strongly when
it’s stronger C1 ligands are present (see section 3.9).
3.9 The role of KIR2DL2 ligands
KIR2DL2 binds HLA group C1 molecules and, with weaker affinity, C2 molecules [76].
We hypothesised that KIR2DL2 -dependent enhancement of HLA-mediated immunity
would be greatest in individuals bearing one or two copies of the group C1 ligand.
This required further stratification of the cohorts and only the size of the HCV cohort
allowed for such calculations. We indeed found that in KIR2DL2+ individuals who
had at least one C1 ligand, the effect of B*57 was protective (OR=0.34, p=0.008,
B57+=31, N=332) and it was slightly weakened when KIR2DL2 was present without
its C1 ligand (OR=0.4, p=0.38, B57+=18, N=76). However, the differences are very
small and certainly not conclusive, possibly because KIR2DL2 also binds C2 molecules
and so there are no people in which KIR2DL2 does not have a ligand.
3.10 The role of other KIRs
3.10.1 Other Inhibitory KIRs
It seems unlikely that KIR2DL2 behaves fundamentally differently to other inhibitory
KIRs. The effect of KIR2DL2 may be most apparent because KIR2DL2 is present at
informative frequencies and its C1 and C2 ligands are ubiquitous [76]. We addressed
the role of other inhibitory KIRs in 3 ways. (i) We studied the effect of individual
118
inhibitory KIRs (section 3.7.2), (ii) we investigated whether the number of inhibitory
KIR:ligands had a cumulative effect and (iii) we examined the role of the group A KIR
haplotypes which are dominated by inhibitory KIRs but do not contain KIR2DL2
(section 3.10.3). We found little evidence that the other inhibitory KIRs enhanced
HLA class I-mediated immunity but this may be due to small cohort sizes and masking
by the dominant KIR2DL2 effect.
We quantified the magnitude of KIR inhibitory signals per individual by count-
ing the number of inhibitory KIR that were present with their ligand (KIR2DL1:C2,
KIR2DL2:C1, KIR2DL3:C1, KIR3DL1:Bw4I [78, 224, 225]). Then, we stratified the
cohort for ‘low inhibitory signal’, count<1, and ‘high inhibitory signal’, count<2. The
cut-off (count<1, count<2) was chosen so that there were similar numbers of indi-
viduals in the two strata. In HCV infection, we found that B*57 had an increased
protective effect for individuals within the ‘high inhibitory signal’ category (OR=0.48,
p=0.03, B57+=50, N=518) but the effect was not significant for‘low inhibitory signal’
individuals (OR=0.65, p=0.3, B57+=34, N=262). In HTLV-1 infection, the protec-
tive effect of C*08 was pronounced among individuals with a‘high inhibitory signal’
(OR=0.30, p=0.02, C*08+=29, N=242) but absent in the ‘low inhibitory signal’ group
(OR=0.98, p=0.97, C*08+=29, N=160). Similarly, the detrimental impact of B*54
was increased in individuals that possessed more inhibitory KIRs with their ligands
(OR= 5.31, p=0.001, B54+=43, N=242) compared to individuals with a ‘low in-
hibitory signal’ score (OR= 1.58, p=0.36, B54+=42, N=160). However, the effect of
the ‘high inhibitory KIR signal’ could be attributable to KIR2DL2 as the ‘high in-
hibitory signal’ group is heavily enriched for individuals with KIR2DL2 (especially for
the HTLV-1 cohort). Additionally, the observation (discussed in section 3.10.3) that
the KIR haplotype AA does not have an enhancing effect suggests that even if other
inhibitory signals may contribute, KIR2DL2 is necessary for a detectable effect.
119
3.10.2 Activatory KIRs
We found no evidence that activating KIR were enhancing HLA class I-restricted
immunity. Haplotype B, the more activatory KIR haplotype, enhanced HLA class
I associations but this was only true if the haplotype contained KIR2DL2 (section
3.10.3). We found no evidence that the cumulative presence of activating KIR enhanced
HLA class I restricted immunity. And, as far as it was possible to separate KIR2DL2
and KIR2DS2, which are in tight linkage disequilibrium, the enhancement of HLA class
I restricted immunity appeared to be attributable to KIR2DL2 rather than KIR2DS2
(section 3.7.2).
KIR2DL2 is usually present on haplotypes which contain multiple activatory KIRs
(section 3.10.3). We therefore also explored the possibility that the cumulative pres-
ence of multiple stimulatory KIR (rather than any one individual KIR) is driving the
KIR2DL2 effect. In HCV infection, for each individual, we counted the number of stim-
ulatory receptors of the KIR B haplotype (KIR2DS1, KIR2DS2, KIR2DS3, KIR2DS5,
KIR3DS1 ). Then we simultaneously investigated 1) the effect of B*57 combined with
1 or 2 stimulatory receptors, 2) the effect of B*57 in the presence 3, 4 or 5 stimulatory
receptors and 3) the effect of B*57 in the presence of KIR2DL2. Using backward elim-
ination, we found that only the B*57-KIR2DL2 had a significant effect on the outcome
of infection. Similar analysis for viral load was not possible because of limited cohort
sizes. In HTLV-1 infection, we could apply the same approach only for C*08 and
the status variable because of limiting numbers. The stimulatory KIRs available in
the cohort were KIR2DS2, KIR2DS3 and KIR3DS1 so we considered individuals with
only 1 KIR stimulatory receptor or with 2-3 stimulatory receptors. We found that, as
for B*57 in HCV, using backward elimination only C*08-KIR2DL2 had a significant
impact on outcome. These two results taken together suggest that KIR2DL2 rather
than the cumulative presence of stimulatory receptors of KIR-B haplotype enhance
the HLA class I-mediated immunity (further discussed in section 3.10.3).
120
No KIR3DS1 effect in HTLV-1 infection
We examined more closely the role of the activatory receptor KIR3DS1 in HTLV-1 in-
fection. KIR3DS1, together with its putative ligand HLA-Bw480I, has been associated
with lower viral load and slower progression to AIDS in HIV infection [85]. A suggested
mechanism driving this association is that the activation of KIR3DS1+ NK cells may
not depend on an HIV-specific signal but occur as a direct or indirect consequence of
retroviral infection [2]. Under this explanation, it would be expected that the ability
of KIR3DS1 to suppress viral load in HIV might also be true in HTLV-1 infection, by
lowering the proviral load and protecting from HAM/TSP development. We tested
this hypothesis in the current HTLV-1 Japanese cohort but we found no significant
association of KIR3DS1+Bw480I with either disease outcome or proviral load (Tables
3.6 and 3.7). This result was further confirmed in a Brazilian cohort in [2].
Odds Ratio p-value % in ACs % in HAM/TSP
KIR3DS1 1.19 0.5 37.8% 36.5%
Bw480I 0.79 0.4 35.6% 39.6%
3DS1+Bw480I 1.44 0.3 13.9% 13.5%
Table 3.6: Analysis of KIR3DS1 and HLA-Bw480I on HTLV-1 disease outcome. We
have 180 ACs and 222 HAP/TSP individuals in the cohort.
ACs HAM/TSP
Slope p-value Slope p-value
KIR3DS1 -0.25 0.09 -0.096 0.27
KIR3DS1+Bw480I -0.3 0.1 -0.169 0.12
Table 3.7: Analysis of KIR3DS1 and HLA-Bw480I on HTLV-1 proviral load. We
have 180 ACs and 222 HAP/TSP individuals in the cohort.
121
3.10.3 KIR haplotypes
Two broad groups of KIR haplotypes have been defined: A haplotypes and B haplo-
types [226]. KIR-A haplotypes are dominated by inhibitory KIRs (having 2 activating
and 5 inhibitory KIRs); KIR-B haplotypes which are less tightly defined, are consid-
ered more activatory (having up to 7 activating KIRs and on average 5 inhibitory
KIRs). KIR-A haplotypes do not have KIR2DL2, B haplotypes can but do not always
include KIR2DL2. We investigated whether different haplotypes were associated with
a different enhancement of the HLA class I associations with infection outcome.
For each individual, the following rules for attributing a KIR haplotype were used.
AA: only KIR2DL1, 3DL1, 2DL3 and 2DS4 are present; AB: all the KIRs in haplotype
A were present as well as one or more of KIR2DL2, 2DS2, 2DS3, 2DS5 or 3DS1 and
BB: not all the KIRs in haplotype A were present and at least one of the KIRs in
haplotype B were present.
We found that the AA haplotypes did not have an effect either on outcome of
infection (Table 3.9), nor on viral burden for any of the three HLA molecules studied
(data not shown). The KIR-AB haplotypes enhanced the effect of both protective and
detrimental molecules, for all three HLA molecules studied. The fact that AB but not
AA enhances HLA associations suggests that it is the B haplotype which is responsible
for the AB enhancement. Unfortunately, there are insufficient numbers of individuals
with BB to draw any conclusions about BB haplotypes.
Alternatively, to investigate the enhancement conveyed by the B haplotype we split
the B+ individuals into groups with and without KIR2DL2 (Table 3.9). This clearly
showed that the haplotype B-enhancement was absent when KIR2DL2 was absent
and present when KIR2DL2 was present. We conclude that a haplotype analysis
offers little additional information, KIR-A and KIR-B simply being imperfect markers
for absence or presence of KIR2DL2.
122
KIR Haplotype
AA AB BB
OR
(p-value, allele carriers, cohort size)
HCV B*57 status 0.73 0.57 0.36
(p=0.5, n=22, N=252) (p=0.1, n=44, N=405) (p=0.07, n=18, N=125)
HTLV-1 C*08 status 0.70 0.38 -
(p=0.54, n=28, N=202) (p=0.07, n=30, N=190) -
HTLV-1 B*54 status 1.23 5.35 -
(p=0.65, n=46, N=202) (p=0.003, n=37, N=190) -
Table 3.8: The role of KIR haplotypes. We had less than 5 observations for BB
haplotypes in the HTLV-1 cohort. AA: only KIR2DL1, 3DL1, 2DL3 and 2DS4 are
present; AB: all the KIRs in haplotype A were present as well as one or more of
KIR2DL2, 2DS2, 2DS3, 2DS5 or 3DS1 and BB: not all the KIRs in haplotype A were
present and at least one of the KIRs in haplotype B were present.
KIR Haplotype
B+(AB or BB) B+KIR2DL2+ B+KIR2DL2-
OR
(p-value, allele carriers, cohort size)
HCV B*57 status 0.51 0.40 1.23
(p=0.02, n=62, N=530) (p=0.007, n=49, N=408) (p=0.75, n=13, N=122)
HTLV-1 C*08 status 0.35 0.16 0.69
(p=0.053, n=30, N=200) (p=0.02, n=14, N=102) (p=0.64, n=16, N=98)
HTLV-1 B*54 status 6.25 12.50 2.78
(p=0.001, n=39, N=200) (p=0.004, n=21, N=102) (p=0.18, n=18, N=98)
Table 3.9: KIR-B haplotype enhancement is not seen in the absence of KIR2DL2.
123
3.11 Potential mechanism
The antiviral effector population which is enhanced by KIR2DL2 could be either NK
(Figure 3.4(a)) cells or T cells (Figures 3.4(b) and 3.4(c)). Three observations indicate
that the mechanism is more likely to be T cell-mediated. First, strong binding of the
HBZ viral peptide by multiple HLA class I molecules was associated with asymptomatic
status and this effect was enhanced by KIR2DL2. Although NK cells can exhibit some
peptide dependence, such strong protein specificity is more consistent with T cells.
Second, the KIR2DL2 enhancement could not be attributed to direct binding between
KIR2DL2 and any of the 3 HLA class I molecules investigated. Third, there is no
KIR2DL2 main effect (with or without ligand) on disease outcome or viral burden, an
observation which would have implied a direct NK-mediated effect.
It has been demonstrated that CD8+ T cells which express inhibitory KIRs (Figure
3.4(b)) have elevated levels of Bcl-2 and are less susceptible to activation induced cell
death (AICD) . Of particular interest are reports [79,80,227,228] that inhibitory KIRs
on CD8+ T cells promote the survival of a subset of memory phenotype CD8+ αβ T
cells with enhanced cytolytic potential (Tm1 [229]) by reducing activation-induced cell
death. Tm1 cells have been described in both HTLV-1 and HCV infections, where they
constitute a minority of virus-specific CD8+ T cells but the majority of perforin-bright
cells [230, 231]. Consistent with our findings, these studies have shown that the HLA
molecule that restricts the T cell whose survival is promoted was independent of the
HLA-C molecules that ligated the KIR [79,229]. We suggest that KIR2DL2 may bind
its HLA-C ligands so that when the CD8+ T cell is activated by engagement of its
TCR by the cognate pMHC complex the CD8+ T cell is less likely to undergo AICD.
However, we should also note that there are evidence in the literature suggesting that
the expression of KIRs on CD8+ T cells might suppress their responsiveness [232].
Nevertheless, an extended -even if suppressed- response might result in an enhanced
124
overall net-effect on outcome.
Importantly, Ugolini et al. [79] proposed that inhibitory KIRs promote T cell sur-
vival by increasing the activation threshold of T cells. This may explain why the
HLA-A*02 protective effect in HTLV-1 is not significantly enhanced by KIR2DL2.
A*02 molecules bind peptides significantly more strongly than other alleles (Appendix
Figures B.1 and B.2) and the immunodominant HTLV-1 peptide Tax 11-19 is bound
exceptionally strongly. Therefore, even if the T cell activation threshold were increased,
the strength of signalling may remain above the threshold and consequently the A*02
protective effect cannot be enhanced.
Alternatively, KIR2DL2 in addition to other iKIRs expressed on NK cells can sup-
press their activation and as a result influence the CD8+ T cell mediated response
(Figure 3.4(c)). Recent data, in MCMV infection, show that NK cells can negatively
regulate the duration and effectiveness of virus-specific CD4+ and CD8+ T cell re-
sponses by limiting exposure of T cells to infected antigen-presenting cells [233]. In
another murine study of LCMV infection, activated NK cells cytolytically eliminate
activated CD4+ T cells that affect CD8 T-cell function and exhaustion [234]. More-
over, in a similar experiment, the absence of an inhibitory receptor on NK cells led
to the lysis of activated but not naive CD8+ T cells in a perforin-dependent manner
in vitro and in vivo [235]. Further evidence supporting NK tuning of T cell responses
have been found in [236]. Hence, one can postulate that the expression of the strong
inhibitory receptor, KIR2DL2, on NK cells might limit NK cell activation allowing
CD8+ T cells to have an enhanced response. Interestingly, NK cells have also been
shown to kill activated T cells and this killing is reduced by inhibitory KIR [237,238].
Under both hypotheses, the HLA molecule whose protective/detrimental effects are
enhanced is the HLA molecule that binds the TCR not the HLA that binds KIR2DL2.
As a result, If the CD8+ T cell is restricted by a protective HLA class I molecule (e.g.
B*57 in HCV or C*08 in HTLV-1) it will survive for longer than in a person who is
KIR2DL2+ and its protective effects will be enhanced. Similarly, if the CD8+ T cell
125
is restricted by a detrimental HLA molecule (e.g. B*54 in HTLV-1) it will also survive
for longer and its detrimental effects will be enhanced.
Both these postulated mechanisms could explain 3 striking features of our ob-
servations: i) that KIR2DL2, a receptor typically associated with innate immunity,
enhances HLA class I-associated immunity ii) that the effect cannot be explained by
KIR2DL2 binding the enhanced HLA molecules directly and iii) that both protective
and detrimental HLA effects are enhanced.
126
infected cell
HLA
KIR2DL2
(a) KIR2DL2 acting on NK cells
HLA
TCR
infected cell
CTL
KIR2DL2
HLA-C
(b) KIR2DL2 acting directly on CD8+ T cells
HLA
TCR
infected cell
CTL
HLA-C
infected cell
KIR2DL2
(c) KIR2DL2 acting indirectly on CD8+ T cells
Figure 3.4: Potential mechanisms of KIR2DL2 impact on HLA class I-mediated im-
munity. The antiviral effector population which is enhanced by KIR2DL2 could be
either NK cells (a) or T cells (b,c). KIR2DL2 can be expressed directly on NK cells
or T cells or it might be expressed on NK cells but mediate the response of T cells.
127
3.12 Discussion
We show that KIR2DL2 enhanced several independent HLA class I-mediated effects
in two unrelated viral infections. In HTLV-1 infection, KIR2DL2 enhanced the pro-
tective and detrimental effects of HLA-C*08 and B*54 respectively on disease sta-
tus. KIR2DL2 also enhanced the association between C*08 and low proviral load in
ACs and between B*54 and high proviral load in HAM/TSP patients. Additionally,
KIR2DL2 enhanced the protective effect of HBZ binding by multiple HLA molecules.
Strikingly, on stratifying by KIR2DL2, we observed, for the first time, a protective
effect of C*08 on pvl in HAM/TSP patients and explained the lack of impact of B*54
on pvl in ACs. In HCV infection, KIR2DL2 enhanced the protective effect of B*57 on
spontaneous clearance and the association between B*57 and low viral load in chronic
carriers; a ‘dose effect’ with KIR2DL2 copy number was also observed. This progres-
sive effect is consistent with reports of an association between KIR gene copy number
and the frequency of cell-surface expression of the respective KIR molecule [239,240].
There are two mechanisms by which KIR2DL2 could act: it could enhance either
NK-mediated or T cell-mediated immunity. That is, NK cell killing of virus-infected
cells could be altered by KIR2DL2 expression or, alternatively, the virus-specific CD8+
T cell response could be modified by KIR2DL2 expression either directly on the CD8+
T cells or indirectly on the NK cells. However, our findings indicate that it is the T cell
response that is more likely to be enhanced. First, strong binding of HBZ viral peptides
via multiple different HLA-A and B molecules was associated with asymptomatic status
[185] and this protective effect was enhanced by KIR2DL2. KIR2DL2 is not known
to bind HLA-A or-B molecules (with the exception of B*4601 and B*7301) [76, 78,
223] so it is unlikely that the enhancement of the protective effect of binding HBZ
by KIR2DL2 is due to direct binding between KIR2DL2 and HBZ peptide in the
context of HLA-A and -B molecules. Furthermore, although NK cells exhibit peptide
128
dependence [241], it is hard to reconcile protein-specificity via multiple HLA molecules
with an NK cell-mediated mechanism. Second, the KIR2DL2 enhancement could not
be explained by binding between KIR2DL2 and any of the 3 HLA class I molecules
investigated. Additionally, there is no KIR2DL2 main effect on disease outcome or
viral burden for HCV or HTLV-1 infection. One further observation also suggests
a T cell-mediated mechanism. Two protective genotypes in HCV infection that are
postulated to operate via innate immune mechanisms [220] (namely a SNP upstream
of IL28B and KIR2DL3-HLA-C1 ) had no impact on viral load in chronic infection
[83, 102]. The authors hypothesised that this was because innate barriers offer little
protection once overcome. In contrast, the KIR2DL2/B*57 effect that we report here
had a significant impact on viral load: again, this is perhaps more consistent with
adaptive immunity.
Our results indicate that KIR2DL2 enhances HLA class I-restricted CD8+ T cell-
mediated adaptive immunity. KIRs expressed on both NK cells and CD8+ T cells have
been reported to shape adaptive immunity [81,82,218]. We postulate that, in the face
of chronic antigen stimulation, protective T cells survive longer if they carry KIR2DL2
and therefore exert stronger protection. Likewise, T cells restricted by HLA alleles
associated with increased disease susceptibility also survive for longer in the presence
of KIR2DL2 and so are more detrimental. Hence, KIR2DL2 enhances both protective
and detrimental HLA class I associations.
Alternatively, it is known that NK cells kill activated T cells and that this killing
is reduced by inhibitory KIR [237, 238]. Furthermore, there are recent evidence sup-
porting NK tuning of T cell responses in murine studies of MCMV and LCMV in-
fections [233, 234, 236]. So again, in this context, T cells restricted by protective and
detrimental HLA class I molecules may survive longer in the presence of inhibitory
KIR that suppress NK cell activity and thus their protective and detrimental effect
would be enhanced.
Why does KIR2DL2 enhance T cell responses whereas the other inhibitory KIRs
129
apparently do not? The effect of KIR2DL2 may be most apparent because KIR2DL2
is present at informative frequencies and its C1 and C2 ligands are ubiquitous; i.e.
unlike the other KIR every individual carries a KIR2DL2 ligand.
It will be important to determine whether inhibitory KIRs play a similar role in en-
hancing CD8+ and possibly CD4+ T cell-mediated immunity to other pathogens and
in autoimmune disease. KIR-expressing virus-specific CD8+ T cells have been reported
in other chronic infections including HIV-1, CMV and EBV [232, 242, 243]. Further-
more, in HIV-1 infection, high expression alleles of an inhibitory KIR, KIR3DL1, in
the context of HLA-Bw4I have been associated with slow progression to AIDS [86].
In order to explain protection by an inhibitory KIR the authors proposed a model
based on NK cell development. Our results suggest an alternative explanation, i.e.
that KIR3DL1 enhances protective HLA-B-restricted responses to HIV-1 (B*57 is a
Bw4I allele).
In contrast to previous studies of KIR genotype, which investigated the antiviral
action of NK cells, we investigated the impact of KIRs on HLA class I-mediated an-
tiviral immunity. We find a clear and consistent effect of KIR2DL2. The effect sizes
are substantial: KIR2DL2 carriers with B*57 are 2 times more likely to clear HCV
infection spontaneously; if they fail to clear the virus they have a viral load that is
reduced by 4 logs. Until now, the advantages offered by inhibitory KIRs in virus in-
fections have been unclear. Our data support an alternative role in which inhibitory
KIRs enhance both beneficial and detrimental T cell-mediated immunity in persistent
viral infection.
130
Chapter 4
Do KIRs affect CD8+ T cell se-
lection pressure? - Study design
The study design presented here has been included in an MRC grant proposal sub-
mitted by Dr. Becca Asquith in 2012. The proposal received full funding and will be
implemented by our group.
4.1 Aim
Our aim was to design a study that can help us understand whether KIRs can influence
CD8+ T cell selection pressure as suggested by our findings in Chapter 3.
4.2 Introduction
In Chapter 3 we show that the KIR2DL2 enhancement was observed for multiple
independent HLA class I molecules in both HCV and HTLV-1 infections (Figure3.3).
In HCV infection, KIR2DL2 enhanced the protective effect of B*57 on spontaneous
clearance and also on viral burden in chronic carriers. Importantly, we also observed
a progressive effect with increasing KIR2DL2 copy number. In HTLV-1 infection,
KIR2DL2 enhanced the protective effect of HLA-C*08 on disease status and proviral
load in ACs; and also enhanced the detrimental effect of B*54 and its association with
a high proviral load in HAM/TSP patients. Furthermore, KIR2DL2 enhanced the
protective effect of binding the HTLV-1 viral protein HBZ by multiple HLA-A and -B
molecules.
131
Our results have a number of striking features. Firstly, KIR2DL2 -enhancement
is not mediated by direct binding between KIR2DL2 and the enhanced HLA class I
molecule. Although HLA-C*08, as a group C1 molecule, is expected to bind KIR2DL2,
the most frequent subtype in our cohort (C*08:01, 88%) binds KIR2DL2 very weakly
(comparable to background [223]) and HLA-B*54 and HLA-B*57 are not expected to
bind KIR2DL2 [76, 223]; also, KIR2DL2 is not expected to bind the HLA-A and B
molecules that bind the HTLV-1 viral protein HBZ although it enhances its protective
effect. Secondly, KIR2DL2 does not have a direct impact on disease outcome or viral
burden in any of the examined associations. Thirdly, both beneficial and detrimental
HLA class I associations with infection outcome are enhanced.
Many associations between KIR-HLA interactions and infection outcome have been
reported [83,89,244]. In each case the KIR-HLA effect was linked to direct NK killing
where NK lysis of virus-infected cells is modulated by inhibitory and activatory KIR-
HLA signals (Figure 3.4(a)). Our observations are more consistent with a CD8+ T
cell response that can be modulated directly or indirectly by inhibitory KIRs (Figures
3.4(b) and 3.4(c)). Inhibitory KIRs directly expressed on CD8+ T cells could increase
their survival by upregulating pro-survival molecules like Bcl-2 and by reducing acti-
vation induced cell death (AICD) [79, 227, 228]. Indirectly, it is known that NK cells
help downmodulate the acute response by killing activated CD8+ T cells and that
this killing is reduced by inhibitory KIRs expressed on NK cells [234, 237, 245]. In
both scenarios, CD8+ T cells restricted by ‘protective’ and ‘detrimental’ HLA class I
molecules would be expected to survive longer in the presence of inhibitory KIRs and
thus their effect would be enhanced. Hence, we postulate that KIR2DL2 consistently
enhances the strength of the CD8+ T cell response in a beneficial or harmful manner
depending on the HLA class I restriction.
As a natural next step, we want to investigate this postulated mechanism further.
Specifically, we want test the hypothesis that the CD8+ T cell response is enhanced by
KIR2DL2 by examining whether the CD8+ T cell response is stronger in KIR2DL2+
132
individuals. We plan to use two independent methods: 1) viral sequencing to quantify
selection pressure and 2) modelling to quantify dynamics. We define a ‘strong’ CD8+
T cell response as one that exerts a large anti-viral pressure (by lytic and/or non-lytic
mechanisms). The selection pressure can be quantified by the presence of non-coding
changes in viral epitope regions. Here, we present only the study design of the viral
sequencing approach and its implications.
4.3 Description of study design
Our aim is to design a framework to quantify how HCV-specific T cell selection pressure
is altered in the presence of KIR2DL2 and other inhibitory KIRs. Our study has two
parts: 1) the amplification and massive parallel sequencing of targeted HCV genomic
regions from a longitudinal cohort of acutely HCV-infected individuals and 2) the
extensive bioinformatics analysis of the high-throughput data in order to quantify
CD8+ T cell selection pressure.
Briefly, in the first part of the study (Figure 4.1) our collaborators (see State-
ment of Collaboration) will synthesise cDNA from viral RNA, followed by nested PCR
amplification (PCR1 and PCR2 - 35 cycles each) using genotype-specific degenerate
primers. Phusion high fidelity polymerase will be used for the amplification process.
The PCR product which consists of 3 amplicons (Figure 4.2) will be cleaved and lig-
ated to bridge primers using the Illumina Nextera transposon technology. This will be
followed by sequencing on the Illumina HiSeq platform.
For the bioinformatics analysis, we will follow a pipeline of data processing (Fig-
ure 4.3) in order to perform thorough quality control of the viral sequences, reliable
sequence alignment and quantification of CD8+ T cell selection pressure. We will
quantify the selection pressure based on both experimentally-defined and predicted
epitopes which are most likely to be presented by each individuals HLA-A and B
alleles and determine the ratio of non-synonymous to synonymous changes (dn/ds)
133
in these peptides. The dn/ds ratio will be compared for the same peptide between
KIR2DL2+ and KIR2DL2- HLA class I matched individuals with CD4+ T cell count
as a covariate (as a control for HIV coinfection). In each step of the analysis the
appropriate software (Figure 4.3) or customised scripts will be used.
To this day, in a pilot study our collaborators have successfully sequenced viral
genomes from 11 HCV-infected individuals which we used in order to construct and
partially implement a preliminary version of this pipeline. Specifically, we investigate
the error sources involved in the study, the quality of the bases and the reads, the
sequence alignment and the available coverage. The full dataset will be analysed by
my successor.
Serum
RNA extraction
RNA
cDNA
RT‐PCR
Primers
Amplicon x3
Amplification
PCR1 – 35 cycles
Amplification
PCR2 – 35 cycles
Amplicon x3
Illumina Nextera strategy
(chopping, tagging, adapter)
Sequencing
(Illumina HiSeq)
Figure 4.1: Experimental setup followed for obtaining the NGS HCV data
134
Figure 4.2: The HCV regions covered by the three amplicons sequenced in this study.
Note: This figure is adapted by a schematic provided by Dr Heather Niederer.
Figure 4.3: Bioinformatics pipeline for post-processing the HCV NGS data
135
4.4 Data
4.4.1 Longitudinal HCV cohort
We have 34 patients co-infected with HCV and HIV monitored prior to treatment.
HCV and HIV viral load measurements and total CD4+ and CD8+ T cell counts are
available in the cohort. Most of the patients are infected with genotype 1a and few
with genotype 1b. For each individual, we have 2-6 timepoints with HCV viral load
>10,000 RNA copies/ml over a period of 6-12 months (Figure 4.4). In total, we have
161 samples which are both KIR2DL2 and HLA typed. We are currently in the process
of recruiting 20 more patients. Based on this information we estimate we will have
sufficient power (80%) to detect a small to moderate effect size [221](d=0.28, a=0.05
two-tailed).
4.4.2 Predicted viral peptides
We used the NetMHCPan epitope prediction algorithm and the H77 HCV genotype
1a sequence (Appendix A) in order to identify the most epitope dense regions of the
HCV genome. The HCV proteins NS3, NS5A and NS5B had the most strong binding
epitopes. For these regions, we obtained the top 5 strong binding peptides for each
HLA-A and -B molecule of each individual in the cohort. We obtained in total 45
predicted peptides (Figure 4.5) which will be used in order to quantify CD8+ T cell
selection pressure. In the list of peptides we also include the ones that differ at 1
position relevant to the consensus. All the predicted peptides will be considered with
caution and studied in parallel to experimentally-defined epitopes available in the
literature.
136
24
68
KIR2DL2+
Months
Lo
g1
0 v
ira
l lo
ad
0 3 6 9 12 15 18
24
68
KIR2DL2−
Months
Lo
g1
0 v
ira
l lo
ad
0 3 6 9 12
Figure 4.4: Longitudinal HCV viral load data from 34 patients coinfected with HCV
and HIV. There are 22 KIR2DL2+ individuals and 12 KIR2DL2- individuals in the
cohort. Each color-symbol combination represents the viral load measurements in
RNA copies/ml (log10 scale) for each individual.
137
(a) NS3 and NS5A
(b) NS5B
Figure 4.5: Predicted HCV epitopes for viral proteins NS3, NS5A and NS5B. The
schematics are obtained using the software Geneious.
138
4.5 NGS data post-processing
In order to construct and assess a preliminary version of the bioinformatics pipeline, we
use 11 sequenced samples (a single time point from 11 HCV-infected individuals). We
focus mainly on the quality of the bases, the depth of coverage and the overall reads, the
alignment of the sequences and the challenges involved in identifying genuine sequence
variation in NGS data.
4.5.1 Error sources
When generating NGS data there are many processes involved that can generate errors
in the sequenced data. These errors can be randomly distributed across the data (e.g.
base calling errors) or they can be propagated systematically in the data analysis
pipeline (e.g. PCR errors). In Figure 4.6 we present the main sources of error involved
in our project. These are: 1) reverse transcription, 2) PCR amplification, 3) sequencing
and 4) contamination. Stringent steps are taken in order to minimise these errors
during the data generation. For reverse transcription and PCR amplification, the
highest fidelity enzymes are used (superscript 3 and phusion high fidelity respectively).
For base calling the Illumina protocol is applied and an extensive quality control is
performed (see below). Finally for limiting contamination, the data are processed in
batches and a cleaning step takes place between pre- and post- PCR product handling;
also a water solution is used as a control.
All the above errors occur in processes taking place before the bioinformatics anal-
ysis and can of course hinder SNP identification. Further errors in the context of
‘variant calling’ can also be introduced downstream of the bioinformatics pipeline and
specifically at the sequence alignment step; we address this issue separately later on
in this chapter.
139
Figure 4.6: Errors arising during the NGS data generation.
140
4.5.2 Quality control
Firstly, we perform a quality control on the pilot data. We use the standalone software
FastQC 1 in order to examine the confidence that we should have in the base calling
process. In all the 11 pilot datasets we observe that the quality of the base calling
drops significantly as we approach the end of the reads (Figure 4.7(a)). Hence, we
decide to apply two ‘cleaning’ procedure in order to improve the quality of the data:
a) trim a part of the end of each read and b) set a requirement on the percentage of
bases in each read that have a Phred quality score ≥ Q20 (i.e. ≤ 1 error in 100 bps).
For this, we use the standalone software FastX 2. Here, we show the results for one
sample but similar results are obtained for all 11 samples.
There is a trade-off between quality and quantity, since we want to achieve the best
possible sequence quality whilst maintaining the maximum possible amount of data.
When we trim the last 50 bps from each read (Figure 4.7(b)), the quality improves
remarkably but we loose ≈ 33% of the bps. When we set the percentage of bases
with quality ≥ Q20 to i) 80%, ii) 90% and iii) 100% per read (Figure 4.8(a)-4.8(c))
we loose approximately 15%, 27% and 80% of the total bases respectively. Since the
bulk of the ‘bad’ quality bps is found at the end of the reads, we decide to combine
the two ‘cleaning’ filters by trimming the last 10 bps of each read and setting the
percentage of bases with Phred quality score ≥ Q20 for each read to 80% (Figure 4.9).
We estimate that the combination of the two base-trimming processes results in the
loss of approximately 25% of the total sequenced bases which still allows for a high
overall coverage.
Next, we looked at the nucleotide frequencies per bp across all reads. We find
that overall the HCV sequence reads have a higher GC content (≈ 60%) compared
to AT content (≈ 40%) as also reported in the literature [246]. We observe no other
1http://www.bioinformatics.babraham.ac.uk/projects/fastqc2http://molecularevolution.org/software/genomics/fastx
141
(a) No Filter (b) Trim last 50bps
Figure 4.7: Base calling quality control (Filter a). The unfiltered data (a) and the
data after trimming the last 50 bps from each read (b). The trimming improves the
quality remarkably but results in the loss of approximately 33% of the sequenced bps.
biases for any specific nucleotide at any read position apart from the first 10 bps
(Figure 4.10). This is an extended signature for sites of transposase-catalyzed adaptor
insertion that as we also verified, weakly resembles the insertion preference of the
native Tn5 transposase (AGNTYWRANCT, where N is any nucleotide, R is A or G,
W is A or T, and Y is C or T) [247].
142
(a) 80% Q20 (b) 90% Q20
(c) 100% Q20
Figure 4.8: Base calling quality control (Filter b). We set the percentage of bases with
quality ≥ Q20 to i) 80% (a), ii) 90% (b) and iii) 100% (c). We loose approximately
15%, 27% and 80% of the total sequenced bps, respectively.
143
Figure 4.9: Optimal base calling quality filter. As an optimal approach, we combine
the two ‘cleaning’ filters by trimming the last 10 bps of each read and setting the
percentage of bases with Phred quality score ≥ Q20 for each read to 80%. This
approach leads to the loss of approximately 25% of the total sequenced bps.
144
Figure 4.10: Adapter sequence bias
4.5.3 Sequence alignment and coverage
Having improved the quality of the sequence data, our next aim is align them to a
reference sequence in order to obtain the coverage per bp, i.e. the number of bases
aligning to each position of the HCV amplicon that we sequence. We obtain the
reference sequence by creating a consensus sequence from 78 available HCV 1a genotype
sequences available in GenBank. However, we should note that when all the timepoints
per patient are sequenced, all the timepoints will be aligned to the first timepoint of
each patient. Only the first timepoint will be aligned to this consensus sequence;
hence, the analysis will be performed on a different consensus per patient so that the
identification of variants involves less uncertainty.
We perform the alignment using the Mosaik software 3. One of the most important
parameters for the alignment configuration is the percentage of bases per read that
can be mismatched to the consensus. This parameter should be carefully chosen since
3http://bioinformatics.bc.edu/marthlab/Mosaik
145
if it too strict true variants would not be identified but if it is too large, the quality
of the alignment drops. In order to obtain an estimate of this parameter we align the
78 HCV genotype 1a sequences and calculate the percentage of the main variant (the
second most common nucleotide) at each position. We find that at each position of the
HCV genotype 1a genomes and for the regions that we are investigating (Figure 4.2),
the second commonest nucleotide is present at an average frequency of approximately
11% (Figure 4.11). In other words, at each position approximately 8/78 sequences
have the same nucleotide to each other but different to the consensus. We use this
value for the mismatch parameter. The algorithm applies a high gap penalty which
we set at the default value since we want to limit the presence of indels that can mask
the presence of true variants.
Next, we examined the depth of coverage that we could obtain using this high-
throughput approach. As shown in Figures 4.12, the depth of coverage that was
achieved varied across the three amplicons. For some samples all the amplicons had a
high coverage (> 20000, Figure 4.12(a)) while in some other cases only one or two out
of the three amplicons provided a high coverage (Figure 4.12(b)). In the 11 samples
analysed, the third amplicon that covers the NS5B region was consistently having
the highest coverage (A3: 11/11; Figure 4.13) and this deteriorated for the other two
amplicons (A1: 8/11 and A2: 6/11 were amplified successfully). This trend implies
that we will have a better overall coverage of the NS5B region and hence, we might
be more powered to detect variants occurring in this part of the HCV genome. The
failure to amplify some of the amplicons can be potentially explained by two main
factors: 1) the high sequence variation even within HCV genotype 1a that required
the use of ’degenerate’ primers and 2) the lack of a complete cDNA extension over
the whole 7kb region which might imply that amplicons A1 and A2 have less viruses
represented in the cDNA than A3. However, despite several strategies applied by our
collaborators in order to improve the amplification of amplicons A1 and A2 such as the
repeat of cDNA synthesis and the use of different primers, none provided a substantial
146
1000 1500 2000 2500 3000
01
02
03
04
05
06
0
Position
% M
ain
va
ria
nt
3000 3500 4000 4500 5000
01
02
03
04
05
06
0
Position
% M
ain
va
ria
nt
5000 5500 6000 6500
01
02
03
04
05
06
0
Position
% M
ain
va
ria
nt
6500 7000 7500 8000
01
02
03
04
05
06
0
Position
% M
ain
va
ria
nt − mean=11.3%
Figure 4.11: Sequence variation of 78 HCV genotype 1a strains.
147
improvement. Although, this will limit the amount of viral variants that we can infer,
it will still allow for enough coverage to test our hypothesis since we also have multiple
timepoints per patient.
148
0 1000 2000 3000 4000 5000 6000 7000
010
000
3000
050
000
Position
Cove
rage
(a) Full amplicon coverage
0 1000 2000 3000 4000 5000 6000 7000
01
00
00
30
00
05
00
00
70
00
0
Position
Cov
era
ge
0 1000 2000 3000 4000 5000 6000 7000
0e
+0
04
e+
04
8e
+0
4
Position
Cov
era
ge
(b) Partial amplicon coverage
Figure 4.12: Coverage of HCV amplicons. For some patient-timepoint samples all the
amplicons were amplified (a) while for some others only some of them were amplified
(b). For the alignment shown here the MOSAIK software is used; the consensus
sequence is obtained based on the 78 HCV genotype 1a strains available on GenBank
and the mismatch parameter is set to 10% per read.
149
A1 A2 A3
0e+0
04e
+04
8e+0
4
Amplicons
Cov
erag
e pe
r bp
Figure 4.13: Coverage per HCV amplicon. For the alignment the MOSAIK software
is used; the consensus sequence is obtained based on the 78 HCV genotype 1a strains
available on GenBank and the mismatch parameter is set to 10% per read.
150
4.6 Future work
The main aim of this project is to investigate how KIRs, and in particular KIR2DL2,
can shape HLA class I-mediated immune control in HCV infection. We propose to
address this question by studying the difference in selection pressure exerted on epitope
regions of the HCV genome between KIR2DL2+ and KIR2DL2- individuals who are
HLA class I matched. The available data include a cohort of 34 HCV/HIV coinfected
individuals with multiple high-throughput sequenced timepoints.
In a pilot study that we performed on 11 sequenced timepoints (not from the same
individual) we identified error sources, explored the quality of the base calling, applied
‘cleaning filters to improve this quality and also examined how we can obtain an optimal
sequence alignment. However, other possibilities can be explored in order to further
improve the alignment such as the use of an ambiguous reference sequence (based on
the IUPAC) code (Appendix Figures C.1, C.2 and C.3) and the local realignment of
areas of particular interest such as the epitope regions.
In order to study selection pressure, the next step is to be able to reliably identify
SNPs present in the amplicons and study whether they result in coding changes or
not. To do this we need to use several ‘variant calling’ algorithms in order to identify
true variants and set a reasonable threshold for defining a SNP as true positive. The
high coverage obtained should help in reducing the uncertainty involved in ‘variant’
and allow for detection of low frequency SNPs. However, it is important to note that
this is highly associated to the primers we used and the main genotype studied - 1a.
Therefore, if a patient is infected with genotypes other than 1a that differ significantly
from it, we will not not able to capture this variation. Once variants have been
detected, the selection pressure in both predicted and experimentally defined HCV
epitopes can be quantified using the ‘dn/ds’ ratio.
Additionally, using the available longitudinal cohort and the model described in
151
[248], we can examine the viral escape dynamics in KIR2DL2+ and KIR2DL2- in-
dividuals. Of course, the next-generation sequencing of this longitudinal HCV/HIV
coinfected cohort provides a unique set of data that can help in exploring hypothe-
ses not only involving KIR- and HLA-mediated immunity but also other important
immunological questions.
152
Chapter 5
Lytic and non-lytic CD8+ T cell
responses
The analysis and findings presented in this chapter have been published in Elemans
M., Seich al Basatena N.-K. et al., PLoS Comput. Biol. 7(9), 2011 [3].
5.1 Aim
Our aim was to investigate whether the experimental data presented in [42] best sup-
ports a lytic or non-lytic mechanism of CD8+ T cell control.
5.2 Introduction
There are many factors that suppress viral load in HIV/SIV infection: 1) CD8+ control
of infected cells, 2) virus cytopathicity, 3) innate immunity, 4) target cell limitation,
5) antibody responses and others. Here, we focus on the CD8+ cellular arm which
is initiated once the peptide-major histocompatiblity complex molecules (pMHC) dis-
played on the surface of cells presenting antigen are recognised by the T cell receptor
(TCR).
In SIV infection, one of the most direct and strong evidence of the importance of
CD8+ T cells in controlling SIV infection in vivo is the observation that, on depleting
CD8+ T cells and for both acute and chronic infection, SIV-1 viral load increases by
0.5 to 1 log [41, 166,249].
153
There are two main CD8+ T cell mechanisms that can mediate viral growth in
HIV/SIV infection [250]: (i) direct, lytic killing and (ii) indirect, viral suppression
by secretion of soluble factors. For direct killing, the immunological synapse (IS)
that forms at the contact site of the T cell and the infected target facilitates the
transportation of lytic granules (which can contain the apoptotic protein called perforin
and granzymes) to target cells [251]. For indirect killing, production of antiviral soluble
factors such cytokines and chemokines (e.g. IFN-γ and RANTES) can potentially
prevent infection of uninfected cells by free virus [56] or decrease production of viral
particles by infected cells [58,252,253]. Interestingly, in [254] using a high-throughput
single-cell assay, the authors found that the majority of antigen-stimulated CD8+ T
cells that secrete IFN-γ did not exhibit cytotoxic responses, indicating that lytic and
non-lytic effector functions might be independently regulated. In addition, in [255],
using multiple lytic and non-lytic markers of CD8+ T cell activity it is shown that
these responses can be functionally segregated in vivo. Hence, it can be suggested that
CD8+ T cells might choose effector function depending on viral agent, site of infection
and pathogenicity [256].
Recently, ground-breaking studies reported that following CD8+ cell depletion in
SIV-infected macaques, viral load robustly increased, however the lifespan of SIV-
infected cells was unaltered [42, 167]. The two groups used ART to block further
rounds of infection to study the turnover of SIV-infected cells. They reported that
when the lifespan of productively infected cells was measured there was no detectable
difference between control macaques with an intact CD8+ T cell response and CD8+
T cell-depleted macaques. This unexpected result led to the suggestion that SIV might
be controlled primarily via non-lytic mechanisms.
The aim of this project was to explore further, using an extensive set of ODE models
that describe both lytic and non-lytic effector function, whether the experimental data
presented in [42] best supports a lytic or non-lytic mechanism of CD8+ T cell control.
154
5.3 Methods
5.3.1 Experimental data
The experimental data were produced by collaborators for the study published in [42]
and are briefly described below.
Ten rhesus macaques infected with SIVmac239 were divided into two groups. Group
A consisted of 5 CD8+ T cell-depleted animals during early chronic phase. Group B
included 5 CD8+ T cell-depleted animals during the late chronic phase. Antiretroviral
therapy (ART) was administered to all animals in both phases. To deplete CD8+ T
cells, the OKT8F was given for 3 consecutive days (Group A, days 5860 after infection;
Group B, days 177179). ART (PMPA and FTC) was given for 28 consecutive days
during both early and late chronic infection (starting at d63 and d168 for group A,
and d63 and d182 for group B). Plasma viraemia was quantified by real-time reverse-
transcriptase PCR. CD8+ and CD4+ T cell counts were measured using multicolor
flow cytometric analysis. For almost all points for which viral load was recorded,
CD4+ and CD8+ T cells measurements were available . These studies were approved
by the Emory University and University of Pennsylvania Institutional Animal Care
and Use Committees. All this experimental work was performed by our collaborators
and further details can be found in [42].
155
5.3.2 Lytic models of infection
Classic model
The basic lytic model of HIV infection describes the dynamics of uninfected target
cells (T ), productively infected cells (T ∗) and free virus V .
T = λ− βTV − δTT (5.1a)
T ∗ = βTV − δIT∗ − κET ∗ (5.1b)
V = pT ∗ − cV (5.1c)
Where λ is the influx of uninfected CD4+ cells (cells d−1), β is the infection rate,
δT and δI are the death rate of uninfected and infected CD4 cells respectively, κ is the
CD8+ killing rate of infected CD4+ T cells, p is the production rate of free virions
and c is the clearance rate of free virions (all measured per day). In all the models
presented here, the fraction of CD8+ T cells, E, is given by the empirical function
calculated by a linear interpolation between time points.
We also considered a model with two populations of productively infected cells
(e.g. CD4+ T cells and macrophages), T ∗ and M∗ that follow different death rates,
δI and δM respectively.
T = λ− βTV − δTT (5.2a)
T ∗ = βTV − δIT∗ − κET ∗ (5.2b)
M = λ− βMV − δTM (5.2c)
M∗ = βMV − δMM∗ − κEM∗ (5.2d)
V = pTT∗ + pMM∗ − cV (5.2e)
Early killing model
In [257] an alternative lytic model is described where CD8+ T cells limit their killing to
the phase prior to viral production. Two populations of uninfected cells are included
156
in the model; recently infected cells which are susceptible to CD8+ T cell killing,
I∗ and productively infected cells that evade CD*+ T cell killing via MHC class I
downregulation, P ∗. Uninfected CD4+ T cells (T ) are still described by 5.1a but the
rest of the dynamics are altered as follows:
I∗ = βTV − γI∗ − δII∗ − κEI∗ (5.3a)
P ∗ = γI∗ − δpP∗ (5.3b)
V = pP ∗ − cV (5.3c)
Where γ is the transition rate from I∗ to the P ∗ population of infected cells and
δp is the death rate of the P ∗ population (d−1).
Late killing model
Another known lytic model described in [258] suggests that infected cells produce only
few virions early in infection and the majority of virions is produced at a later stage
just before cell death. This cytopathic effect of the viral infection can be captured by
two populations of infected cells; recently infected cells that do not produce virions
and die at a negligible rate, L∗, and productively infected cells, A∗. The dynamics are
given by:
L∗ = βTV − γL∗ − τL∗ − δIL∗ (5.4a)
A∗ = τL∗ − δAA∗ − κEA∗ (5.4b)
V = pA∗ − cV (5.4c)
Where τ is the transition rate from L∗ to the A∗ population of infected cells and
δI and δA are the death rates of the L∗ and A∗ populations (both d−1) respectively.
157
5.3.3 Non-lytic models of infection
Blocking infection model
The non-lytic models are similar to an extent with the lytic models regarding the
dynamics of the uninfected and infected cells and the viral production but do not
include direct CD8+ T cell killing. Instead, CD8+ T cells decrease the infection rate,
β, or the virion production rate, p, by a fraction u:
1
1 + ηE(t)(5.5)
The dynamics of a non-lytic model where viral infection is decreased can be de-
scribed by:
T = λ−(
1
1 + ηE
)
βTV − δTT (5.6a)
T ∗ =
(
1
1 + ηE
)
βTV − δIT∗ (5.6b)
V = pT ∗ − cV (5.6c)
We also fit a biphasic non-lytic infection model, similar to Eqs. 5.2:
T = λT −(
1
1 + ηNE
)
βTV − δTT (5.7a)
T ∗ =
(
1
1 + ηNE
)
βTV − δIT∗ (5.7b)
M = λM −(
1
1 + ηNE
)
βMV − δTM (5.7c)
M∗ =
(
1
1 + ηNE
)
βMV − δMM∗ (5.7d)
V = pTT∗ + pMM∗ − cV (5.7e)
158
Blocking production model
The dynamics of a non-lytic model where viral production is decreased can be described
by:
T = λ− βTV − δTT (5.8a)
T ∗ = βTV − δIT∗ (5.8b)
V =
(
1
1 + ηE
)
pT ∗ − cV (5.8c)
We also fit a biphasic non-lytic production model:
T = λT − βTV − δTT (5.9a)
T ∗ = βTV − δIT∗ (5.9b)
M = λM − βMV − δTM (5.9c)
M∗ = βMV − δSM∗ (5.9d)
V =
(
1
1 + ηE
)
(pTT∗ + pMM∗)− cV (5.9e)
5.3.4 Model fitting and selection
Model fitting The four different lytic and non-lytic models were fitted to data of
virus load and total CD4+ T cell population, scaled to obtain equal mean. E(t) was
obtained by linear interpolation of experimental data. ART-treatment was simulated
by setting infection rate β = 0. To fit the models to the data we used the pseudorandom
algorithm in the modFit-function of the FME-package in the statistical software R.
Model selection The small sample (second-order) bias-adjusted Akaike Informa-
tion Criterion (AICc) [259, 260] was used to compare the fit of the models. AICc
adjusts for differences in number of parameters (5.10). The model with the lowest
AICc is considered to describe the experimental data best and the comparison be-
tween the models is defined by ∆i (Equation 5.11). As a rule of thumb, we assumed
159
considerable support for models with an AICc within two of the lowest; if models differ
by three to seven AICc units from the minimum AICc there is some support for the
model with the higher AICc while a difference larger than 10 suggests that the model
is very unlikely to describe the underlying data [260]. Note: This part of the study
was performed by M. Elemans.
AICc = Nln
(
SSR
N
)
+ 2K +2K(K + 1)
N −K − 1(5.10)
where N is the data sample size, K is the number of parameters and SSR is the
residual sum of squares of the fitted model.
∆i = AICci −min(AICc) (5.11)
where AICci is the AICc for model i and min(AICc) is the minimum AICc value
of all the models fitted to the data.
5.4 Model comparison
In summary, we fitted 4 lytic and 4 non-lytic models given in Table 5.1:
Although the quality of the fits is rather poor (Figure 5.1), our results show clear
and consistent support for the non-lytic model in which CD8+ T cells reduce infection;
none of the lytic models receive any support for most data sets. The best-fitting
non-lytic model (non-lytic model ii, in which non-lytic factors reduced new infection
events and there are two populations of productively infected cells) was compared with
each of the lytic models in turn. Furthermore the non-lytic model in which infection
was reduced performed consistently better than the non-lytic model in which virion
production was reduced though the differences in performance were not as large as for
the comparison between lytic and non-lytic models (Table 5.2 and Figure 5.2).
The reason why the lytic models fail can be seen from studying the equations. In
the lytic models the rate of post-ART decline in viral load is determined by infected cell
160
Model name Description Equations
Lytic i (Li) Classic model 5.1
Lytic ii (Lii) Biphasic classic model 5.2
Lytic iii (Liii) Early killing model 5.3
Lytic iv (Liv) Late killing model 5.4
Non-Lytic i (NLi) Blocking infection model 5.6
Non-Lytic ii (NLii) Biphasic blocking infection model 5.7
Non-Lytic iii (NLiii) Blocking production model 5.8
Non-Lytic iv (NLiv) Biphasic blocking production model 5.9
Table 5.1: Models fitted to experimental data.
Models AICc mean difference p-value
NLii vs Li 15 0.043
NLii vs Lii 26 0.028
NLii vs Liii 40 0.018
NLii vs Liv 39 0.018
NLii vs NLiii 10 0.018
Table 5.2: The best fitting model, the biphasic non-lytic model where infection of
new cells is blocked by non-lytic factors, was compared with each of the lytic models
in turn as well as with the biphasic non-lytic model where viral production is blocked.
The AICc consistently provided support for the non-lytic model of blocking infection.
All the models fitted are summarised in Table 5.1. All the tests were performed using
a paired two-tailed Mann-Whitney test.
161
death due to viral toxicity and CD8+ T cell killing. Under the CD8+ T cell depletion
regime post-ART decline in viral load is solely determined by infected cell death due to
viral toxicity. Thus, to predict the similar post-ART decline under the two treatment-
regimes lytic models need to attribute a small role to CD8+ cells. This small role
of CD8+ T cells is poorly compatible with the increase in viral load following CD8+
T cell depletion in the absence of ART. Therefore, lytic models cannot accurately fit
both post-ART decline and the increase in viral load upon depletion whereas non-lytic
models can. Consequently models with a non-lytic component are likely to consistently
outperform similar models with a lytic component.
We conclude that although self-consistent hypotheses can be constructed in which
CD8+ T cells exert their antiviral effects by lysis without a detectable impact on
infected cell lifespan, these models are poorly predictive and a non-lytic model provides
a better explanation of the viral load and CD4+ T cell dynamics.
162
Figure 5.1: Experimental viral load and percentage CD4+ T cells (filled dots on left
column and right column respectively) and the best fitting non-lytic model (non-lytic
model ii; solid lines) and the best-fitting lytic model (lytic model i; dashed lines).
Number of parameters: 10 and 8 respectively. Model selection is based on AICc which
controls for the number of parameters in the model. More fits can be found in the
supplementary material of [3]. For animals in Group B, data for CD4+ and CD8+ T
cells was only available from 45 days after infection, hence viral load data before this
time point have not been fitted. Note: This figure was produced by M. Elemans.
163
Figure 5.2: The AICc of the model minus the AICs of the best fitting model for
that animal is shown. A large difference represents a poor fit. Overall the model
in which CD8+ T cells operate via a non-lytic mechanism which reduces infection
(non-lytic model ii) offers the best fit in 6 out of 7 cases, a lytic model only offers the
best fit in 1/7 cases. The 4 lytic models are i) the basic model ii) an extension of
the basic model to include two populations of productively infected cells iii) a model
following Klenerman et al. [258] in which SIV is cytopathic and iv) a model following
Althaus et al. [257] in which CD8+ T cell killing is limited to the early non-productive
stage of the viral lifecycle. The 4 non-lytic control models were: i) a model in which
non-lytic factors reduced new infection events ii) an extension of model i to include
two populations of productively infected cells iii) a model in which non-lytic factors
reduce virion production iv) an extension of model iii to include two populations of
productively infected cells. The values are given in Appendix Figure D.1. Note: This
figure was produced by M. Elemans.
164
5.5 Discussion
SIV-infection of rhesus macaques is one of the most widely used animal models of
HIV-1 infection. Therefore, the observation that the increase of viral load that followed
CD8+ T cell depletion in SIV-infected macaques was not ‘accompanied’ by an increase
of infected cell lifespan [42,167] raises important questions not only in SIV but also in
HIV-1 infection. This observation is reminiscent of the findings of the Klenerman et
al. study which found that CD8+ T cell-mediated lysis can reduce viral load but has
small effects on the half-life of infected cells. [258].
A number of hypotheses can potentially explain this striking observation of the
Klatt et al. and Wong et al. findings. Firstly, the CD8+ T cell control might be
primarily non-lytic. Secondly, the CD8+ T cell control may be lytic but ART treatment
impairs CD8+ T cell function. Thirdly, CD8+ T cell lytic killing might occur early and
prior to viral production [257] or late and just before the cell would die anyway [258].
Fourthly, the CD8+ T cell control is lytic but the measurements of infected cell lifespan
are accurate enough to discern a difference.
Our study finds evidence that a non-lytic mechanism, especially via blocking in-
fection, explains best the SIV dynamics during acute and chronic infection across all
animals studied. The models include a low number of parameters and require the fit-
ting of both viral load and CD4+ T cell count data. As a consequence, the divergence
between prediction and observation (particularly for CD4+ T cell count) was large.
However, the model comparison presented here is based on the AIC and selects for
the model that approaches better the underlying ‘true’ dynamics amongst all models
fitted [259].
In [3], the rest of the above hypotheses are also explored. The authors find that
ART treatment can impair CD8+ T cell function but the difference between control
and CD8+ T cell depleted animals remains small. In addition, they find that both
165
the early and late CD8+ T cell killing models provide qualitative consistent results
but only for relatively narrow model parameter ranges. Interestingly, the authors find
that the limited accuracy of the data could offer a plausible explanation for the lack
of difference.
Hence, taking all of the above observations together, we conclude that although
the lack of an effect on the lifespan of infected cells found in [42, 167] cannot exclude
a lytic CD8+ T cell effector function, the models of non-lytic control provided better
overall predictions for the time course of infection and a potential reason why no
difference in lifespan was detected between control and depleted animals. In [261], the
authors use a biologically-directed modelling approach to study the Klatt et al. data.
Their study involves a high number of biological parameters (which make the model
hard to track) and suggests that aspects of the HIV lifecycle such as the eclipse phase
and an exponential pattern of virion production can explain the lack of difference in
infected cell lifespan between CD8+ T cell depleted and control SIV-infected macaques.
However, in the models studied here we have considered such properties for the lytic
effector function but the non-lytic response could still better explain the data although
we could not exclude a lytic mechanism. It is afterall likely that these two effector
mechanisms act in parallel. Importantly, the findings in [3] shows that there is a clear
need for more accurate experimental measurements before we can draw more robust
conclusions regarding the mechanism of CD8+ T cell control in HIV/SIV infection.
166
Chapter 6
A Cellular Automaton model of
CD8+ T cell responses
6.1 Aim
Our aim was to develop and implement a Cellular Automaton model that captures
both temporal and spatial aspects of HIV/SIV dynamics and can be used to model
both lytic and non-lytic CD8+ T cell responses.
6.2 Introduction
In order to study HIV/SIV dynamics including CD8+ T cell viral suppression and
its effect on viral escape, we develop and implement a 3D Cellular Automaton (CA)
model. A CA is an individual-based computer simulation of a system that evolves
in time on a multidimensional (usually 2D or 3D) lattice of nodes. A CA model
1) simulates both spatial and temporal characteristics of the system, 2) can be used
to study both the individual as well as the population behaviour and 3) can be easily
redefined to study different aspects of the underlying biological processes. Agent-based
models (ABMs) - which include CA models - have been used to study HIV infection
in multiple studies [262–264]. However, they can be computationally demanding and
require a substantial knowledge of the underlying biological principles. A more detailed
description of the successes and challenges of agent-based models is given in [265].
The CA model that we developed incorporates rules on cell motility, the reproduc-
167
tion and death of infected cells, the eclipse phase of viral production, CD8+ T cell
killing and conjugate formation, as well as the lifespan and influx of uninfected cells.
The timescale of the model can integrate fast and slow processes and can therefore
be used to simulate the course of the infection during both acute and chronic phases.
Using this approach, we can produce in silico experiments of viral escape dynamics
under both lytic and non-lytic CD8+ T cell effector functions.
6.3 Model description
We use a three-dimensional Cellular Automaton (CA) to simulate the spread of HIV/SIV
infection in a small portion of the spleen (roughly 0.05 − 0.5%). The CA (see Figure
6.1) is a lattice which consists of nodes and edges and is updated at every timestep (30
sec). Each node of the lattice represents a cell or part of the cell and has 26 neighbours.
The grid is governed by toroidal boundary conditions where a cell leaving one side of
the lattice reappears on the opposite side.
The cell populations included in the model are given in Figure 6.2. CD4+ T
cells can be uninfected, infected with the wild-type strain or infected with a variant
strain of the virus. The CD8+ T cell population that we explicitly model represents
a population of all CD8+ T cell clones which are specific for a single epitope. The
model also includes macrophages, the reticular network of the spleen and a generic
splenocyte population.
Most cells -apart from macrophages (MΦs) that occupy four nodes- are represented
by one node on the lattice; this is to reflect the difference between the diameter of T
cell which is 7µm [266] and that of macrophages which is calculated to be 10− 16µm
and has been implemented as such in [267]. An additional set of nodes represents
the Reticular Network (RN), a rigid cellular structure, found in the spleen. These
nodes are immobile and act as spatial obstacles to the overall movement of cells.
Only a small percentage of nodes is considered to be unoccupied since the spleen is
168
a dense organ (≈ 1%). The nodes that are left after setting the known frequencies
of the specific cell populations within biological ranges (see Appendix Tables E.1 and
E.2) are considered to be unspecified splenocytes. The initialisation values of all the
populations included in the simulations are given in Appendix Table E.1. The cellular
automaton is implemented (by the author of this thesis) in C++.
Figure 6.1: A snapshot of the 3D cellular automaton model. The different coloured
nodes represent different cell populations considered in the model (see Figure 6.2).
Figure 6.2: Cell populations considered in the 3D cellular automaton model.
169
6.4 Model assumptions
Modelling the dynamics of HIV/SIV infection demands the incorporation of many
complex rules. However, focusing on the specific questions that we are aiming to
address and in order to speed-up the simulations and ease the interpretation of our
results, we allowed for multiple assumptions to be included in the model as long as
these would not alter the outcome of the tested hypotheses. The following assumptions
were made (the exact parameter values used are discussed in the relevant sections):
1. The same motility parameters apply for all cell types since we mostly focus on
T cells for which the parameters have been experimentally defined in multiple
studies [268–273].
2. No free virus is included in the model. Viral spread occurs via an increased
probability of infection of neighbouring uninfected cells. This probability is set
based on the HIV viral reproduction rate which has been estimated experimen-
tally [274].
3. We do not make any distinction between activated and non-activated CD4+ T
cells. However, there is evidence in the literature that activated CD4+ T cells
can be the main targets of HIV-1 [275].
4. We only model a single HIV/SIV CD8+ T cell response specific for a single viral
epitope and which is present at a magnitude reported for the chronic phase of
infection. Because we focus on the dynamics of infection after set point viral
load we do not consider proliferation or contraction of this CD8+ T population.
Although epitope escape can lead to decreased antigen stimulation of the spe-
cific CD8+ T cell population and consequently to the decline of the response,
this is expected to happen with a lag of a few months [276]. Additionally, the
estimates of CD8+ T cell decline range from: 0.0002-0.015 d−1 [277–280] which
170
correspond to an average specific CD8+ T cell lifespan of 200 days. In the latter
studies, the rate of specific CD8+ T cell decline is estimated following HAART
when loss of antigenic stimulus is much greater than following escape. Hence,
these estimates of the rate of specific CD8+ T cell decline might be an over-
estimate of the rate following viral escape on which we focus here. The above
observations taken together strongly suggest that during the timescale that we
simulate viral escape (<3 months) the assumption of an approximately constant
epitope-specific CD8+ T cell population should be valid.
5. There is no fitness cost associated with the escape mutations.
6. We do not consider superinfection of infected cells. The infected CD4+ T cells
are either infected with the wild-type or the variant strain. Findings regarding
superinfection remain controversial in the literature [281,282].
6.5 T cell motility
The use of two-photon microscopy has allowed major advancements in the in vivo
study of the motility patterns of cells. Quantities that characterise motility such
as speed, motility coefficient and displacement have been calculated experimentally
for different cell types and under different experimental settings in multiple studies
[268–273]. In summary, the average speed of non-activated T cells is reported to be
10− 15 µm/min with a maximum of 25 µm/min [268,269], the motility coefficient is
50−100 µm2/min and the displacement is calculated to be proportional to the square
root of time (d ∝√t). Most of these measurements have been in the lymph nodes; for
the spleen, they are suggested to be comparable but lower [283, 284]. Interestingly, T
cell motility is argued to resemble a random; in [285], it is shown that this resemblance
can be dictated by the lymph node environment rather than an intrinsic motility
program.
171
To simulate the motility of cells with the CA model we implement the rules of cell
motility used in [267] (see Appendix E.1 for rule description). The two main parame-
ters which calibrate the motility of the cells are the speed and the motility coefficient.
In our model simulations, the average T cell speed is approximately 9 µm/min with
a maximum of 25 µm/min and the average T cell motility coefficient is 75 µm2/min
(Figure 6.3). Additionally, we indeed find that under this motility rules, the displace-
ment is proportional to the square root of time (Figure 6.4) suggesting a ‘random walk’
behaviour. As expected, the mean CD8+ T cell speed is reduced (Figure 6.5) because
of conjugate formation when infected cells are introduced in the model; this is also
consistent with experimental observations [283,286].
0 50 100 150 200
05
1015
2025
Time (min)
CD
8+ T
cel
l spe
ed (µ
m m
in−1
)
0 50 100 150 200
05
1015
2025
Time (min)
CD
8+ T
cel
l spe
ed (µ
m m
in−1
)
0 50 100 150 200
05
1015
2025
Time (min)
CD
8+ T
cel
l spe
ed (µ
m m
in−1
)
0 50 100 150 200
05
1015
2025
Time (min)
CD
8+ T
cel
l spe
ed (µ
m m
in−1
)
Figure 6.3: The speed of four individual simulated CD8+ T cells in the CA model is
shown. The mean speed is approximately 9 µm/min and the mean motility coefficient
is 75 µm2/min. These values agree with the experimental measurements reported in
the literature [268–273].
172
0 10 20 30
02
00
40
06
00
80
0
dt ( min)
Me
an
CD
8+
T c
ell
dis
pla
cem
en
t (µ
m)
Figure 6.4: The mean displacement of CD8+ T cells with respect to√time is shown.
The straight line suggests that the CD8+ T cell movement resembles a random walk.
173
No Rec 1/1000 1/100 1/10 All Rec
02
46
81
0
Successful CD8+ T cell target−scans
CD
8+
T c
ell s
pe
ed (
µm
min
−1)
Figure 6.5: The mean CD8+ T cell speed decreases as the ‘probability of successful
scan’, i.e. recognition followed by CD8+ T cell activation, of infected targets increases.
Each boxplot represents the results obtained by the simulation of 50 epitope-specific
CD8+ T cells. Abbreviations: No Rec=CD8+ T cells do not recognise any infected
target that they meet, All Rec=CD8+ T cells recognise all infected targets that they
meet; the rest represent intermediate cases.
174
6.6 CD4+ T cell influx
In order to obtain a sustained infection we also need to simulate an influx of CD4+ T
cells. This influx can include both proliferation and inflow of cells to the spleen but
since we are modelling a closed system we only focus on proliferation. To define the
probability of additional CD4+ T cell appearing at the site of infection (here being
the spleen) we use the estimates of CD4+ T cell proliferation rates reported in [287].
In the latter study deuterated glucose is used to label DNA of proliferating cells in 4
healthy and 7 treatment-naive HIV+ subjects. By applying a mathematical model the
authors estimate that the mean CD4+ proliferation rate in HIV+ subjects is 0.025 d−1
when their mean CD4+ T cell count is approximately 400 µl−1, i.e. about 33% of a
normal count. As an estimate of the influx probability of CD4+ T cells appearing in
the grid we use a Hill function (Equation 6.1 and Figure 6.6) where j is calculated
such that the probability of entering the grid, pinflux when the CD4+ T cell count
is decreased by 67% provides a CD4+ T cell proliferation rate of 0.025 d−1 and n is
chosen to provide a plausible change of pinflux with respect to the percentage of CD4+
T cells, U, at any given timepoint. We note that pinflux → 0 when CD4+ T cell
count goes to normal, i.e. 100% of the initial population. This formulation simulates
a homeostatic influx mechanism for the replenishment of CD4+ T cells which allows
for a sustained viral infection in the CA model.
pinflux =Un
Un + jn(6.1)
The CD4+ T cells entering the grid with a probability, pinflux, can be uninfected
or infected (either with the wild-type or the variant strain). Equations 6.2-6.4 describe
the set of probabilities which define whether the CD4+ T cell that entered the grid will
be uninfected, pu or infected (with the wild-type, pw, or the variant, pv strain). These
probabilities depend on the current population of uninfected, Nu, wild-type infected,
Nw and variant infected, Nv, cells present on the grid as well as their respective
175
0.0 0.2 0.4 0.6 0.8 1.0
0.0
0.2
0.4
0.6
0.8
1.0
U
pin
flux
Figure 6.6: The influx of CD4+ T cells is modelled based on a Hill function (Equation
6.1) of order n = −5. This type of influx can be described as homeostatic.
176
lifespans, Lu, Lw and Lv. We consider Lw = Lv.
pw =NwLw
NuLupu (6.2)
pv =NvLv
NuLupu (6.3)
pu =NuLu
NuLu +NwLw +NvLv(6.4)
This fraction of newly introduced infected cells can represent reservoirs of latently
infected CD4 T cells which can be important in maintaining active infection [288,289].
Additionally, it can reflect the small proportion of circulating infected CD4+ T cells
that enter the spleen at any given time [290].
6.7 CD4+ T cell reproduction and death
6.7.1 Reproductive rate
In the context of within host viral infection, the basic reproductive ratio, R0, is defined
as the number of secondary infected cells produced by a primary infected cell during
its lifetime; this definition assumes that uninfected target cells are not limiting. In
this model we consider R0 = 6 which agrees with estimates reported in the literature
during acute HIV infection [274,291] and therefore before CD8+ T cell responses take
effect. At this viral reproduction rate, we record a peak in infected cell population
15− 20 days post infection (dpi).
6.7.2 Death rate
Since we are mostly interested in the dynamics of CD4+ T cells (uninfected and
infected) during chronic HIV+ infection and specifically in the comparison of lytic
and non-lytic responses, we only allow for reproduction and death of CD4+ T cells in
the model. The rest of the cell populations remain of constant size during the course
of the infection.
177
For uninfected CD4+ T cells, we set the death rate based on the proliferation rate
estimated in [287]. As shown in [292] the death rate measured in [287] is not that of
all CD4+ cells but only of labelled cells, which are not a representative sample of the
whole. However, the proliferation rate measured is the proliferation rate of the whole
CD4+ population. Hence, because at equilibrium the proliferation equals the death
rate and since we are mainly interested in the chronic phase of the infection, we set
the death rate of uninfected CD4+ T cells to 0.025 d−1 which equates to a lifespan of
40 days.
For infected CD4+ T cells, we consider two phases: 1) An eclipse phase which
represents a viral delay of 1 day [293,294] and 2) a productively infected phase. Dur-
ing the eclipse phase, infected cells can neither be recognised by CD8+ T cells nor
infect uninfected CD4+ T cells and die with the same rate as uninfected cells. In the
productively infected phase, they die at an exponential rate with a mean of 1 d−1 (i.e.
they have a mean lifespan of 1 day) as estimated in [295] (Figure 6.7). In the case
where cytotoxic CD8+ T lymphocytes have to search for infected target cells to de-
liver their lethal hit, a constant rate of death and, therefore, exponentially distributed
lifespans of the cells may serve as appropriate descriptions of the virus dynamics [296].
In summary, infected CD4+ T cells die at a rate of 0.025 d−1 during the eclipse phase
and at a rate of 1 d−1 at the viral productive phase. The latter death rate estimates
refer to total death which includes both CD8+ T cell control and viral cytotopathicity.
Focusing on CD8+ T cell control, we assume that CD8+ T cells as part of the
second line of defence do not immediately start to cope with the infection. We set
the time to CD8+ T cell appearance to 10 days. The delay in CD8+ T cell responses
is affected by multiple factors such as the presentation of epitopes to naive CD8+ T
cells in local lymph, the time needed for clonal expansion and the migration to the
site of infection. We set the containment of the infection attributable to CD8+ T
cells to 30% [3, 248, 297]. The total death rate of productively infected CD4+ T cells
(attributed to all factors) is 1 d−1, hence, we assume an average death rate of 0.7 d−1
178
before CD8+ T cell activation (occurring after 10 days) and an average death rate of
0.95 d−1 after activation, allowing a small proportion of the death rate (<= 5%) to
be attributable to the single epitope-specific CD8+ T cell population that is explicitly
included in the model. In general, as mentioned in [298], most current models for
CD8+ T cell killing in HIV assume the additive property [248, 299, 300]. To allow for
stochastic effects, the lifespan of infected cells, which is the reciprocal of death rate, is
sampled for each infected cell from a normal distribution (mean±sd) of: 1.4±0.25 days
before introducing CD8+ T cell in the model and 1.05±0.25 days after activation (not
including the death rate attributable to the explicitly modelled single epitope-specific
CD8+ T cell population).
Time (days)
Wild
−ty
pe
infe
cte
d c
ells
fre
qu
en
cy
0 2 4 6 8 10
0.0
0.2
0.4
0.6
Eclipse phase (1 day)Producing phase (mean~1 day)
Figure 6.7: The lifespan of wild-type infected cells. During the eclipse phase of viral
infection which lasts 1 day, the infected cells die at the same rate as the uninfected
cells while at the viral productive phase they die at a rate of 1 d−1.
179
6.8 CD8+ T cell effector function
6.8.1 Conjugate formation
We explicitly model the formation of conjugates between the single epitope-specific
CD8+ T cell population included in the model and its target cells, i.e. the wild-type
infected CD4+ T cells. Once a CD8+ T cell encounters a target cell (that ‘presents’
its cognate antigen) it must make the decision of ‘acting or moving on’; we call the
time this takes the scanning time. The time that it takes a T cell to scan its target
is estimated to be 7 ± 2 min [301] and this process is explicitly incorporated in the
model. In general, the decision to act is related to the characteristics of the TCR-
pMHC interaction and the level of pMHC expression on the surface of the target
cell [302]. This process is simulated in the model via a probability which defines the
decision of the CD8+ T cell upon encounter to which we refer as probability of successful
scan and is discussed below (see section 6.8.2). If the CD8+ T cell recognises the cell
as infected then the interaction time begins. The conjugates remain immobile during
both scanning and interaction time. This is a good approximation of the underlying
biological process; in [286, 301] the conjugates of T cells with B cells and granuloma
cells respectively end up in complete immobility of the complex within minutes after
recognition.
In addition, two more important aspects of the CD8+ T cell and target cell en-
counter are incorporated in the model: 1) multiple CD8+ T cells can kill the same
target [303, 304] and 2) multiple targets can be killed by a single CD8+ T cell as
observed in vitro [305]. The latter is further supported by the configuration of the
contact between TCR and APCs; during recognition of a target cell the microtubule-
organising center (MTOC) of the CD8+ T cell polarises towards the immunological
(IS). The high motility of the MTOC allows for rapid switch of polarization between
targets [306]. However, as found in [304] most of the CD8+ T cells in conjugate for-
180
mation did not have more than four target cells bound to them, an observation that
we implement in the model.
In summary, at the end of the interaction time between the CD8+ T cell and
its target(s) the CD8+ T cell can respond as follows: 1) during a lytic response,
deliver the lethal hit and therefore eliminate the target or 2) in a non-lytic response,
‘secrete’ soluble factors affecting a specified area or 3) disassociate from the conjugate
formation and move on to another target (Figure 6.8). The lytic and non-lytic models
are simulated separately with disassociation always being the alternative option to
responding after scanning.
6.8.2 Target recognition by CD8+ T cells
The effectiveness of a CD8+ T cell response is obviously dependent on their ability
to successfully survey potentially infected cells and recognise them as targets. The
recognition is dictated by a series of factors such as: 1) the level of antigenic stimulation
[307, 308], 2) the structural rearrangements of TCR-binding [309], 3) the confinement
time of TCR-pMHC interaction [310], 4) the formation of the peptide-MHC complex
[311] and others. We simulate the stochasticity of this process by setting a probability
of successful scan of an infected target once the initial scanning time is completed. If
the target is successfully scanned, the probability of lysis or secretion of soluble factors
is set to 1. The probability of successful scan is one of the parameters that allows us
to vary the immune control exerted by the specific CD8+ T cell population included
in our simulations.
6.8.3 Lytic and non-lytic CD8+ T cell viral suppression
The lytic response can be divided into at least three stages [312]: 1) CD8+ T cells
survey potential target cells and recognise a subgroup of them, 2) they then form a
conjugate (see section 6.8.1) with their target in order to deliver the lethal hit and
181
Figure 6.8: The CD+8 T cell first scans wild-type infected cells and then decides
between the following: 1) react to the infected cell (lyse or secrete soluble factors) or
2) disassociate from the target cell and move on to the next target. The lytic and non-
lytic models are simulated separately with disassociation always being the alternative
option to responding after scanning. Abbreviations: T=uninfected cells, WT=wild-
type infected cell, VAR=variant infected cells, CD8=single epitope-specific CD8+ T
cells
182
3) once the target lyses, they continue hunting for new targets. All these stages are
explicitly included in the model for the epitope-specific CD8+ T cell population. Only
one CD8+ T cell ‘lethal hit’ can be delivered per target cell as observed in [313]. The
conjugate formation in the model is set to have a 30 min duration time, in agreement
with measurements obtained via intravital multiphoton microscopy [301,305] and also
taking into account the disintegration time of the complex [314]. Of note, a CD8+ T
cell might need to ‘rearm’ before moving on to their next target [315, 316] or pause
before identifying a new target [301] therefore increasing the duration of this process
(we vary this parameter in order to investigate its effect).
The non-lytic response can be summarised in the following steps: 1) CD8+ T
cells survey potential target cells and recognise a subgroup of them, 2) they bind on
a conjugate with their target and 3) they start secreting soluble factors while being
in a conjugate, suppressing therefore viral production or infection. We model the
suppression of viral infection by not allowing the infection of uninfected CD4+ T cells
that ‘interact’ with the soluble factors. In a similar way, we model the suppression
of viral production by not allowing infected cells to infect other cells once the former
‘interact’ with the secreted factors. The duration of the conjugate is set to 10min as
reported in [301] when no lysis is observed. The in vivo area of diffusivity of soluble
factors is poorly quantified, we therefore simulate the following different secretion
patterns (see Figure 6.10): 1) localised effect, where only the wild-type infected cell in
conjugate with the CD8+ T cell loses its ability to infect new targets for a given interval
(called the duration of the effect), 2) a polarised (grid radius, r = 1) effect where all
the cells (infected or uninfected depending on the type of viral suppression) present at
the immediate area on the one ‘side’ (in grid terms) of the CD8+ T cell are affected
and 3) a diffusive pattern where all the cells equidistantly found in the immediate or
broader area (r = 1, 2) around the CD8+ T cell are affected by the soluble factor.
The duration of the effect depends on the soluble factor under consideration. Here
we focus on RANTES as a case study. The recycling time for the CCR5 receptors
183
after interaction with RANTES has a half-life of 6-9 hrs [317] so we set the total
duration of the RANTES ‘protective effect’ on uninfected CD4+ T cells to 10 hrs;
after that period the uninfected cells can again be infected either with the wild-type
or the variant strain. We consider a full protection of the uninfected targets unless
otherwise stated. The ‘radius’ of non-lytic effect which can clearly influence the viral
escape is closely investigated using the different simulated patterns. A summary chart
of all the non-lytic models simulated is given in Figure 6.9.
Figure 6.9: Non-lytic models simulated with the 3D cellular automaton model.
For lytic killing, the target cells are lysed at the time that the conjugate is detached.
For non-lytic responses, the secretion of soluble factors is immediately stopped after
conjugate resolution. This is supported by experimental observations. In [318] they
show that IFN-γ production from activated CD8+ T cells is terminated immediately
after the contact of the T cell and its cognate antigen is disrupted. This can be
a regulatory mechanism of preventing excessive cytokine production which can be
destructive for the host. In the same study [318], in vitro experiments indicate that
peptide-pulsed spleen cells cause the production of cytokines (IFN-γ and TNF-α) by
LCMV-specific CD8+ T cells within 30 mins and in vivo, no cytokine production was
observed in the absence of stimulation.
184
(a) Localised (b) Polarised (r=1)
(c) Diffusive (r=1) (d) Diffusive (r=2)
Figure 6.10: Secretion patterns of soluble factors by CD8+ T cells under a non-lytic
response. The red nodes represent the activated CD8+ T cell and the green nodes
depict the area that is affected by soluble factors secreted by this CD8+ T cell.
185
6.9 Quantifying CD8+ T cell killing and viral
escape
We quantify CD8+ T cell killing and viral escape on the population level using two
different models: 1) a model presented in [248] which assumes mass-action killing, i.e.
killing is proportional to infected target cells and effector cells and 2) a model that we
describe here and assumes that the rate of killing of the effector cells can eventually
saturate in terms of target cells. Using the simulations of the CA model, we can also
calculate the actual killing rate in the simulated datasets and use this information to
investigate which of the models better predicts the CD8+ T cell killing rate.
6.9.1 Simulated killing rate
The CA model is implemented such that we record the exact number of wild-type
infected cells killed by the epitope-specific CD8+ T cell population at each timestep
(30 sec). Hence, using the simulated data we can readily calculate the absolute number
of targets killed per day by the epitope-specific CD8+ T cell population, ns, as well
as the ‘actual’ killing rate per day, ks, as follows:
ns = # WT Infected cells killed (day) (6.5)
ks =ns
#WT Infected cells (day)(6.6)
6.9.2 Mass-action killing model
First, we use the model developed in [248] to calculate the escape rate of the variant
infected cells which under the assumption of mass-action killing equals the epitope-
specific CD8+ T cell killing rate (see below). Based on this model, CD4+ T cells
productively infected with wild-type virus, y, replicate at rate a, die at rate b (including
186
the death attributable to specific-CD8+ T cells recognising epitopes other than the
escape epitope) and are also killed by specific-CD8+ T cells recognising the escape
epitope at rate k. CD4+ cells productively infected with an escape variant, x, replicate
at rate a′ and die (including death by specific CD8+ T cells recognising epitopes other
than the escape epitope) at rate b. Hence,
y = ay − by − ky (6.7)
x = a′x− bx (6.8)
If p(t) is the proportion of productively infected cells infected with the variant viral
sequences, the following holds [248]:
p(t) =x(t)
x(t) + y(t)=
1
ge−ct + 1(6.9)
where g = y(0)x(0) , t=0 is the time when the variant is introduced in the viral popula-
tion and c = k − (a− a′), is the escape rate of the variant viral strain. If we consider
no fitness cost for the replication ability of the variant virus, a = a′, then there is no
difference in replication rate between the escape variant and the wild-type, a− a′ = 0
and hence c = k. It should be noted that k quantifies the overall efficiency of the
epitope-specific CD8+ T cell population.
In the simulations, viral escape is quantified during the chronic phase of the in-
fection (starting at 50 dpi). Equation 6.9 can be used to estimate the variant escape
rate irrespective of the assumptions made in the model described by Equations 6.7-6.8
but it can only be used as an estimate of the killing rate if the mass-action killing
assumption holds.
6.9.3 Saturated killing model
Then, we also formulate a model where the killing rate of CD8+ T cells saturates after
a specific infected target cell threshold is exceeded. The same saturation killing term
187
is also discussed in [319]. This model can be described as follows:
y = ay − by − vmaxy
km + y(6.10)
x = a′x− bx (6.11)
where the variables y and x and the parameters a, b and a′ are the same quantities
described in the previous model. Additionally, vmax is the maximum number of infected
cells that can be killed per day when y → ∞ and km is the number or infected target
cells that would give half the maximum number killed per day, vmax.
The Equations 6.10-6.11 can not readily be rewritten in a similar format to Equa-
tions 6.9 but can still be fitted to ‘escape data’ as described by the ratio p(t) = x(t)x(t)+y(t) .
In addition, because we assume no fitness cost for the variant viral strain we set again
a′ = a. Hence, we can easily reduce the above model in the following form:
y = fy − vmaxy
km + y(6.12)
x = fx (6.13)
where f = a − b. This model (reduced or not) includes two main different sub-
models of killing depending on the availability of infected target cells. If y << km then
the model describes mass-action killing whilst if y >> km then the number of infected
cells killed per day saturates and converges to vmax (Figure 6.11).
Using the estimates of the wild-type population, y, estimated when fitting the
model to the data, we can obtain the average killing rate per day by calculating the
average of the quantity vmax
km+yover all the fitted values of y. This provides an estimate
of the average killing rate when assuming a saturated killing process.
188
0.0
0.2
0.4
0.6
0.8
1.0
Infected cells (y)
Num
ber
of in
fect
ed c
ells
kill
ed b
y C
D8+
T c
ells
per
day
km 1000 2000 3000 4000
Massaction Saturation
vmax
Figure 6.11: Schematic of a model which includes saturated epitope-specific CD8+ T
cell killing per day. This model includes two different sub-models of killing depending
on the availability of infected cells (see Equation 6.12). If y << km then we have
mass-action killing but if y >> km then the number of infected cells killed per day
converges to a maximum value, vmax.
189
6.10 Discussion
We have developed and implemented a complex agent-based model that captures both
acute and chronic phase dynamics of HIV/SIV infection. The model includes detailed
rules of T cell motility, viral production, death rate of infected and uninfected CD4+
T cells, conjugate formation of CD8+ T cells and infected targets and CD8+ T cell
response that can be manifested lytically or non-lytically.
Importantly, the set-up of the model allows for the future study of numerous addi-
tional aspects of HIV/SIV infection such as fitness cost of variant strains, simulation of
multiple epitope-specific CD8+ T cell populations, combination of lytic and non-lytic
CD8+ T cell responses, early killing by CD8+ T cells [257], superinfection of infected
cells and others.
We next use this model to study the CD8+ T cell killing process on a population
level, investigate the efficiency of lytic and non-lytic CD8+ T cell responses and explore
‘viral escape’ dynamics under these different CD8+ T cell effector functions.
190
Chapter 7
CD8+ T cell effector function and
viral escape
7.1 Aim
Our aim was to investigate whether non-lytic CD8+ T cell responses can drive the
growth of escape variants, identify factors involved in their outgrowth and quantify
the relationship between viral control and the outgrowth rate.
7.2 Introduction
It is widely accepted that CD8+ T cells suppress replication of human (HIV) and
simian (SIV) immunodeficiency virus but this process is not yet fully understood.
CD8+ T cells can control viral replication via lytic or non-lytic effector mechanisms.
Lytic mechanisms require direct contact with the infected target cell and result in
apoptosis. Non-lytic mechanisms are mediated by soluble factors and reduce viral
production by infected cells or viral infection of uninfected cells. Strong evidence
that CD8+ T cell-secreted chemokines can play an important role in inhibiting HIV
infection come from in vitro experiments [58] involving RANTES, MIP-1α and MIP-1β
which bind CCR5 and act as competitive inhibitors of CCR5-mediated HIV/SIV entry.
Further findings in the same direction have been reported by multiple studies [320,321]
and extended to other soluble factors reviewed in [50].
Recently, ground-breaking studies reported that following CD8+ T cell depletion
191
in SIV-infected macaques, viral load robustly increased, however the lifespan of SIV-
infected cells remained unaltered [42, 167]. These unexpected results led to the sug-
gestion that SIV might be controlled via non-lytic mechanisms; a suggestion which
was further studied and corroborated in [3] (see Chapter 5). In the latter study, ODE
models of CD8+ T cells with lytic and non-lytic effector mechanisms were fitted to the
experimental data obtained in [42]. The non-lytic models were found to be a better
predictor of viral load measurements.
Both HIV and SIV are characterised by the evolution of viral epitope mutants that
can escape CD8+ T cell responses. The virus mutates at the base substitution rate
(10−4 − 10−3 per bp [322]) defined by the error-prone replication process. Cells can
be infected with the variant virus strains that have not lost their replicative capacity.
Additionally, variants of an epitope can arise from point mutations in or around the
epitope presented by MHC class I molecules to CD8+ T cells. These mutations may
lead to impaired recognition of the variant infected cells resulting in a higher immune
pressure exerted on the wild-type virus. Eventually, the variant infected cells might
outgrow the wild-type infected cells; a process termed as ‘viral escape’. The outcome of
escaping a CD8+ T cell response is given by the trade-off between evading the CD8+
T cell response and the cost that this poses on viral fitness. Hence, evading a CD8+
T cell attack does not necessarily provide a growth advantage to the virus and this is
indicated both by the emergence of compensatory mutations as well as by reversion to
the wild-type in MHC-mismatched hosts. Nevertheless, there are many studies in the
literature reporting the rapid or slow successful escape of specific epitopes from CD8+
T cell control during primary [323–326] and chronic infection [164,165,276,327].
Whilst it is clear that under a lytic mechanism an escape variant has a fitness
advantage compared to the wild-type, it is less obvious whether this holds in the face
of non-lytic control where both wild-type and variant infected cells would be similarly
affected by soluble factors. Here, we use the Cellular Automaton model described
in Chapter 6 to explore whether non-lytic CD8+ T cell responses can result in viral
192
escape and how the escape rate compares to the one attained via a lytic response.
We also investigate the killing process under a lytic mechanism and the efficiency of
non-lytic mechanisms in controlling infection.
7.3 Lytic CD8+ response and viral escape
7.3.1 The CA model captures HIV/SIV dynamics
Before using the CA model to simulate the different CD8+ T cell effector functions,
we investigate the correspondence of the model dynamics to the observed HIV/SIV
dynamics described in multiple studies in the literature.
The model has a typical viral expansion phase, with a viraemic peak reached after
approximately two to three weeks (Figure 7.1). The infected CD4+ T cells in the first
days post infection (dpi) grow at a rate close to 1.5 d−1 which is within the 1-2 d−1
range that has been reported in other studies [274,328,329]. Close to the peak there is
a large depletion of CD4+ T cells and the uninfected target cells become limited. The
immune control introduced at day 10 post infection together with the target limitation
result in the slowing down of viral growth and the subsequent decline to a viral set-
point within a few weeks. The proportion of infected CD4+ T cells at the steady-state
(> 30 dpi) is of the order of 10−3 − 10−2 in compliance with experimentally observed
values [290].
7.3.2 Probability of recognition by CD8+ T cells and
killing rate
We model different levels of specific CD8+ T cell pressure in the absence of variant
virus. We simulate CD8+ T cell pressure expressed in terms of different epitope-specific
CD8+ T cell killing rates. We calculate the average simulated killing rate observed
193
0 50 100 150
01
23
45
Days
Siz
e o
f ce
ll p
op
ula
tion
s (lo
g1
0 s
cale
)
Wild−typeUninfected
Figure 7.1: Dynamics of uninfected CD4+ T cells and wild-type infected CD4+ T
cells over a course of 150 days when no variant infected cell population is present. The
bars represent the 95% central range values. The sizes of the populations (y-values)
are given on a log10 scale.
194
over 25-50 dpi using Equation 6.6. Given that we consider a constant CD8+ T cell
population in magnitude and we have calibrated the cell motility to biological ranges,
three parameters in the model could influence the killing rate: 1) the probability of
successful scan (recognition of infected target cells which leads to cytolysis or secretion
of non-lytic factors), 2) the scanning time and 3) the duration of the killing process.
Our aim is to simulate an epitope-specific CD8+ T cell population that can kill 1%-5%
of the infected cells per day, in the range estimated in [248] for HIV specific CD8+ T
cell clones recognising one epitope. We next explore how these three parameters can
affect the killing rate.
First, we vary the probability of successful scan. Initially we set this probability
to 1, i.e. each‘interaction’ of a CD8+ T cell with a wild-type infected CD4+ T cell
leads to a response (lytic in this case) but we find that this leads to an unrealistically
high killing rate (> 20 fold clearance per day). Interestingly, we find that if a CD8+
T cell has a probability as small as 0.001 to successfully scan its target upon collision
then the killing rate achieved is on average 1% per day (Figure 7.2) and for an epitope-
specific CD8+ T cell population to kill daily 100% of the infected cells, a probability of
successful scan of 0.025 suffices (Figure 7.2 inset). In other words, the epitope-specific
CD8+ T cell population that constitutes 0.5% of the total splenocyte population can
kill 1% of the infected CD4+ T cell population within one day. By increasing the
probability of successful scan to 0.002 and 0.003 we observed a median killing rate of
3% and 5% respectively (Figure 7.2). A probability of successful scan of 0.001 implies
that a CD8+ T cell will attempt 1000 scanning events on average before it successfully
recognises and responds to the antigenic stimulus (500 and 330 infected targets for
probabilities 0.002 and 0.003 respectively).
Scanning time is restricted to a few minutes, is not expected to vary significantly
and is already modelled as a distribution (7± 2 min [301]) hence, we do not vary this
parameter further.
It has been estimated in multiple studies that the CD8+ T cell killing process takes
195
approximately 30 min [251,301,305]. However, a CD8+ T cell might, for example, need
to ‘recharge’ before it is able to kill again and this time can be additive to the killing
time [315,316]. So, to explore the robustness of the killing rate against the duration of
killing, we vary the parameter in the range of 15 to 120 min. No substantial effect of
the killing duration on killing rate is observed although there is a trend for the killing
rate to decrease slightly with the increasing duration of killing (Figure 7.3). The very
limited effect is perhaps not surprising since even the increase of the killing time by
4-fold would still result in a killing process that evolves in a fast timescale.
The fact that the probability of successful scan has a large impact on the killing rate
while the duration of killing does not, implies that the process of efficiently identifying
rather than rapidly eliminating a target defines the killing rate of infected CD4+ T
cells.
7.3.3 Higher CD8+ T cell killing leads to faster viral es-
cape
Next, we investigated the dynamics of infection when a variant infected population
(seeded at a magnitude which is approximately 30% of the wild-type population) is
included in the model. The variant infected cells appear at a random age and occupy
an available free space on the grid. The variant population is introduced at 50 dpi
when the steady-state has been reached and the specific-CD8+ T cell population is not
expected to expand and contract substantially (as it does during the acute phase). We
model single epitope-specific CD8+ T cell lytic responses averaging to three different
killing rates, ks: (i) 1%, (ii) 3% and (iii) 5% per day. We run 25 simulations for each
response (not all result in escape) by using different ‘initial random seeds’ (Figure 7.4)
and observe that the variant population outgrows the wild-type population at a faster
rate as the CD8+ T cell pressure increases (Figure 7.6). We fit Equation 6.9 [248]
to the simulated datasets in order to estimate the variant escape rate (Figure 7.5).
196
0.001 0.002 0.003
0.00
0.02
0.04
0.06
0.08
0.10
0.12
Probability of successful scan
CD
8+ T
cel
l killi
ng ra
te (p
er d
ay) b
efor
e ’e
scap
e ph
ase’
(25−
50 d
pi)
0.01 0.025 0.05
0.0
0.5
1.0
1.5
2.0
2.5
Figure 7.2: Probability of successful scan by epitope-specific CD8+ T cells and the
corresponding observed killing rates. The dotted line represents the lower and upper
bounds of the single epitope-specific CD8+ T cell killing rate that we aim at simulating
(1 − 5%). The inset plot shows the killing rates recorded for higher probabilities of
successful scan. The killing rate is calculated based on 25 simulations using Equation
6.6 and is averaged over 25-50 dpi.
197
15 30 60 90 120
0.00
00.
005
0.01
00.
015
0.02
00.
025
0.03
0
Duration of CD8+ T cell killing (min)
CD
8+ T
cel
l kill
ing
rate
(pe
r da
y) b
efor
e ’e
scap
e ph
ase’
(25
−50
dpi
)
Figure 7.3: Duration CD8+ T cell killing and the corresponding observed killing rates.
The killing rate is calculated based on 10-25 simulations in each case using Equation
6.6 and is averaged over 25-50 dpi. The probability of successful scan is set to 0.001
for all the simulations shown here. Varying the killing duration showed no statistically
significant differences in killing rate compared to a killing duration of 30 mins (p>0.05)
apart from the case where the killing duration was increased to 120 mins (p=0.03);
however, even in this case the mean difference in killing rate was only 0.2%.
198
Importantly, when we increase the grid size of the CA this trend as well as the killing
and escape rate estimates remain similar (p > 0.05, Appendix Figure F.1).
The model used to estimate the escape rate assumes a mass-action specific-CD8+
T cell killing term which we investigate closer. Given the lack of replicative cost for the
variant population, we expect that the killing rate and the variant escape rate would
be approximately equal (see section 6.9.2). However, as it is readily seen in Figure
7.6, whilst the escape and killing rates are indeed significantly (p=0.002) positively
correlated, they are not of equal magnitude. This interesting result can potentially
be attributed to aspects of the CA model that are not included in the ODE model
described by Equations 6.7-6.8. The main differences are the inclusion of an ‘eclipse’
phase, the CD4+ T cell influx of uninfected cells and potentially the mass-action killing
assumption considered in the ODE description. When we introduce an eclipse phase
or a homeostatic influx (instead of a steady influx) of uninfected cells to the ODE
model the estimated escape rate is unchanged, indicating that neither of these two
model aspects can explain the observed difference. Hence, the form of the killing term
remains the most probable explanation of the observed discrepancy and we explore
this further.
199
0 50 100 150
01
23
45
ks=1%
Days
Size
of c
ell p
opula
tions
(log
10 sc
ale)
(a)
0 50 100 150
01
23
45
ks=3%
Days
Size
of c
ell p
opula
tions
(log
10 sc
ale)
(b)
0 50 100 150
01
23
45
ks=5%
Days
Size
of c
ell p
opula
tions
(log
10 sc
ale)
Wild−typeVariantUninfected
(c)
Figure 7.4: Dynamics of uninfected, wild-type and variant infected CD4+ T cell populations over a course
of 150 days when the variant infected cell population is introduced at 50 dpi at an initial magnitude of 30% of
the wild-type. We model three levels of epitope-specific CD8+ T cell killing: i) ks ≈ 1% (a), ii) ks ≈ 3% (b)
and iii) ks ≈ 5% (c) of infected cells killed per day. We run 25 simulations in each case. Only the simulations
that lead to escape (> 10 in each case) are shown. The bars represent the 95% central range values. The sizes
of the populations (y-values) are given on a log10 scale.
200
0.0
0.2
0.4
0.6
0.8
1.0
ks=1%
Days
Fra
ctio
n o
f va
ria
nt
infe
cte
d c
ells
50 65 80 95 110 130
0.0
0.2
0.4
0.6
0.8
1.0
ks=3%
Days
Fra
ctio
n o
f va
ria
nt
infe
cte
d c
ells
50 65 80 95 110 130
0.0
0.2
0.4
0.6
0.8
1.0
ks=5%
Days
Fra
ctio
n o
f va
ria
nt
infe
cte
d c
ells
50 65 80 95 110 130
Figure 7.5: We estimate the variant escape rate in the simulated datasets using Equa-
tion 6.9 and under three levels of epitope-specific CD8+ T cell killing rate: (i) ks ≈ 1%,
ii) ks ≈ 3% and iii) ks ≈ 5% of infected cells killed per day. The variant infected cell
population is seeded at 50 dpi. We run 25 simulations in each case. Only the simula-
tions that lead to escape (> 10 in each case) are shown.
201
1% 3% 5%
0.0
20
.04
0.0
60
.08
0.1
00
.12
Median specific CD8+ T cell killing rate (per day) before ’escape phase’
Est
ima
ted
esc
ap
e r
ate
pe
r d
ay
Figure 7.6: Higher CD8+ T cell killing leads to faster viral escape. We model three
levels of epitope-specific CD8+ T cell selection pressure: (i) ks ≈ 1%, ii) ks ≈ 3% and
iii) ks ≈ 5% of infected cells killed per day. We run 25 simulations in each case. Only
the simulations that lead to escape (> 10 in each case are shown). The killing rate is
estimated over 25-50 dpi using Equation 6.6. The escape rate is estimated by fitting
Equation 6.9 to the simulated datasets.
202
7.3.4 CD8+ T cell killing rate is higher during ‘escape
phase’
Next, we investigated more closely the CD8+ T cell response under a lytic mechanism.
We studied two quantities: 1) the absolute number of wild-type infected cells killed by
the epitope-specific CD8+ T cell population per day and 2) the killing rate, described
by Equations 6.5 and 6.6 respectively. As shown in Figure 7.7, the killing rate per day
is not constant and it increases after the variant-infected population starts outgrowing
the wild-type population. We call the phase after the variant population appears the
‘escape phase’. Given that the number of effector cells in our model is constant, this
increase can be explained by two main hypotheses: i) the number of wild-type infected
cells killed per day by the epitope-specific CD8+ T cell population increases over the
course of infection or 2) the number of wild-type cells killed is approximately constant
but the overall wild-type population decreases and therefore the fraction of wild-type
infected cells killed increases. By calculating the absolute number of wild-type cells
killed per day by the epitope-specific CD8+ T cells, we find that the number of cells
killed remains approximately constant over time (Figure 7.8). This confirms our second
hypothesis; the variant population outgrows the wild-type during the ‘escape phase’
and because the wild-type population shrinks the killing rate increases but the absolute
killing does not.
Interestingly, the fact that the number of wild-type infected cells killed per day
by the specific CD8+ T cells are almost constant although the number of wild-type
infected cells varies a lot during this period (Figure 7.4), implies that there is an
upper threshold to the number that the CD8+ T cells can kill and therefore the killing
process might be better explained by a saturated killing term. In addition, as it is
readily shown in Figure 7.7 we obtain different killing rate estimates depending on
whether we perform the calculation before or during the ‘escape phase’ since during
the escape phase the variant strain gradually replaces the wild-type strain.
203
0 50 100 150
0.0
0.2
0.4
0.6
0.8
Days
Fra
ctio
n of
wild
−ty
pe in
fect
ed c
ells
kill
ed (
per
day)
k~5%k~3%k~1%
CTLs
Set−point
Variant
Figure 7.7: CD8+ T cell killing rate is higher during ‘escape phase’. The three time-
points labelled are: CTLs=The day at which the HIV-1 epitope-specific CD8+ T cell
population is introduced in the model, Set-point=The approximate day that steady-
state is attained, Variant=The day that the variant infected population is introduced
in the model.
204
0 50 100 150
02
46
810
Days
Num
ber
of w
ild−
type
infe
cted
cel
ls k
illed
(pe
r da
y)
k~5%k~3%k~1%CTLs
Set−point
Variant
Figure 7.8: The absolute number of wild-type infected cells killed per day is ap-
proximately constant. The three timepoints labelled are: CTLs=The day at which
the HIV-1 epitope-specific CD8+ T cell population is introduced in the model, Set-
point=The approximate day that steady-state is attained, Variant=The day that the
variant infected population is introduced in the model.
205
7.3.5 Saturation term better estimates killing rate
The form of the killing function is important for determining the dynamics of the
immune response. Usually, in models of immune control the killing process is defined
by a mass-action term [330] (e.g. Equation 6.7) which assumes that the killing of
infected cells is proportional to both infected target cells and effector cells. In the CA
model the number of epitope-specific CD8+ T cells, i.e. the effector cells is assumed
constant over time because we focus on the chronic phase of infection. However, we find
that if we increase the number of effector cells then the killing rate increases and that
the relation between the two quantities is consistent with a mass-action behaviour in
terms of effector cells (Appendix Figure F.2); however, the data are noisy so we cannot
draw further conclusions at this point. So, we decide to explore if mass-action also
holds in terms of infected target cells. We have already shown (Figure 7.8) that the
number of wild-type infected cells killed during the expansion and contraction of the
infected target cells is approximately constant. This suggests that the mass-action
killing term is not the most appropriate term to describe the killing process under
a lytic mechanism. As it has been argued in other studies, limiting aspects such as
finding, recognising and adhering to the target can create a saturation effect similar
to Michaelis-Menten enzyme-substrate kinetics [139,319,330].
In order to thoroughly address whether a mass-action term or a saturated killing
term better explains the underlying killing process, we use the simulated ‘escape data’
that capture the HIV/SIV dynamics and fit the two different models (Equations 6.9
-which under the mass-action assumption and lack of fitness cost is equivalent to 6.7-
6.8- and 6.12-6.13 respectively). Both models provide satisfactory fits (Figures 7.5
and 7.9) but when we use the AICc (Equation 5.10) to select the best fitting model,
we find that the model which assumes saturated CD8+ T cell killing (in terms of
infected target cells) explains the data significantly better (p=0.001, paired Wilcoxon
test; Figure 7.10). In addition, when we compare the average killing rate estimated
206
using the two models before and after the ‘escape phase’ we find that the saturated
killing model predicts the killing rate more accurately during the ’escape phase’ while
the mass-action model underestimates it substantially (Figure 7.11). We also find that
both models overestimate the killing rate before the ‘escape phase’ (Figure 7.11). This
is expected since the killing rate increases during the ‘escape phase’ and both models
are fitted over the period before and during the ‘escape phase’. However, the killing
rate that relates to the escape rate is expected to be better described by the ‘escape
phase’ estimate, making therefore the saturation killing model a better selection.
By exploring the values of the infected target cells needed in order to start transi-
tioning from mass-action killing to saturated killing, km (Figure 6.11), we find that the
number of infected CD4+ T cells is already higher than this value when the specific-
CD8+ T cell population is introduced in the simulation (10 dpi) and remains higher
for most of the timecourse. This suggests that the killing of the specific CD8+ T cell
population is better described by a saturated rather that a mass-action killing term
during the timecourse of the simulations. However, in the case of the higher specific
CD8+ T cell killing rate (≈ 3− 5%) we observe that the absolute number of infected
target cells decreases towards the end of the simulated timecourse (Figure 7.8). This
is expected because only at this killing rate level, the number of available wild-type
infected targets drops below the km value (then the wild-type infected cells are almost
entirely replaced by the variant infected population within the 3 months simulation)
and therefore at this particular case the killing rate is better described by a mass-action
term.
207
0.0
0.2
0.4
0.6
0.8
1.0
ks=1%
Days
Fra
ctio
n o
f va
ria
nt
infe
cte
d c
ells
50 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 135
0.0
0.2
0.4
0.6
0.8
1.0
ks=3%
Days
Fra
ctio
n o
f va
ria
nt
infe
cte
d c
ells
50 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 135
0.0
0.2
0.4
0.6
0.8
1.0
ks=5%
Days
Fra
ctio
n o
f va
ria
nt
infe
cte
d c
ells
50 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 13550 65 80 95 115 135
Figure 7.9: We use the saturated killing model (Equation 6.11) in order to estimate
the killing rate in the simulated datasets under the three levels of epitope-specific
CD8+ T cell selection pressure (i) ks ≈ 1%, ii) ks ≈ 3% and iii) ks =≈ 5%. We run 25
simulations in each case. Only the simulations that lead to escape (> 10 in each case)
are shown.
208
Simulated Datasets
AIC
c
−8
00
−6
00
−4
00
−2
00
0
Simulated Datasets
AIC
c
−8
00
−6
00
−4
00
−2
00
0
1 3 5 7 9 12 15 18 21 24 27 30 33 36 39 42 45 48 51
Mass actionSaturation
Figure 7.10: The AICc of the fitted models over different simulated datasets. We fitted
both the mass-action and saturated killing models to 51 simulated ‘escape datasets’ and
calculated the bias-adjusted AICc which penalises for the number of model parameters
used. A lower AICc implies a better fit to the data. The saturated killing model has
a significantly lower AICc to the mass-action model (p=0.001, paired Wilcoxon test).
It is also readily observed that the light blue bars that show the AICc of the saturated
killing model provide a better fit for most of the datasets (35 out of 51).
209
0.00 0.02 0.04 0.06 0.08 0.10
0.0
00
.02
0.0
40
.06
0.0
80
.10
Before ’escape phase’
Specific−CD8+ T cell killing rate
Est
ima
ted
kill
ing
ra
te
y=xMass actionSaturation
0.00 0.10 0.20 0.30
0.0
00
.05
0.1
00
.15
0.2
00
.25
0.3
00
.35
During ’escape phase’
Specific−CD8+ T cell killing rate
Est
ima
ted
kill
ing
ra
te
Figure 7.11: Correlation of simulated and estimated killing rate using a mass-action
or a saturated killing term. The estimates before ‘escape phase’ are calculated over
25-50 dpi and the estimates during ‘escape phase’ are calculated over 50-150 dpi (if the
wild-type infected cell population is eliminated before 150 dpi then the last positive
value defines the end of the phase).
210
7.4 Non-lytic CD8+ response and viral escape
7.4.1 Non-lytic suppression leads to slower viral escape
dynamics
Although it is readily understood how a lytic CD8+ T cell response can drive the
expansion of a variant strain, it is not obvious how the dynamics of escape will be
reshaped under a non-lytic control. Here, we use the CA model in order to explore
the dynamics of a non-lytic CD8+ T cell response which blocks viral infection or viral
production and how it relates to the rate of outgrowth of the variant infected CD4+
T cells.
The term ‘non-lytic control’ can imply the effect of many different soluble factors
which act in turn or simultaneously. We model -as a case study- the secretion of
RANTES for which the viral suppressive effect has been studied [317] (see Chapter
6). The simulation of the non-lytic effect that blocks infection can be summarised as
follows: We consider that the ‘protective effect’ is conveyed to all uninfected CD4+
T cells that travel through the area of secretion (polarised or diffusive, see Chapter
6) formed around the activated epitope-specific CD8+ T cells. This effect lasts for
10 hrs (as measured for RANTES) and totally abolishes the possibility that these
uninfected targets become infected. Once this period expires the formerly ‘protected’
cells become again susceptible to infection. In a similar way, a non-lytic effect that
blocks production can be modelled by abrogating for a given time the infectiveness of
infected cells that cross the area of secretion (see Chapter 6).
Using the CA model, we simulate a non-lytic single-epitope specific CD8+ T cell
response mechanism and introduce a variant infected population as described for the
lytic mechanism at 50 dpi. We set all the other parameters of the model to the exact
same values as in the model of the lytic response. This allows us to study the two
211
different control mechanisms under exactly the same parameter space with only the
viral control being different. Hence, we can observe how the CD8+ T cell pressure in
each case shapes the viral escape dynamics. We find that in the case of a non-lytic
response that blocks infection or production (for 10hrs), the viral escape (estimated
using Equation 6.9) is slower, on average 1% per day and it does not increase with
increasing CD8+ T cell control or increasing radius of secretion (Figure 7.12).
In light of this finding, we first examine whether the number of uninfected CD4+ T
cells ‘protected’ from blocking infection are increasing as both the radius of secretion
and the CD8+ T cell pressure (measured by the probability of successful scan) are
increasing. We find that in both cases, as expected, there is a clear increase of protected
uninfected targets (Figure 7.13). Similarly, we find that under a non-lytic response
that blocks viral production from infected CD4+ T cells, the number of cells ‘blocked’
increases with both increasing radius of secretion and higher CD8+ T cell pressure
(Appendix Figure F.3).
Next, in order to understand whether this slower escape rate can be attributed to
the lack of a strong immune control or to intrinsic properties of a non-lytic response, we
quantified the immune control in both the epitope-specific lytic and non-lytic responses
(blocking infection or production) in terms of the number of new infections prevented
per day. This quantity can be calculated by simulating the infection dynamics of two
models: a) one with and b) one without the epitope-specific CD8+ T cell population,
and by then calculating the number of new infections produced per day in each case.
We do this for 40-50 dpi, just before the variant infected cell population is introduced in
the simulations. The difference of new infections per day between these two models is
the number of infections prevented by the epitope-specific CD8+ T cell population. We
find that the number of infections prevented under a non-lytic pressure is comparable
to a lytic killing pressure of 1% per day (Figure 7.14 and Appendix Figure F.4); this
is also in agreement with our observation that a non-lytic control leads to a lower
set-point of productively infected cells during the steady-state compared to a lytic
212
control of 1% per day (Figure 7.15 and Appendix Figure F.5). Although the immune
control is comparable for the low lytic response (1% killing per day) and the non-lytic
response, the lytic control results in a significantly faster escape rate (approx 3.5% vs
1.2%, p<0.05; Figure 7.12). Interestingly, the lytic immune control is higher compared
to the non-lytic when we increase the probability of successful CD8+ T cell scanning
but the non-lytic control remains approximately unchanged (Figure 7.12). In line with
the above findings, we also observe that for the lytic control, the escape rate is positive
for ≈ 80%) of the overall simulations (with this number increasing with probability of
successful scan), while for the non-lytic response that blocks infection we only have an
average of 40% of the simulations returning a positive estimate and for the non-lytic
response that blocks viral production although this percentage goes up to 60% it is
still lower than for lytic control and in neither of the latter cases there is a trend with
increasing probability of successful scan (Appendix Table F.1).
Therefore, the lack of immune control cannot explain the slower dynamics of escape
that we observe when the CD8+ T cell viral suppression is non-lytic. This suggests that
intrinsic properties of the non-lytic control are responsible for the slower variant growth
over the wild-type. In the case of blocking infection, the fact that the uninfected targets
are ‘protected’ from both wild-type and variant infection limits the advantage of the
variant over the wild-type even when the effect of the soluble factor is not extended
to a wide area around the activated specific CD8+ T cell (e.g. polarised pattern).
Similarly, the fact that both wild-type and variant infected cells can be ‘silenced’ by
soluble factors obviously restricts the advantage of the variant population.
To summarise, we observe slower and less frequent viral escape under a non-lytic
compared to a lytic control mechanism which remains similarly slow for the different
non-lytic secretion patterns (polarised and diffusive) and despite of an increase in the
probability of successful scan by the effector cells; this is observed both when the viral
production or the viral infection are abrogated.
213
0.0
00
.04
0.0
80
.12
Probability of Successful Scan
Est
ima
ted
esc
ap
e r
ate
pe
r d
ay
0.001 0.002 0.0030.001 0.002 0.003
LyticNLi−P1NLi−D1NLi−D2NLp−TONLp−P1NLp−D1NLp−D2
Figure 7.12: Escape rate is slower when the CD8+ T cell selection pressure is exerted
in a non-lytic manner. We simulated different secretion patterns of non-lytic factors
and different levels of selection pressure in each case. We estimate the variant escape
rate in the simulated datasets using Equation 6.9. The ‘protective effect’ lasts for
10 hrs and it totally safeguards the cells from infection. Abbreviations: NLi: Non-
lytic model - blocking infection of uninfected CD4+ T cells, NLp: Non-lytic model -
blocking viral production from infected CD4+ T cells, TO=Target Only, P1=Polarised
secretion (r=1), D1=Diffusive secretion (r=1) and D2=Diffusive secretion (r=2).
214
0.001 0.002 0.003
2040
6080
NLi − Polarised (r=1)
Num
ber
of u
ninf
ecte
d C
D4+
T c
ells
`pr
otec
ted’
non
−ly
tical
ly
0.001 0.002 0.003
5010
015
020
025
030
0
NLi − Diffusive (r=1)
Num
ber
of u
ninf
ecte
d C
D4+
T c
ells
`pr
otec
ted’
non
−ly
tical
ly
0.001 0.002 0.003
400
600
800
1000
1200
1400
NLi − Diffusive (r=2)
Num
ber
of u
ninf
ecte
d C
D4+
T c
ells
`pr
otec
ted’
non
−ly
tical
ly
Probability of successful scan
Figure 7.13: Number of uninfected CD4+ T cells ‘protected’ from infection under a
non-lytic CD8+ T cell response manifested in a polarised or diffusive secretion pat-
tern. We show the cumulative number at 50 dpi. Abbreviations: NLi: Non-lytic
model - blocking infection of uninfected CD4+ T cells, P1=Polarised secretion (r=1),
D1=Diffusive secretion (r=1) and D2=Diffusive secretion (r=2).
215
Lytic NLi−P1 NLi−D1 NLi−D2
050
100
150
200
250
pss=0.001
Num
ber
of in
fect
ions
pre
vent
ed b
y th
e ep
itope
−sp
ecifi
c ef
fect
ors
Type of epitope−specific CD8+ T cell control
Lytic killing~1%
Lytic NLi−P1 NLi−D1 NLi−D2
050
100
150
200
250
pss=0.002
Num
ber
of in
fect
ions
pre
vent
ed b
y th
e ep
itope
−sp
ecifi
c ef
fect
ors
Type of epitope−specific CD8+ T cell control
Lytic killing~3%
Lytic NLi−P1 NLi−D1 NLi−D2
050
100
150
200
250
pss=0.003
Num
ber
of in
fect
ions
pre
vent
ed b
y th
e ep
itope
−sp
ecifi
c ef
fect
ors
Type of epitope−specific CD8+ T cell control
Lytic killing~5%
Figure 7.14: New infections prevented under a non-lytic CD8+ T cell response that
blocks infection. We show the number of infections prevented by the epitope-specific
CD8+ T cell clones for 40-50 dpi, just before the variant infected cell population
is introduced in the simulations. Abbreviations: NLi: Non-lytic model - blocking
infection of uninfected CD4+ T cells, P1=Polarised secretion (r=1), D1=Diffusive
secretion (r=1) and D2=Diffusive secretion (r=2).
216
Lytic NLi−P1 NLi−D1 NLi−D2
0.00
00.
005
0.01
00.
015
0.02
00.
025
0.03
0
pss=0.001
% P
rodu
ctiv
ely
wild
−ty
pe in
fect
ed c
ells
afte
r se
t−po
int
Lytic killing~1%
Lytic NLi−P1 NLi−D1 NLi−D2
0.00
00.
005
0.01
00.
015
0.02
00.
025
0.03
0
pss=0.002
% P
rodu
ctiv
ely
wild
−ty
pe in
fect
ed c
ells
afte
r se
t−po
int
Lytic killing~3%
Lytic NLi−P1 NLi−D1 NLi−D2
0.00
00.
005
0.01
00.
015
0.02
00.
025
0.03
0
pss=0.003
% P
rodu
ctiv
ely
wild
−ty
pe in
fect
ed c
ells
afte
r se
t−po
int
Lytic killing~5%
Type of epitope−specific CD8+ T cell control
Figure 7.15: Set-point of productively infected cells. We show the percentage of pro-
ductively infected wild-type cells for 40-50 dpi, just before the variant infected cell
population is introduced in the simulations and after the steady-state has been at-
tained. Here, we present the results for the lytic control and the non-lytic control that
blocks viral infection (see Appendix F for non-lytic response that blocks viral produc-
tion). Abbreviations: NLi: Non-lytic model - blocking infection of uninfected CD4+ T
cells, P1=Polarised secretion (r=1), D1=Diffusive secretion (r=1) and D2=Diffusive
secretion (r=2).
217
7.4.2 Cluster formation of infected cells
Using the CA model, we showed that under a comparable immune control, the variant
infected cells outgrow the wild-type infected cells at a slower rate when the CD8+ T cell
effector mechanism is non-lytic. However, the reason why the variant infected cells have
an advantage over the wild-type cells -even if limited- is not immediately understood.
The non-lytic response, either the one that blocks viral infection or the one that blocks
viral production, affects both infected populations; so under the assumption of spatially
well-mixed cell populations, neither would be expected to have an advantage. We
further explored whether there are spatial patterns that can explain this advantage of
the variant population under a non-lytic control.
Although on average there are approximately similar numbers of wild-type and
variant infected cells around infected CD4+ T cells of any of the two types, the overall
distribution is highly skewed toward the presence of cells of the same type (Figure 7.16).
The percentage of wild-type infected cells around a wild-type type cell is predominantly
>= 50% while the percentage of variant infected cells is mainly <= 50%. The same is
true for the infected cells surrounding the variant infected cells, indicating the presence
of clusters of wild-type and variant infected cells. The existence of such clusters rejects
the well-mixed assumption and suggests that spatial patterns can provide an advantage
to the variant infected cells since only the wild-type infected cells -found usually close
to other wild-type cells -can trigger the secretion of soluble factors by a CD8+ T cell
that would suppress viral spread. If a wild-type infected cell activates the CD8+ T
cell and this in turn secretes factors that block infection in the surrounding area, it
would be more likely to affect wild-type infected cells and thus the variant infected
cells have an advantage even though they are equally susceptible to non-lytic factors.
Nevertheless, this process cannot have an extensive effect because of the fast timescale
of T cell motility that forces the spatial effects to quickly mix. However, this is a
dynamic process and the repetitive formation of such clusters during the spread of the
218
infection is likely to provide a net whilst limited advantage to the variant infected cell
population that can explain outgrowth albeit at a slower speed.
WT/WT WT/VAR WT/ALL
0.0
0.2
0.4
0.6
0.8
1.0
Wild−type clusters
Spatial location
Fre
quen
cy o
f wild
−ty
pe in
fect
ed c
ells
p<0.001
VAR/VAR VAR/WT VAR/ALL
0.0
0.2
0.4
0.6
0.8
1.0
Variant clusters
Spatial location
Fre
quen
cy o
f var
iant
infe
cted
cel
ls
p<0.001
Figure 7.16: The percentage of wild-type infected cells around a wild-type cells is
significantly higher than the percentage of variant infected cells that surround it.
The same is true for variant infected cells. We track simulated wild-type and vari-
ant infected CD4+ T cells (≈ 500 cells) at multiple timepoints. Abbreviations:
WT/WT=Wild-type infected cells in the immediate area surrounding a wild-type in-
fected cell, WT/VAR=Wild-type infected cells in the immediate area surrounding a
variant infected cell, WT/ALL=Wild-type infected cells on the whole grid, VAR/VAR,
VAR/WT and VAR/ALL are defined in the same way.
7.4.3 The efficiency of the non-lytic control mechanism
Although we find that CD8+ T cells that act non-lytically can lead to viral escape,
we also observe that non-lytic control results in a lower number of ‘prevented’ cell
219
infections and this does not increase either with the radius of secretion or the increase
of the successful encounters with infected cells (Figure 7.14). Therefore, we explore
further the efficiency of the non-lytic control not in terms of viral escape but in terms
of limiting the growth of the wild-type.
There are four main quantities in the CA model that could potentially strengthen
the non-lytic response: 1) the radius of secretion, 2) the probability of successful scan,
3) the duration of the effect of the soluble factor and 4) the size of the CD8+ T
cell population. We vary each parameter separately by setting all other parameters
at a constant value. We already saw that the first two parameters did not have a
substantial effect on boosting the response. Additionally, when we double the effect
of the soluble factor from 10 to 20 hrs, we find no consistent substantial impact on
controlling the spread of the wild-type either by blocking infection (Figure 7.17) or
by blocking production (data not shown). Finally, we increase the size of the effector
population from 0.5% to 1.5% of the splenocytes. Although, the number of cells that
are ‘protected’ by the soluble factors increases (Figure 7.18), again we observe no
consistent substantial effect on the strength (in terms of infections prevented) of the
non-lytic control (Figure 7.19); we also see no effect on the set-point of the infected cell
population. This result becomes even clearer when we juxtapose it to the lytic control
where the increase of the size of the effector population results in a stronger immune
pressure both in terms of infections prevented (Figure 7.19) and set-point infected cell
population (data not shown). Our findings taken together suggest that the non-lytic
control can be less efficient in limiting the infection compared to lytic control and
might require a very large number of effector cells in order to provide strong viral
suppression.
220
050
100
150
200
Type of epitope−specific CD8+ T cell control
Num
ber
of in
fect
ions
pre
vent
ed b
y th
e ep
itope
−sp
ecifi
c ef
fect
ors
NLi−P1 NLi−D1 NLi−D2
SE= 10 hrsSE= 20 hrs
Figure 7.17: New infections prevented under a non-lytic CD8+ T cell response that
blocks infection for varying duration of the effect of the soluble factor (SE). We show
the number of infections prevented by the epitope-specific CD8+ T cell clones for 40-50
dpi, just before the variant infected cell population is introduced in the simulations.
We run 10 simulations for each case. The probability of successful scan is set to 0.002
in all the simulations. Abbreviations: NLi: Non-lytic model - blocking infection of
uninfected CD4+ T cells, P1=Polarised secretion (r=1), D1=Diffusive secretion (r=1)
and D2=Diffusive secretion (r=2).
221
NLi−P1 NLi−D1 NLi−D2
050
010
0015
0020
0025
00
Type of epitope−specific CD8+ T cell control
Num
ber
of u
ninf
ecte
d C
D4+
T c
ells
`pr
otec
ted’
non
−ly
tical
ly
NLi−P1 NLi−D1 NLi−D2
050
010
0015
0020
0025
00
NLi−P1 NLi−D1 NLi−D2
050
010
0015
0020
0025
00
C= 0.5%C= 0.75%C= 1.5%
Figure 7.18: Number of uninfected CD4+ T cells ‘protected’ from infection under a
non-lytic CD8+ T cell response manifested in a polarised or diffusive secretion pattern
for different sizes of the epitope-specific effector cell population, C, as a percentage
of the total splenocyte population. We show the cumulative number at 50 dpi. We
run 10 simulations for each case. The probability of successful scan is set to 0.002
in all the simulations. Abbreviations: NLi: Non-lytic model - blocking infection of
uninfected CD4+ T cells, P1=Polarised secretion (r=1), D1=Diffusive secretion (r=1)
and D2=Diffusive secretion (r=2).
222
050
100
150
200
250
300
Type of epitope−specific CD8+ T cell controlNum
ber
of in
fect
ions
pre
vent
ed b
y ep
itope
−sp
ecifi
c ef
fect
ors
Lytic NLi−P1 NLi−D1 NLi−D2
Lytic−C=0.5%Lytic−C=1.5%NLi−C=0.5%NLi−C=1.5%
Figure 7.19: New infections prevented under a non-lytic CD8+ T cell response that
blocks infection. We show the number of infections prevented by different sizes of the
epitope-specific CD8+ T cell clones, C, for 40-50 dpi, just before the variant infected
cell population is introduced in the simulations. We run 10 simulations for each case.
The probability of successful scan is set to 0.002 in all the simulations. Abbreviations:
NLi: Non-lytic model - blocking infection of uninfected CD4+ T cells, P1=Polarised
secretion (r=1), D1=Diffusive secretion (r=1) and D2=Diffusive secretion (r=2).
223
7.5 Discussion
Emerging evidence suggests that CD8+ T cells can control HIV viral burden through
multiple mechanisms [37, 42, 58, 167, 331] which can be broadly classified as lytic and
non-lytic. Although, the importance of non-lytic effects in HIV infection is shown by
multiple studies (reviewed in [50]) little is known about the efficiency of such a response
compared to the lytic mechanism. Quantifying the CD8+ T cell lytic and non-lytic
immune control would require a complex set of in vivo experimental data that would
capture spatial and temporal aspects of the system. To address these questions in
silico we construct a 3D cellular automaton model of HIV/SIV dynamics. Our aim
is two-fold: 1) to investigate the efficiency of lytic and non-lytic responses and 2) to
explore whether non-lytic control can drive viral escape. Whilst it is readily understood
how direct killing of wild-type infected cells can result in variant outgrowth, this is
not as obvious under a non-lytic immune suppression mechanism that blocks either
viral infection or viral production and which can equally affect wild-type and variant
infected cells.
In summary, the 3D cellular automaton model includes an extensive set of rules
on cell motility, infection spread and cell death that can recapitulate experimentally
observed dynamics of HIV/SIV infection and can be used for the explicit simulation
of both lytic and non-lytic epitope-specific CD8+ T cell responses.
For a lytic response, the model allows us to study closely the CD8+ T cell killing
term. We find that the killing process is better explained by a model that incorporates
both mass-action killing and saturated killing. This implies that CD8+ T cell killing
can be substantially restricted by factors other than the availability of infected cells.
Specifically, we find that the killing rate is predominantly limited by the time that the
CD8+ T cell needs in order to successfully identify its infected target and less restricted
by the duration of killing per se or the number of targets. Interestingly, we find that
224
biologically plausible CD8+ T cell killing rates can be attained for a small probability
of successful identification of infected cells. Furthermore, it has been suggested by
other studies that a mass-action term is only adequate for well-mixed populations at
intermediate effector:target ratios [332]. In the simulations however, we observe the
formation of clusters that do not align with a well-mixed hypothesis and could possibly
result in the break-down of the mass-action behaviour in terms of infected targets; at
this point though, this is only a potential explanation. Nevertheless, in terms of effector
cells, our findings are consistent with a mass-action behaviour, i.e. we cannot exclude
that the killing rate increases proportionally to the size of the CD8+ T cell population.
Additionally, we study the relationship between immune control and viral escape
dynamics. We compare a lytic mechanism with a non lytic mechanism that either
blocks viral infection or inhibits viral production. Our findings show that a non-lytic
control mechanism that blocks viral infection (such as in the case of RANTES) or
restricts viral production (such as in the case of IFN-γ) results in significantly slower
escape rate even when the lytic and non-lytic immune pressures are comparable in
terms of both number of infections prevented and set-point dynamics. Importantly,
we find that although the number of uninfected cells which are ‘protected’ by infection
increases as the radius of secretion of soluble factor increases, the escape rate remains
relatively constant and is significantly slower than the escape rate observed under
a lytic mechanism. In line with the latter observation, in [298] the authors use a
theoretical ODE model which also shows that if CD8+ T cells control virus replication
non-lytically the rate of viral escape can be reduced although numerically the reduction
can be dismal; however, the latter model is deterministic and assumes that non-lytic
mechanisms can drive viral escape while the CA model allows us to formally prove
this.
Our simulations further indicate that the advantage which the escape mutant has
under a non-lytic control can be potentially attributed to the formation of clusters of
infected cells. A wild-type cell that triggers the secretion of soluble factors is more likely
225
to be surrounded by other wild-type cells rather than variant cells, allowing therefore
an outgrowth advantage to the variant strain. Interestingly, patches of infected cell
clusters have been reported in in situ hybridization studies of SIV-infected cells [333].
Focusing on non-lytic control, we find that the efficiency of a non-lytic response
in terms of infections prevented at steady-state is lower compared to the efficiency
of a lytic response. We show that an increase of the duration of the non-lytic effect
(from 10hrs to 20hrs) does not result in a stronger response, neither does an increase
by three-fold of the effector population. This implies that although many uninfected
cells are ‘protected’ or many infected cells are ’blocked’, if the infected cells are not
eliminated by the effector cells, there are still enough available targets to sustain the
set-point viral burden. Hence, potentially very high numbers of effectors cells are
needed for an efficient non-lytic suppression.
Our results are quantitatively and qualitatively very similar for non-lytic responses
of blocking infection and blocking production. This is not surprising since in the
model we focus on the chronic stage of infection where under the quasi-steady state
assumption that virus production is proportional to the number of infected cells, the
two types of suppression are expected to have identical dynamics. This can also be
readily shown by the ODE models that describe the two different effector mechanisms
(see Appendix F.5) and have been discussed in other studies [3].
Of course, since our model is theoretical it relies on parameter estimates which
have been previously published. We vary many of these parameters in our simulations
in order to check for the robustness of our results. In some cases, such as the motility
behaviour of T cells and the lifespan of infected cells the estimates have been corrobo-
rated by many studies. However, there are other parameters in the model that are not
that well-defined such as the effect and radius of secretion of soluble factors. Here, we
have varied the radius of the effect of the non-lytic response in order to examine how
it impacts the immune control dynamics. Regarding the actual effect of the soluble
factor, e.g. if it totally abrogates viral infection or production, little is known for most
226
of the cytokines and chemokines. Therefore, we have assumed a fully abrogating effect
but as more data become available the model can be adjusted easily for the study of
other factors and different effect sizes.
It is worth noting that the CA model can easily be used to investigate many aspects
of the HIV/SIV dynamics. Specifically, it would be interesting to study the effect of
the CD4+ T cell response on the course of the infection. In [334], the authors find
evidence that the CD4+ T cell effector function can influence HIV/SIV dynamics
during primary infection; although this would not alter the conclusions of this study
since we mainly focus on chronic infection, it would be interesting to quantify -using the
CA- the impact of a CD4+ T cell-mediated antiviral effect on the post-peak vireamic
decline.
Our findings have implications for future studies of HIV progression and vaccine-
induced responses. The appearance of escape mutants has been associated with loss of
viral control and disease progression [335]. Our results indicate that non-lytic control
can indeed slow down the appearance of escape mutants. In addition, in theory, a non
cytolytic effector mechanism has three main advantages 1) can act on more than one
target 2) can suppress without eliminating potentially vital cell populations and 3) can
exhibit a ‘bystander’ protective effect [50]. Therefore, it can be argued that the tipping
of balance towards non-lytic responses can be proven a strong weapon in the fight of
HIV infection; however, an efficient non-lytic response mechanism that will not only
lead to slower viral escape but also restrict the growth of the infected cell population
might require a large number of effector cells. On the other hand, a lytic response can
result in faster escape but also higher immune control suggesting that an appropriate
combination of both responses can possibly offer the most efficient viral containment.
227
Chapter 8
General Discussion
As in every research project, some questions were answered and many more exciting
ones arose. In this last part of the thesis, I would like to briefly summarise the findings
presented and discuss their relevance and implications for further studies.
There are many persistent infections that have afflicted the human population
over the years. Arguably, the best solution to persistent infections is the existence of
prophylactic vaccines. However, for many viruses such as HCV, HTLV-1 and HIV this
has not yet been possible. A key question that still remains largely unanswered is what
differs between individuals that progress to persistent infection and those who remain
asymptomatic or even successfully clear the virus (as in the case of HCV). The fact that
persistent infection develops though complex interactions between the immune system
and the pathogen makes it hard to decipher. Hence, we need to concentrate on a few
key players at a time and try to understand their role in this interplay and how they fit
in the answer to this question. Here, we focus mainly on CD8+ T cells because they
have been repeatedly shown to play a major role in many viral infections providing
therefore a valuable means of revealing potential drivers that lead to persistent disease.
Both the quality and the quantity of CD8+ T cell response to viral infection have
been associated with viral control. However, when studying the role of CD8+ T cells
in viral infection, one main problem occurs: causality. It is not possible, in the lack
of elaborate longitudinal data, to determine whether the CD8+ T cell response is the
cause or the consequence of variations in viral burden. One of the strongest evidence
for the importance of CD8+ T cells in controlling viral infection is the association
between HLA class I molecules and viral infection load and outcome. Hence, in order to
228
address the issue of causality, we studied the impact of HLA class I molecules, for which
the direction of causality is unequivocal, on disease manifestation. In the context of
HCV infection we tried to understand which HLA class I-related factors could explain
why some individuals resolve the infection while others develop persistent disease.
We found that although there were molecules that had a protective effect, namely
B*57 and C*01, and there was a weak heterozygote advantage for HLA-B alleles,
the overall predicted breadth, protein specificity and epitope specificity of the HLA
class I molecules of individuals that resolved or persisted did not show any significant
differences. This made us question the magnitude of the role of HLA class I molecules
in determining the outcome of HCV infection. We quantified the effect using the
explained fraction - a measure that estimates how much a given factor explains the
heterogeneity of an outcome - and found that the HLA class I molecules associated
with HCV infection outcome (for B*57, C*01 and C*04 ) could explain a total of 3%
of the outcome whilst in HTLV-1 the fraction was close to 6% (for A*02, B*54 and
C*08 together) and even larger for HIV infection where for B*57 alone the explained
fraction was approximately 5%. This suggests that the most protective molecule in
HIV or HTLV-1 infection might provide a larger benefit to the host compared to the
most protective molecule in HCV infection or -and not mutually exclusive- the most
detrimental molecule in HCV infection might not be as detrimental as in HIV or
HTLV-1 infections. The fact that B*57 is a ‘protective’ molecule in both HIV and
HCV infection can potentially form a basis for experimentally testing this finding by
comparing the B*57-restricted CD8+ T cell response against HCV and HIV peptides
in order to understand under a different benchmark what defines a strong ‘protective
molecule’.
Next, we wanted to investigate whether there were other genetic factors that could
potentially enhance this response. For this, we turned to Killer Cell Immunoglobulin-
like receptors (KIRs) who are ligated by HLA class I molecules and have been associ-
ated mainly with NK cell responses. Interestingly, we found that a specific inhibitory
229
KIR, namely KIR2DL2, could enhance both protective and detrimental HLA class
I-mediated immunity. This was true for molecules involved in both HCV and HTLV-1
infections suggesting a more general underlying mechanism. Perhaps, the most sur-
prising indication was that KIR2DL2 appears to enhance the downstream effect of
CD8+ T cells rather than NK cells (although it might still be expressed on NK cells
alone). This introduces a novel role for KIRs in adaptive immunity which should be
explored further. We postulate that the expression of iKIRs directly on NK or CD8+
T cells can eventually increase CD8+ T cell survival. Ligation of iKIRs on CD8+ T
cells can upregulate pro-survival molecules and reduce activation induced cell death
(AICD) [79,80,227,228]. Alternatively, ligation of iKIRs on NK cells can downregulate
the killing of activated T cells [237, 238]. In both cases, CD8+ T cells would survive
longer in the presence of inhibitory KIRs and thus their protective and detrimental
associations would be enhanced. In order to better understand the underlying mecha-
nism we propose to study the effect of KIR2DL2 (and also other KIRs) on CD8+ T cell
selection pressure. The latter can be quantified using sequence data and estimating the
number of coding changes within epitope regions. If inhibitory KIRs are found to tune
CD8+ T cell responses it would provide a new way of modulating viral control. Of
particular interest are studies suggesting that strategies to block inhibitory receptors
such as iKIRs may be of potential use in increasing the efficacy of immunotherapy and
this has already been achieved in simple mice models of cancer [336,337]. Additionally,
it would be important to test whether HLA class I-mediated protection against other
viral infections or viral-associated malignancies can be enhanced by KIRs.
Until now our study focused on the immunogenetics of persistent viral infections,
in terms of HLA and KIR genes, with a main focus on their impact on CD8+ T cell
responses. However, we also aimed at investigating the different manifestations of these
responses (lytic versus non-lytic activities) and how they relate to viral dynamics. For
this part of the study, we chose to concentrate on HIV/SIV infection since recent studies
[42, 167] of SIV infection challenged the cytolytic role of CD8+ T cells and pointed
230
towards a non-lytic control that can lead to viral decline without reducing the lifespan
of productively infected cells. These two effector functions differ in that the secretion
of soluble factors (non-lytic response) can have both local and systemic consequences,
whereas cell lysis is restricted to the elimination of the direct target. So, the first
question that rises is whether a non-lytic response could explain the findings of these
recent studies and what is the potential anti-viral efficiency of non-lytic mechanisms
compared to lytic mechanisms? If we also take into account the observation that
HIV and SIV escape the CD8+ T cell control frequently, some additional important
questions are posed. Is the consistent observation of viral escape proof that HIV-1-
specific CD8+ T cells lyse infected cells? If CD8+ T cells control SIV primarily by
non-lytic mechanisms then why is viral escape observed?
On a first approach, we studied whether lytic and non-lytic responses can explain
the data observed in [42]. We focused on non-lytic effector functions that block infec-
tion of uninfected cells or inhibit viral production by infected cells. We fitted a number
of ODE models to the experimental data obtained from SIV-infected macaques and
found some evidence that non-lytic control, particularly the one that blocks viral in-
fection, can better explain the data and describe the underlying CD8+ T cell response.
Nevertheless, as discussed in [3] more accurate measurements of immunological data
are needed before firmer conclusions are drawn.
It should be emphasised that lytic and non-lytic responses are not mutually exclu-
sive. The inherent ability of viruses to induce variable antigenic stimulation may be a
determinant of the lytic versus non-lytic virus-specific CD8+ T cell responses [338,339].
Additionally, different markers expressed on the cell surface can also regulate antiviral
CD8+ T cell effector activities [340]. However, in order to understand how these two
different functions alter viral dynamics it would be ideal to study them separately. This
cannot easily be tested in vivo, it can however be explored in silico. Hence, we con-
structed a detailed computational model, namely a 3D cellular automaton, so that we
can simulate lytic and non-lytic CD8+ T cell responses and study how they modulate
231
virus-infected cell dynamics. Our main aim was to understand how non-lytic responses
compare in efficiency to lytic responses and whether a non-lytic response can result in
the outgrowth of the cell population that is infected with a variant strain; this is after
all one of the strongest consequences of CD8+ T cells acting in a lytic manner. Our
approach although theoretical is based on many biological parameters and processes
that have been quantified experimentally. As in every theoretical model assumptions
needed to be made in order to ease the analysis as well as the interpretation of the
results. The 3D cellular automaton model of HIV/SIV dynamics produced quantita-
tive and qualitative results that are very consistent with experimental observations.
Importantly, the flexibility of this model allows for the future study of other important
questions such as target cell competition, early CD8+ T cell killing, simulation of other
non-lytic soluble factors and CD8+ T cell dysfunction.
We first studied the CD8+ T cell lytic response and found that for a CD8+ T cell
to eliminate 1-5% of the infected targets per day, it needs to successfully recognise on
average one infected target every 1000 down to 300 encounters respectively. It would
be interesting to investigate whether this observation could be corroborated by the use
of an ODE model that takes into account the encounter rate of CD8+ T cells with
their infected targets. We also found that a model of lytic killing that is described by a
saturating behaviour in terms of infected targets can better explain the killing process.
This suggests that there are limiting factors involved in CD8+ T cell killing that would
cause the mass-action behaviour in terms of the infected targets to collapse; however,
we find that a higher number of effector cells results in a higher killing rate and this
relationship can be consistent with a mass-action behaviour in terms of effector cells.
Our findings further suggest that one of the factors that might restrict killing in terms
of infected targets can be the need to identify the target before actually eliminating it
while the duration of killing per se has a dismal impact on the killing rate. Although
important two-photon microscopy studies of CD8+ T cell killing [301, 305] allow the
measurement of the time that CD8+ T cells spent in delivering the lethal hit to antigen-
232
pulsed cells and record the time they browse the pulsed or unpulsed targets, they do
not capture information on the time that a CD8+ T cell takes before it successfully
locates its target. Based on our findings, this type of knowledge would facilitate a
better understanding of what may lead to the restriction of the CD8+ T cell killing
rate. Interestingly, other studies have also indicated a saturation killing term as a
better description of a lytic CD8+ T cell response and although there is still a debate
on the subject it can be potentially explained by the fact that many studies are being
performed on different viruses and under different experimental setups.
Next, we investigated the efficiency of the two different CD8+ T cell effector mech-
anisms and how they can shape the dynamics of viral escape. Clearly, production of
soluble factors could lead to viral escape only if it is MHC-dependent; but even if this
is the case, can viral escape be observed? We found that lytic CD8+ T cells could re-
producibly drive rapid viral escape. Escape from non-lytic CD8+ T cells (both those
that block infection and those that block production) was also repeatedly observed
however it was less frequent and slower than in the lytic case. We also found that
the formation of infected cell clusters of the same type (wild-type or variant) could
explain the advantage of the variant strain under a non-lytic mechanism. Overall, we
saw that the non-lytic control is less efficient than the lytic control in terms of infec-
tions prevented and set-point viral burden. Furthermore, the non-lytic effect did not
strengthen with increasing number of effector cells (while the lytic effect did) or when
the duration of the abrogating effect (blocking infection or production) was extended.
These findings taken together suggest that viral escape is consistent with both lytic
and non-lytic CD8+ T cell responses. However, a non-lytic response although it can
lead to slower and less frequent viral escape compared to a lytic response, it might
not be as efficient in controlling the viral burden and may require very large numbers
of effector cells in order to exhibit high viral suppression. Our findings also provide a
potential reconciliation between studies that report rapid and slow escape by showing
that the underlying CD8+ T cell effector function can be a main driver of the observed
233
differences. It would be very important, especially for future T cell-based vaccine re-
search, to explore such a possibility experimentally by recording longitudinally the
epitope-specific CD8+ T cell lytic and non-lytic profile along with the escape rate of
different viral epitopes.
As a concluding remark, I would like to note that a large amount of this work was
the result of a productive collaboration between theoreticians and experimentalists
and can be considered as a small indication of how these disciplines can work along-
side each other. Hopefully, it also shows how this can accommodate the advancing of
our knowledge of the immune system and biology as a whole. As J. Cohen insight-
fully said [341] ‘Mathematics Is Biology’s Next Microscope, Only Better; Biology Is
Mathematics’ Next Physics, Only Better’.
234
Appendices
235
A Appendix A
A.1 The HCV genome and gene products
Appendix Figure A.1: The HCV genome and gene products. HCV is positive-sense
single-stranded RNA virus. This viral genome contains a single open reading frame
(ORF) encoding a polyprotein, processed into at least 10 proteins; three structural
proteins and seven non-structural proteins [342]. The image is taken from [342].
236
A.2 The HCV-1a protein sequence (isolate H77)
>F
MSTNPKPQRKPNVTPTVAHRTSSSRVAVRSLVEFTCCRAGALDWVCARRGRLPSGRNLEVDVSLSPRHVG
PRAGPGLSPGTLGPSMAMRVAGGRDGSCLPVALGLAGAPQTPGVGRAIWVRSSIPLRAASPTSWGTYRSS
APLLEALPGPWRMASGFWKTA
>core
MSTNPKPQRKTKRNTNRRPQDVKFPGGGQIVGGVYLLPRRGPRLGVRATRKTSERSQPRGRRQPIPKARR
PEGRTWAQPGYPWPLYGNEGCGWAGWLLSPRGSRPSWGPTDPRRRSRNLGKVIDTLTCGFADLMGYIPLV
GAPLGGAARALAHGVRVLEDGVNYATGNLPGCSFSIFLLALLSCLTVPASA
>E1
YQVRNSSGLYHVTNDCPNSSIVYEAADAILHTPGCVPCVREGNASRCWVAVTPTVATRDGKLPTTQLRRH
IDLLVGSATLCSALYVGDLCGSVFLVGQLFTFSPRRHWTTQDCNCSIYPGHITGHRMAWDMMMNWSPTAA
LVVAQLLRIPQAIMDMIAGAHWGVLAGIAYFSMVGNWAKVLVVLLLFAGVDA
>E2
ETHVTGGSAGRTTAGLVGLLTPGAKQNIQLINTNGSWHINSTALNCNESLNTGWLAGLFYQHKFNSSGCP
ERLASCRRLTDFAQGWGPISYANGSGLDERPYCWHYPPRPCGIVPAKSVCGPVYCFTPSPVVVGTTDRSG
APTYSWGANDTDVFVLNNTRPPLGNWFGCTWMNSTGFTKVCGAPPCVIGGVGNNTLLCPTDCFRKHPEAT
YSRCGSGPWITPRCMVDYPYRLWHYPCTINYTIFKVRMYVGGVEHRLEAACNWTRGERCDLEDRDRSELS
PLLLSTTQWQVLPCSFTTLPALSTGLIHLHQNIVDVQYLYGVGSSIASWAIKWEYVVLLFLLLADARVCS
CLWMMLLISQAEA
>P7
ALENLVILNAASLAGTHGLVSFLVFFCFAWYLKGRWVPGAVYAFYGMWPLLLLLLALPQRAYA
>NS2
LDTEVAASCGGVVLVGLMALTLSPYYKRYISWCMWWLQYFLTRVEAQLHVWVPPLNVRGGRDAVILLMCV
VHPTLVFDITKLLLAIFGPLWILQASLLKVPYFVRVQGLLRICALARKIAGGHYVQMAIIKLGALTGTYV
YNHLTPLRDWAHNGLRDLAVAVEPVVFSRMETKLITWGADTAACGDIINGLPVSARRGQEILLGPADGMV
SKGWRLL
237
>NS3
APITAYAQQTRGLLGCIITSLTGRDKNQVEGEVQIVSTATQTFLATCINGVCWTVYHGAGTRTIASPKGP
VIQMYTNVDQDLVGWPAPQGSRSLTPCTCGSSDLYLVTRHADVIPVRRRGDSRGSLLSPRPISYLKGSSG
GPLLCPAGHAVGLFRAAVCTRGVAKAVDFIPVENLETTMRSPVFTDNSSPPAVPQSFQVAHLHAPTGSGK
STKVPAAYAAQGYKVLVLNPSVAATLGFGAYMSKAHGVDPNIRTGVRTITTGSPITYSTYGKFLADGGCS
GGAYDIIICDECHSTDATSILGIGTVLDQAETAGARLVVLATATPPGSVTVSHPNIEEVALSTTGEIPFY
GKAIPLEVIKGGRHLIFCHSKKKCDELAAKLVALGINAVAYYRGLDVSVIPTSGDVVVVSTDALMTGFTG
DFDSVIDCNTCVTQTVDFSLDPTFTIETTTLPQDAVSRTQRRGRTGRGKPGIYRFVAPGERPSGMFDSSV
LCECYDAGCAWYELTPAETTVRLRAYMNTPGLPVCQDHLEFWEGVFTGLTHIDAHFLSQTKQSGENFPYL
VAYQATVCARAQAPPPSWDQMWKCLIRLKPTLHGPTPLLYRLGAVQNEVTLTHPITKYIMTCMSADLEVV
T
>NS4A
STWVLVGGVLAALAAYCLSTGCVVIVGRIVLSGKPAIIPDREVLYQEFDEMEEC
>NS4B
SQHLPYIEQGMMLAEQFKQKALGLLQTASRQAEVITPAVQTNWQKLEVFWAKHMWNFISGIQYLAGLSTL
PGNPAIASLMAFTAAVTSPLTTGQTLLFNILGGWVAAQLAAPGAATAFVGAGLAGAAIGSVGLGKVLVDI
LAGYGAGVAGALVAFKIMSGEVPSTEDLVNLLPAILSPGALVVGVVCAAILRRHVGPGEGAVQWMNRLIA
FASRGNHVSPTHYVPESDAAARVTAILSSLTVTQLLRRLHQWISSECTTPC
>NS5A
SGSWLRDIWDWICEVLSDFKTWLKAKLMPQLPGIPFVSCQRGYRGVWRGDGIMHTRCHCGAEITGHVKNG
TMRIVGPRTCRNMWSGTFPINAYTTGPCTPLPAPNYKFALWRVSAEEYVEIRRVGDFHYVSGMTTDNLKC
PCQIPSPEFFTELDGVRLHRFAPPCKPLLREEVSFRVGLHEYPVGSQLPCEPEPDVAVLTSMLTDPSHIT
AEAAGRRLARGSPPSMASSSASQLSAPSLKATCTANHDSPDAELIEANLLWRQEMGGNITRVESENKVVI
LDSFDPLVAEEDEREVSVPAEILRKSRRFARALPVWARPDYNPPLVETWKKPDYEPPVVHGCPLPPPRSP
PVPPPRKKRTVVLTESTLSTALAELATKSFGSSSTSGITGDNTTTSSEPAPSGCPPDSDVESYSSMPPLE
GEPGDPDLSDGSWSTVSSGADTEDVVCC
>NS5B
SMSYSWTGALVTPCAAEEQKLPINALSNSLLRHHNLVYSTTSRSACQRQKKVTFDRLQVLDSHYQDVLKE
238
VKAAASKVKANLLSVEEACSLTPPHSAKSKFGYGAKDVRCHARKAVAHINSVWKDLLEDSVTPIDTTIMA
KNEVFCVQPEKGGRKPARLIVFPDLGVRVCEKMALYDVVSKLPLAVMGSSYGFQYSPGQRVEFLVQAWKS
KKTPMGFSYDTRCFDSTVTESDIRTEEAIYQCCDLDPQARVAIKSLTERLYVGGPLTNSRGENCGYRRCR
ASGVLTTSCGNTLTCYIKARAACRAAGLQDCTMLVCGDDLVVICESAGVQEDAASLRAFTEAMTRYSAPP
GDPPQPEYDLELITSCSSNVSVAHDGAGKRVYYLTRDPTTPLARAAWETARHTPVNSWLGNIIMFAPTLW
ARMILMTHFFSVLIARDQLEQALNCEIYGACYSIEPLDLPPIIQRLHGLSAFSLHSYSPGEINRVAACLR
KLGVPPLRAWRHRARSVRARLLSRGGRAAICGKYLFNWAVRTKLKLTPIAAAGRLDLSGWFTAGYSGGDI
YHSVSHARPRWFWFCLLLLAAGVGIYLLPNR
A.3 The HTLV-1 protein sequence
The HTLV-1 reference strain is from [343] with the exception of HBZ, which was more
recently described in [344].
>Gag
MGQIFSRSASPIPRPPRGLAAHHWLNFLQAAYRLEPGPSSYDFHQLKKFLKIALETPARICPINYSLLAS
LLPKGYPGRVNEILHILIQTQAQIPSRPAPPPPSSPTHDPPDSDPQIPPPYVEPTAPQVLPVMHPHGAPP
NHRPWQMKDLQAIKQEVSQAAPGSPQFMQTIRLAVQQFDPTAKDLQDLLQYLCSSLVASLHHQQLDSLIS
EAETRGITGYNPLAGPLRVQANNPQQQGLRREYQQLWLAAFAALPGSAKDPSWASILQGLEEPYHAFVER
LNIALDNGLPEGTPKDPILRSLAYSNANKECQKLLQARGHTNSPLGDMLRACQTWTPKDKTKVLVVQPKK
PPPNQPCFRCGKAGHWSRDCTQPRPPPGPCPLCQDPTHWKRDCPRLKPTIPEPEPEEDALLLDLPADIPH
PKNFIGGEV
>Env
MGKFLATLILFFQFCPLIFGDYSPSCCTLTIGVSSYHSKPCNPAQPVCSWTLDLLALSADQALQPPCPNL
VSYSSYHATYSLYLFPHWTKKPNRNGGGYYSASYSDPCSLKCPYLGCQSWTCPYTGAVSSPYWKFQHDVN
FTQEVSRLNINLHFSKCGFPFSLLVDAPGYDPIWFLNTEPSQLPPTAPPLLPHSNLDHILEPSIPWKSKL
LTLVQLTLQSTNYTCIVCIDRASLSTWHVLYSPNVSVPSSSSTPLLYPSLALPAPHLTLPFNWTHCFDPQ
IQAIVSSPCHNSLILPPFSLSPVPTLGSRSRRAVPVAVWLVSALAMGAGVAGGITGSMSLASGKSLLHEV
DKDISQLTQAIVKNHKNLLKIAQYAAQNRRGLDLLFWEQGGLCKALQEQCRFPNITNSHVPILQERPPLE
239
NRVLTGWGLNWDLGLSQWAREALQTGITLVALLLLVILAGPCILRQLRHLPSRVRYPHYSLIKPESSL
>Pro
HPTPKKLHRGGGLTSPPTLQQVLPNQDPASILPVIPLDPARRPVIKAQVDTQTSHPKTIEALLDTGADMT
VLPIALFSSNTPLKNTSVLGAGGQTQDHFKLTSLPVLIRLPFRTTPIVLTSCLVDTKNNWAIIGRDALQQ
CQGVLYLPEAKRPPVILPIQAPAVLGLEHLPRPPEISQFPLNQNASRPCNTWSGRPWRQAISNPTPGQGI
TQYSQLKRPMEPGDSSTTCGPLTL
>Pol
GKKAACNLANTGASRPWARTPPKAPRNQPVPFKPERLQALQHLVRKALEAGHIEPYTGPGNNPVFPVKKA
NGTWRFIHDLRATNSLTIDLSSSSPGPPDLSSLPTTLAHLQTIDLRDAFFQIPLPKQFQPYFAFTVPQQC
NYGPGTRYAWKVLPQGFKNSPTLFEMQLAHILQPIRQAFPQCTILQYMDDILLASPSHEDLLLLSEATMA
SLISHGLPVSENKTQQTPGTIKFLGQIISPNHLTYDAVPTVPIRSRWALPELQALLGEIQWVSKGTPTLR
QPLHSLYCALQRHTDPRDQIYLNPSQVQSLVQLRQALSQNCRSRLVQTLPLLGAIMLTLTGTTTVVFQSK
EQWPLVWLHAPLPHTSQCPWGQLLASAVLLLDKYTLQSYGLLCQTIHHNISTQTFNQFIQTSDHPSVPIL
LHHSHRFKNLGAQTGELWNTFLKTAAPLAPVKALMPVFTLSPVIINTAPCLFSDGSTSRAAYILWDKQIL
SQRSFPLPPPHKSAQRAELLGLLHGLSSARSWRCLNIFLDSKYLYHYLRTLALGTFQGRSSQAPFQALLP
RLLSRKVVYLHHVRSHTNLPDPISRLNALTDALLITPVLQLSPAELHSFTHCGQTALTLQGATTTEASNI
LRSCHACRGGNPQHQMPRGHIRRGLLPNHIWQGDITHFKYKNTLYRLHVWVDTFSGAISATQKRKETSSE
AISSLLQAIAHLGKPSYINTDNGPAYISQDFLNMCTSLAIRHTTHVPYNPTSSGLVERSNGILKTLLYKY
FTDKPDLPMDNALSIALWTINHLNVLTNCHKTRWQLHHSPRLQPIPETRSLSNKQTHWYYFKLPGLNSRQ
WKGPQEALQEAAGAALIPVSASSAQWIPWRLLKRAACPRPVGGPADPKEKDLQHHG
>Rof
MPKTRRRPRRSQRKRPPTPWQLPPFSLQGLHLAFQLSSIAINPQLLHFFFPSTMLFRLLSPLSPLALTAL
LLFLLPPSDVSGLLLRPPPAPCLLLFLPFQILSGLLFLLFLPLFFSLPLLLSPSLPITMRFPARWRFLPW
RAPSQPAAAFLF
>P12
MLFRLLSPLSPLALTALLLFLLPPSDVSGLLLRPPPAPCLLLFLPFQILSGLLFLLFLPLFFSLPLLLSP
SLPITMRFPARWRFLPWRAPSQPAAAFLF
>Tof
240
MALCCFAFSAPCLHLRSRRSCSSCFLLATSAAFFSARLLRRAFSSSFLFKYSAVCFSSSFSRSFFRFLFS
SARRCRSRCVSPRGGAFSPGGPRRSRPRLSSSKDSKPSSTASSSSLSFNSSSKDNSPSTNSSTSRSSGHD
TGKHRNSPADTKLTMLIISPLPRVWTESSFRIPSLRVWRLCTRRLVPHLWGTMFGPPTSSRPTGHLSRAS
DHLGPHRWTRYRLSSTVPYPSTPLLPHPENL
>P13
MLIISPLPRVWTESSFRIPSLRVWRLCTRRLVPHLWGTMFGPPTSSRPTGHLSRASDHLGPHRWTRYRLS
STVPYPSTPLLPHPENL
>Rex
MPKTRRRPRRSQRKRPPTPWPTSQGLDRVFFSDTQSTCLETVYKATGAPSLGDYVRPAYIVTPYWPPVQS
IRSPGTPSMDALSAQLYSSLSLDSPPSPPREPLRPSRSLPRQSLIQPPTFHPPSSRPCANTPPSEMDTWN
PPLGSTSQPCLFQTPDSGPKTCTPSGEAPLSACTSTSFPPPSPGPSCPT
>P21
MDALSAQLYSSLSLDSPPSPPREPLRPSRSLPRQSLIQPPTFHPPSSRPCANTPPSEMDTWNPPLGSTSQ
PCLFQTPDSGPKTCTPSGEAPLSACTSTSFPPPSPGPSCPT
>Tax
MAHFPGFGQSLLFGYPVYVFGDCVQGDWCPISGGLCSARLHRHALLATCPEHQITWDPIDGRVIGSALQF
LIPRLPSFPTQRTSKTLKVLTPPITHTTPNIPPSFLQAMRKYSPFRNGYMEPTLGQHLPTLSFPDPGLRP
QNLYTLWGGSVVCMYLYQLSPPITWPLLPHVIFCHPGQLGAFLTNVPYKRIEELLYKISLTTGALIILPE
DCLPTTLFQPARAPVTLTAWQNGLLPFHSTLTTPGLIWTFTDGTPMISGPCPKDGQPSLVLQSSSFIFHK
FQTKAYHPSFLLSHGLIQYSSFHSLHLLFEEYTNIPISLLFNEKEADDNDHEPQISPGGLEPPSEKHFRE
TEV
>HBZ
MAASGLFRCLPVSCPEDLLVEELVDGLLSLEEELKDKEEEKAVLDGLLSLEEESRGRLRRGPPGEKAPPR
GETHRDRQRRAEEKRKRKKEREKEEEKQIAEYLKRKEEEKARRRRRAEKKAADVARRKQEEQERRERKWR
QGAEKAKQHSARKEKMQELGIDGYTRQLEGEVESLEAERRKLLQEKEDLMGEVNYWQGRLEAMWLQ
241
A.4 HLA class I associations with HCV infection out-
come
In Table A.1 we have recorded HLA alleles that have been associated with the outcome
of HCV infection. Other factors, such as differences in the ethnic and geographical
background of the infected individuals have also been shown to influence the outcome
of the infection [28]. Hence, whenever possible we have identified the country where the
study took place and the prevalent HCV genotypes in the given cohort. In addition, we
have taken into account the size of the cohort so that we can build a measure confidence
in the power of each analysis. However, it needs to be emphasised that conclusive
association studies regarding the influence of HLA on infectious disease require large
samples, proper ethnic-background stratification, accurate clinical information, and
use of models that consider other known genetic effects on the disease [16].
242
Allele Effect Country(G) Cohort size Method Mediation References
A*02P USA(NC)/- 105(-/49/56) OR CTLs [28]
D USA(C)/- 105(-/49/56) OR CTLs [28]
A*03
P Ireland/1b 243(-/95/148) OR CTLs [26]
P USA(NC)/- 105(-/49/56) OR CTLs [28]
D Korea/1b,2a 343(206/-/137) RR - [345]
A*34 LVL Taiwan/1b 160 WRST - [202]
A*19 D Saudi/- 268(122/-/146) - - [346]
A*2301 D USA/- 675(-/231/444) OR CTLs [24]
B*56LVL Taiwan/1b 160 WRST - [202]
D Japan/1b,2a,2b 285(172/-/113) OR CTLs [99]
B*4001 HVL Taiwan/1b 160 WRST - [202]
B*14 D Italy/- 606(489/-/117) RR - [347]
B*54 D Japan/1b,2a,2b 1046(916/-/130) OR CTLs [348]
B*55 D Japan/1b,2a,2b 285(172/-/113) OR CTLs [99]
B*70
B*51
B*52 P Japan/1b,2a,2b 285(172/-/113) OR CTLs [99]
B*61
Cw*03
B*35 D Korea/1b,2a 343(206/-/137) RR - [345]
B*46
A*1101 P USA/- 675(-/231/444) OR CTLs [24]
B*57(01,03)
P USA/- 675(-/231/444) OR CTLs [24]
P Ghana/2 104(-/35/37) OR CTLs [25]
P USA/1a,b 758(-/136/622) OR CTLs/NK [98]
Cw*04
D USA/- 675(-/231/444) OR CTLs [24]
D Ireland/1,3,5,2 225(-/86/139) OR CTLs/NK [27]
P Japan/1b,2a,2b 285(172/-/113) OR CTLs [99]
B*07 P Ireland/1b 243(-/95/148) OR CTLs [26]
B*27
Cw*01(02)
P Ireland/1b 243(-/95/148) OR CTLs [26]
P Japan/1b,2a,2b 285(172/-/113) OR CTLs [99]
P USA/- 675(-/231/444) OR CTLs [24]
P USA/1a,b 758(-/136/622) OR CTLs/NK [98]
B*18 D Ireland/1b 243(-/95/148) OR CTLs [26]
B*08D Ireland/1b 243(-/95/148) OR CTLs [26]
P Saudi/- 268(122/-/146) - - [346]
Cw*05 P USA(C/NC)/- 105(-/49/56) OR CTLs [28]
Appendix Table A.1: HLA class I associations with HCV outcome reported in the
literature. Protective (P) or Detrimental (D), Low (LVL) or High (HVL) Viral Load,
Odds Ratio (OR), Relative Risk (RR), Wilcoxon Rank Sum Test (WRST), Caucasians
(C) and Non-Caucasians (NC).
243
A.5 Confounding factors in HCV infection
Variable UK(161) USA(621) Pooled(782)
OR(p-value)
Race(Caucasians) - 0.70(0.24) 0.73(0.29)
Mode(IVDU) 0.11(0.009) 1.18(0.58) 0.13(0.02)
Mode(Transfusion) 0.07(0.009) - 0.08(0.01)
HBV - 3.85(1.7× 10−5) 3.79(1.9× 10−5)
HIV - 0.76(0.15) 0.76(0.15)
rs12979860 1.63(0.18) 3.09(3× 10−9) 2.76(2.5× 10−9)
KIR2DL3/2DL3+HLA-C1C1 2.64(0.03) 1.58(0.07) 1.76(0.009)
Appendix Table A.2: Impact of potential confounding variables on the outcome of
HCV infection.
A.6 The Explained Fraction depends on factor frequency
In order to show that the explained fraction depends on factor frequency, we con-
structed many 2x2 contingency tables keeping the frequency of the outcomes (column
marginals) fixed. We then computed the Odds Ratio, the Explained fraction and the
frequency of the causal factors and plotted the variation of Explained Fraction with
respect to frequency for the same factor effect i.e. for the same OR. In Appendix Figure
A.2, we readily observe that the higher the effect of a causal factor on disease outcome
(OR6=1) the higher the impact of the factor frequency on the explained fraction. We
also observe, that the maximum EF is reached in all cases for factor frequency 0.5.
This is expected because the EF depends on mutual information and we have the
maximum mutual information when the frequencies of the causal factors are equal i.e.
each factor has frequency 0.5.
244
0 5 10 20
0.2
0.6
OR ≈ 0.1
EF
Fre
qu
en
cy
0 5 10 20
0.2
0.6
OR ≈ 0.2
EF
Fre
qu
en
cy
0 5 10 20
0.2
0.6
OR ≈ 0.4
EF
Fre
qu
en
cy
0 5 10 20
0.0
0.4
0.8
OR ≈ 0.6
EF
Fre
qu
en
cy
0 5 10 20
0.2
0.6
1.0
OR ≈ 0.8
EF
Fre
qu
en
cy
0 5 10 20
0.0
0.4
0.8
OR ≈ 1
EF
Fre
qu
en
cy
0 5 10 20
0.2
0.6
OR ≈ 1.5
EF
Fre
qu
en
cy
0 5 10 20
0.0
0.4
0.8
OR ≈ 2
EF
Fre
qu
en
cy
0 5 10 20
0.2
0.6
OR ≈ 4
EF
Fre
qu
en
cy
0 5 10 20
0.2
0.6
OR ≈ 6
EF
Fre
qu
en
cy
0 5 10 20
0.2
0.6
OR ≈ 8
EF
Fre
qu
en
cy
0 5 10 20
0.2
0.6
OR ≈ 10
EF
Fre
qu
en
cy
Appendix Figure A.2: For the same Odds Ratio i.e. the same factor effect, the
Explained Fraction depends on the frequency of the factor.
245
B Appendix B
B.1 False Discovery Rate of cohort stratifications
Result FDR
B*57 status 6%
B*54 status <0.0001%
C*08 status <0.0001%
B*54+C*08 status <0.0001%
B*54 PVL HAM/TSP 15%
C*08 PVL HAM/TSP 4%
C*08 PVL ACs 15%
B*57 VP MHCS 5%
B*57 VL ALIVE 1%
HBZ binding 1%
Appendix Table B.1: False Discovery Rate of cohort stratifications. We examined
how many times we would see odds ratios equal to or more extreme than we observed
in the actual cohorts (FDR).
B.2 Canonical KIR-HLA binding not supported by link-
age effects
HTLV-1: other group C1 HLA alleles do not exhibit the same be-
haviour as C*08
We found that, in HTLV-1 infection, HLA-C*08 was associated with a protective effect:
reducing the risk of HAM/TSP and reducing proviral load in ACs. This effect was
246
enhanced in the presence of KIR2DL2 but reduced/absent in the absence of KIR2DL2.
We hypothesised that the effect of KIR2DL2 on the C*08 protective effect is not
mediated via a canonical, direct KIR-HLA interaction. To test this hypothesis we first
investigated whether the other group C1 alleles exhibited the same behaviour as C*08
in HTLV-1 infection. Virtually all HTLV-1 infected individuals possess at least one
C1 allele (430/432) therefore it was not possible to look at presence or absence of the
C1 ligand instead we looked at the impact of C1 homozygosity. Grouping all the C1
alleles we find no significant association between C1 homozygosity and disease status
either in the whole cohort or in the context of KIR2DL2. Similarly, C1/C1 was not
associated with decreased proviral load in either ACs or HAM/TSP patients; nor did
this become significant in the context of KIR2DL2.
In case the failure to find that the C1 grouping behaved the same as C*08 was
because we were forced to looked at homozygosity for C1 rather than presence or
absence as we had done for C*08 we additionally, investigated the individual C1 alleles
(Table B.2). This confirmed the hypothesis that the protective C*08 effect and its
enhancement by KIR2DL2 we had observed was not exhibited by other group C1
alleles.
HTLV-1: the B*54 effect cannot be attributed to linkage with C*01
B*54, a group Bw6 HLA allele, is not known to bind any KIR and so we tested whether
the observed B*54 effect was attributable to C*01 which is in linkage disequilibrium
with B*54 and does encode molecules which bind KIR2DL2. We found that B*54
rather than C*01 appeared to be the primary gene associated with HAM/TSP whose
detrimental effect was enhanced by KIR2DL2 (see HLA linkage).
247
(a) Disease status
HLA-C1 Allele OR KIR2DL2 OR p-value Allele Cohort size
(whole cohort) Genotype (stratified cohort) Carriers
C*01 0.768 + 0.307 0.028 43 102
(p=0.296) 1.05 0.877 129 300
C*03 1.683 + 1.331 0.561 51 102
(p=0.038) 1.973 0.029 138 300
C*07 0.824 + 0.335 0.07 26 102
(p=0.495) - 1.153 0.686 70 300
C*08 2.126 + 6.249 0.02 14 102
(p=0.032) - 1.504 0.364 44 300
C*12 1.122 + 1.888 0.342 15 102
(p=0.704) - 1 0.993 68 300
C*14 0.576 + 1.447 0.553 21 102
(p=0.073) - 0.429 0.024 66 300
(b) Proviral load
ACs
HLA-C1 Allele Difference in VL KIR2DL2 Difference in VL p-value Allele Cohort size
(whole cohort) Genotype (stratified cohort) Carriers
C*01 -0.164 + -0.358 0.233 15 48
(p=0.234) -0.11 0.516 55 132
C*03 0.113 + -0.047 0.866 25 48
(p=0.407) 0.171 0.313 66 132
C*07 0.222 + 0.408 0.23 10 48
(p=0.187) - 0.146 0.49 28 132
C*08 -0.33 + -0.662 0.053 10 48
(p=0.047) - -0.35 0.093 26 132
C*12 0.07 + -0.134 0.716 8 48
(p=0.675) - 0.154 0.442 30 132
C*14 0.008 + 0.164 0.614 11 48
(p=0.964) - -0.164 0.455 23 132
HAM/TSP
HLA-C1 Allele Change in VL KIR2DL2 Change in VL p- value Allele Cohort size
(whole cohort) Genotype (stratified cohort) Carriers
C*01 0.183 + 0.402 0.011 28 54
(p=0.033) 0.143 0.152 74 168
C*03 -0.169 + -0.047 0.775 26 54
(p=0.049) -0.22 0.028 72 168
C*07 -0.057 + -0.074 0.682 16 54
(p=0.554) - 0.015 0.891 42 168
C*08 -0.173 + -0.943 0.002 4 54
(p=0.208) - -0.087 0.578 24 168
C*12 -0.005 + -0.175 0.472 7 54
(p=0.965) - -0.004 0.973 38 168
C*14 0.085 + -0.019 0.923 10 54
(p=0.408) - 0.068 0.555 43 168
Appendix Table B.2: Impact of HLA-C1 alleles on HTLV-1 disease status and proviral load. No other
HLA-C1 alleles have the same effect with C*08 in the context of KIR2DL2.
248
HCV: the B*57-KIR2DL2 effect cannot be attributed to B*57 linkage
with HLA molecules that do bind KIR2DL2
HLA-B*57 is in linkage with 3 HLA class I alleles: A*01, C*18 and C*06. None
of these HLA molecules is expected to bind KIR2DL2; furthermore A*01, C*06 and
C*18 do not have a significant impact on HCV status either overall or in the context of
KIR2DL2. We therefore conclude that the B*57-KIR2DL2 effect cannot be attributed
to B*57 linkage with HLA molecules that do bind KIR2DL2.
HCV: the B*57-KIR2DL2 effect cannot be attributed to KIR2DL2
linkage with other KIRs that do bind B*57
HLA-B*57 does not bind KIR2DL2 however B*57 does bind KIR3DL1 and possibly
KIR3DS1 both of which are in weak linkage disequilibrium with KIR2DL2 (p=0.04
and p=0.1 respectively). We therefore examined whether the KIR2DL2 enhancement
of the B*57 protective effect could instead be explained by KIR3DL1 or 3DS1. Four
observations argue against this:
1. If B*57 binding of KIR3DL1 or 3DS1 was associated with enhanced immunity
then other HLA-B ligands of KIR3DS1 or KIR3DS1 with similar binding to
B*57, namely Bw4.80I, would be expected to show a similar pattern to B*57.
This was not observed (Table B.3).
2. Examining the enhancement of B*57 by KIR2DL2, KIR3DL1 and KIR3DS1 it
can be seen that the strongest enhancement is by KIR2DL2 (B.4).
3. If we exclude KIR2DL2+ individuals then neither KIR3DL1 nor KIR3DS1 en-
hance B*57 (in KIR2DL2-3DL1+ OR =0.83 p= 0.63. KIR2DL2-3DS1+ OR
=0.98 p= 0.97). However there are only 11 KIR2DL2-3DS1+B*57+ individuals
(35 KIR2DL2-3DL1+B*57+) so loss of significance in the latter case may be
attributable to power. Conversely, if we exclude KIR3DS1+ individuals then
249
KIR HLA ligand OR p-value N KIR+HLA+ Cohort size
3DL1 Bw4 0.94 0.7 485 782
3DL1 Bw4.80I 0.88 0.5 291 782
3DL1 Bw4/Bw4 1.19 0.5 131 782
3DL1 Bw4.80I/Bw4.80I 1.85 0.1 47 782
3DS1 Bw4 0.93 0.7 159 782
3DS1 Bw4.80I 0.8 0.4 91 782
3DS1 Bw4/Bw4 0.94 0.9 45 782
3DS1 Bw4.80I/Bw4.80I 0.61 0.4 12 782
Appendix Table B.3: Only the effect of B*57 and not of alleles with similar binding
is enhanced by KIR2DL2.
Cohort stratification OR for B*57 p-value N B*57+ N B*57-
KIR2DL2+ 0.40 0.007 49 359
KIR2DL2- 0.83 0.631 35 339
KIR3DL1+ 0.56 0.021 80 658
KIR3DL1- 2.5 0.6 4 40
KIR3DS1+ 0.43 0.05 30 214
KIR3DS1- 0.67 0.2 54 484
Appendix Table B.4: KIRs which are known to bind B*57 do not enhance the B*57
protective effect as much as KIR2DL2 does.
250
there is still a trend for KIR2DL2 to enhance B*57 (OR=0.47 p=0.092); there
are 30 KIR2DL2+KIR3DS1- individuals with B*57. There are only 4 individ-
uals who are KIR2DL2+ but KIR3DL1- so excluding KIR3DL1 individuals is
not possible.
4. In a model to predict HCV status in which KIR2DL2 with B*57 and KIR3DL1
with B*57 were both included as covariates (plus confounders) and then non
significant HLA:KIR were removed by backwards stepwise exclusion then only
KIR2DL2 with B*57 remained as a significant predictor. Similarly, in a model
to predict HCV status in which KIR2DL2 with B*57 and KIR3DS1 with B*57
were both included as covariates (plus confounders) then, following backwards
stepwise exclusion, only KIR2DL2 with B*57 remained as a significant predic-
tor.
B.3 A*02 binds peptides strongly
In order to show that A*02 molecules bind peptides significantly more strongly than
other HLA molecules we used two approaches. First, we extracted the list of pos-
itive MHC Binding Assays from the Immune Epitope Database (IEDB) [349] which
contained in total 78202 peptides that bind HLA molecules. We then compared the ex-
perimental affinity measurements of all the A*02xx molecules with the corresponding
values of other frequent HLA molecules (only HLA-A and B molecules were consid-
ered). The frequencies of the HLA molecules were calculated based on available data
for UK and USA populations at the Allele Frequency Net Database [204]. Secondly, we
used the epitope prediction software Metaserver [198] to identify potential epitopes for
both the HCV and HTLV-1 proteomes for 36 and 21 HLA molecules respectively (we
only obtained predicted epitopes for the HLA alleles present in the cohorts included in
the study). Then we compared, as in the previous approach, the predicted affinity of
the A*02xx:peptide complexes which were less than 500 nM IC50 with the predicted
251
values for the rest of the pMHC complexes included in the analysis. Both the above
approaches suggest that A*02 binds peptides significantly more strongly compared to
other HLA molecules. In the first approach (experimentally measured affinities), A*02
bound epitopes significantly more strongly than each of A*01, A*30, B*07, B*35,
B*44 and B*45 (p<0.00001 in each case, Wilcoxon Rank sum, two-tailed see Ap-
pendix Figure B.1). In the second approach (theoretically predicted affinities), A*02
bound epitopes significantly more strongly than the other A alleles considered (HTLV-
1: p=0.015, HCV: p<0.00001), B alleles (HTLV-1: p<0.00001, HCV p<0.00001) and
A and B alleles combined (HTLV-1: p<0.00001, HCV p<0.00001), Appendix Figures
2(a) and 2(b).
A*02 A*01 A*30 B*07 B*35 B*44 B*45
−4−2
02
46
8
IEDB
Frequent HLA molecules
log1
0(E
C50
nM
)
p<0.00001
Appendix Figure B.1: A*02 binds epitopes significantly more strongly than other
HLA molecules which are frequent in UK and USA populations. The binding measure
shown here is the affinity (nM IC50) values obtained from the IEDB database for 16958
peptides.
252
A*02 A−A*02 B A&B−A*02
01
23
45
Predicted HTLV−1 Epitopes
HLA molecules
log1
0(E
C50
nM
)− P
redi
cted p<0.05
p<0.00001
(a) HTLV-1 predicted epitopes
A*02 A−A*02 B A&B−A*02
01
23
4
Predicted HCV Epitopes
HLA molecules
log1
0(E
C50
nM
)−P
redi
cted
p<0.00001
(b) HCV predicted epitopes
Appendix Figure B.2: A*02 binds peptides significantly more strongly than other
HLA molecules based on predicted epitopes from the HTLV-1 and HCV proteomes.
The binding measure shown here is the affinity (nM IC50) values obtained from
Metaserver [198] for 2446 and 3611 peptides, respectively.
253
C Appendix C
C.1 Alignment to ambiguous reference sequence
In Appendix Figure C.1, we show a very small part of a multiple alignment where some
bases of the reads have aligned to ambiguous bases of the reference sequence based on
the IUPAC code. An 85% ambiguous consensus sequence is used. This means that
we first perform a multiple alignment of the 78 HCV genotype 1a strains available on
GenBank and then use both the most common nucleotide and the variants that exceed
>15% frequency at each position in order to define the consensus of that position.
Appendix Figure C.1: Sequence alignment using an 85% ambiguous reference strain.
The alignmemt mismatch parameter is set to 5% per read.
1http://www.hiv.lanl.gov/content/sequence/CONSENSUS/consensus.html
254
3000 4000 5000 6000 7000 8000 9000
010
000
2000
030
000
4000
0
Position
Cove
rage
(a) Full amplicon coverage
3000 4000 5000 6000 7000 8000 9000
0e
+0
02
e+
04
4e
+0
46
e+
04
8e
+0
41
e+
05
Position
Cov
era
ge
3000 4000 5000 6000 7000 8000 9000
05
00
00
10
00
00
15
00
00
Position
Cov
era
ge
(b) Partial amplicon coverage
Appendix Figure C.2: Coverage of HCV amplicons. For some patient-timepoint
samples all the amplicons were amplified (a) while for some others only some of them
were amplified (b). For the alignment shown here the MOSAIK software is used; an
85% ambiguous consensus sequence is provided (i.e. positions that have >15% different
nucleotides are defined by the IUPAC code) based on the 78 HCV genotype 1a strains
available on GenBank. The ambiguous reference sequence is constructed using the
Consensus Maker v2.0.0 software1. The mismatch parameter is set to 5% per read.
255
A1 A2 A3
0e+0
04e
+04
8e+0
4
Amplicons
Cov
erag
e pe
r bp
Appendix Figure C.3: Coverage per HCV amplicon. For the alignment the MOSAIK
software is used; an 85% ambiguous consensus sequence is provided (i.e. positions that
have >15% different nucleotides are defined by the IUPAC code) based on the 78
HCV genotype 1a strains available on GenBank. The ambiguous reference sequence is
constructed using the Consensus Maker v2.0.0 software. The mismatch parameter is
set to 5% per read.
256
D Appendix D
D.1 AICc differences of the fitted models
The Appendix Figure D.1 shows the AICc of the model minus the AICc of the best
fitting model for each animal. As a rule of thumb a difference of < 2 suggests substan-
tial evidence for both models, values between 3 and 7 indicate that the model with the
worse fit has considerably less support, whereas a difference > 10 indicates that the
model with the worse fit is very unlikely.
Appendix Figure D.1: The fit of each model for all the animals. A large difference
represents a poor fit. The best fit model is shaded, comparable models are shown in
bold. For explanation of the models see Table 5.1. The last three animals in the list
were euthanised after early chronic infection, receiving either ART-treatment alone or
the combination of CD8+ T cell depletion and ART. These animals were excluded from
the analysis as, in each case, they had received only half the treatment (i.e. just ART
or just ART/depletion) so there was insufficient data to constrain the fits. Including
these 3 animals in the analysis did not change the result (p=0.028, two tailed paired
Mann-Whitney as before). Note: This table was produced by M. Elemans.
257
E Appendix E
E.1 3D Cellular Automaton - Cell motility rules
The following rules which govern the motility of the cell populations on the grid are
the same with the ones used in [267] and where kindly provided by F. Graw. The
implementation in C++ was performed by N.-K. Seich al Basatena (thesis author).
We distinguish between two types of movement:
I Movement into free space (FS)
II Neighbour swapping between neighboured cells
The cells are marked as swapped/moved in order to avoid moving them twice during
one timestep. In addition, a check is performed in order to establish whether a cell is
able to move or not (e.g. it is bound to a conjugate).
Move into free space:
1. Find an FS in the lattice
2. Determine the neighbouring cells of the FS which are able to move into this node
according to their moving direction
3. If there is more than one cell in (2), choose one of these cells at random, change
the position of the FS and the chosen cell, and mark both cells as moved in order
to avoid to move them twice during one step
4. If there is no cell found in (2), change the preferred moving direction of each
neighbouring cell valid for the next timestep
5. Repeat step (1)-(4) until all FS were updated
Neighbour swapping:
The algorithm is formulated for a CTL2. Movement of the other cell types occurs
2Here we use the terms CTL and CD8+ T cell, as described in Chapter 5, interchangeably.
258
similarly.
1. Find a CTL in the lattice
2. Determine the type of the neighbouring cell to which the moving direction of
the CTL is pointing and proceed obeying the following rules if the cell belongs
to:
(a) splenocytes: swap the position with the CTL if the splenocyte was not
already moved during this timestep, otherwise determine a new moving di-
rection for the CTL valid for the next timestep
(b) CTL: check if the moving direction of this second CTL points to the first
CTL
• if yes then swap both cells
• if no then determine new moving directions for both cells valid for the
next timestep
(c) infected cells: analogous to (b)
(d) uninfected cells: analogous to (b)
(e) RN: determine new moving direction for the CTL valid for the next timestep
(f) FS: swap cells
3. Repeat step (1)-(2) until all CTLs have been updated
The moving direction γ ∈ {1, ..., 26} points to one of the 26 neighbours of a cell. All
cells are attributed with a position in 3D-space. The moving direction is updated as
follows:
1. Translate the moving direction, γ, to cartesian coordinates (x, y, z)T with x, y, z ∈
{−1, 0, 1}. The position of the cell under consideration is (0, 0, 0)T .
259
2. Choose one coordinate ǫ ∈ {x, y, z} at random and change its values as follows:
ǫnew =
ǫ = 1 ⇒
ǫ− 1, p = 0.8
ǫ− 2, p = 0.2
ǫ = 0 ⇒
ǫ+ 1, p = 0.5
ǫ− 1, p = 0.5
ǫ = −1 ⇒
ǫ+ 1, p = 0.8
ǫ+ 2, p = 0.2
3. Replace ǫ by ǫnew and translate the cartesian coordinates to the moving direction
γ.
Based on this algorithm the cells prefer small turning angles. In general, cell movement
involves restructuring of the actin-filament network in the cytoskeleton so cells are
expected to prefer small turning angles [350].
E.2 Model quantities: look-up tables
260
Quantity Value Reference Notes
Initial cell populations
Lattice edge 50 n/a
125000 cells,
0.05− 0.5%
of the splenic
white pulp [351]
Timestep 30 secn/a Integrates micro
and macro scale
Reticular network 20% -
Free space 1%- Spleen is a
densely packed organ
Epitope-specific0.5% [168,352]
Estimated based on
CD8+ T cells ELISPOT assays
CD4+ T cells 15% - Mouse spleen
MΦs & DCs 1% -
Motility Parameters
T cell speed 10− 15 µm/min [268, 269,271,272]In spleen, estimates
are lower [283,284]
T cell motility50− 100 µm2/min [285, 353,354]
coefficient
Appendix Table E.1: CA model initialisation values and motility parameters.
261
Quantity Value Reference Notes
Cell properties
R0 6-8 [274,291]
Total death25− 30% [3,248,297]
attributed to
CD8+ T cells
Lifespan
13− 76 d [355]of uninfected
CD4+ T cells
Lifespan
2 d [295]
Includes 1 d viral
of infected eclipse phase
CD4+ T cells
Eclipse phase
1 d [293, 294]of viral
production
Duration of Half-life:[317]
Case study: RANTES
cytokine effect 6-9 hrs
CD8+ T cell
≈ 10 mins [301]
Ag+ recognition
cytokine secretion with no
duration lytic response
CD8+ T cell7± 2 min−1 [301]
Normal
scanning time distribution
CD8+ T cell
30 mins [251,301,305]lytic killing
duration
Appendix Table E.2: CA model parameters.
262
F Appendix F
F.1 No significant effect of the grid size on the estimated
escape rates
0.001 0.002 0.003
0.00
0.02
0.04
0.06
0.08
0.10
0.12
Grid size = 503
Probability of target cell recognition
Est
imat
ed e
scap
e ra
te
0.001 0.002 0.003
0.00
0.02
0.04
0.06
0.08
0.10
0.12
Grid size = 603
Probability of target cell recognition
Est
imat
ed e
scap
e ra
te
Appendix Figure F.1: The effect of the CA grid size on estimated escape rates under
lytic CD8+ T cell suppression. When comparing the estimated escape rates (or killing
rates - data not shown) no significant difference (p > 0.05) is observed between grid
sizes for any of the CD8+ T cell suppression levels.
263
F.2 Killing rate increases with effector cells size
We vary the size of the epitope-specific CD8+ T cell population in the range of 0.5%-
1.5% of the total splenocyte population (in line with biological ranges for PBMCs) in
order to investigate its effect on killing rate. We find that the killing rate increases as
the effector size increases (Appendix Figure F.2-left). We further explore whether this
increase is proportional and find that for all the different probabilities of successful
scan (pss) the data are consistent with a mass-action behaviour. In other words, the
relative increase in effector cells entails a relative increase in effector cell killing rate
which is not statistically significantly different (p> 0.05; Appendix Figure F.2- right).
However, some of the data are noisy so we cannot draw firmer conclusions at this point.
264
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
Size of epitope−specific CD8+ T cell population (%)
Kill
ing
rate
per
day
0.5 0.75 1 1.5
−−−
pss=0.001pss=0.002pss=0.003
0 1 2 3 4 5 6 7
01
23
45
67
Effector cell ratio
Kill
ing
rate
rat
io
Appendix Figure F.2: We vary the magnitude of the epitope-specific CD8+ T cell
population and we observe that a larger number of effector cells leads to a higher
CD8+ T cell killing rate (left). The relative increase in effector cells is not significantly
different to the relative increase in killing rate, i.e. the slopes of the lines that correlate
the relative increases of the two quantities (right) are not significantly different to the
unit, p> 0.05). Abbreviations: pss=Probability of Successful Scan by CD8+ T cells.
265
F.3 The percentage of simulations for each model that do
not result in viral escape
We also explore the number of simulations that do not result in viral escape as a
potential surrogate measure of CD8+ T cell pressure on the wild-type infected cells.
Models
pss Lytic NLi-P1 NLi-D1 NLi-D2 NLp-TO NLp-P1 NLp-D1 NLp-D2
0.001 33% 57% 33% 76% 50% 43% 35% 57%
0.002 24% 71% 57% 62% 36% 43% 21% 36%
0.003 0% 52% 52% 62% 36% 36% 58% 36%
Appendix Table F.1: The percentage of simulations for each model that do not
result in viral escape. Abbreviations: pss=Probability of Successful Scan by CD8+
T cells, NLi: Non-lytic model - blocking infection of uninfected CD4+ T cells, NLp:
Non-lytic model - blocking viral production from infected CD4+ T cells, TO=Target
Only, P1=Polarised secretion (r=1), D1=Diffusive secretion (r=1) and D2=Diffusive
secretion (r=2).
266
F.4 Quantifying the immune control of a non-lytic re-
sponse that blocks viral production
0.001 0.002 0.003
050
100
150
200
NLp − Polarised (r=1)
Num
ber
of in
fect
ed C
D4+
T c
ells
`bl
ocke
d’ n
on−
lytic
ally
0.001 0.002 0.003
050
100
150
200
250
NLp − Diffusive (r=1)
Num
ber
of in
fect
ed C
D4+
T c
ells
`bl
ocke
d’ n
on−
lytic
ally
0.001 0.002 0.003
010
020
030
040
0
NLp − Diffusive (r=2)
Num
ber
of in
fect
ed C
D4+
T c
ells
`bl
ocke
d’ n
on−
lytic
ally
Probability of successful scan
Appendix Figure F.3: Number of infected CD4+ T cells ‘blocked’ from viral pro-
duction under a non-lytic CD8+ T cell response manifested in a polarised or diffusive
secretion pattern. Abbreviations: NLp: Non-lytic model - blocking viral production
from infected CD4+ T cells, P1=Polarised secretion (r=1), D1=Diffusive secretion
(r=1) and D2=Diffusive secretion (r=2).
267
Lytic
NLp
−TO
NLp
−P
1
NLp
−D
1
NLp
−D
2
0
50
100
150
200
250
pss=0.001
Num
ber
of in
fect
ions
pre
vent
ed b
y th
e ep
itope
−sp
ecifi
c ef
fect
ors
Lytic killing~1%
Type of epitope−specific CD8+ T cell control
Lytic
NLp
−TO
NLp
−P
1
NLp
−D
1
NLp
−D
2
0
50
100
150
200
250
pss=0.002
Num
ber
of in
fect
ions
pre
vent
ed b
y th
e ep
itope
−sp
ecifi
c ef
fect
ors
Lytic killing~3%
Type of epitope−specific CD8+ T cell control
Lytic
NLp
−TO
NLp
−P
1
NLp
−D
1
NLp
−D
20
50
100
150
200
250
pss=0.003
Num
ber
of in
fect
ions
pre
vent
ed b
y th
e ep
itope
−sp
ecifi
c ef
fect
ors
Lytic killing~5%
Type of epitope−specific CD8+ T cell control
Appendix Figure F.4: New infections prevented under a non-lytic CD8+ T cell
response that blocks infection. We show the number of infections prevented by the
epitope-specific CD8+ T cell clones for 40-50 dpi, just before the variant infected cell
population is introduced in the simulations. Abbreviations: NLp: Non-lytic model -
blocking viral production from infected CD4+ T cells, P1=Polarised secretion (r=1),
D1=Diffusive secretion (r=1) and D2=Diffusive secretion (r=2).
268
Lytic
NLp
−TO
NLp
−P
1
NLp
−D
1
NLp
−D
2
0.000
0.005
0.010
0.015
0.020
0.025
0.030
pss=0.001
% P
rodu
ctiv
ely
wild
−ty
pe in
fect
ed c
ells
afte
r se
t−po
int
Lytic killing~1%
Lytic
NLp
−TO
NLp
−P
1
NLp
−D
1
NLp
−D
2
0.000
0.005
0.010
0.015
0.020
0.025
0.030
pss=0.002
% P
rodu
ctiv
ely
wild
−ty
pe in
fect
ed c
ells
afte
r se
t−po
int
Lytic killing~3%
Lytic
NLp
−TO
NLp
−P
1
NLp
−D
1
NLp
−D
2
0.000
0.005
0.010
0.015
0.020
0.025
0.030
pss=0.003
% P
rodu
ctiv
ely
wild
−ty
pe in
fect
ed c
ells
afte
r se
t−po
int
Lytic killing~5%
Type of epitope−specific CD8+ T cell control
Appendix Figure F.5: Set-point of productively infected cells. We show the per-
centage of productively infected wild-type cells for 40-50 dpi, just before the variant
infected cell population is introduced in the simulations and after the steady-state has
been attained. Here, we present the results for the lytic control and the non-lytic
control that blocks viral production. Abbreviations: NLp: Non-lytic model - block-
ing viral production from infected CD4+ T cells, TO= Target Only, P1=Polarised
secretion (r=1), D1=Diffusive secretion (r=1) and D2=Diffusive secretion (r=2).
269
F.5 Equivalence of non-lytic models in chronic infection
It can readily be shown that under a quasi-equilibrium between infected cells and
free virus which holds during the chronic phase of infection, a non-lytic model where
the CD8+ T cells reduce viral production (Equation 5.6) and a non-lytic model
where CD8+ T cells reduce infection of new targets (Equation 5.8) produce the same
dynamics for the population of productively infected cells (T ∗).
The quasi-steady assumption for the non-lytic model where infection of new cells
is hindered results in:
V =p
cT ∗ (1)
while the quasi-steady state assumption for the non-lytic model where viral pro-
duction is blocked results in:
V =
(
1
1 + ηE
)
p
cT ∗ (2)
Considering a constant population of uninfected target cells (S) and substituting
Equations 1 and 2 in the ones governing the behaviour of productively infected cells
(T ∗), Equations 5.6 and 5.8 respectively, they change into:
T ∗ =
((
1
1 + ηE
)
βpS
c− δI
)
T ∗ (3)
where β is the infection rate, p is the production rate of free virions, c is the
clearance rate of free virions and δI is the death rate of productively infected cells (all
d−1). The fraction of CD8+ T cells, E, is given by the empirical function calculated
by a linear interpolation between time points.
270
Bibliography
[1] Nafisa-Katrin Seich Al Basatena, Aidan Macnamara, Alison M Vine, et al. KIR2DL2
Enhances Protective and Detrimental HLA Class I-Mediated Immunity in Chronic Viral
Infection. PLoS Pathogens, 7(10):e1002270, October 2011. PMID: 22022261.
[2] Geraldine M O’Connor, Nafisa-Katrin Seich Al Basatena, Viviana Olavarria, et al. In
Contrast to HIV, KIR3DS1 Does Not Influence Outcome In HTLV-1 Retroviral Infection.
Human immunology, May 2012. PMID: 22609443.
[3] Marjet Elemans, Nafisa-Katrin Seich Al Basatena, Nichole R Klatt, et al. Why Don’t
CD8+ T Cells Reduce the Lifespan of SIV-Infected Cells In Vivo? PLoS Computational
Biology, 7(9):e1002200, September 2011. PMID: 21990968.
[4] A Sette, A Vitiello, B Reherman, et al. The relationship between class I binding affin-
ity and immunogenicity of potential cytotoxic T cell epitopes. Journal of Immunology
(Baltimore, Md.: 1950), 153(12):5586–5592, December 1994. PMID: 7527444.
[5] Alasdair Leslie, David A Price, Pamela Mkhize, et al. Differential selection pressure
exerted on HIV by CTL targeting identical epitopes but restricted by distinct HLA
alleles from the same HLA supertype. Journal of immunology (Baltimore, Md.: 1950),
177(7):4699–4708, October 2006. PMID: 16982909.
[6] Philip J R Goulder and David I Watkins. Impact of MHC class I diversity on immune
control of immunodeficiency virus replication. Nature Reviews. Immunology, 8(8):619–
630, August 2008. PMID: 18617886.
[7] J M Claverie, P Kourilsky, P Langlade-Demoyen, et al. T-immunogenic peptides are
constituted of rare sequence patterns. Use in the identification of T epitopes in the human
immunodeficiency virus gag protein. European journal of immunology, 18(10):1547–1553,
October 1988. PMID: 2461306.
[8] A Sette, S Buus, E Appella, et al. Prediction of major histocompatibility complex
binding regions of protein antigens by sequence pattern analysis. Proceedings of the
National Academy of Sciences of the United States of America, 86(9):3296–3300, May
1989. PMID: 2717617 PMCID: PMC287118.
271
[9] S.l. Lauemller, A. Holm, J. Hilden, et al. Quantitative predictions of peptide binding
to MHC class I molecules using specificity matrices and anchor-stratified calibrations.
Tissue Antigens, 57(5):405414, 2001.
[10] Hiroshi Mamitsuka. A Learning Method of Hidden Markov Models for Sequence Dis-
crimination. Journal of Computational Biology, 3(3):361–373, January 1996.
[11] Claus Lundegaard, Ole Lund, and Morten Nielsen. Prediction of epitopes using neural
network based methods. Journal of Immunological Methods, October 2010. PMID:
21047511.
[12] Laurent Jacob and Jean-Philippe Vert. Efficient peptide-MHC-I binding prediction for
alleles with few known binders. Bioinformatics, page btm611, December 2007.
[13] Claus Lundegaard, Ole Lund, Sren Buus, and Morten Nielsen. Major histocompati-
bility complex class I binding predictions as a tool in epitope discovery. Immunology,
130(3):309–318, July 2010. PMID: 20518827.
[14] Mette Voldby Larsen, Claus Lundegaard, Kasper Lamberth, et al. An integrative ap-
proach to CTL epitope prediction: a combined algorithm integrating MHC class I bind-
ing, TAP transport efficiency, and proteasomal cleavage predictions. European Journal
of Immunology, 35(8):2295–2303, August 2005. PMID: 15997466.
[15] Thomas Stranzl, Mette Voldby Larsen, Claus Lundegaard, and Morten Nielsen. NetCTL-
pan: pan-specific MHC class I pathway epitope predictions. Immunogenetics, 62(6):357–
368, June 2010. PMID: 20379710.
[16] Mary Carrington and Stephen J O’Brien. The influence of HLA genotype on AIDS.
Annual Review of Medicine, 54:535–551, 2003. PMID: 12525683.
[17] Rob Boer, Jos Borghans, Michiel Boven, Can Kemir, and Franz Weissing. Heterozygote
advantage fails to explain the high degree of polymorphism of the MHC. Immunogenetics,
55(11):725–731, February 2004.
[18] Jos A M Borghans, Joost B Beltman, and Rob J De Boer. MHC polymorphism under
host-pathogen coevolution. Immunogenetics, 55(11):732–739, February 2004. PMID:
14722687.
272
[19] Jason L Kubinak, James S Ruff, Cornelius Whitney Hyzer, Patricia R Slev, and Wayne K
Potts. Experimental viral evolution to specific host MHC genotypes reveals fitness and
virulence trade-offs in alternative MHC types. Proceedings of the National Academy
of Sciences of the United States of America, 109(9):3422–3427, February 2012. PMID:
22323587.
[20] S. G. E. Marsh, E. D. Albert, W. F. Bodmer, et al. Nomenclature for factors of the HLA
system, 2010. Tissue Antigens, 75(4):291–455, 2010.
[21] S G E Marsh, E D Albert, W F Bodmer, et al. An update to HLA nomenclature, 2010.
Bone Marrow Transplantation, 45(5):846–848, May 2010. PMID: 20348972.
[22] Mary Carrington, George W. Nelson, Maureen P. Martin, et al. HLA and HIV-1: Het-
erozygote Advantage and B*35-Cw*04 Disadvantage. Science, 283(5408):1748–1752,
March 1999.
[23] Katie J. M. Jeffery, Koichiro Usuku, Sarah E. Hall, et al. HLA alleles determine hu-
man T-lymphotropic virus-I (HTLV-I) proviral load and the risk of HTLV-I-associated
myelopathy. Proceedings of the National Academy of Sciences of the United States of
America, 96(7):3848–3853, March 1999.
[24] Chloe L. Thio, Xiaojiang Gao, James J. Goedert, et al. HLA-Cw*04 and Hepatitis C
Virus Persistence. J. Virol., 76(10):4792–4797, May 2002.
[25] Wing Chia-Ming Chuang, Francis Sarkodie, Colin J Brown, et al. Protective effect of
HLA-B57 on HCV genotype 2 infection in a West African population. Journal of Medical
Virology, 79(6):724–733, June 2007. PMID: 17546694.
[26] Susan M McKiernan, Richard Hagan, Michael Curry, et al. Distinct MHC class I and II
alleles are associated with hepatitis C viral clearance, originating from a single source.
Hepatology (Baltimore, Md.), 40(1):108–114, July 2004. PMID: 15239092.
[27] Liam J. Fanning, Elizabeth Kenny-Walsh, and Fergus Shanahan. Persistence of hepatitis
C virus in a white population: Associations with human leukocyte antigen class 1. Human
Immunology, 65(7):745–751, July 2004.
[28] Jane Wang, Xin Zheng, Xiaogang Ke, et al. Ethnic and geographical differences in HLA
associations with the outcome of hepatitis C virus infection. Virology Journal, 6(1):46,
2009.
273
[29] A V Hill, C E Allsopp, D Kwiatkowski, et al. Common west African HLA antigens
are associated with protection from severe malaria. Nature, 352(6336):595–600, August
1991. PMID: 1865923.
[30] Rashmi Singh, Rashmi Kaul, Anil Kaul, and Khalid Khan. A comparative review of HLA
associations with hepatitis B and C viral infections across global populations. World
Journal of Gastroenterology: WJG, 13(12):1770–1787, March 2007. PMID: 17465466.
[31] Thi Phuong Lan Nguyen, Mihoko Kikuchi, Thi Que Huong Vu, et al. Protective and
enhancing HLA alleles, HLA-DRB1*0901 and HLA-A*24, for severe forms of dengue
virus infection, dengue hemorrhagic fever and dengue shock syndrome. PLoS Neglected
Tropical Diseases, 2(10):e304, 2008. PMID: 18827882.
[32] S C L Gough and M J Simmonds. The HLA Region and Autoimmune Disease: Asso-
ciations and Mechanisms of Action. Current Genomics, 8(7):453–465, November 2007.
PMID: 19412418.
[33] William R. Heath and Francis R. Carbone. CROSS-PRESENTATION IN VIRAL
IMMUNITY AND SELF-TOLERANCE. Nature Reviews Immunology, 1(2):126–134,
November 2001.
[34] Zabrina L Brumme and P Richard Harrigan. The impact of human genetic variation
on HIV disease in the era of HAART. AIDS Reviews, 8(2):78–87, June 2006. PMID:
16848275.
[35] David G. Bowen and Christopher M. Walker. Mutational escape from CD8+ T cell
immunity: HCV evolution, from chimpanzees to man. J. Exp. Med., 201(11):1709–1714,
June 2005.
[36] Philip J R Goulder and David I Watkins. HIV and SIV CTL escape: implications
for vaccine design. Nature Reviews. Immunology, 4(8):630–640, August 2004. PMID:
15286729.
[37] C M Walker, D J Moody, D P Stites, and J A Levy. CD8+ lymphocytes can con-
trol HIV infection in vitro by suppressing virus replication. Science (New York, N.Y.),
234(4783):1563–1566, December 1986. PMID: 2431484.
274
[38] M Kannagi, L V Chalifoux, C I Lord, and N L Letvin. Suppression of simian immun-
odeficiency virus replication in vitro by CD8+ lymphocytes. Journal of Immunology
(Baltimore, Md.: 1950), 140(7):2237–2242, April 1988. PMID: 2965185.
[39] R A Koup, J T Safrit, Y Cao, et al. Temporal association of cellular immune responses
with the initial control of viremia in primary human immunodeficiency virus type 1
syndrome. Journal of Virology, 68(7):4650–4655, July 1994. PMID: 8207839.
[40] Naglaa H. Shoukry, Arash Grakoui, Michael Houghton, et al. Memory CD8+ T Cells
Are Required for Protection from Persistent Hepatitis C Virus Infection. J. Exp. Med.,
197(12):1645–1655, June 2003.
[41] X Jin, D E Bauer, S E Tuttleton, et al. Dramatic rise in plasma viremia after CD8(+)
T cell depletion in simian immunodeficiency virus-infected macaques. The Journal of
Experimental Medicine, 189(6):991–998, March 1999. PMID: 10075982.
[42] Nichole R. Klatt, Emi Shudo, Alex M. Ortiz, et al. CD8+ Lymphocytes Control Viral
Replication in SIVmac239-Infected Rhesus Macaques without Decreasing the Lifespan
of Productively Infected Cells. PLoS Pathog, 6(1):e1000747, January 2010.
[43] Robert Thimme, Stefan Wieland, Carola Steiger, et al. CD8+ T Cells Mediate Viral
Clearance and Disease Pathogenesis During Acute Hepatitis B Virus Infection. Journal
of Virology, 77(1):68–76, January 2003.
[44] Lauren E Yauch, Raphal M Zellweger, Maya F Kotturi, et al. A protective role for
dengue virus-specific CD8+ T cells. Journal of Immunology (Baltimore, Md.: 1950),
182(8):4865–4873, April 2009. PMID: 19342665.
[45] S J Brodie, D A Lewinsohn, B K Patterson, et al. In vivo migration and function of
transferred HIV-1-specific cytotoxic T cells. Nature Medicine, 5(1):34–41, January 1999.
PMID: 9883837.
[46] Paul Bolitho, Ilia Voskoboinik, Joseph A Trapani, and Mark J Smyth. Apoptosis in-
duced by the lymphocyte effector molecule perforin. Current Opinion in Immunology,
19(3):339–347, June 2007. PMID: 17442557.
[47] J. A. Trapani. Granzymes, cytotoxic granules and cell death: the early work of Dr. Jurg
Tschopp. Cell Death & Differentiation, 19(1):21–27, November 2011.
275
[48] E Rouvier, M F Luciani, and P Golstein. Fas involvement in Ca(2+)-independent T cell-
mediated cytotoxicity. The Journal of experimental medicine, 177(1):195–200, January
1993. PMID: 7678113.
[49] Michele Barry and R Chris Bleackley. Cytotoxic T lymphocytes: all roads lead to death.
Nature reviews. Immunology, 2(6):401–409, June 2002. PMID: 12093006.
[50] Anthony L DeVico and Robert C Gallo. Control of HIV-1 infection by soluble factors of
the immune response. Nature Reviews. Microbiology, 2(5):401–413, May 2004. PMID:
15100693.
[51] Y Shirazi and P M Pitha. Interferon alpha-mediated inhibition of human immunodefi-
ciency virus type 1 provirus synthesis in T-cells. Virology, 193(1):303–312, March 1993.
PMID: 8438572.
[52] L G Guidotti, S Guilhot, and F V Chisari. Interleukin-2 and alpha/beta interferon down-
regulate hepatitis B virus gene expression in vivo by tumor necrosis factor-dependent
and -independent pathways. Journal of virology, 68(3):1265–1270, March 1994. PMID:
8107192.
[53] P N Gilles, G Fey, and F V Chisari. Tumor necrosis factor alpha negatively regulates
hepatitis B virus gene expression in transgenic mice. Journal of virology, 66(6):3955–
3960, June 1992. PMID: 1583737.
[54] Michael G. Katze, Yupeng He, and Michael Gale. Viruses and interferon: a fight for
supremacy. Nature Reviews Immunology, 2(9):675–687, September 2002.
[55] Nunen, Andeltje B, Oscar Pontesilli, et al. Suppression of hepatitis B virus replication
mediated by hepatitis Ainduced cytokine production. Liver, 21(1):45–49, February 2001.
[56] Jay A Levy. The search for the CD8+ cell anti-HIV factor (CAF). Trends in Immunology,
24(12):628–632, December 2003. PMID: 14644135.
[57] R T Bailer, A Holloway, J Sun, et al. IL-13 and IFN-gamma secretion by activated T cells
in HIV-1 infection associated with viral suppression and a lack of disease progression.
Journal of immunology (Baltimore, Md.: 1950), 162(12):7534–7542, June 1999. PMID:
10358209.
276
[58] F Cocchi, A L DeVico, A Garzino-Demo, et al. Identification of RANTES, MIP-1 alpha,
and MIP-1 beta as the major HIV-suppressive factors produced by CD8+ T cells. Science
(New York, N.Y.), 270(5243):1811–1815, December 1995. PMID: 8525373.
[59] A Garzino-Demo, A L DeVico, K E Conant, and R C Gallo. The role of chemokines
in human immunodeficiency virus infection. Immunological reviews, 177:79–87, October
2000. PMID: 11138787.
[60] Ralf Geiben-Lynn. Anti-human immunodeficiency virus noncytolytic CD8+ T-cell re-
sponse: a review. AIDS Patient Care and STDs, 16(10):471–477, October 2002. PMID:
12437858.
[61] A J Leslie, K J Pfafferott, P Chetty, et al. HIV evolution: CTL escape mutation
and reversion after transmission. Nature Medicine, 10(3):282–289, March 2004. PMID:
14770175.
[62] A Erickson, Y Kimura, S Igarashi, et al. The Outcome of Hepatitis C Virus Infection
Is Predicted by Escape Mutations in Epitopes Targeted by Cytotoxic T Lymphocytes.
Immunity, 15(6):883–895, December 2001.
[63] Rebecca P Payne, Henrik Klverpris, Jonah B Sacha, et al. Efficacious early antiviral
activity of HIV Gag- and Pol-specific HLA-B 2705-restricted CD8+ T cells. Journal of
virology, 84(20):10543–10557, October 2010. PMID: 20686036.
[64] Thomas C. Friedrich, Christopher A. Frye, Levi J. Yant, et al. Extraepitopic Com-
pensatory Substitutions Partially Restore Fitness to Simian Immunodeficiency Virus
Variants That Escape from an Immunodominant Cytotoxic-T-Lymphocyte Response.
Journal of Virology, 78(5):2581–2585, January 2004.
[65] Hayley Crawford, Julia G. Prado, Alasdair Leslie, et al. Compensatory Mutation Par-
tially Restores Fitness and Delays Reversion of Escape Mutation Within the Immun-
odominant HLA-B*5703-Restricted Gag Epitope in Chronic Human Immunodeficiency
Virus Type 1 Infection. Journal of Virology, 81(15):8346–8351, January 2007.
[66] Rika Draenert, Todd M Allen, Yang Liu, et al. Constraints on HIV-1 evolution and
immunodominance revealed in monozygotic adult twins infected with the same virus.
The Journal of experimental medicine, 203(3):529–539, March 2006. PMID: 16533886.
277
[67] Joerg Timm, Bin Li, Marcus G. Daniels, et al. Human leukocyte antigen-associated
sequence polymorphisms in hepatitis C virus reveal reproducible immune responses and
constraints on viral evolution. Hepatology, 46(2):339–349, 2007.
[68] Thomas C Friedrich, Elizabeth J Dodds, Levi J Yant, et al. Reversion of CTL escape-
variant immunodeficiency viruses in vivo. Nature medicine, 10(3):275–281, March 2004.
PMID: 14966520.
[69] P J Goulder, C Brander, Y Tang, et al. Evolution and transmission of stable CTL escape
mutations in HIV infection. Nature, 412(6844):334–338, July 2001. PMID: 11460164.
[70] Sylvie Le Gall, Lars Erdtmann, Serge Benichou, et al. Nef Interacts with the Subunit of
Clathrin Adaptor Complexes and Reveals a Cryptic Sorting Signal in MHC I Molecules.
Immunity, 8(4):483–495, April 1998.
[71] George B Cohen, Rajesh T Gandhi, Daniel M Davis, et al. The Selective Downregulation
of Class I Major Histocompatibility Complex Proteins by HIV-1 Protects HIV-Infected
Cells from NK Cells. Immunity, 10(6):661–671, June 1999.
[72] Maria Cristina Mingari, Alessandro Moretta, and Lorenzo Moretta. Regulation of KIR
expression in human T cells: a safety mechanism that may impair protective T-cell
responses. Immunology Today, 19(4):153–157, April 1998.
[73] Daniel E. Speiser, Mikal J. Pittet, Danila Valmori, et al. In Vivo Expression of Natu-
ral Killer Cell Inhibitory Receptors by Human MelanomaSpecific Cytolytic T Lympho-
cytes. The Journal of Experimental Medicine, 190(6):775–782, September 1999. PMID:
10499916 PMCID: 2195637.
[74] Daniel E. Speiser, Danila Valmori, Donata Rimoldi, et al. CD28-negative cytolytic
effector T cells frequently express NK receptors and are present at variable proportions in
circulating lymphocytes from healthy donors and melanoma patients. European Journal
of Immunology, 29(6):1990–1999, June 1999.
[75] Eric O. Long and Sumati Rajagopalan. HLA class I recognition by killer cell Ig-like
receptors. Seminars in Immunology, 12(2):101–108, April 2000.
[76] Achim K Moesta, Paul J Norman, Makoto Yawata, et al. Synergistic polymorphism at
two positions distal to the ligand-binding site makes KIR2DL2 a stronger receptor for
278
HLA-C than KIR2DL3. Journal of Immunology (Baltimore, Md.: 1950), 180(6):3969–
3979, March 2008. PMID: 18322206.
[77] C A Mller, G Engler-Blum, V Gekeler, et al. Genetic and serological heterogeneity of
the supertypic HLA-B locus specificities Bw4 and Bw6. Immunogenetics, 30(3):200–207,
1989. PMID: 2777338.
[78] Carlos Vilches and Peter Parham. KIR: Diverse, Rapidly Evolving Receptors of Innate
and Adaptive Immunity. Annual Review of Immunology, 20(1):217–251, April 2002.
[79] S Ugolini, C Arpin, N Anfossi, et al. Involvement of inhibitory NKRs in the survival of
a subset of memory-phenotype CD8+ T cells. Nature Immunology, 2(5):430–435, May
2001. PMID: 11323697.
[80] N T Young, M Uhrberg, J H Phillips, L L Lanier, and P Parham. Differential expres-
sion of leukocyte receptor complex-encoded Ig-like receptors correlates with the tran-
sition from effector to memory CTL. Journal of Immunology (Baltimore, Md.: 1950),
166(6):3933–3941, March 2001. PMID: 11238638.
[81] Franca Gerosa, Barbara Baldani-Guerra, Carla Nisii, et al. Reciprocal activating in-
teraction between natural killer cells and dendritic cells. The Journal of Experimental
Medicine, 195(3):327–333, February 2002. PMID: 11828007.
[82] Diego Piccioli, Silverio Sbrana, Emiliano Melandri, and Nicholas M. Valiante. Contact-
dependent Stimulation and Inhibition of Dendritic Cells by Natural Killer Cells. The
Journal of Experimental Medicine, 195(3):335–341, February 2002. PMID: 11828008
PMCID: 2193592.
[83] Salim I. Khakoo, Chloe L. Thio, Maureen P. Martin, et al. HLA and NK Cell Inhibitory
Receptor Genes in Resolving Hepatitis C Virus Infection. Science, 305(5685):872–874,
August 2004.
[84] Antonio LpezVzquez, Luis Rodrigo, Jess MartnezBorra, et al. Protective Effect of
the HLABw4I80 Epitope and the Killer Cell ImmunoglobulinLike Receptor 3DS1 Gene
against the Development of Hepatocellular Carcinoma in Patients with Hepatitis C Virus
Infection. The Journal of Infectious Diseases, 192(1):162–165, July 2005.
279
[85] Maureen P Martin, Xiaojiang Gao, Jeong-Hee Lee, et al. Epistatic interaction between
KIR3DS1 and HLA-B delays the progression to AIDS. Nature Genetics, 31(4):429–434,
August 2002. PMID: 12134147.
[86] Maureen P Martin, Ying Qi, Xiaojiang Gao, et al. Innate partnership of HLA-B and
KIR3DL1 subtypes against HIV-1. Nature Genetics, 39(6):733–740, June 2007. PMID:
17496894.
[87] D. Middleton, A. S. Diler, A. Meenagh, C. Sleator, and P. A. Gourraud. Killer
immunoglobulin-like receptors (KIR2DL2 and/or KIR2DS2) in presence of their ligand
(HLA-C1 group) protect against chronic myeloid leukaemia. Tissue Antigens, 73(6):553–
560, 2009.
[88] Jill Hollenbach, Martha Ladner, Koy Saeteurn, et al. Susceptibility to Crohns disease is
mediated by KIR2DL2/KIR2DL3 heterozygosity and the HLA-C ligand. Immunogenet-
ics, 61(10):663–671, October 2009.
[89] Maureen P. Martin, George Nelson, Jeong-Hee Lee, et al. Cutting Edge: Susceptibility to
Psoriatic Arthritis: Influence of Activating Killer Ig-Like Receptor Genes in the Absence
of Specific HLA-C Alleles. J Immunol, 169(6):2818–2822, September 2002.
[90] Salim I. Khakoo and Mary Carrington. KIR and disease: a model system or system of
models? Immunological Reviews, 214(1):186–201, 2006.
[91] Lynn B. Dustin and Charles M. Rice. Flying Under the Radar: The Immunobiology of
Hepatitis C. Annual Review of Immunology, 25(1):71–99, April 2007.
[92] Ryosuke Suzuki, Tetsuro Suzuki, Koji Ishii, Yoshiharu Matsuura, and Tatsuo Miyamura.
Processing and Functions of Hepatitis C Virus Proteins. Intervirology, 42(2-3):145–152,
1999.
[93] L. Krekulov, V. ehk, and L. Riley. Structure and functions of hepatitis C virus proteins:
15 years after. Folia Microbiologica, 51(6):665–680, November 2006.
[94] Maria Buti, Yoav Lurie, Natalia G. Zakharova, et al. Randomized trial of peginterferon
alfa-2b and ribavirin for 48 or 72 weeks in patients with hepatitis C virus genotype 1
and slow virologic response. Hepatology, 9999(9999):NA, 2010.
280
[95] Olav Dalgard, Kristian Bjro, Kjell Block Hellum, et al. Treatment with pegylated inter-
feron and ribavarin in HCV infection with genotype 2 or 3 for 14 weeks: A pilot study.
Hepatology, 40(6):1260–1265, December 2004.
[96] Eleanor Barnes, Antonella Folgori, Stefania Capone, et al. Novel Adenovirus-Based
Vaccines Induce Broad and Sustained T Cell Responses to HCV in Man. Science Trans-
lational Medicine, 4(115):115ra1, January 2012.
[97] Benot Callendret and Christopher M. Walker. Will There Be a Vaccine to Protect
Against the Hepatitis C Virus? Gastroenterology, 142(6):1384–1387, May 2012.
[98] Mark H Kuniholm, Andrea Kovacs, Xiaojiang Gao, et al. Specific human leukocyte anti-
gen class I and II alleles associated with hepatitis C virus viremia. Hepatology (Baltimore,
Md.), 51(5):1514–1522, May 2010. PMID: 20169624.
[99] Yasuteru Kondo, Koju Kobayashi, Tomoo Kobayashi, et al. Distribution of the HLA
class I allele in chronic hepatitis C and its association with serum ALT level in chronic
hepatitis C. The Tohoku Journal of Experimental Medicine, 201(2):109–117, October
2003. PMID: 14626512.
[100] Joaqun Ziga, Viviana Romero, Jos Azocar, et al. Protective KIR-HLA interactions
for HCV infection in intravenous drug users. Molecular Immunology, 46(13):2723–2727,
August 2009.
[101] Dongliang Ge, Jacques Fellay, Alexander J. Thompson, et al. Genetic variation in IL28B
predicts hepatitis C treatment-induced viral clearance. Nature, 461(7262):399–401, 2009.
[102] David L. Thomas, Chloe L. Thio, Maureen P. Martin, et al. Genetic variation in IL28B
and spontaneous clearance of hepatitis C virus. Nature, 461(7265):798–801, October
2009.
[103] Yasuhito Tanaka, Nao Nishida, Masaya Sugiyama, et al. Genome-wide association of
IL28B with response to pegylated interferon-alpha and ribavirin therapy for chronic
hepatitis C. Nature Genetics, 41(10):1105–1109, October 2009. PMID: 19749757.
[104] Vijayaprakash Suppiah, Max Moldovan, Golo Ahlenstiel, et al. IL28B is associated with
response to chronic hepatitis C interferon-alpha and ribavirin therapy. Nature Genetics,
41(10):1100–1104, October 2009. PMID: 19749758.
281
[105] Andri Rauch, Zoltn Kutalik, Patrick Descombes, et al. Genetic Variation in IL28B Is As-
sociated With Chronic Hepatitis C and Treatment Failure: A Genome-wide Association
Study. Gastroenterology, January 2010. PMID: 20060832.
[106] S Knapp, L J Yee, A J Frodsham, et al. Polymorphisms in interferon-induced genes and
the outcome of hepatitis C virus infection: roles of MxA, OAS-1 and PKR. Genes and
Immunity, 4(6):411–419, September 2003. PMID: 12944978.
[107] Susanne Knapp, BranwenJ.W. Hennig, AngelaJ. Frodsham, et al. Interleukin-10 pro-
moter polymorphisms and the outcome of hepatitis C virus infection. Immunogenetics,
55(6):362–369, 2003.
[108] Perdita Wietzke-Braun, Larissa-Bettina Manhardt, Albert Rosenberger, et al. Spon-
taneous elimination of hepatitis C virus infection: a retrospective study on demo-
graphic, clinical, and serological correlates. World Journal of Gastroenterology: WJG,
13(31):4224–4229, August 2007. PMID: 17696252.
[109] C Goulding, A Murphy, G MacDonald, et al. The CCR5-32 mutation: impact on disease
outcome in individuals with hepatitis C infection from a single source. Gut, 54(8):1157
–1161, 2005.
[110] Helmut M Diepolder. New insights into the immunopathogenesis of chronic hepatitis C.
Antiviral Research, 82(3):103–109, June 2009. PMID: 19428600.
[111] A Weiner, A L Erickson, J Kansopon, et al. Persistent hepatitis C virus infection in
a chimpanzee is associated with emergence of a cytotoxic T lymphocyte escape vari-
ant. Proceedings of the National Academy of Sciences of the United States of America,
92(7):2755–2759, March 1995. PMID: 7708719 PMCID: 42297.
[112] Silvana Gaudieri, Andri Rauch, Lawrence P. Park, et al. Evidence of Viral Adaptation
to HLA Class I-Restricted Immune Pressure in Chronic Hepatitis C Virus Infection. J.
Virol., 80(22):11094–11104, November 2006.
[113] Christoph Neumann-Haefelin, Susan McKiernan, Scott Ward, et al. Dominant influ-
ence of an HLA-B27 restricted CD8+ T cell response in mediating HCV clearance and
evolution. Hepatology (Baltimore, Md.), 43(3):563–572, March 2006. PMID: 16496339.
282
[114] Christian Markus Lange, Kirsten Roomp, Anette Dragan, et al. HLA class I allele
associations with HCV genetic variants in patients with chronic HCV genotypes 1a or
1b infection. Journal of Hepatology, 53(6):1022–1028, December 2010. PMID: 20800922.
[115] Marianne Ruhl, Torben Knuschke, Kevin Schewior, et al. The CD8+ T-Cell Response
Promotes Evolution of Hepatitis C Virus Nonstructural Proteins. Gastroenterology,
March 2011. PMID: 21376049.
[116] Daniel Yerly, David Heckerman, Todd M. Allen, et al. Increased Cytotoxic T-
Lymphocyte Epitope Variant Cross-Recognition and Functional Avidity Are Associated
with Hepatitis C Virus Clearance. J. Virol., 82(6):3147–3153, March 2008.
[117] Antonella Folgori, Stefania Capone, Lionello Ruggeri, et al. A T-cell HCV vaccine
eliciting effective immunity against heterologous virus challenge in chimpanzees. Nature
Medicine, 12(2):190–197, February 2006. PMID: 16462801.
[118] J Bukh, R H Miller, and R H Purcell. Genetic heterogeneity of hepatitis C virus:
quasispecies and genotypes. Seminars in Liver Disease, 15(1):41–63, February 1995.
PMID: 7597443.
[119] Avidan U. Neumann, Nancy P. Lam, Harel Dahari, et al. Hepatitis C Viral Dynamics
in Vivo and the Antiviral Efficacy of Interferon- Therapy. Science, 282(5386):103–107,
October 1998.
[120] Jean-Michel Pawlotsky. Hepatitis C virus genetic variability: pathogenic and clinical
implications. Clinics in Liver Disease, 7(1):45–66, February 2003. PMID: 12691458.
[121] M Nagai, K Usuku, W Matsumoto, et al. Analysis of HTLV-I proviral load in 202
HAM/TSP patients and 243 asymptomatic HTLV-I carriers: high proviral load strongly
predisposes to HAM/TSP. Journal of Neurovirology, 4(6):586–593, December 1998.
PMID: 10065900.
[122] J. H. Richardson, A. J. Edwards, J. K. Cruickshank, P. Rudge, and A. G. Dalgleish. In
Vivo Cellular Tropism of Human T-Cell Leukemia Virus Type 1. Journal of Virology,
64(11):5682–5687, January 1990.
[123] M Nakagawa, K Nakahara, Y Maruyama, et al. Therapeutic trials in 200 patients with
HTLV-I-associated myelopathy/ tropical spastic paraparesis. Journal of neurovirology,
2(5):345–355, October 1996. PMID: 8912211.
283
[124] Graham P Taylor, Peter Goon, Yoshitaka Furukawa, et al. Zidovudine plus lamivudine in
Human T-Lymphotropic Virus type-I-associated myelopathy: a randomised trial. Retro-
virology, 3:63, 2006. PMID: 16984654.
[125] Stphane Olindo, Gildas Belrose, Nicolas Gillet, et al. Safety of long-term treatment
of HAM/TSP patients with valproic acid. Blood, 118(24):6306–6309, December 2011.
PMID: 21998206.
[126] Philippe V. Afonso, Mourad Mekaouche, Franck Mortreux, et al. Highly Active An-
tiretroviral Treatment Against STLV-1 Infection Combining Reverse Transcriptase and
HDAC Inhibitors. Blood, 116(19):3802–3808, November 2010.
[127] Katie J. M. Jeffery, Asna A. Siddiqui, Mike Bunce, et al. The Influence of HLA Class I
Alleles and Heterozygosity on the Outcome of Human T Cell Lymphotropic Virus Type
I Infection. J Immunol, 165(12):7278–7284, December 2000.
[128] Alison M Vine, Aviva D Witkover, Alun L Lloyd, et al. Polygenic control of human
T lymphotropic virus type I (HTLV-I) provirus load and the risk of HTLV-I-associated
myelopathy/tropical spastic paraparesis. The Journal of infectious diseases, 186(7):932–
939, October 2002. PMID: 12232833.
[129] Alison M Vine, Adrian G Heaps, Lambrini Kaftantzi, et al. The role of CTLs in persistent
viral infection: cytolytic gene expression in CD8+ lymphocytes distinguishes between
individuals with a high or low proviral load of human T cell lymphotropic virus type
1. Journal of immunology (Baltimore, Md.: 1950), 173(8):5121–5129, October 2004.
PMID: 15470056.
[130] E Hanon, S Hall, G P Taylor, et al. Abundant tax protein expression in CD4+ T cells
infected with human T-cell lymphotropic virus type I (HTLV-I) is prevented by cytotoxic
T lymphocytes. Blood, 95(4):1386–1392, February 2000. PMID: 10666215.
[131] Becca Asquith, Angelina J Mosley, Anna Barfield, et al. A functional CD8+ cell assay
reveals individual variation in CD8+ cell antiviral efficacy and explains differences in
human T-lymphotropic virus type 1 proviral load. The Journal of general virology,
86(Pt 5):1515–1523, May 2005. PMID: 15831965.
[132] S Niewiesk, S Daenke, C E Parker, et al. The transactivator gene of human T-cell
leukemia virus type I is more variable within and between healthy carriers than patients
284
with tropical spastic paraparesis. Journal of virology, 68(10):6778–6781, October 1994.
PMID: 8084014.
[133] M D Lairmore, B Roberts, D Frank, et al. Comparative biological responses of rabbits
infected with human T-lymphotropic virus type I isolates from patients with lympho-
proliferative and neurodegenerative disease. International journal of cancer. Journal
international du cancer, 50(1):124–130, January 1992. PMID: 1345820.
[134] F Ibrahim, L Fiette, A Gessain, et al. Infection of rats with human T-cell leukemia
virus type-I: susceptibility of inbred strains, antibody response and provirus location.
International journal of cancer. Journal international du cancer, 58(3):446–451, August
1994. PMID: 8050826.
[135] N Murata, E Hakoda, H Machida, et al. Prevention of human T cell lymphotropic
virus type I infection in Japanese macaques by passive immunization. Leukemia: official
journal of the Leukemia Society of America, Leukemia Research Fund, U.K, 10(12):1971–
1974, December 1996. PMID: 8946939.
[136] Miles P Davenport, Ruy M Ribeiro, Dennis L Chao, and Alan S Perelson. Predicting
the impact of a nonsterilizing vaccine against human immunodeficiency virus. Journal
of virology, 78(20):11340–11351, October 2004. PMID: 15452255.
[137] Christophe Fraser, T Dirdre Hollingsworth, Ruth Chapman, Frank de Wolf, and
William P Hanage. Variation in HIV-1 set-point viral load: epidemiological analysis
and an evolutionary hypothesis. Proceedings of the National Academy of Sciences of the
United States of America, 104(44):17441–17446, October 2007. PMID: 17954909.
[138] Dan H Barouch and Bette Korber. HIV-1 vaccine development after STEP. Annual
review of medicine, 61:153–167, 2010. PMID: 20059334.
[139] Rob J De Boer. Understanding the failure of CD8+ T-cell vaccination against
simian/human immunodeficiency virus. Journal of Virology, 81(6):2838–2848, March
2007. PMID: 17202215.
[140] Susan P Buchbinder, Devan V Mehrotra, Ann Duerr, et al. Efficacy assessment of a
cell-mediated immunity HIV-1 vaccine (the Step Study): a double-blind, randomised,
placebo-controlled, test-of-concept trial. Lancet, 372(9653):1881–1893, November 2008.
PMID: 19012954.
285
[141] Supachai Rerks-Ngarm, Punnee Pitisuttithum, Sorachai Nitayaphan, et al. Vaccination
with ALVAC and AIDSVAX to prevent HIV-1 infection in Thailand. The New England
journal of medicine, 361(23):2209–2220, December 2009. PMID: 19843557.
[142] Sihame Benmira, Vish Bhattacharya, and Matthias L Schmid. An effective HIV vaccine:
A combination of humoral and cellular immunity? Current HIV research, 8(6):441–449,
September 2010. PMID: 20636279.
[143] Jacques Fellay, Kevin V. Shianna, Dongliang Ge, et al. A Whole-Genome Association
Study of Major Determinants for Host Control of HIV-1. Science, 317(5840):944–947,
August 2007.
[144] Cyril Dalmasso, Wassila Carpentier, Laurence Meyer, et al. Distinct genetic loci control
plasma HIV-RNA and cellular HIV-DNA levels in HIV-1 infection: the ANRS Genome
Wide Association 01 study. PloS one, 3(12):e3907, 2008. PMID: 19107206.
[145] Sophie Limou, Sigrid Le Clerc, Cdric Coulonges, et al. Genomewide association study
of an AIDS-nonprogression cohort emphasizes the role played by HLA genes (ANRS
Genomewide Association Study 02). The Journal of infectious diseases, 199(3):419–426,
February 2009. PMID: 19115949.
[146] Sigrid Le Clerc, Sophie Limou, Cdric Coulonges, et al. Genomewide Association Study
of a Rapid Progression Cohort Identifies New Susceptibility Alleles for AIDS (ANRS
Genomewide Association Study 03). Journal of Infectious Diseases, 200(8):1194–1201,
January 2009.
[147] C Costello, J Tang, C Rivers, et al. HLA-B*5703 independently associated with slower
HIV-1 disease progression in Rwandan women. AIDS (London, England), 13(14):1990–
1991, October 1999. PMID: 10513667.
[148] S A Migueles, M S Sabbaghian, W L Shupert, et al. HLA B*5701 is highly associated
with restriction of virus replication in a subgroup of HIV-infected long term nonprogres-
sors. Proceedings of the National Academy of Sciences of the United States of America,
97(6):2709–2714, March 2000. PMID: 10694578.
[149] Marcus Altfeld, Marylyn M Addo, Eric S Rosenberg, et al. Influence of HLA-B57 on
clinical presentation and viral control during acute HIV-1 infection. AIDS (London,
England), 17(18):2581–2591, December 2003. PMID: 14685052.
286
[150] Photini Kiepiela, Alasdair J. Leslie, Isobella Honeyborne, et al. Dominant influence of
HLA-B in mediating the potential co-evolution of HIV and HLA. Nature, 432(7018):769–
775, December 2004.
[151] Xiaojiang Gao, Arman Bashirova, Astrid K N Iversen, et al. AIDS restriction HLA
allotypes target distinct intervals of HIV-1 pathogenesis. Nat Med, 11(12):1290–1292,
December 2005.
[152] Henrik N Kloverpris, Anette Stryhn, Mikkel Harndahl, et al. HLA-B*57 Micropolymor-
phism shapes HLA allele-specific epitope immunogenicity, selection pressure, and HIV
immune control. Journal of Virology, 86(2):919–929, January 2012. PMID: 22090105.
[153] X Gao, G W Nelson, P Karacki, et al. Effect of a single amino acid change in MHC class
I molecules on the rate of progression to AIDS. The New England Journal of Medicine,
344(22):1668–1675, May 2001. PMID: 11386265.
[154] M R Klein, I P Keet, J D’Amaro, et al. Associations between HLA frequencies and
pathogenic features of human immunodeficiency virus type 1 infection in seroconverters
from the Amsterdam cohort of homosexual men. The Journal of infectious diseases,
169(6):1244–1249, June 1994. PMID: 8195600.
[155] Brinda Emu, Elizabeth Sinclair, Hiroyu Hatano, et al. HLA class I-restricted T-cell
responses may contribute to the control of human immunodeficiency virus infection, but
such responses are not always necessary for long-term virus control. Journal of Virology,
82(11):5398–5407, June 2008. PMID: 18353945.
[156] Florencia Pereyra, Marylyn M Addo, Daniel E Kaufmann, et al. Genetic and immuno-
logic heterogeneity among persons who control HIV infection in the absence of therapy.
The Journal of Infectious Diseases, 197(4):563–571, February 2008. PMID: 18275276.
[157] Edward T Mee, Neil Berry, Claire Ham, et al. Mhc haplotype H6 is associated with
sustained control of SIVmac251 infection in Mauritian cynomolgus macaques. Immuno-
genetics, 61(5):327–339, May 2009. PMID: 19337730.
[158] E T Mee, N Berry, C Ham, et al. Mhc haplotype M3 is associated with early control of
SHIVsbg infection in Mauritian cynomolgus macaques. Tissue Antigens, 76(3):223–229,
September 2010. PMID: 20403147.
287
[159] J Tang, C Costello, I P Keet, et al. HLA class I homozygosity accelerates disease
progression in human immunodeficiency virus type 1 infection. AIDS research and human
retroviruses, 15(4):317–324, March 1999. PMID: 10082114.
[160] Shelby L O’Connor, Jennifer J Lhost, Ericka A Becker, et al. MHC heterozygote advan-
tage in simian immunodeficiency virus-infected mauritian cynomolgus macaques. Science
Translational Medicine, 2(22):22ra18, March 2010. PMID: 20375000.
[161] Geraldine M O’Connor, Eriko Yamada, Andy Rampersaud, et al. Analysis of binding
of KIR3DS1*014 to HLA suggests distinct evolutionary history of KIR3DS1. Journal
of immunology (Baltimore, Md.: 1950), 187(5):2162–2171, September 2011. PMID:
21804024.
[162] Salix Boulet, Rujun Song, Philomena Kamya, et al. HIV Protective KIR3DL1 and HLA-
B Genotypes Influence NK Cell Function Following Stimulation with HLA-Devoid Cells.
J Immunol, 184(4):2057–2064, February 2010.
[163] Mary Carrington and Bruce D Walker. Immunogenetics of spontaneous control of HIV.
Annual review of medicine, 63:131–145, 2012. PMID: 22248321.
[164] S Koenig, A J Conley, Y A Brewah, et al. Transfer of HIV-1-specific cytotoxic T lym-
phocytes to an AIDS patient leads to selection for mutant HIV variants and subsequent
disease progression. Nature Medicine, 1(4):330–336, April 1995. PMID: 7585062.
[165] Mark J Geels, Marion Cornelissen, Hanneke Schuitemaker, et al. Identification of sequen-
tial viral escape mutants associated with altered T-cell responses in a human immun-
odeficiency virus type 1-infected individual. Journal of Virology, 77(23):12430–12440,
December 2003. PMID: 14610167.
[166] J E Schmitz, M J Kuroda, S Santra, et al. Control of viremia in simian immunodeficiency
virus infection by CD8+ lymphocytes. Science (New York, N.Y.), 283(5403):857–860,
February 1999. PMID: 9933172.
[167] Joseph K. Wong, Matthew C. Strain, Rodin Porrata, et al. In Vivo CD8+ T-Cell
Suppression of SIV Viremia Is Not Mediated by CTL Clearance of Productively Infected
Cells. PLoS Pathog, 6(1):e1000748, January 2010.
288
[168] Photini Kiepiela, Kholiswa Ngumbela, Christina Thobakgale, et al. CD8+ T-cell re-
sponses to different HIV proteins have discordant associations with viral load. Nat Med,
13(1):46–53, January 2007.
[169] Marybeth Daucher, David A Price, Jason M Brenchley, et al. Virological outcome after
structured interruption of antiretroviral therapy for human immunodeficiency virus in-
fection is associated with the functional profile of virus-specific CD8+ T cells. Journal
of Virology, 82(8):4102–4114, April 2008. PMID: 18234797.
[170] M M Addo, X G Yu, A Rathod, et al. Comprehensive epitope analysis of human im-
munodeficiency virus type 1 (HIV-1)-specific T-cell responses directed against the entire
expressed HIV-1 genome demonstrate broadly directed responses, but no correlation to
viral load. Journal of Virology, 77(3):2081–2092, February 2003. PMID: 12525643.
[171] Ingrid M M Schellens, Jos A M Borghans, Christine A Jansen, et al. Abundance of early
functional HIV-specific CD8+ T cells does not predict AIDS-free survival time. PloS
One, 3(7):e2745, 2008. PMID: 18648514.
[172] Ann Chahroudi, Steven E Bosinger, Thomas H Vanderford, Mirko Paiardini, and Guido
Silvestri. Natural SIV hosts: showing AIDS the door. Science (New York, N.Y.),
335(6073):1188–1193, March 2012. PMID: 22403383.
[173] Donald L Sodora, Jonathan S Allan, Cristian Apetrei, et al. Toward an AIDS vaccine:
lessons from natural simian immunodeficiency virus infections of African nonhuman pri-
mate hosts. Nature medicine, 15(8):861–865, August 2009. PMID: 19661993.
[174] Rachel V Duyne, Aarthi Narayanan, Kylene K-Hall, et al. Humanized mouse models of
HIV-1 latency. Current HIV research, 9(8):595–605, December 2011. PMID: 22211664.
[175] Stephen J O’Brien, Jennifer L Troyer, Meredith A Brown, et al. Emerging viruses in the
Felidae: shifting paradigms. Viruses, 4(2):236–257, February 2012. PMID: 22470834.
[176] Dan H. Barouch. Challenges in the Development of an HIV-1 Vaccine. Nature,
455(7213):613–619, October 2008. PMID: 18833271 PMCID: PMC2572109.
[177] Marjet Elemans, Nafisa-Katrin Seich al Basatena, and Becca Asquith. The Efficiency of
the Human CD8+ T Cell Response: How Should We Quantify It, What Determines It,
and Does It Matter? PLoS Comput Biol, 8(2):e1002381, February 2012.
289
[178] Arthur Y Kim, Thomas Kuntzen, Joerg Timm, et al. Spontaneous control of HCV is
associated with expression of HLA-B 57 and preservation of targeted epitopes. Gastroen-
terology, 140(2):686–696.e1, February 2011. PMID: 20875418.
[179] Peter Hraber, Carla Kuiken, and Karina Yusim. Evidence for human leukocyte antigen
heterozygote advantage against hepatitis C virus infection. Hepatology, 46(6):1713–1721,
2007.
[180] M Thursz, R Yallop, R Goldin, C Trepo, and H C Thomas. Influence of MHC class
II genotype on outcome of infection with hepatitis C virus. The HENCORE group.
Hepatitis C European Network for Cooperative Research. Lancet, 354(9196):2119–2124,
December 1999. PMID: 10609818.
[181] Elizabeth Trachtenberg, Bette Korber, Cristina Sollars, et al. Advantage of rare HLA
supertype in HIV disease progression. Nat Med, 9(7):928–935, July 2003.
[182] Katie J. M. Jeffery and Charles R. M. Bangham. Do infectious diseases drive MHC
diversity? Microbes and Infection, 2(11):1335–1341, September 2000.
[183] Shaomin Tian, Robert Maile, Edward J. Collins, and Jeffrey A. Frelinger. CD8+ T
Cell Activation Is Governed by TCR-Peptide/MHC Affinity, Not Dissociation Rate. The
Journal of Immunology, 179(5):2952–2960, January 2007.
[184] Jos A. M. Borghans, Anne Mlgaard, Rob J. de Boer, and Can Kemir. HLA Alleles
Associated with Slow Progression to AIDS Truly Prefer to Present HIV-1 p24. PLoS
ONE, 2(9):e920, 2007.
[185] Aidan MacNamara, Aileen Rowan, Silva Hilburn, et al. HLA Class I Binding of HBZ
Determines Outcome in HTLV-1 Infection. 6(9), September 2010. PMID: 20886101
PMCID: 2944806.
[186] D Vlahov, J C Anthony, A Munoz, et al. The ALIVE study, a longitudinal study of
HIV-1 infection in intravenous drug users: description of methods and characteristics of
participants. NIDA Research Monograph, 109:75–100, 1991. PMID: 1661376.
[187] J J Goedert, C M Kessler, L M Aledort, et al. A prospective study of human im-
munodeficiency virus type 1 infection and the development of AIDS in subjects with
290
hemophilia. The New England Journal of Medicine, 321(17):1141–1148, October 1989.
PMID: 2477702.
[188] M W Hilgartner, S M Donfield, A Willoughby, et al. Hemophilia growth and devel-
opment study. Design, methods, and entry data. The American Journal of Pediatric
Hematology/Oncology, 15(2):208–218, May 1993. PMID: 8498644.
[189] M. Bunce, C. M. O’Neill, M. C. N. M. Barnardo, et al. Phototyping: comprehensive DNA
typing for HLA-A, B, C, DRB1, DRB3, DRB4, DRB5 & DQB1 by PCR with 144 primer
mixes utilizing sequence-specific primers (PCR-SSP). Tissue Antigens, 46(5):355–367,
November 1995.
[190] George W Nelson and Stephen J O’Brien. Using mutual information to measure the
impact of multiple genetic factors on AIDS. Journal of Acquired Immune Deficiency
Syndromes (1999), 42(3):347–354, July 2006. PMID: 16763524.
[191] Solomon Kullback. Information Theory and Statistics. Dover Publications, July 1997.
[192] Xingdong Yang and Xinglong Yu. An introduction to epitope prediction methods and
software. Reviews in Medical Virology, 19(2):77–96, March 2009. PMID: 19101924.
[193] David Heckerman, Carl Kadie, and Jennifer Listgarten. Leveraging information across
HLA alleles/supertypes improves epitope prediction. Journal of Computational Biology:
A Journal of Computational Molecular Cell Biology, 14(6):736–746, August 2007. PMID:
17691891.
[194] Ilka Hoof, Bjoern Peters, John Sidney, et al. NetMHCpan, a method for MHC class I
binding prediction beyond humans. Immunogenetics, 61(1):1–13, January 2009.
[195] Huynh-Hoa Bui, John Sidney, Bjoern Peters, et al. Automated generation and evaluation
of specific MHC binding predictive tools: ARB matrix applications. Immunogenetics,
57(5):304–314, June 2005.
[196] Nebojsa Jojic, Manuel Reyes-Gomez, David Heckerman, Carl Kadie, and Ora Schueler-
Furman. Learning MHC I–peptide binding. Bioinformatics, 22(14):e227–235, July 2006.
[197] Magdalini Moutaftsi, Bjoern Peters, Valerie Pasquetto, et al. A consensus epitope pre-
diction approach identifies the breadth of murine TCD8+-cell responses to vaccinia virus.
Nat Biotech, 24(7):817–819, July 2006.
291
[198] Aidan MacNamara, Ulrich Kadolsky, Charles R. M. Bangham, and Becca Asquith. T-Cell
Epitope Prediction: Rescaling Can Mask Biological Variation between MHC Molecules.
PLoS Comput Biol, 5(3):e1000327, March 2009.
[199] S Buus, S L Lauemller, P Worning, et al. Sensitive quantitative predictions of peptide-
MHC binding by a ’Query by Committee’ artificial neural network approach. Tissue
Antigens, 62(5):378–384, November 2003. PMID: 14617044.
[200] Morten Nielsen, Claus Lundegaard, Peder Worning, et al. Reliable prediction of T-cell
epitopes using neural networks with novel sequence representations. Protein Science: A
Publication of the Protein Society, 12(5):1007–1017, May 2003. PMID: 12717023.
[201] John Sidney, Bjoern Peters, Nicole Frahm, Christian Brander, and Alessandro Sette.
HLA class I supertypes: a revised and updated classification. BMC Immunology, 9:1–1.
PMID: 18211710 PMCID: 2245908.
[202] Li-Yu Wang, Hans Hsienhong Lin, Tsung-Dao Lee, et al. Human leukocyte antigen phe-
notypes and hepatitis C viral load. Journal of Clinical Virology, 32(2):144–150, February
2005.
[203] Keyur Patel, Suzanne Norris, Lauralynn Lebeck, et al. HLA class I allelic diversity and
progression of fibrosis in patients with chronic hepatitis C. Hepatology, 43(2):241–249,
2006.
[204] D. Middleton, L. Menchaca, H. Rood, and R. Komerofsky. New allele frequency database:
http://www.allelefrequencies.net. Tissue Antigens, 61(5):403–407, May 2003.
[205] E A Popov, B N Levitan, L P Alekseev, O A Pronina, and S V Suchkov. [Immuno-
genetic HLA markers of chronic viral hepatitis]. Terapevticheski Arkhiv, 77(2):54–59,
2005. PMID: 15807454.
[206] Robert Thimme, David Oldach, Kyong-Mi Chang, et al. Determinants of Viral Clearance
and Persistence during Acute Hepatitis C Virus Infection. J. Exp. Med., 194(10):1395–
1406, November 2001.
[207] Susan Smyk-Pearson, Ian A Tester, Dennis Lezotte, et al. Differential antigenic hierarchy
associated with spontaneous recovery from hepatitis C virus infection: implications for
292
vaccine design. The Journal of infectious diseases, 194(4):454–463, August 2006. PMID:
16845628.
[208] Bruce D Walker. Elite control of HIV Infection: implications for vaccines and treat-
ment. Topics in HIV Medicine: A Publication of the International AIDS Society, USA,
15(4):134–136, September 2007. PMID: 17720999.
[209] Peter Parham and Tomoko Ohta. Population Biology of Antigen Presentation by MHC
Class I Molecules. Science, 272(5258):67–74, April 1996.
[210] Philip J. R. Goulder, Yanhua Tang, Stephen I. Pelton, and Bruce D. Walker. HLA-
B57-Restricted Cytotoxic T-Lymphocyte Activity in a Single Infected Subject toward
Two Optimal Epitopes, One of Which Is Entirely Contained within the Other. J. Virol.,
74(11):5291–5299, June 2000.
[211] Geraldine M A Gillespie, Rupert Kaul, Tao Dong, et al. Cross-reactive cytotoxic T
lymphocytes against a HIV-1 p24 epitope in slow progressors with B*57. AIDS (London,
England), 16(7):961–972, May 2002. PMID: 11953462.
[212] Huabiao Chen, Zaza M. Ndhlovu, Dongfang Liu, et al. TCR clonotypes modulate the
protective effect of HLA class I molecules in HIV-1 infection. Nature Immunology, 2012.
[213] Cesar Oniangue-Ndza, Thomas Kuntzen, Michael Kemper, et al. Compensatory Muta-
tions Restore the Replication Defects Caused by CTL Escape Mutations in HCV Poly-
merase. Journal of Virology, August 2011. PMID: 21880756.
[214] G S Ogg, X Jin, S Bonhoeffer, et al. Quantitation of HIV-1-specific cytotoxic T lympho-
cytes and plasma load of viral RNA. Science (New York, N.Y.), 279(5359):2103–2106,
March 1998. PMID: 9516110.
[215] G R Kaufmann, J J Zaunders, P Cunningham, and D A Cooper. Phenotypic analysis of
CD8+ T lymphocytes in a cohort of HIV type 1-infected patients treated with saquinavir,
ritonavir, and two nucleoside analogs for 1 year, and association with plasma HIV type
1 RNA. AIDS Research and Human Retroviruses, 15(11):963–972, July 1999. PMID:
10445808.
[216] Ryuji Kubota, Kousuke Hanada, Yoshitaka Furukawa, et al. Genetic stability of human
T lymphotropic virus type I despite antiviral pressures by CTLs. Journal of Immunology
(Baltimore, Md.: 1950), 178(9):5966–5972, May 2007. PMID: 17442981.
293
[217] B Rehermann, K M Chang, J G McHutchison, et al. Quantitative analysis of the periph-
eral blood cytotoxic T lymphocyte response in patients with chronic hepatitis C virus
infection. Journal of Clinical Investigation, 98(6):1432–1440, September 1996. PMID:
8823309 PMCID: 507570.
[218] David H Raulet. Interplay of natural killer cells and their receptors with the adaptive
immune response. Nature Immunology, 5(10):996–1002, October 2004. PMID: 15454923.
[219] N Wagtmann, S Rajagopalan, C C Winter, M Peruzzi, and E O Long. Killer cell
inhibitory receptors specific for HLA-C and HLA-B identified by direct binding and by
functional transfer. Immunity, 3(6):801–809, December 1995. PMID: 8777725.
[220] Golo Ahlenstiel, Maureen P Martin, Xiaojiang Gao, Mary Carrington, and Barbara Re-
hermann. Distinct KIR/HLA compound genotypes affect the kinetics of human antiviral
natural killer cell responses. The Journal of Clinical Investigation, 118(3):1017–1026,
March 2008. PMID: 18246204.
[221] Jacob Cohen. Statistical Power Analysis for the Behavioral Sciences. Routledge, July
1988.
[222] Paul J Norman, Mark A Cook, B Sean Carey, et al. SNP haplotypes and allele frequencies
show evidence for disruptive and balancing selection in the human leukocyte receptor
complex. Immunogenetics, 56(4):225–237, July 2004. PMID: 15185041.
[223] C C Winter, J E Gumperz, P Parham, E O Long, and N Wagtmann. Direct binding
and functional transfer of NK cell inhibitory receptors reveal novel patterns of HLA-C
allotype recognition. Journal of Immunology (Baltimore, Md.: 1950), 161(2):571–577,
July 1998. PMID: 9670929.
[224] William Henry Carr, Marcelo Jorge Pando, and Peter Parham. KIR3DL1 polymorphisms
that affect NK cell inhibition by HLA-Bw4 ligand. Journal of Immunology (Baltimore,
Md.: 1950), 175(8):5222–5229, October 2005. PMID: 16210627.
[225] M Cella, A Longo, G B Ferrara, J L Strominger, and M Colonna. NK3-specific natural
killer cells are selectively inhibited by Bw4-positive HLA alleles with isoleucine 80. The
Journal of Experimental Medicine, 180(4):1235–1242, October 1994. PMID: 7931060.
294
[226] M Uhrberg, N M Valiante, B P Shum, et al. Human diversity in killer cell inhibitory
receptor genes. Immunity, 7(6):753–763, December 1997. PMID: 9430221.
[227] Asma Gati, Nadia Guerra, Catherine Gaudin, et al. CD158 receptor controls cytotoxic
T-lymphocyte susceptibility to tumor-mediated activation-induced cell death by inter-
fering with Fas signaling. Cancer Research, 63(21):7475–7482, November 2003. PMID:
14612548.
[228] Neil T. Young and Markus Uhrberg. KIR expression shapes cytotoxic repertoires: a
developmental program of survival. Trends in Immunology, 23(2):71–75, February 2002.
[229] N Anfossi, V Pascal, E Vivier, and S Ugolini. Biology of T memory type 1 cells. Im-
munological Reviews, 181:269–278, June 2001. PMID: 11513148.
[230] K Bieganowska, P Hllsberg, G J Buckle, et al. Direct analysis of viral-specific CD8+
T cells with soluble HLA-A2/Tax11-19 tetramer complexes in patients with human T
cell lymphotropic virus-associated myelopathy. Journal of Immunology (Baltimore, Md.:
1950), 162(3):1765–1771, February 1999. PMID: 9973440.
[231] Paula Bonorino, Vincent Leroy, Tania Dufeu-Duchesne, et al. Features and distribution of
CD8 T cells with human leukocyte antigen class I-specific receptor expression in chronic
hepatitis C. Hepatology (Baltimore, Md.), 46(5):1375–1386, November 2007. PMID:
17668887.
[232] Galit Alter, Suzannah Rihn, Hendrik Streeck, et al. Ligand-independent exhaustion of
killer immunoglobulin-like receptor-positive CD8+ T cells in human immunodeficiency
virus type 1 infection. Journal of Virology, 82(19):9668–9677, October 2008. PMID:
18579582.
[233] Daniel M Andrews, Marie J Estcourt, Christopher E Andoniou, et al. Innate Immunity
Defines the Capacity of Antiviral T Cells to Limit Persistent Infection. The Journal of
Experimental Medicine, May 2010.
[234] Stephen N. Waggoner, Markus Cornberg, Liisa K. Selin, and Raymond M. Welsh. Nat-
ural killer cells act as rheostats modulating antiviral T cells. Nature, advance online
publication, November 2011.
295
[235] Stephen N Waggoner, Ruth T Taniguchi, Porunelloor A Mathew, Vinay Kumar, and
Raymond MWelsh. Absence of mouse 2B4 promotes NK cell-mediated killing of activated
CD8+ T cells, leading to prolonged viral persistence and altered pathogenesis. The
Journal of Clinical Investigation, 120(6):1925–1938, June 2010. PMID: 20440077.
[236] Emilie Narni-Mancinelli, Baptiste N Jaeger, Claire Bernat, et al. Tuning of natural
killer cell reactivity by NKp46 and Helios calibrates T cell responses. Science (New
York, N.Y.), 335(6066):344–348, January 2012. PMID: 22267813.
[237] Katrina Soderquest, Thierry Walzer, Biljana Zafirova, et al. Cutting edge: CD8+ T
cell priming in the absence of NK cells leads to enhanced memory responses. Journal of
Immunology (Baltimore, Md.: 1950), 186(6):3304–3308, March 2011. PMID: 21307295.
[238] Sumiko Takao, Takayuki Ishikawa, Kouhei Yamashita, and Takashi Uchiyama. The rapid
induction of HLA-E is essential for the survival of antigen-activated naive CD4 T cells
from attack by NK cells. Journal of Immunology (Baltimore, Md.: 1950), 185(10):6031–
6040, November 2010. PMID: 20952676.
[239] Geraldine M O’Connor, Kieran J Guinan, Rodat T Cunningham, et al. Functional
polymorphism of the KIR3DL1/S1 receptor on human NK cells. Journal of immunology
(Baltimore, Md.: 1950), 178(1):235–241, January 2007. PMID: 17182560.
[240] Makoto Yawata, Nobuyo Yawata, Monia Draghi, et al. Roles for HLA and KIR poly-
morphisms in natural killer cell repertoire selection and modulation of effector function.
The Journal of Experimental Medicine, 203(3):633 –645, March 2006.
[241] L. Fadda, G. Borhis, P. Ahmed, et al. Peptide antagonism as a mechanism for NK cell
activation. Proceedings of the National Academy of Sciences, May 2010.
[242] Kathryn Poon, Damien Montamat-Sicotte, Nicole Cumberbatch, Andrew J McMichael,
and Margaret F C Callan. Expression of leukocyte immunoglobulin-like receptors and
natural killer receptors on virus-specific CD8+ T cells during the evolution of Epstein-
Barr virus-specific immune responses in vivo. Viral immunology, 18(3):513–522, 2005.
PMID: 16212530.
[243] Lars T van der Veken, Maria Diez Campelo, Menno A W G van der Hoorn, et al.
Functional analysis of killer Ig-like receptor-expressing cytomegalovirus-specific CD8+
296
T cells. Journal of immunology (Baltimore, Md.: 1950), 182(1):92–101, January 2009.
PMID: 19109139.
[244] Maureen P Martin and Mary Carrington. Immunogenetics of viral infections. Current
Opinion in Immunology, 17(5):510–516, October 2005.
[245] Cristina Cerboni, Michele Ardolino, Angela Santoni, and Alessandra Zingoni. Detuning
CD8+ T lymphocytes by down-regulation of the activating receptor NKG2D: role of
NKG2D ligands released by activated T cells. Blood, 113(13):2955 –2964, March 2009.
[246] Xu Tan, Zhi John Lu, Geng Gao, et al. Tiling genomes of pathogenic viruses identi-
fies potent antiviral shRNAs and reveals a role for secondary structure in shRNA effi-
cacy. Proceedings of the National Academy of Sciences of the United States of America,
109(3):869–874, January 2012. PMID: 22219365.
[247] Andrew Adey, Hilary G Morrison, Asan, et al. Rapid, low-input, low-bias construction
of shotgun fragment libraries by high-density in vitro transposition. Genome Biology,
11(12):R119, December 2010.
[248] Becca Asquith, Charles T. T Edwards, Marc Lipsitch, and Angela R McLean. Inefficient
Cytotoxic T LymphocyteMediated Killing of HIV-1Infected Cells In Vivo. PLoS Biol,
4(4):e90, March 2006.
[249] Eun-Young Kim, Ronald S Veazey, Roland Zahn, et al. Contribution of CD8+ T cells to
containment of viral replication and emergence of mutations in Mamu-A*01-restricted
epitopes in Simian immunodeficiency virus-infected rhesus monkeys. Journal of Virology,
82(11):5631–5635, June 2008. PMID: 18367519.
[250] Nitin K Saksena, Jing Qin Wu, Simon J Potter, John Wilkinson, and Bin Wang. Human
immunodeficiency virus interactions with CD8+ T lymphocytes. Current HIV Research,
6(1):1–9, January 2008. PMID: 18288969.
[251] Jane C. Stinchcombe, Giovanna Bossi, Sarah Booth, and Gillian M. Griffiths. The
Immunological Synapse of CTL Contains a Secretory Domain and Membrane Bridges.
Immunity, 15(5):751–761, November 2001.
[252] D A Price, A K Sewell, T Dong, et al. Antigen-specific release of beta-chemokines by
anti-HIV-1 cytotoxic T lymphocytes. Current Biology: CB, 8(6):355–358, March 1998.
PMID: 9512422.
297
[253] Ludwig Wagner, Otto O. Yang, Eduardo A. Garcia-Zepeda, et al. [beta]-Chemokines
are released from HIV-1-specific cytolytic T-cell granules complexed to proteoglycans.
Nature, 391(6670):908–911, February 1998.
[254] Navin Varadarajan, Boris Julg, Yvonne J Yamanaka, et al. A high-throughput single-
cell analysis of human CD8 T cell functions reveals discordance for cytokine secretion
and cytolysis. The Journal of Clinical Investigation, 121(11):4322–4331, November 2011.
PMID: 21965332.
[255] J K Sandberg, N M Fast, and D F Nixon. Functional heterogeneity of cytokines and
cytolytic effector molecules in human CD8+ T lymphocytes. Journal of Immunology
(Baltimore, Md.: 1950), 167(1):181–187, July 2001. PMID: 11418647.
[256] Qing Han, Neda Bagheri, Elizabeth M Bradshaw, et al. Polyfunctional responses by
human T cells result from sequential release of cytokines. Proceedings of the National
Academy of Sciences, 109(5):1607–1612, January 2012.
[257] Christian L Althaus and Rob J De Boer. Implications of CTL-mediated killing of HIV-
infected cells during the non-productive stage of infection. PloS One, 6(2):e16468, 2011.
PMID: 21326882.
[258] P Klenerman, R E Phillips, C R Rinaldo, et al. Cytotoxic T lymphocytes and viral
turnover in HIV type 1 infection. Proceedings of the National Academy of Sciences of
the United States of America, 93(26):15323–15328, December 1996. PMID: 8986810.
[259] Kenneth P. Burnham and David R. Anderson. Model selection and multimodel inference:
a practical information-theoretic approach. Springer, 2002.
[260] Clifford M Hurvich and Chih-Ling Tsai. Bias of the corrected AIC criterion for under-
fitted regression and time series models. Biometrika, 78(3):499–509, September 1991.
[261] W. David Wick and Otto O. Yang. Biologically-Directed Modeling Reflects Cytolytic
Clearance of SIV-Infected Cells In Vivo in Macaques. PLoS ONE, 7(9):e44778, September
2012.
[262] R M Zorzenon dos Santos and S Coutinho. Dynamics of HIV infection: a cellular
automata approach. Physical Review Letters, 87(16):168102, October 2001. PMID:
11690248.
298
[263] Peter Sloot, Fan Chen, and Charles Boucher. Cellular automata model of drug therapy
for HIV infection. LECTURE NOTES IN COMPUTER SCIENCE, 2493:282—293, 2002.
[264] E. Burkhead, J. Hawkins, and D. Molinek. A Dynamical Study of a Cellular Automata
Model of the Spread of HIV in a Lymph Node. Bulletin of Mathematical Biology,
71(1):25–74, January 2009.
[265] Amy L Bauer, Catherine A A Beauchemin, and Alan S Perelson. Agent-based mod-
eling of host-pathogen systems: The successes and challenges. Information Sciences,
179(10):1379–1389, April 2009. PMID: 20161146.
[266] S P Preston, S L Waters, O E Jensen, P R Heaton, and D I Pritchard. T-cell motility
in the early stages of the immune response modeled as a random walk amongst targets.
Physical Review. E, Statistical, Nonlinear, and Soft Matter Physics, 74(1 Pt 1):011910,
July 2006. PMID: 16907130.
[267] Frederik Graw and Roland R. Regoes. Investigating CTL Mediated Killing with a 3D
Cellular Automaton. 5(8), August 2009. PMID: 19696876 PMCID: 2715871.
[268] Mark J Miller, Sindy H Wei, Ian Parker, and Michael D Cahalan. Two-photon imaging
of lymphocyte motility and antigen response in intact lymph node. Science (New York,
N.Y.), 296(5574):1869–1873, June 2002. PMID: 12016203.
[269] Mark J Miller, Sindy H Wei, Michael D Cahalan, and Ian Parker. Autonomous T cell
trafficking examined in vivo with intravital two-photon microscopy. Proceedings of the
National Academy of Sciences of the United States of America, 100(5):2604–2609, March
2003. PMID: 12601158.
[270] Mark J Miller, Arsalan S Hejazi, Sindy H Wei, Michael D Cahalan, and Ian Parker. T
cell repertoire scanning is promoted by dynamic dendritic cell behavior and random T
cell motility in the lymph node. Proceedings of the National Academy of Sciences of the
United States of America, 101(4):998–1003, January 2004. PMID: 14722354.
[271] Alexandre Boissonnas, Luc Fetler, Ingrid S Zeelenberg, Stphanie Hugues, and Sebastian
Amigorena. In vivo imaging of cytotoxic T cell infiltration and elimination of a solid
tumor. The Journal of Experimental Medicine, 204(2):345–356, February 2007. PMID:
17261634.
299
[272] Tim Worbs, Thorsten R Mempel, Jasmin Blter, Ulrich H von Andrian, and Reinhold
Frster. CCR7 ligands stimulate the intranodal motility of T lymphocytes in vivo. The
Journal of Experimental Medicine, 204(3):489–495, March 2007. PMID: 17325198.
[273] JacksonG. Egen, AntonioGigliotti Rothfuchs, CarlG. Feng, et al. Intravital Imaging
Reveals Limited Antigen Presentation and T Cell Effector Function in Mycobacterial
Granulomas. Immunity, 34(5):807–819, May 2011.
[274] Ruy M Ribeiro, Li Qin, Leslie L Chavez, et al. Estimation of the initial viral growth
rate and basic reproductive number during acute HIV-1 infection. Journal of Virology,
84(12):6096–6102, June 2010. PMID: 20357090.
[275] M Stevenson, T L Stanwick, M P Dempsey, and C A Lamonica. HIV-1 replication is
controlled at the level of T cell activation and proviral integration. The EMBO Journal,
9(5):1551–1560, May 1990. PMID: 2184033 PMCID: PMC551849.
[276] Beth D. Jamieson, Otto O. Yang, Lance Hultin, et al. Epitope Escape Mutation and
Decay of Human Immunodeficiency Virus Type 1-Specific CTL Responses. The Journal
of Immunology, 171(10):5372–5379, November 2003.
[277] S A Kalams, P J Goulder, A K Shea, et al. Levels of human immunodeficiency virus type
1-specific cytotoxic T-lymphocyte effector and memory responses decline after suppres-
sion of viremia with highly active antiretroviral therapy. Journal of virology, 73(8):6721–
6728, August 1999. PMID: 10400770.
[278] G S Ogg, X Jin, S Bonhoeffer, et al. Decay kinetics of human immunodeficiency virus-
specific effector cytotoxic T lymphocytes after combination antiretroviral therapy. Jour-
nal of virology, 73(1):797–800, January 1999. PMID: 9847391.
[279] M Altfeld, E S Rosenberg, R Shankarappa, et al. Cellular immune responses and viral
diversity in individuals treated during acute and early HIV-1 infection. The Journal of
experimental medicine, 193(2):169–180, January 2001. PMID: 11148221.
[280] J P Casazza, M R Betts, L J Picker, and R A Koup. Decay kinetics of human im-
munodeficiency virus-specific CD8+ T cells in peripheral blood after initiation of highly
active antiretroviral therapy. Journal of virology, 75(14):6508–6516, July 2001. PMID:
11413318.
300
[281] Andreas Jung, Reinhard Maier, Jean-Pierre Vartanian, et al. Recombination: Multiply
infected spleen cells in HIV patients. Nature, 418(6894):144, July 2002. PMID: 12110879.
[282] Lina Josefsson, Martin S King, Barbro Makitalo, et al. Majority of CD4+ T cells from
peripheral blood of HIV-1-infected individuals contain only one HIV DNA molecule.
Proceedings of the National Academy of Sciences of the United States of America,
108(27):11199–11204, July 2011. PMID: 21690402.
[283] Taiki Aoshi, Bernd H Zinselmeyer, Vjollca Konjufca, et al. Bacterial entry to the splenic
white pulp initiates antigen presentation to CD8+ T cells. Immunity, 29(3):476–486,
September 2008. PMID: 18760639.
[284] Marc Bajnoff, Nicolas Glaichenhaus, and Ronald N Germain. Fibroblastic reticular cells
guide T lymphocyte entry into and migration within the splenic T cell zone. Journal
of Immunology (Baltimore, Md.: 1950), 181(6):3947–3954, September 2008. PMID:
18768849.
[285] Joost B Beltman, Athanasius F M Mare, Jennifer N Lynch, Mark J Miller, and Rob J
de Boer. Lymph node topology dictates T cell migration behavior. The Journal of
Experimental Medicine, 204(4):771–780, April 2007. PMID: 17389236.
[286] JacksonG. Egen, AntonioGigliotti Rothfuchs, CarlG. Feng, et al. Intravital Imaging
Reveals Limited Antigen Presentation and T Cell Effector Function in Mycobacterial
Granulomas. Immunity, 34(5):807–819, May 2011.
[287] H Mohri, A S Perelson, K Tung, et al. Increased turnover of T lymphocytes in HIV-
1 infection and its reduction by antiretroviral therapy. The Journal of Experimental
Medicine, 194(9):1277–1287, November 2001. PMID: 11696593.
[288] Justin Bailey, Joel N Blankson, Megan Wind-Rotolo, and Robert F Siliciano. Mecha-
nisms of HIV-1 escape from immune responses and antiretroviral drugs. Current Opinion
in Immunology, 16(4):470–476, August 2004. PMID: 15245741.
[289] B Ramratnam, J E Mittler, L Zhang, et al. The decay of the latent reservoir of
replication-competent HIV-1 is inversely correlated with the extent of residual viral repli-
cation during prolonged anti-retroviral therapy. Nature Medicine, 6(1):82–85, January
2000. PMID: 10613829.
301
[290] A T Haase. Population biology of HIV-1 infection: viral and CD4+ T cell demographics
and dynamics in lymphatic tissues. Annual Review of Immunology, 17:625–656, 1999.
PMID: 10358770.
[291] M A Stafford, L Corey, Y Cao, et al. Modeling plasma virus concentration during pri-
mary HIV infection. Journal of Theoretical Biology, 203(3):285–301, April 2000. PMID:
10716909.
[292] Becca Asquith, Christophe Debacq, Derek C Macallan, Luc Willems, and Charles R M
Bangham. Lymphocyte kinetics: the interpretation of labelling data. Trends in Im-
munology, 23(12):596–601, December 2002. PMID: 12464572.
[293] Narendra M Dixit, Martin Markowitz, David D Ho, and Alan S Perelson. Estimates
of intracellular delay and average drug efficacy from viral load data of HIV-infected
individuals under antiretroviral therapy. Antiviral Therapy, 9(2):237–246, April 2004.
PMID: 15134186.
[294] A S Perelson, A U Neumann, M Markowitz, J M Leonard, and D D Ho. HIV-1 dynamics
in vivo: virion clearance rate, infected cell life-span, and viral generation time. Science
(New York, N.Y.), 271(5255):1582–1586, March 1996. PMID: 8599114.
[295] Martin Markowitz, Michael Louie, Arlene Hurley, et al. A novel antiviral intervention
results in more accurate assessment of human immunodeficiency virus type 1 replication
dynamics and T-cell decay in vivo. Journal of Virology, 77(8):5037–5038, April 2003.
PMID: 12663814.
[296] Christian L Althaus, Anneke S De Vos, and Rob J De Boer. Reassessing the human
immunodeficiency virus type 1 life cycle through age-structured modeling: life span of
infected cells, viral generation time, and basic reproductive number, R0. Journal of
Virology, 83(15):7659–7667, August 2009. PMID: 19457999.
[297] Becca Asquith and Angela R McLean. In Vivo CD8+ T Cell Control of Immunodeficiency
Virus Infection in Humans and Macaques. Proceedings of the National Academy of
Sciences, 104(15):6365–6370, October 2007.
[298] Vitaly V. Ganusov, Nilu Goonetilleke, Michael K.P. Liu, et al. Fitness costs and diversity
of CTL response determine the rate of CTL escape during the acute and chronic phases
of HIV infection. J. Virol., pages JVI.00655–11, August 2011.
302
[299] Christian L. Althaus and Rob J. De Boer. Dynamics of Immune Escape during HIV/SIV
Infection. PLoS Comput Biol, 4(7):e1000103, July 2008.
[300] Vitaly V Ganusov and Rob J De Boer. Estimating Costs and Benefits of CTL Escape
Mutations in SIV/HIV Infection. PLoS Computational Biology, 2(3):e24, March 2006.
PMID: 16604188.
[301] Thorsten R Mempel, Mikael J Pittet, Khashayarsha Khazaie, et al. Regulatory T cells
reversibly suppress cytotoxic T cell function independent of effector differentiation. Im-
munity, 25(1):129–141, July 2006. PMID: 16860762.
[302] Huan Zheng, Bo Jin, Sarah E Henrickson, et al. How antigen quantity and quality deter-
mine T-cell decisions in lymphoid tissue. Molecular and Cellular Biology, 28(12):4040–
4051, June 2008. PMID: 18426917.
[303] A S Perelson, C A Macken, E A Grimm, L S Roos, and B Bonavida. Mechanism of
cell-mediated cytotoxicity at the single cell level. VIII. Kinetics of lysis of target cells
bound by more than one cytotoxic T lymphocyte. Journal of Immunology (Baltimore,
Md.: 1950), 132(5):2190–2198, May 1984. PMID: 6609191.
[304] D Zagury, J Bernard, P Jeannesson, N Thiernesse, and J C Cerottini. Studies on the
mechanism of T cell-mediated lysis at the single effector cell level. I. Kinetic analysis
of lethal hits and target cell lysis in multicellular conjugates. Journal of Immunology
(Baltimore, Md.: 1950), 123(4):1604–1609, October 1979. PMID: 314465.
[305] Aurelie Wiedemann, David Depoil, Mustapha Faroudi, and Salvatore Valitutti. Cyto-
toxic T lymphocytes kill multiple targets simultaneously via spatiotemporal uncoupling
of lytic and stimulatory synapses. Proceedings of the National Academy of Sciences of
the United States of America, 103(29):10985–10990, July 2006. PMID: 16832064.
[306] Jane C Stinchcombe and Gillian M Griffiths. Secretory mechanisms in cell-mediated cy-
totoxicity. Annual Review of Cell and Developmental Biology, 23:495–517, 2007. PMID:
17506701.
[307] Omer Dushek, Milos Aleksic, Richard J Wheeler, et al. Antigen potency and maximal
efficacy reveal a mechanism of efficient T cell activation. Science Signaling, 4(176):ra39,
2011. PMID: 21653229.
303
[308] Pablo A Gonzlez, Leandro J Carreo, Daniel Coombs, et al. T Cell Receptor Binding
Kinetics Required for T Cell Activation Depend on the Density of Cognate Ligand on
the Antigen-Presenting Cell. Proceedings of the National Academy of Sciences of the
United States of America, 102(13):4824–4829, March 2005.
[309] Michelle Krogsgaard, Nelida Prado, Erin J Adams, et al. Evidence that structural
rearrangements and/or flexibility during TCR binding can contribute to T cell activation.
Molecular Cell, 12(6):1367–1378, December 2003. PMID: 14690592.
[310] Milos Aleksic, Omer Dushek, Hao Zhang, et al. Dependence of T Cell Antigen Recog-
nition on T Cell Receptor-Peptide MHC Confinement Time. Immunity, 32(2):163–174,
February 2010.
[311] Adam S Chervin, Jennifer D Stone, Phillip D Holler, et al. The Impact of TCR-Binding
Properties and Antigen Presentation Format on T Cell Responsiveness. The Journal of
Immunology, 183(2):1166–1178, July 2009.
[312] Roland R Regoes, Andrew Yates, and Rustom Antia. Mathematical models of cytotoxic
T-lymphocyte killing. Immunology and Cell Biology, 85(4):274–279, June 2007. PMID:
17420769.
[313] A S Perelson and G I Bell. Delivery of lethal hits by cytotoxic T lymphocytes in mul-
ticellular conjugates occurs sequentially but at random times. Journal of Immunology
(Baltimore, Md.: 1950), 129(6):2796–2801, December 1982. PMID: 6982942.
[314] C A Macken and A S Perelson. A multistage model for the action of cytotoxic T lym-
phocytes in multicellular conjugates. Journal of Immunology (Baltimore, Md.: 1950),
132(4):1614–1624, April 1984. PMID: 6607943.
[315] S Isaaz, K Baetz, K Olsen, E Podack, and G M Griffiths. Serial killing by cytotoxic T
lymphocytes: T cell receptor triggers degranulation, re-filling of the lytic granules and
secretion of lytic proteins via a non-granule pathway. European Journal of Immunology,
25(4):1071–1079, April 1995. PMID: 7737276.
[316] Arun T Pores-Fernando, Roslyn A Bauer, Georjeana A Wurth, and Adam Zweifach.
Exocytic responses of single leukaemic human cytotoxic T lymphocytes stimulated by
agents that bypass the T cell receptor. The Journal of Physiology, 567(3):891 –903, 2005.
304
[317] N Signoret, A Pelchen-Matthews, M Mack, A E Proudfoot, and M Marsh. Endocytosis
and recycling of the HIV coreceptor CCR5. The Journal of Cell Biology, 151(6):1281–
1294, December 2000. PMID: 11121442.
[318] M K Slifka, F Rodriguez, and J L Whitton. Rapid on/off cycling of cytokine produc-
tion by virus-specific CD8+ T cells. Nature, 401(6748):76–79, September 1999. PMID:
10485708.
[319] D M Callewaert, D F Johnson, and J Kearney. Spontaneous cytotoxicity of cultured
human cell lines mediated by normal peripheral blood lymphocytes. III. Kinetic param-
eters. Journal of immunology (Baltimore, Md.: 1950), 121(2):710–717, August 1978.
PMID: 681753.
[320] O O Yang, S A Kalams, A Trocha, et al. Suppression of human immunodeficiency
virus type 1 replication by CD8+ cells: evidence for HLA class I-restricted triggering
of cytolytic and noncytolytic mechanisms. Journal of Virology, 71(4):3120–3128, April
1997. PMID: 9060675.
[321] Ralf Geiben-Lynn, Mischo Kursar, Nancy V Brown, et al. Noncytolytic Inhibition of
X4 Virus by Bulk CD8+ Cells from Human Immunodeficiency Virus Type 1 (HIV-
1)-Infected Persons and HIV-1-Specific Cytotoxic T Lymphocytes Is Not Mediated by
-Chemokines. Journal of Virology, 75(17):8306–8316, January 2001.
[322] M Nowak. HIV mutation rate. Nature, 347(6293):522, October 1990. PMID: 2215679.
[323] P Borrow, H Lewicki, X Wei, et al. Antiviral pressure exerted by HIV-1-specific cytotoxic
T lymphocytes (CTLs) during primary infection demonstrated by rapid selection of CTL
escape virus. Nature Medicine, 3(2):205–211, February 1997. PMID: 9018240.
[324] D A Price, P J Goulder, P Klenerman, et al. Positive selection of HIV-1 cytotoxic
T lymphocyte escape variants during primary infection. Proceedings of the National
Academy of Sciences of the United States of America, 94(5):1890–1895, March 1997.
PMID: 9050875.
[325] Nicola A Jones, Xiping Wei, Darren R Flower, et al. Determinants of human immunod-
eficiency virus type 1 escape from the primary CD8+ cytotoxic T lymphocyte response.
The Journal of Experimental Medicine, 200(10):1243–1256, November 2004. PMID:
15545352.
305
[326] Philip A Mudd, Adam J Ericsen, Benjamin J Burwitz, et al. Escape from CD8(+)
T cell responses in Mamu-B*00801(+) macaques differentiates progressors from elite
controllers. Journal of immunology (Baltimore, Md.: 1950), 188(7):3364–3370, April
2012. PMID: 22387557.
[327] R E Phillips, S Rowland-Jones, D F Nixon, et al. Human immunodeficiency virus ge-
netic variation that can escape cytotoxic T cell recognition. Nature, 354(6353):453–459,
December 1991. PMID: 1721107.
[328] S J Little, A R McLean, C A Spina, D D Richman, and D V Havlir. Viral dynam-
ics of acute HIV-1 infection. The Journal of Experimental Medicine, 190(6):841–850,
September 1999. PMID: 10499922.
[329] M A Nowak, A L Lloyd, G M Vasquez, et al. Viral dynamics of primary viremia and
antiretroviral therapy in simian immunodeficiency virus infection. Journal of virology,
71(10):7518–7525, October 1997. PMID: 9311831.
[330] S S Pilyugin and R Antia. Modeling immune responses with handling time. Bulletin of
mathematical biology, 62(5):869–890, September 2000. PMID: 11016088.
[331] Kevin O Saunders, Cavin Ward-Caviness, Robert J Schutte, et al. Secretion of MIP-1
and MIP-1 by CD8(+) T-lymphocytes correlates with HIV-1 inhibition independent of
coreceptor usage. Cellular immunology, 266(2):154–164, 2011. PMID: 21030011.
[332] Andrew Yates, Frederik Graw, Daniel L. Barber, et al. Revisiting Estimates of CTL
Killing Rates In Vivo. PLoS ONE, 2(12):e1301, December 2007.
[333] Qingsheng Li, Lijie Duan, Jacob D Estes, et al. Peak SIV replication in resting memory
CD4+ T cells depletes gut lamina propria CD4+ T cells. Nature, 434(7037):1148–1152,
April 2005. PMID: 15793562.
[334] Alexandra M. Ortiz, Nichole R. Klatt, Bing Li, et al. Depletion of CD4+ T cells abro-
gates post-peak decline of viremia in SIV-infected rhesus macaques. Journal of Clinical
Investigation, 121(11):4433–4445, November 2011.
[335] P J Goulder, R E Phillips, R A Colbert, et al. Late escape from an immunodominant
cytotoxic T-lymphocyte response associated with progression to AIDS. Nature Medicine,
3(2):212–217, February 1997. PMID: 9018241.
306
[336] C Y Koh, B R Blazar, T George, et al. Augmentation of antitumor effects by NK cell
inhibitory receptor blockade in vitro and in vivo. Blood, 97(10):3132–3137, May 2001.
PMID: 11342440.
[337] Caroline Sola, Pascale Andr, Cline Lemmers, et al. Genetic and antibody-mediated
reprogramming of natural killer cell missing-self recognition in vivo. Proceedings of the
National Academy of Sciences of the United States of America, 106(31):12879–12884,
August 2009. PMID: 19561305.
[338] S Valitutti, S Mller, M Dessing, and A Lanzavecchia. Different responses are elicited in
cytotoxic T lymphocytes by different levels of T cell receptor occupancy. The Journal of
experimental medicine, 183(4):1917–1921, April 1996. PMID: 8666949.
[339] Michael R Betts, David A Price, Jason M Brenchley, et al. The functional profile of
primary human antiviral CD8+ T cell effector activity is dictated by cognate peptide
concentration. Journal of immunology (Baltimore, Md.: 1950), 172(10):6407–6417, May
2004. PMID: 15128832.
[340] Matthew M Hufford, Taeg S Kim, Jie Sun, and Thomas J Braciale. Antiviral CD8+ T cell
effector activities in situ are regulated by target cell type. The Journal of Experimental
Medicine, 208(1):167–180, January 2011. PMID: 21187318.
[341] Joel E Cohen. Mathematics is biology’s next microscope, only better; biology is math-
ematics’ next physics, only better. PLoS biology, 2(12):e439, December 2004. PMID:
15597117.
[342] Shigeaki Ishii and Margaret Koziel. Immune responses during acute and chronic infection
with hepatitis C virus. Clinical Immunology, 128(2):147, 133, 2008.
[343] M Seiki, S Hattori, Y Hirayama, and M Yoshida. Human adult T-cell leukemia virus:
complete nucleotide sequence of the provirus genome integrated in leukemia cell DNA.
Proceedings of the National Academy of Sciences of the United States of America,
80(12):3618–3622, June 1983. PMID: 6304725.
[344] Yorifumi Satou, Jun-ichirou Yasunaga, Mika Yoshida, and Masao Matsuoka. HTLV-I
basic leucine zipper factor gene mRNA supports proliferation of adult T cell leukemia
cells. Proceedings of the National Academy of Sciences of the United States of America,
103(3):720–725, January 2006. PMID: 16407133.
307
[345] Seung Kew Yoon, Joon Yeol Han, Chul-Woo Pyo, et al. Association between human
leukocytes antigen alleles and chronic hepatitis C virus infection in the Korean popula-
tion. Liver International, 25(6):1122–1127, 2005.
[346] Alaa Hadhoud, Azza M Abdulaziz, Lubna Al Menawi, et al. The relationship between
HLA typing and HCV infection and outcome of renal transplantation in HCV positive
patients. Experimental and Clinical Transplantation: Official Journal of the Middle East
Society for Organ Transplantation, 1(1):19–25, June 2003. PMID: 15859903.
[347] Claudio Zavaglia, Cristina Bortolon, Gabriella Ferrioli, et al. HLA typing in chronic type
B, D and C hepatitis. Journal of Hepatology, 24(6):658–665, June 1996.
[348] N Kuzushita, N Hayashi, T Moribe, et al. Influence of HLA haplotypes on the clinical
courses of individuals infected with hepatitis C virus. Hepatology (Baltimore, Md.),
27(1):240–244, January 1998. PMID: 9425943.
[349] Alessandro Sette, Ward Fleri, Bjoern Peters, et al. A Roadmap for the Immunomics of
Category A-C Pathogens. Immunity, 22(2):155–161, February 2005.
[350] Anne J Ridley, Martin A Schwartz, Keith Burridge, et al. Cell migration: integrating
signals from front to back. Science (New York, N.Y.), 302(5651):1704–1709, December
2003. PMID: 14657486.
[351] Daniel L. Barber, E. John Wherry, and Rafi Ahmed. Cutting Edge: Rapid In Vivo
Killing by Memory CD8 T Cells. J Immunol, 171(1):27–31, July 2003.
[352] Nicole Frahm, Karina Yusim, Todd J Suscovich, et al. Extensive HLA class I allele
promiscuity among viral CTL epitopes. European Journal of Immunology, 37(9):2419–
2433, September 2007. PMID: 17705138.
[353] Thorsten R Mempel, Sarah E Henrickson, and Ulrich H Von Andrian. T-cell priming by
dendritic cells in lymph nodes occurs in three distinct phases. Nature, 427(6970):154–159,
January 2004. PMID: 14712275.
[354] Mark J Miller, Olga Safrina, Ian Parker, and Michael D Cahalan. Imaging the single cell
dynamics of CD4+ T cell activation by dendritic cells in lymph nodes. The Journal of
Experimental Medicine, 200(7):847–856, October 2004. PMID: 15466619.
308
[355] Derek C Macallan, Becca Asquith, Andrew J Irvine, et al. Measurement and modeling
of human T cell kinetics. European Journal of Immunology, 33(8):2316–2326, August
2003. PMID: 12884307.
309