Review of Hanna et.al

“Direct cell reprogramming is a

stochastic process amenable to

acceleration”

Nature, 462, 595-601 (2009)

Bradly Alicea

(http://www.msu.edu/~aliceabr/)

Main Predictions

I. Reprogramming of somatic cells is a continuous stochastic process where nearly

all somatic donor cells have the ability to give rise to iPS cells

* given continuous passaging and expression of four factors.

II. The extent of doxycycline selection or number of cell divisions achieved before

a given clonal population generates iPS cells varies widely.

III. The ‘elite’ model of reprogramming is not an accurate representation of the

reprogramming process.

* most lineage committed B cells or monocytes are able to generate iPS cells rather than only

a small fraction of putative somatic stem cells present in the donor cell population.

IV. Somatic cells reprogram with different latencies that cannot be predicted on the

basis of doxycycline selection or proliferation rate.

* supports the existence of a stochastic mechanism.

Outline of Model

Competing models for reprogramming (stochastic vs. deterministic):

1) stochastic: transformation occurs according to a variable latency.

* time from trigger to transformation is variable (cell cycle c = m transformations).

2) deterministic: transformation occurs according to a uniform latency.

* time from trigger to transformation is uniform.

Elite models argue that only a subset (1/n) of cells will reprogram (innate ability).

* elite model is independent of stochasticity vs. uniformity (independent mechanisms?)

Outline of Model (con’t)

Three components to the stochastic model of reprogramming:

1) basic assumption: given N cells, one-step reprogramming process occurs with

constant cell-intrinsic rate k. “Popcorn” metaphor.

2) latency: interval tp defined as time between tn (when first cell in population N is

reprogrammed) and tn+1 when daughter cells grow to reach detection threshold.

From “first settler” to viable colony.

3) scaling: at any tn, population of cells in a well, N(t), scales at rate which first

reprogramming event takes place (determines slope). Cumulative PDF:

Donor cells and initial population

1) B-cell lineage committed cells can be

efficiently cloned as single cells immediately

after isolation (homogeneous starting

population).

2) NGFP1 iPS cell line generated by infecting

fibroblasts from NANOG-GFP knock-in mice.

* NANOG-GFP fibroblasts infected with

doxycycline-inducible lentiviral vector with

four factors, injected into host blastocysts to

generate secondary chimeras.

* NGFP1-derived pre-B-cells from secondary

chimera single-cell sorted into individual wells

and exhibited high cloning efficiency (more

than 80%) under doxycyclene selection.

Donor cells and initial population

(con’t) 3) Cells monitored weekly for reactivation of NANOG-GFP (late event during

reprogramming).

* reprogramming efficiency defined as long-term potential of cell to generate

daughter cells (iPS).

Minimal threshold for detection:

* detection of > 0.5% for GFP+ cells per well (allowed for stable derivation of

GFP+ iPS cells upon doxycycline withdrawl.

Conversion:

* GFP+ cells detected at 2 weeks, 92% of wells had GFP+ cells at 18 weeks.

* neither transgene expression levels nor increased proliferation rate underlie well-

to-well variation in latency of reprogramming.

Assumptions

Example: MacArthur et.al, PLoS One, 3(8), e3086.

Computational approaches to gene expression include adding

noise (stochastic element) to model.

* non-specific noise in expression of four factors, other genes can trigger

reprogramming.

Black function: Oct4, Sox2. Blue function: NANOG. Red function: lineage-

specific master genes, σ: parameter value for amplitude of noise (same for

every gene).

Hypothesis: Cellular reprogramming can be driven by noise.

* noise in the form of transcriptional variance and other

stochastic processes can trigger, drive reprogramming process in

vitro.

* presence of Oct4, Sox2, and NANOG suppress differentiation

genes and activate stem cell genes (modules).

Assumptions (con’t) There are many different potential outcomes

of reprogramming (iPS, piPS):

Stemness = what do the diversity of induced

stem cells types have in common?

* pluripotency, gene regulation profiles.

* multi-stability (ability to change state in

response to environmental, viral cues).

Switch that governs this transformation

may be stochastic:

* Two factors activate their own expression,

mutually repress each other (all-or-nothing

response).

* Weiner process (additive) = stochasticity.

At σ = 0, switch between fate at rate r. MacArthur et.al Nat Rev Cell Biology, 10, 673 (2009).

Assumptions (con’t) Stemness is maintained by a network centered on NANOG, which is controlled by

the four factors:

Left: protein-protein interaction network of genes upregulated when cell is in a

“stem” state (based on ChIP experiments).

Right: genes on A side of figure downregulated, genes on B side upregulated when

middle box (3 of 4 factors + NANOG) circuit is activated.

A B

Genetic Perturbation Tests Hypothesis: survival curves (time evolution) of various genetic perturbations should

be statistically significant.

Null hypothesis: the survival curves do not differ across groups.

* tested using Mantel-Cox (non-parametric) test.

* latency datasets were fit to several potential univariate PDFs using MLE.

* χ2 measure was used to assess best fit. Gamma distribution was best.

Simulation:

* hybrid model used to evolve size of β-cell and iPS population.

* Gillespie algorithm (stochastic) applied to Poisson growth dynamics for

small population sizes, transition dynamics.

* deterministic evolution of population sizes: t = .001 when the probability

of generating a new cell in time t exceeded 0.1 (mutation rate).

Genetic Perturbation Tests (con’t)

Population-averaged doubling times (td) dervied from overexpression lines on Doxycycline. Rescaling time by td = number of cell divisions that occur during latency period (each cell division – opportunity to transform).

Genetic Perturbation Tests (con’t)

Introduction of defined genetic perturbations:

* p53 inhibition (will it influence reprogramming efficiency of selected iPS cells?).

* p53 reduces initial apoptosis after infection.

* used NGFP1 iPS cells (siRNA for p53 introduced - lentivirus).

Effect of p53 on proliferation rate = how many cell divisions occur for NGFP1 and

NGFP1-p53 populations during latency period?

Effective population size:

Neff is the number of cells by which the culture expands

between passages. Implicit measure of population

doubling.

e.g. t = 0 (passage 1) and t = 1 (passage 2).

* rate of population expansion (potential).

Genetic Perturbation Tests (con’t)

Scaling separates out effects of intrinsic ability to reprogram and contributions of

population size.

* rate of cell division ~ population size ~ reprogramming ability.

Mechanisms behind acceleration Two modes (with a switch between them):

A) “cell-division, rate-dependent” mode:

* cumulative prob. for successful reprogramming higher.

* can be achieved earlier, ~ augmentation in cell division rate.

B) “cell-division, rate-independent” mode:

* reprogramming acceleration occurs over a lower average # of cell divisions.

* different genetic perturbations can favor one mode or the other.

1) cell division could amplify # of daughter cells from

differentiated cells. Each resulting cell can become iPS

cell at prob(p).

2) nuclear changes at cell division ~ acquisition of

epigenetic markers which stabilize pluripotency.

Potentially interesting

evolutionary dynamics

w.r.t small population

size and rate of

expansion.

Bistable

switch

Mechanisms behind acceleration

(con’t)

Example of scaling (more instances of transformation with more cells):

N = 103 time to reach > 90% reprogrammed cells in well longer.

N = 106 time to reach > 90% reprogrammed cells in well shorter.

Mechanisms behind acceleration

(con’t) Simplest scenario:

* one-step rate-limiting transition characterized by a cell-intrinsic rate, which does

not describe reprogramming behavior before and soon after transgene induction.

Perhaps there are multiple modes of reprogramming acceleration:

* arrive at this model by considering: 1) closely monitoring transgene induction, 2)

plating efficiency, 3) cell proliferation, 4) changes in population size across

experiment.

There is a yet-to-be-defined rate-limiting, continuous stochastic mechanism

(according to model):

* function of cell division before fully reprogrammed state is attained.

* results support “all with variable latency” model (neither “elite” nor

“deterministic”).

* might iPS cells arise preferentially from a precursor (progenitor or adult

SC)?

Conclusions Conclusions:

* reprogramming of somatic cells is a continuous stochastic process, all cells can

give rise to iPS cells (given enough time).

* latency = 8-10d (minimum). However, time of doxycycline exposure, # of cell

divisions before iPS cells generated varies greatly w.r.t. experimental conditions.

* NANOG overexpression accelerates reprogramming kinetics via cell-intrinsic

mechanisms. Independent of altered cell proliferation rate.

Issues:

1) no decay function (leakiness).

2) are there other appropriate

computational techniques (discrete,

evolutionary dynamics)?

3) should this a one-step or multi-step

model (intermediate cell types)?