NEURODEVELOPMENT

Reprogramming of avian neural crestaxial identity and cell fateMarcos Simoes-Costa1,2 and Marianne E. Bronner1

Neural crest populations along the embryonic body axis of vertebrates differ indevelopmental potential and fate, so that only the cranial neural crest can contribute to thecraniofacial skeleton in vivo. We explored the regulatory program that imbues the cranialcrest with its specialized features. Using axial-level specific enhancers to isolate andperform genome-wide profiling of the cranial versus trunk neural crest in chick embryos,we identified and characterized regulatory relationships between a set of cranial-specifictranscription factors. Introducing components of this circuit into neural crest cells ofthe trunk alters their identity and endows these cells with the ability to give rise tochondroblasts in vivo. Our results demonstrate that gene regulatory circuits that supportthe formation of particular neural crest derivatives may be used to reprogram specificneural crest–derived cell types.

Neural crest cells are characterized by theirmultipotency and migratory ability. Dur-ing embryonic development, the neuralcrest differentiates intomultiple cell types,including chondrocytes and osteocytes,

melanocytes, and neurons and glia of the peri-pheral nervous system (1, 2). Neural crest stemcells are retained postnatally in the skin and peri-pheral nerves, providing a potential target forreplacement therapy in regenerative medicine(3, 4). However, not all neural crest populationsalong the body axis are alike. Quail-chick graftingexperiments demonstrated that the cranial andtrunk neural crest differ in developmental po-tential: Whereas the cranial neural crest formsmuch of the craniofacial skeleton, the trunk crestfails to contribute to skeletal lineages, even whengrafted in vivo to the head (1). Approaches forengineering and replacing specific cell types de-pend on a better understanding of the molecularmechanisms that underlie the establishment ofspecific cell types during embryonic development.We took advantage of differences in neural crestsubpopulations (1, 5–7) to identify the regula-tory circuit that controls commitment of thecranial neural crest to a chondrocytic fate.Expression of neural crest specifier genes such

as FoxD3 and Sox10 is controlled by enhancersspecific to particular axial levels, driving onsetof their transcription in either the head or trunkneural crest but not both (8, 9). The enhancersare activated by different inputs, suggesting thatneural crest specification is driven by distinct ge-netic programs in different subpopulations (2, 9).In order to identify the transcriptional programthat endows the cranial neural crest with itsability to give rise to ectomesenchyme (the car-tilage and bone of the face), we used the FoxD3enhancers NC1 and NC2 (9) (Fig. 1, A to C), ac-

tive in the cranial and trunk neural crest, respec-tively, to isolate pure populations of neural crestcells for comparative transcriptional profiling.Embryos were electroporated with expressionvectors driving green fluorescent protein (GFP)expression under the control of these enhancers,and early-migrating cranial and trunk neural crestcells were obtained by fluorescence-activatedcell sorting (FACS) (10). RNA-seq analysis com-paring these two populations identified 216genes that were enriched in the cranial neural

crest relative to the trunk (Fig. 1, D and E, anddatabase S1), including 16 transcription factors(listed in Fig. 1F). We confirmed the expressionof these regulators in the cranial neural crest byin situ hybridization; whereas six genes wereexpressed throughout the cranial neural crest(fig. S1, A to F), the remainder were detected inspecific subsets of cells (fig. S1, G to L). Of these,we focused on the first group, which were ex-pressed in all cranial neural crest cells, includ-ing Brain-SpecificHomeobox Protein 3C (Brn3c),LIMHomeobox Protein 5 (Lhx5), Diencephalon/Mesencephalon Homeobox 1 (Dmbx1), Transcrip-tion Factor AP-2 Beta (Tfap2b), SRY Box 8 (Sox8),and the V-Ets Avian Erythroblastosis Virus E26Oncogene Homolog 1 (Ets1).Analysis of the spatiotemporal expression of

these cranial-specific regulators demonstratedthat Brn3c, Lhx5, and Dmbx1 were first detectedin the anterior regions of gastrula-stage embryosat Hamburger Hamilton stage 4 (HH4) and per-sisted through stages of neural crest specification(Fig. 2, A and E). These early cranial-specific geneswere down-regulated after the neural crest delam-inated from the neural tube. The onset of Tfap2b,Sox8, and Ets1 expression was observed later atHH7 and HH8, in neural crest progenitors re-siding within the cranial neural folds (Fig. 2, Band E). These genes were maintained in themigratory neural crest cells during later stages ofdevelopment (HH10 to 14; Fig. 2, B and E). Co-localization of neural plate border markers MshHomeobox 1 (Msx1) or Paired Box 7 (Pax7) with

1570 24 JUNE 2016 • VOL 352 ISSUE 6293 sciencemag.org SCIENCE

1Division of Biology and Biological Engineering, CaliforniaInstitute of Technology, Pasadena, CA 91125, USA.2Department of Molecular Biology and Genetics, CornellUniversity, Ithaca, NY 14853-2703, USA.*Corresponding author. Email: mbronner@caltech.edu

216 genes

Fold enrichment

Lhx5En1

FoxI2Dmbx1

Dlx6Otx2

Lmo3Sox8RbpjlEya2Rfx5

Grhl2Tfap2bZnf160

RcorGanf

0 2 4 6 8

5

-30 30

10

15

327 genes

Trunk NC Cranial NC

Cra

nia

l

NC1:eGFP

HH9

HH14

NC2:eGFP

Tru

nk

Enzymes

Transportersand carrierproteins

Nucleicacid metabolism

CellAdhesion

Receptors

Cytoskeletalproteins

Others Signalingmolecules

Transcriptionfactors 8%

Sta

tistic

al s

igni

fican

ce-lo

g(pV

alue

)

Log2 (Fold Enrichment)

Fig. 1. Identification of cranial-specific regulators by comparative transcriptomics. (A) Diagramdepicting dorsal view of chick embryo. (B and C) Embryos electroporated with FoxD3 axial-specificenhancers NC1 and NC2, active in cranial and trunk neural crest (NC), respectively. (D) Comparativetranscriptome analysis of FACS-sorted cranial and trunk neural crest populations identified 216 craniallyenriched genes. (E) Summary of gene ontology analysis for the cranial-enriched genes. (F) Enrichmentlevels of transcription factors expressed in the cranial neural crest.

RESEARCH | REPORTSon A

ugust 23, 2020

http://science.sciencemag.org/

Dow

nloaded from

Brn3c, Lhx5, and Dmbx1 showed that the latterare expressed by an anterior subset of the neuralcrest progenitors (Fig. 2C). To verify that thecranial regulators mark the territory that con-tains cranial neural crest precursors, we ana-lyzed the fate map of the neural plate border atthe three-somite stage, using focal injections ofa vital lipophilic dye (CM-DiI) to label cells alongthe anterior-posterior axis. The injected embryoswere cultured until the 12-somite stage (HH11-),when the labeled cellular progeny were scoredwith respect to their fate as cranial or vagal/trunk neural crest cells. The results show thatthe domain of expression of the early regulators

(Brn3c, Lhx5, and Dmbx1) demarcates the ter-ritory that contains the progenitors of the cranialneural crest in the early neurula (HH8-) (Fig. 2D).We asked whether cranial-specific regulators

are part of a transcriptional circuit that underliescranial identity by knocking down each regulatorindividually and assaying for changes in expres-sion levels of the other five genes. This was donevia bilateral electroporations (11), with controltransfections on the left side and function-blockingmorpholinos or dominant negative constructs onthe right side of the same embryo (Fig. 3, A to D).Transfected embryos were analyzed by in situhybridization (Fig. 3, A to H) and quantitative

polymerase chain reaction (qPCR) (Fig. 3I) foreffects on putative targets; loss of the target genein the experimental side of the embryo indicatedthe existence of a regulatory link between thetwo transcription factors. The diagram in Fig. 3Jcontains interactions confirmed by both qPCRand in situ hybridization.Morpholino knockdownefficiency was validated by in vivo translation-blocking assays (fig. S2).By testing 25 putative regulatory links, we

found that the early and late cranial-specificgenes constitute different hierarchical levels of agene regulatory network (Fig. 3J). Brn3c, whichis placed at the top of the circuit, is necessary for

SCIENCE sciencemag.org 24 JUNE 2016 • VOL 352 ISSUE 6293 1571

Fig. 2. Spatial and temporal expression of cra-nial neural crest transcription factors. (A and B)Dorsal views of embryos after in situ hybridizationfor cranial neural crest–specific transcription fac-tors reveals expression at early (A) or later (B)stages. (C) Double in situ hybridization revealsthat cranial regulator Dmbx1 is expressed in theanterior neural plate border, whereas Msx1 is ex-pressed along the entire neural axis. (D) Fate mapof neural crest progenitors (red and green dots)at stage HH8- confirms cranial specific expressionof Dmbx1 (purple). (E) Diagram summarizing thetiming of the expression of early and late cranialregulators.

Msx1Dmbx1

I

M

L

O

NR

V

Z

A’

W

X

P

S

B’

T

HH8-HH8-

Trunk/Vagal

Cranial

Dmbx1 expression

50% 100%0%

AH

D

E

F

Q

GU

B

C’

C

Dmbx1

Dmbx1

HH6 HH6 HH6

HH8HH8HH8

Lhx5

Lhx5

Brn3c

Brn3c

HH9 HH9

Sox8 Tfap2b

HH10HH10

Sox8 Tfap2b Ets1

Ets1

Ear

ly c

rani

al N

C g

enes

Late

cra

nial

NC

gen

es

HH10

HH9

A B C D E F G H I J

1

2

3

4

5

6

7

8

9

10

11

12

Dmbx1

Earlycranial NCgenes

Latecranial NCgenes

HH embryonic stage

Tfap2b

Ets1Sox8

Lhx5Brn3c

4 5 6 7 7+ 8- 8 8+ 9- 9 9+ 10- 1010+11-

Fig. 3. Cranial-specific transcriptional circuit un-derlying neural crest axial identity. (A to D) Whole-mount dorsal views of embryos after morpholino(Mo) targeted to the indicated transcription fac-tor was transfected to the right side (green) andcontrol morpholino (CoMo) to the left side (blue)of each embryo. (E to H) Same embryos as aboveafter in situ hybridization for the indicated down-stream transcription factor; blue arrows indicatetranscript down-regulation. (I) Comparing controlto loss-of-function (LOF) neural folds by qPCR re-veals differential regulation of downstream targetswith significant changes indicated by an asterisk(Student’s t test, P < 0.05). (J) Diagram summariz-ing cranial-specific gene regulatory circuit deline-ated by functional assays. (K) ChIP demonstratesdirect association of cranial specific transcriptionfactors with promoters of their downstream targets.No enrichment was observed for intergenic negativecontrol regions (NCRs) or when the procedure wasperformed with mock-transfected embryos (seematerials and methods for more information). IGG,immunoglobulin G. Error bars in (I) and (K) representstandard deviation.

Lhx5Mo

Lhx5Mo

Dmbx1Mo

Tfap2bMoCoMo CoMo CoMo CoMo

Tfap2b Tfap2b Ets1Sox8

ControlNeuralFold

LOFNeuralFold

Brn3c

Dmbx1

Tfap2b

Ets1

Lhx5

Neural plate border

Premigratory NC

Migratory NC

Ectomesenchymaldifferentiation

Sox8

Fol

d E

nric

hmen

t ove

r IG

G

0

1

2

3

4

6

Brn3c BrnIP:

Target: Dmbx1 Tfap2b Tfap2b Ets1

Lhx5 Dmbx1

IPMock

NCR

Tfap2b

Loss of function analysis TF-Promoter interactions

Nor

mal

ized

RN

A A

bund

ance

Tfap2b LOF

Brn3c LOF

Sox8 LOF

Lhx5 LOF Dmbx1 LOF

*

0.0

0.0

0.5

0.5

1.0

1.0

Lhx5

Brn3c

Dmbx1

Tfap2b

Sox8

Ets

RESEARCH | REPORTSon A

ugust 23, 2020

http://science.sciencemag.org/

Dow

nloaded from

the activation of Dmbx1 in the anterior neuralplate border (Fig. 3I and fig. S3). Subsequently,Lhx5 andDmbx1 drive expression of Tfap2b andSox8 in the dorsal neural folds (Fig. 3, A to C, Eto G, and I). Finally, Tfap2b activates expressionof Ets1 as the neural crest becomes specified(Fig. 3, D, H, and I). Chromatin immunopreci-pitation (ChIP) experiments performed in micro-dissected neural crest cells showed the associationof the cranial-specific transcription factors withpromoters of predicted downstream target genes,suggesting that these regulatory links are direct(Fig. 3K and fig. S4).Sox8, Tfap2b, and Ets1 are all retained in the

migrating cranial crest cells as they move ven-trally to give rise to the facial mesenchyme dur-ing stages HH10 to 14. To investigate whetherthese genes play a role in the differentiation ofneural crest cells into chondroblasts, we assayedthe effects of disrupting the terminal module ofthe circuit on the expression of markers of chon-drocytic differentiation. We found that Ets1 wasrequired for the expression of ALX Homeobox 1(Alx1, also known as Cartilage Paired-ClassHomeoprotein 1) in the facial mesenchyme (fig.S5), indicating a link between cranial identityand chondrocytic differentiation. Thus, cranial-specific regulators are part of a transcriptionalcircuit that conveys regulatory information fromthe anterior neural plate border to the late mi-gratory neural crest.

To test whether this cranial-specific regula-tory circuit could be used to manipulate neuralcrest identity, we employed neural crest axial-specific enhancers as reporters of axial level iden-tity. We electroporated expression constructs inthe trunkneural tube of stageHH10 embryos, andfound that transfection of the late cranial-specificfactors (Sox8, Tfap2b, and Ets1) robustly acti-vated the cranial enhancer Sox10E2 (9, 11) inthe trunk neural crest [n = 15 out of 15 (15/15)embryos] (Fig. 4, A to E), consistent with a shiftfrom trunk to cranial identity. Other axial-specific enhancers were similarly affected; trunk-specific enhancers NC2 and Sox10E1 were re-pressed after electroporation of the late factors(fig. S6). Early cranial-specific factors (Brn3c,Lhx5, and Dmbx1), or individual late factors,were unable to activate cranial enhancers in thetrunk. To identify changes in the regulatory stateof the reprogrammed trunk neural crest, we iso-lated transfected cells by FACS and analyzed theirexpression profile by qPCR, focusing on tran-scription factors involved in craniofacial differ-entiation. The results revealed elevated expressionof chondrocytic genes Runt-Related TranscriptionFactor 2 (Runx2) and Alx1, in the trunk Sox10E2+

cells as compared with native trunk crest (Fig. 4F).The reprogrammed trunk Sox10E2+ cells also dis-played a loss of genes enriched in the trunkneural crest, such as Developing Brain Homeo-box 1 (Dbx2) and Hairy And Enhancer Of Split 6

(Hes6) (Fig. 4G). This confirmed that the re-programmed cells adopt a cranial-like expressionprofile, and raised the possibility that these cellsmight display augmented chondrocytic potential.Finally, we tested whether this cranial neural

crest circuit could reprogram not only enhanceractivity and expression of axial-specific neuralcrest genes, but also cell fate, so that repro-grammed trunk neural crest cells could differen-tiate into craniofacial cartilage. We cotransfectedthree constructs encoding late transcription fac-tors Sox8, Tfap2b, and Ets1 into the posteriorepiblast of HH5 chicken embryos transgenic forGFP by electroporating DNA in the region pos-terior to the Hensen’s node. The GFP+ trunk neu-ral folds then were microdissected at HH11 andimmediately transplanted to the cranial regionsof HH9 wild-type chick embryos (fig. S7). Thegrafted embryos were incubated until host em-bryonic day 7 (E7), by which time endogenouscartilage cells have differentiated. The fate ofdonor tissue was assayed using markers for neu-ronal, melanocytic, and chondrocytic differenti-ation. By E7, wild-type (host) and reprogrammed(donor) trunk neural crest migrated to the prox-imal part of the jaw. As observedwith chick-quailchimeras (5–7), we found that wild-type ormock-transfected trunk neural crest cells grafted intothe cephalic region gave rise to neurons andmelanocytes but were unable to differentiateinto chondroblasts (n = 0/5 embryos) (Fig. 4, H

1572 24 JUNE 2016 • VOL 352 ISSUE 6293 sciencemag.org SCIENCE

Sox10E2

Sox10E2

Sox10E2

+ Sox8+ Tfap2b

+ Ets1

+RFP

WT

HH18

HH18

REP

1010

10

10

10

Are

a::G

FP

Area::7AAD10

10

10

10

10

0

0

1

1

2

2

3

3

4

4

1010

10

10

10

Are

a::G

FP

Area::7AAD10

10

10

10

10

0

0

1

1

2

2

3

3

4

4

GFP+

GFP+

Rep. graft

Col9a Col9a

Nor

mal

ized

exp

ress

ion

Nor

mal

ized

exp

ress

ion

Runx2

Dbx2 Hes6

Alx1

WT Trunk NCRep Trunk NC

Cranial NC

0

0

0.8

1.5

0.4

0.2

0.5

0.6

1.0

1.0

WT

Gra

ftR

EP

Gra

ft +

Sox

8, T

fap2

b, E

ts1

Col9a Col9a

Fig. 4. Reprogramming of neural crest axial identity and fate. (A) Diagramof chick embryo electroporated with the Sox10E2 enhancer, active only inmigrating cranial neural crest. (B) Transfection of trunk neural crest cells witha control RFP expression vector shows no Sox10E2 activity. (C) Repro-gramming of trunk neural crest cells with cranial-specific regulators Sox8,Tfap2b, and Ets1 results in robust activity of the cranial Sox10E2 enhancerin the trunk region (n = 15/15 embryos). (D and E) Flow cytometry anal-ysis of dissociated embryonic trunks shows a large number of Sox10E2+

trunk neural crest cells after reprogramming. (F and G) Reprogrammed(Rep) trunk neural crest display increased expression of the chondrocytic

genes Runx2 and Alx1, whereas trunk genes Dbx2 and Hes6 are stronglydown-regulated. Error bars represent standard deviation. (H and I) Histo-logical sections of E7 embryonic heads show that wild-type (WT) trunk neuralcrest cells (GFP+, green) fail to form cartilage (Col9a+, red) after transplan-tation to the cranial region (n = 0/5 embryos). (J and K) Inset of (H) and (I),showing the absence of GFP+ chondrocytes. (L and M) Reprogrammed trunkneural crest cells (expressing Sox8, Tfap2b, and Ets1) from GFP donorembryos transplanted to the head form ectopic cartilage nodules. (N and O)Inset of (L) and (M), showing chondrocytes derived from trunk neural crestcells (GFP+ and Col9a+).

RESEARCH | REPORTSon A

ugust 23, 2020

http://science.sciencemag.org/

Dow

nloaded from

to K). The same result was observed with trunkneural crest transfected with the early cranial-specific factors (n = 0/6 embryos). Reprogrammedtrunk neural crest, however, acquired chondrogenicpotential and formed ectopic cartilage nodules(n = 4/7 embryos) (Fig. 4, L to O) in the proximaljaw. Thus, introducing components of the cranial-specific transcriptional circuit is sufficient to re-program trunk neural crest cells and to drive themto adopt an additional cartilaginous fate. Theseresults definitively show that the cranial-specificregulatory circuit (Fig. 3J) we have defined conferschondrocytic potential to the trunk neural crest.The development and differentiation of neu-

ral crest cells are controlled by a complex generegulatory network, composed of transcriptionfactors, signalingmolecules, and epigeneticmod-ifiers (12, 13). We have expanded the knowncranial neural crest gene regulatory network byidentifying transcriptional interactions specificto the cranial crest and absent from other sub-populations. By linking anterior identity in thegastrula to the expression of drivers of chondro-cytic differentiation, we have identified a cranial-specific circuit (Fig. 3J) that endows the neuralcrest with its potential to differentiate into thecraniofacial skeleton of vertebrates. Our resultshighlight how transcriptional circuits can be re-wired to alter progenitor cell identity and fateduring embryonic development.

REFERENCES AND NOTES

1. N. Le Douarin, The Neural Crest (Cambridge Univ. Press, 1982).2. M. Simões-Costa, M. E. Bronner, Genome Res. 23, 1069–1080 (2013).3. T. Uesaka, M. Nagashimada, H. Enomoto, J. Neurosci. 35,

9879–9888 (2015).4. V. Dyachuk et al., Science 345, 82–87 (2014).5. C. S. Le Lièvre, N. M. Le Douarin, J. Embryol. Exp. Morphol. 34,

125–154 (1975).6. C. S. Le Lievre, G. G. Schweizer, C. M. Ziller, N. M. Le Douarin,

Dev. Biol. 77, 362–378 (1980).7. P. Y. Lwigale, G. W. Conrad, M. Bronner-Fraser, Development

131, 1979–1991 (2004).8. P. Betancur, M. Bronner-Fraser, T. Sauka-Spengler, Proc. Natl.

Acad. Sci. U.S.A. 107, 3570–3575 (2010).9. M. S. Simões-Costa, S. J. McKeown, J. Tan-Cabugao,

T. Sauka-Spengler, M. E. Bronner, PLOS Genet. 8, e1003142 (2012).10. M. Simões-Costa, J. Tan-Cabugao, I. Antoshechkin,

T. Sauka-Spengler, M. E. Bronner, Genome Res. 24, 281–290 (2014).11. M. Simões-Costa, M. Stone, M. E. Bronner, Dev. Cell 34,

544–554 (2015).12. M. Simões-Costa, M. E. Bronner, Development 142, 242–257 (2015).13. S. A. Green, M. Simoes-Costa, M. E. Bronner, Nature 520,

474–482 (2015).

ACKNOWLEDGMENTS

We thank J. Tan-Cabugao, M. Stone, B. Jun, and D. S. E. Koo fortechnical assistance. The Caltech Millard and Muriel Jacobs Geneticsand Genomics Laboratory provided sequencing and bioinformaticssupport. We are indebted to D. Perez, K. Beadle, and R. Diamond forcell-sorting assistance at the Caltech Flow Cytometry Cell SortingFacility. We also thank M. Barembaum for the Sox8 and Tfap2bexpression constructs. This work was supported by NIH grantsDE024157 and HD037105 to M.E.B. M.S.-C. was funded by a fellowshipfrom the Pew Fellows Program in Biomedical Sciences and by NIH grantK99DE024232. The supplementary materials contain additional data.

SUPPLEMENTARY MATERIALS

www.sciencemag.org/content/352/6293/1570/suppl/DC1Materials and MethodsFigs. S1 to S7References (14–26)Database S1

18 January 2016; accepted 23 May 201610.1126/science.aaf2729

ETHICS

The social dilemma ofautonomous vehiclesJean-François Bonnefon,1 Azim Shariff,2* Iyad Rahwan3†

Autonomous vehicles (AVs) should reduce traffic accidents, but they will sometimes haveto choose between two evils, such as running over pedestrians or sacrificing themselvesand their passenger to save the pedestrians. Defining the algorithms that will help AVsmake these moral decisions is a formidable challenge. We found that participants in sixAmazon Mechanical Turk studies approved of utilitarian AVs (that is, AVs that sacrificetheir passengers for the greater good) and would like others to buy them, but they wouldthemselves prefer to ride in AVs that protect their passengers at all costs. The studyparticipants disapprove of enforcing utilitarian regulations for AVs and would be less willingto buy such an AV. Accordingly, regulating for utilitarian algorithms may paradoxicallyincrease casualties by postponing the adoption of a safer technology.

The year 2007 saw the completion of the firstbenchmark test for autonomous driving inrealistic urban environments (1, 2). Sincethen, autonomous vehicles (AVs) such asGoogle’s self-driving car covered thousands

of miles of real-road driving (3). AVs have thepotential to benefit the world by increasing traf-fic efficiency (4), reducing pollution (5), and elim-inating up to 90% of traffic accidents (6). Not allcrashes will be avoided, though, and some crasheswill require AVs tomake difficult ethical decisionsin cases that involve unavoidable harm (7). Forexample, the AV may avoid harming several pe-destrians by swerving and sacrificing a passerby,or the AV may be faced with the choice of sacri-ficing its own passenger to save one or morepedestrians (Fig. 1).Although these scenarios appear unlikely, even

low-probability events are bound to occur withmillions of AVs on the road. Moreover, even ifthese situations were never to arise, AV program-ming must still include decision rules about whatto do in such hypothetical situations. Thus, thesetypes of decisions need be made well before AVsbecome a global commodity. Distributing harm isa decision that is universally considered to fallwithin the moral domain (8, 9). Accordingly, thealgorithms that control AVs will need to embedmoral principles guiding their decisions in situa-tions of unavoidable harm (10). Manufacturersand regulators will need to accomplish three po-tentially incompatible objectives: being consistent,not causing public outrage, and not discouragingbuyers.However, pursuing these objectives may lead

to moral inconsistencies. Consider, for example,the case displayed in Fig. 1A, and assume that

the most common moral attitude is that the AVshould swerve. This would fit a utilitarian moraldoctrine (11), according to which themoral courseof action is to minimize casualties. But considerthen the case displayed in Fig. 1C. The utilitariancourse of action, in that situation, would be forthe AV to swerve and kill its passenger, but AVsprogrammed to follow this course of actionmightdiscourage buyers who believe their own safetyshould trump other considerations. Even thoughsuch situations may be exceedingly rare, theiremotional saliency is likely to give them broadpublic exposure and a disproportionate weightin individual and public decisions about AVs. Toalign moral algorithms with human values, wemust start a collective discussion about the ethicsof AVs—that is, the moral algorithms that we arewilling to accept as citizens and to be subjectedto as car owners. Thus, we initiate the data-drivenstudy of driverless car ethics, inspired by themeth-ods of experimental ethics (12).We conducted six online surveys (n = 1928 total

participants) between June and November 2015.All studies were programmed on Qualtrics surveysoftware and recruited participants (U.S. resi-dents only) from the Amazon Mechanical Turk(MTurk) platform, for a compensation of 25 centseach. Studies described in the experimental ethicsliterature largely rely onMTurk respondents, withrobust results, even though MTurk respondentsare not necessarily representative of the U.S. pop-ulation (13, 14). A possible concern with MTurkstudies is that some participants may already befamiliar with testingmaterials, particularlywhenthesematerials are used bymany research groups.However, this concern does not apply to our test-ing materials, which have never been used in apublished MTurk study to date.In all studies, participants provided basic demo-

graphic information. Regression analyses (see sup-plementary materials) showed that enthusiasmfor self-driving cars was consistently greater foryounger, male participants. Accordingly, all sub-sequent analyses included age and sex as co-variates. The last item in every study was an easyquestion (e.g., howmany pedestrianswere on the

SCIENCE sciencemag.org 24 JUNE 2016 • VOL 352 ISSUE 6293 1573

1Toulouse School of Economics, Institute for Advanced Study inToulouse, Center for Research in Management, CNRS,University of Toulouse Capitole, Toulouse, France. 2Departmentof Psychology, University of Oregon, Eugene, OR 97403, USA.3The Media Lab, Massachusetts Institute of Technology,Cambridge, MA 02139, USA.*Present address: Department of Psychology and Social Behavior,4201 Social and Behavioral Sciences Gateway, University ofCalifornia, Irvine, Irvine, CA 92697, USA. †Corresponding author.Email: irahwan@mit.edu

RESEARCH | REPORTSon A

ugust 23, 2020

http://science.sciencemag.org/

Dow

nloaded from

Reprogramming of avian neural crest axial identity and cell fateMarcos Simoes-Costa and Marianne E. Bronner

DOI: 10.1126/science.aaf2729 (6293), 1570-1573.352Science 

, this issue p. 1570Scienceembryonic neural crest cells carry subspecialties that are defined but malleable.Ectopic expression of some of these factors caused trunk neural crest cells to function as cranial neural crest cells. Thus,expression of a handful of transcription factors identifies cranial neural crest cells as distinct from trunk neural crest cells.

show, studying chick embryos, that not all neural crest cells are alike. Theet al.and peripheral nerves. Simoes-Costa Neural crest cells wander far and wide through the developing vertebrate embryo to build tissues such as the jaw

Versatile embryonic neural crest cells

ARTICLE TOOLS http://science.sciencemag.org/content/352/6293/1570

MATERIALSSUPPLEMENTARY http://science.sciencemag.org/content/suppl/2016/06/22/352.6293.1570.DC1

REFERENCES

http://science.sciencemag.org/content/352/6293/1570#BIBLThis article cites 24 articles, 9 of which you can access for free

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAAS.ScienceScience, 1200 New York Avenue NW, Washington, DC 20005. The title (print ISSN 0036-8075; online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Copyright © 2016, American Association for the Advancement of Science

on August 23, 2020

http://science.sciencem

ag.org/D

ownloaded from