复旦大学：《人类进化 Human Evolution》经典论文_携带受选择EDAR变异在小鼠的表达建模揭示近期人类进化 2013 Cell Modeling Recent Human Evolution in Mice by Expression of a Selected EDAR Variant

Cell Modeling recent Human Evolution in Mice by Expression of a selected EDAR Variant Yana G. Kamberov, 1, 2,3,5,6,7, 16 Sijia Wang, 5, 7, 16, 18 Jingze Tan, 9 Pascale Gerbault, 10 Abigail Wark, 1 Longzhi Tan, 5 unhao Mao 8.21 Asa Schachar, 5, 7 Madeline Paymer, 5,7 Elizabeth Hostetter, Elizabeth byrne, Melissa Burnett, 24.S20 Yajun Yang, 9 Shilin Li,Kun Tang, 13 Hua Chen, 14 Adam Powell, 1 Yuval Itan, 10, 19 Dorian Fuller, 12 Jason Lohmueller ndrew P. McMahon 8,22 Mark G. Thomas 10 Daniel E. Lieberman 6,17 Li Jin 9, 13, 17, Clifford J. Tabin 1, 17 Bruce A. Morgan 2,317, and Pardis C. Sabeti5, 7, 15,17,* 1Department of Genetics Harvard Medical School, Boston, MA 02115, USA cUtaneous Biology Research Center 4Department of Dermatology Massachusetts General Hospital, Boston, MA 02114, USA SThe Broad Institute of Harvard and MIT, cambridge, MA 02142, USA 6Department of Human Evolutionary Biology 7Center for Systems Biology, Department of Organismic and Evolutionary Biology aDepartment of Molecular and cellular Biology ity, Cambridge, MA 02138, US 9MOE Key Laboratory of Contemporary Anthropology, Fudan University, Shanghai 200433, China 10Department of Genetics, Evolution and Environment 11UCL Genetics Institute(UGI ste of University College London, London WC1H OPY, UK 13CAS-MPG Partner Institute for Computational Biology, Shanghai Institutes for Biological Sciences, Chinese Academy of Science, 14Department of Epidemiology Department of Immunology and Infectious Diseases Harvard School of Public Health, Boston, MA 02115, USA 1 These authors contributed equally to this work 17These authors contributed equally to this work and are cosenior authors 18Present address: Max Planck-CAS Paul Gerson Unna Research Group on Dermatogenomics, CAS-MPG Partner Institute for Computational Biology, Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences, Shanghai 200031, China 19Present address: St. Giles Laboratory of Human Genetics of Infectious Diseases, Rockefeller Branch, The Rockefeller University, New York, NY, USA 20Present address: Department of Systems Biology Harvard Medical School, Boston, MA 02115, USA 2Present address: Department of Cancer Biology, University of Massachusetts Medical School, Worcester, MA 01605, USA 22Present address: Department of Stem Cell Biology and Regenerative Medicine, Broad-CIRM Center, Keck School Of Medicine University of Southern Califomia, CA 90089, USA Correspondence:lijinfudan@gmail.com(L.J.)bruce.morgan@cbrc2.mgh.harvard.edu(BA.M),pardis@broadinstituteorg(PC..) ttp/ dx. doi. oro/10.1016/ce.2013.01.016 SUMMARY its direct biological significance and potential adaptive role remain unclear. We generated a An adaptive variant of the human Ectodysplasin knockin mouse model and find that, as in humans receptor, EDARV370A, is one of the strongest candi- hair thickness is increased in EDAR370A mice We lates of recent positive selection from genome- identify new biological targets affected by the muta- wide scans. We have modeled EDAR370A in mice tion, including mammary and eccrine glands and characterized its phenotype and evolutionary Building on these results, we find that EDAR370A origins in humans. Our computational analysis is associated with an increased number of active suggests the allele arose in central China approxi- eccrine glands in the Han Chinese. This interdisci mately 30,000 years ago. Although EDAR370A plinary approach yields unique insight into the has been associated with increased scalp hair thick generation of adaptive variation among modern ness and changed tooth morphology in humans, humans Cell 152, 691-702, February 14, 2013@2013 Elsevier Inc. 691

Modeling Recent Human Evolution in Mice by Expression of a Selected EDAR Variant Yana G. Kamberov,1,2,3,5,6,7,16 Sijia Wang,5,7,16,18 Jingze Tan,9 Pascale Gerbault,10 Abigail Wark,1 Longzhi Tan,5 Yajun Yang,9 Shilin Li,9 Kun Tang,13 Hua Chen,14 Adam Powell,11 Yuval Itan,10,19 Dorian Fuller,12 Jason Lohmueller,5,20 Junhao Mao,8,21 Asa Schachar,5,7 Madeline Paymer,5,7 Elizabeth Hostetter,5 Elizabeth Byrne,5 Melissa Burnett,2,4 Andrew P. McMahon,8,22 Mark G. Thomas,10 Daniel E. Lieberman,6,17 Li Jin,9,13,17, * Clifford J. Tabin,1,17 Bruce A. Morgan,2,3,17, * and Pardis C. Sabeti5,7,15,17, * 1Department of Genetics 2Department of Dermatology Harvard Medical School, Boston, MA 02115, USA 3Cutaneous Biology Research Center 4Department of Dermatology Massachusetts General Hospital, Boston, MA 02114, USA 5The Broad Institute of Harvard and MIT, Cambridge, MA 02142, USA 6Department of Human Evolutionary Biology 7Center for Systems Biology, Department of Organismic and Evolutionary Biology 8Department of Molecular and Cellular Biology Harvard University, Cambridge, MA 02138, USA 9MOE Key Laboratory of Contemporary Anthropology, Fudan University, Shanghai 200433, China 10Department of Genetics, Evolution and Environment 11UCL Genetics Institute (UGI) 12Institute of Archaeology University College London, London WC1H 0PY, UK 13CAS-MPG Partner Institute for Computational Biology, Shanghai Institutes for Biological Sciences, Chinese Academy of Science, Shanghai 200031, China 14Department of Epidemiology 15Department of Immunology and Infectious Diseases Harvard School of Public Health, Boston, MA 02115, USA 16These authors contributed equally to this work 17These authors contributed equally to this work and are cosenior authors 18Present address: Max Planck-CAS Paul Gerson Unna Research Group on Dermatogenomics, CAS-MPG Partner Institute for Computational Biology, Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences, Shanghai 200031, China 19Present address: St. Giles Laboratory of Human Genetics of Infectious Diseases, Rockefeller Branch, The Rockefeller University, New York, NY, USA 20Present address: Department of Systems Biology Harvard Medical School, Boston, MA 02115, USA 21Present address: Department of Cancer Biology, University of Massachusetts Medical School, Worcester, MA 01605, USA 22Present address: Department of Stem Cell Biology and Regenerative Medicine, Broad-CIRM Center, Keck School Of Medicine, University of Southern California, CA 90089, USA *Correspondence: lijin.fudan@gmail.com (L.J.), bruce.morgan@cbrc2.mgh.harvard.edu (B.A.M.), pardis@broadinstitute.org (P.C.S.) http://dx.doi.org/10.1016/j.cell.2013.01.016 SUMMARY An adaptive variant of the human Ectodysplasin receptor, EDARV370A, is one of the strongest candi￾dates of recent positive selection from genome￾wide scans. We have modeled EDAR370A in mice and characterized its phenotype and evolutionary origins in humans. Our computational analysis suggests the allele arose in central China approxi￾mately 30,000 years ago. Although EDAR370A has been associated with increased scalp hair thick￾ness and changed tooth morphology in humans, its direct biological significance and potential adaptive role remain unclear. We generated a knockin mouse model and find that, as in humans, hair thickness is increased in EDAR370A mice. We identify new biological targets affected by the muta￾tion, including mammary and eccrine glands. Building on these results, we find that EDAR370A is associated with an increased number of active eccrine glands in the Han Chinese. This interdisci￾plinary approach yields unique insight into the generation of adaptive variation among modern humans. Cell 152, 691–702, February 14, 2013 ª2013 Elsevier Inc. 691

Cell INTRODUCTION et al., 2008). This finding suggested that a pre-existing mouse overexpressed, Humans are unique among primates in having colonized nearly might provide insight into 370A's phenotypic consequences every corner of the world; consequently, niche 008). Indeed, trans- pressures likely helped shape the phenotypic varia present-day g, enlarged globin-B a P. falcipa 2005);muta digest milk (Enattah genes dr 2005) Althor s.We tated the chara tional e ces. affect a by exp ther of car ghlights the variation in idat g other candidate those w S persist unknowr difficu native to t date adap A SNP in 51 worldwide particularl of 370A. Haplotype hum subtle p ype adapti g plasin A et al., 200 of the se East Asian ar scalp hair th 200 2009;Park quantify tained whe phenotypes. The bioch that the variant Structural models Domain(DD) required transducer EDARADD we performed a pression of 370A has bee NFkB signaling in vitro re sent-day Han Chinese 62e,69-7 February 1420120e

INTRODUCTION Humans are unique among primates in having colonized nearly every corner of the world; consequently, niche-specific selective pressures likely helped shape the phenotypic variation currently evident in Homo sapiens. Identifying the genetic variants that underlie regional adaptations is thus central to understanding present-day human diversity, yet only a few adaptive traits have been elucidated. These include mutations in the Hemo￾globin-B and Duffy antigen genes, driving resistance to P. falciparum and P. vivax malaria, respectively (Kwiatkowski, 2005); mutations in lactase allowing some adult humans to digest milk after the domestication of milk-producing livestock (Enattah et al., 2002); and mutations in SLC24A5 and other genes driving variation in skin pigmentation (Lamason et al., 2005). Although breakthroughs in genomic technology have facili￾tated the identification of hundreds of candidate genetic variants with evidence of recent positive natural selection, validation and characterization of putative genetic adaptations requires func￾tional evidence linking genotypes to phenotypes that could affect an organism’s fitness (Akey, 2009). This is made difficult by experimental challenges in isolating the phenotypic effects of candidate loci and by methodological limitations on the phenotypes that can be readily assessed in humans. Accord￾ingly, the best-characterized human adaptive alleles are typically those whose phenotypic outcomes are easily measured and strongly related to known genetic variation, such as lactase persistence or skin pigmentation. Many genes, however, have unknown or pleiotropic effects, making their adaptive advantage difficult to uncover (Sivakumaran et al., 2011). A promising alter￾native to tackle these difficulties is to study the effects of candi￾date adaptive alleles in animal models. Although such models, particularly using mice, have been used extensively to study human disease alleles, they have not been used to model the subtle phenotypic changes expected to result from human adaptive variation. A compelling candidate human adaptive allele to emerge from genome-wide scans is a derived coding variant of the Ectodys￾plasin A (EDA) receptor (EDAR), EDARV370A (370A) (Sabeti et al., 2007; Grossman et al., 2010). Computational fine-mapping of the selection signal and the restricted occurrence of 370A in East Asian and Native American populations have led to sugges￾tions that 370A was selected in Asia (Bryk et al., 2008). In support of this hypothesis, 370A was shown to associate with increased scalp hair thickness and incisor tooth shoveling in multiple East Asian populations (Fujimoto et al., 2008a, 2008b; Kimura et al., 2009; Park et al., 2012). However, because association studies quantify correlation rather than causation, it remains to be ascer￾tained whether 370A is the genetic change driving the observed phenotypes. The biochemical properties of 370A support the possibility that the variant directly causes the associated phenotypes. Structural models predict that V370A lies in the EDAR Death Domain (DD) required for interaction with the downstream signal transducer EDARADD (Sabeti et al., 2007). Moreover, overex￾pression of 370A has been reported to upregulate downstream NFkB signaling in vitro relative to 370V (Bryk et al., 2008; Mou et al., 2008). This finding suggested that a pre-existing mouse model, in which the ancestral 370V allele is overexpressed, might provide insight into 370A’s phenotypic consequences (Headon and Overbeek, 1999; Mou et al., 2008). Indeed, trans￾genic mice expressing multiple copies of 370V have thicker hair shafts as seen in humans with the 370A allele (Fujimoto et al., 2008a, 2008b; Mou et al., 2008). In addition, these animals exhibit increased mammary gland branching, enlarged mammary glands and hyperplastic sebaceous and Meibomian glands that secrete hydrophobic films as a barrier to water loss in the skin and eyes, respectively (Chang et al., 2009). These latter phenotypes led to the proposal that the 370A variant may have been selected in response to cold and arid environ￾mental conditions (Chang et al., 2009). Evaluating which forces may have contributed to the spread of 370A requires knowledge of both the environmental context in which this variant was selected and its phenotypic effects. We therefore employed a multi-disciplinary approach to test the role of 370A in recent human evolution. This included modeling to reconstruct the evolutionary history of 370A, and a knockin mouse model to examine its direct phenotypic consequences. Analysis of the mouse knockin revealed phenotypes not previously reported in human genetic studies, which we further characterized in a Han Chinese cohort. This work highlights the utility of modeling nonpathological human genetic variation in mice, providing a framework for assessing other candidate adaptive human alleles. RESULTS Single Origin of 370A in Central China Using both newly generated and publicly available data, we examined 280 SNPs flanking the 370A SNP in 51 worldwide populations in order to assess the origin of 370A. Haplotype analysis supports a single origin of the derived allele (Figure 1A), with the mutation lying on a unique, nearly unbroken haplotype extending more than 100 kb among both East Asians and Native Americans (Figure S1 available online). To estimate the allele’s geographic and temporal origin, we performed more than one million spatially explicit demic forward simulations modeling the appearance and spread of 370A in Asia (Itan et al., 2009) (Modeling the Origins and Spread of 370A in an Approximate Bayesian Computation Framework). We used approximate Bayesian computation (ABC) (Beaumont et al., 2002) to compare simulated to observed allele frequencies and to estimate key evolutionary and demographic parameters (Fagundes et al., 2007; Itan et al., 2009; Ray et al., 2010). This analysis estimated the 370A allele originated in central China (Figure 1B) between 13,175 and 39,575 years BP (95% credible interval), with a mode of 35,300 years BP and a median of 30,925 years BP. The estimated selection coefficient has a 95% credible interval between 0.030 and 0.186, with a mode of 0.122 and a median of 0.114 (Figures S2, S3, and S4, and Tables S1, S2, S3, and S4, and Modeling the Origins and Spread of 370A in an Approximate Bayesian Computation Framework). As a separate calculation of the age of 370A, we performed a maximum likelihood inference analysis using the allele frequency spectrum of 1,677 nearby SNPs in present-day Han Chinese 692 Cell 152, 691–702, February 14, 2013 ª2013 Elsevier Inc

Cell Figure 1. Origins of 370A (A) Haplotype distribution of the genomic region surrounding V370A, based on 24 SNPs covering 139 kb. The six most common haplotypes are shown, and the haining low-frequency haplotypes grouped as"other" The chimpanzee allele was assumed to be ancestral Derived alleles are in dark blue, except for the ariant which is red (B) The approximate posterior probability density for the geographic origin of 370A obtained by ABC simulation. The heat map was generated usil kemel density estimation of the latitude and longitude co tes from the top 5,000(0. 46%)of 1, 083, 966 simulations. Red color represents the highest probability, and lue the lowest See also Figures S1, S2, s3, S4, S5, and S7 and Tables s1, S2, s3, and S4 stimating Selection Time of 370A using the Coalescent-Based this change and concomitant rough coat phenotype suggest Allele Frequency Spectrum)(Chen, 2012). This method provided this model has limited utility(Mou et al., 2008). In contrast imilar estimates of the allele age (95% confidence interval: 370A knockin mice exhibit a smooth hair coat with all four 34, 775-38, 208 years BP; maximum likelihood estimation hair types that are normally found in the mouse pelage(Sund [MLE]: 36, 490 years BP)and selection intensity(95%confidence berg, 1994; Mikkola, 2011)(Figure 2C and data not shown) interval: 0.0657 7-0.0831;MLE:0.0744; Figure S5) We evaluated the sizes of both the awl and achene hair types in the mouse coat by scoring medulla cell number across the Generation of 370A Mouse Mode hair shafts(Sundberg, 1994; Enshell-Seijffers et aL., 2010)(Fig o test the biological consequence of 370A, we evaluated its ure 3A). Our analysis revealed that Edar genotype was signifi efficiency to drive a phenotypic change in vivo. In humans cantly associated with hair size(MANOVA, p=0.034 and p and mice, loss-of-function mutations in the genes coding for 0.027 for awl and achene hairs respectively, Table S5). 370A the ligand EDA-A1, EDAR, and EDARADD lead to strikingly homozygous mutant mice had more of the thickest, four-cell imilar phenotypes characterized by defective hair development, awl hairs and fewer three-cell hairs than 370V homozygotes absence of eccrine glands, and missing or misshapen teeth(Mik- (p=0.007 and p= 0.005 for four- and three-cell hairs respec- ola, 2008, 2011: Cluzeau et aL., 2011). The conserved role of the tively)(Figure 3B and Table S5). Similarly, 370A homozygotes Ectodysplasin pathway in the development of ectodermally had a higher proportion of thicker achenes than 370V and derived organs(Gruneberg, 1971; Kondo et aL., 2001; Colosimo 370v/370A animals(p=0.007 and p=0.007, respectively, et aL., 2005: Mikkola, 2008, 2011)suggested that a 370A mouse Table S5 and Figure 3C). The 370A mouse thus recapitulates knockin model would be an accurate system in which to isolate the associated human phenotype of increased hair thickness, and examine the effects of the derived allele confirming that the mutation is causal, and demonstrating the The DDs of mouse and human EDAR are identical in sequence, model's utility for accurately characterizing the allele's biolog- with mice natively expressing the 370V allele(Figure 2A). To ical effects construct 370A knockin mice, we used homologous recombina- tion in embryonic stem cells to introduce the T1326C point 370A Does Not Increase Meibomian Gland Size mutation into the endogenous murine Edar locus resulting in aA previous study overexpressing 370V in mice found an V370A substitution in the encoded protein(Figures 2B and S6). increase in Meibomian gland size, leading to speculation on 370A mice were born at expected Mendelian ratios, appeared the adaptive benefit of 370A(Chang et al., 2009). To evaluate healthy, and did not exhibit altered fertility or longevity compared the effect of 370A on Meibomian gland size, we measured the to wild-type littermates(Figure 2C and data not shown) total gland area of the upper and lower eyelids of 370V, 370V/ 370A, and 370A mice. No significant difference in gland area 370A Increases Hair Thickness in Mice yas observed between the different genotypes(MANOVA, p In humans, 370A is associated with increased scalp hair thick- 0.244: Figure 4; Table S5). Similarly, we found no detectable ness(Fujimoto et aL., 2008a, 2008b). Mice that overexpress the change in the size of the related sebaceous glands of the skin 370V allele also have thicker hairs, but the larger magnitude of (data not shown). Cell 152, 691-702, February 14, 2013@2013 Elsevier Inc. 693

(Estimating Selection Time of 370A using the Coalescent-Based Allele Frequency Spectrum) (Chen, 2012). This method provided similar estimates of the allele age (95% confidence interval: 34,775–38,208 years BP; maximum likelihood estimation [MLE]: 36,490 years BP) and selection intensity (95% confidence interval: 0.0657–0.0831; MLE: 0.0744; Figure S5). Generation of 370A Mouse Model To test the biological consequence of 370A, we evaluated its sufficiency to drive a phenotypic change in vivo. In humans and mice, loss-of-function mutations in the genes coding for the ligand EDA-A1, EDAR, and EDARADD lead to strikingly similar phenotypes characterized by defective hair development, absence of eccrine glands, and missing or misshapen teeth (Mik￾kola, 2008, 2011; Cluzeau et al., 2011). The conserved role of the Ectodysplasin pathway in the development of ectodermally derived organs (Gru¨ neberg, 1971; Kondo et al., 2001; Colosimo et al., 2005; Mikkola, 2008, 2011) suggested that a 370A mouse knockin model would be an accurate system in which to isolate and examine the effects of the derived allele. The DDs of mouse and human EDAR are identical in sequence, with mice natively expressing the 370V allele (Figure 2A). To construct 370A knockin mice, we used homologous recombina￾tion in embryonic stem cells to introduce the T1326C point mutation into the endogenous murine Edar locus resulting in a V370A substitution in the encoded protein (Figures 2B and S6). 370A mice were born at expected Mendelian ratios, appeared healthy, and did not exhibit altered fertility or longevity compared to wild-type littermates (Figure 2C and data not shown). 370A Increases Hair Thickness in Mice In humans, 370A is associated with increased scalp hair thick￾ness (Fujimoto et al., 2008a, 2008b). Mice that overexpress the 370V allele also have thicker hairs, but the larger magnitude of this change and concomitant rough coat phenotype suggest this model has limited utility (Mou et al., 2008). In contrast 370A knockin mice exhibit a smooth hair coat with all four hair types that are normally found in the mouse pelage (Sund￾berg, 1994; Mikkola, 2011) (Figure 2C and data not shown). We evaluated the sizes of both the awl and auchene hair types in the mouse coat by scoring medulla cell number across the hair shafts (Sundberg, 1994; Enshell-Seijffers et al., 2010) (Fig￾ure 3A). Our analysis revealed that Edar genotype was signifi- cantly associated with hair size (MANOVA, p = 0.034 and p = 0.027 for awl and auchene hairs respectively, Table S5). 370A homozygous mutant mice had more of the thickest, four-cell awl hairs and fewer three-cell hairs than 370V homozygotes (p = 0.007 and p = 0.005 for four- and three-cell hairs respec￾tively) (Figure 3B and Table S5). Similarly, 370A homozygotes had a higher proportion of thicker auchenes than 370V and 370V/370A animals (p = 0.007 and p = 0.007, respectively, Table S5 and Figure 3C). The 370A mouse thus recapitulates the associated human phenotype of increased hair thickness, confirming that the mutation is causal, and demonstrating the model’s utility for accurately characterizing the allele’s biolog￾ical effects. 370A Does Not Increase Meibomian Gland Size A previous study overexpressing 370V in mice found an increase in Meibomian gland size, leading to speculation on the adaptive benefit of 370A (Chang et al., 2009). To evaluate the effect of 370A on Meibomian gland size, we measured the total gland area of the upper and lower eyelids of 370V, 370V/ 370A, and 370A mice. No significant difference in gland area was observed between the different genotypes (MANOVA, p = 0.244; Figure 4; Table S5). Similarly, we found no detectable change in the size of the related sebaceous glands of the skin (data not shown). Figure 1. Origins of 370A (A) Haplotype distribution of the genomic region surrounding V370A, based on 24 SNPs covering
139 kb. The six most common haplotypes are shown, and the remaining low-frequency haplotypes grouped as ‘‘Other.’’ The chimpanzee allele was assumed to be ancestral. Derived alleles are in dark blue, except for the 370A variant which is red. (B) The approximate posterior probability density for the geographic origin of 370A obtained by ABC simulation. The heat map was generated using 2D kernel density estimation of the latitude and longitude coordinates from the top 5,000 (0.46%) of 1,083,966 simulations. Red color represents the highest probability, and blue the lowest. See also Figures S1, S2, S3, S4, S5, and S7 and Tables S1, S2, S3, and S4. Cell 152, 691–702, February 14, 2013 ª2013 Elsevier Inc. 693

Cel GGCGGTNCG TG A GGCGGCCG TG Human RMLSSTYNSEKAVKTWRHLAESFGLKRDEIGGMTD louse RMLSSTYNSEKAVKTWRHLAESFGLKRDEIGGMTD Human GMQLFDRISTAGYSIPELLTKLVQIERLDAVESLCADIL Mouse GMQLFDRISTAGYSIPELLTKLVQIERLDAVESLCADIL Targeting Vector ecombination 370V 70V/370A370A Figure 2. Generation of the 370A Mouse EDAR DD. 370V is in bold with superscript asterisk. xons 11 and 12 and contains the T1326C mutation (red line). The targeting vector also contained a neomycin (Neo) reseda.dar geno TArgeting strategy for the introduction of the 370A mutation into the mouse Edar locus. The construct spans part of the edar ge cassette flanked by LoxP sites purple arrows)inserted into the EDAR 3 untranslated region (UTR. Neo was excised by breeding to a ubiquitously expressing B-Actin Cre line The final genomic tructure of the knockin Edar locus is shown with exons as boxes, coding sequence in dark gray, UTR in black and all other genomic sequence as a black line Diphtheria toxin A selection cassette(DTA) is shown in blue and vector sequence in light gray. (C) Appearance of 370V, 370v/370A and 370A animals and confirmation of the T1326C (underlined)substitution. See also Figure Mammary Gland Branching and Fat Pad Size Are Altered icantly affected by genotype (ANCOVA, p=0.045) in 370A Mice gland area was significantly affected by body weight The mammary gland and surrounding stromal tissue, the by genotype(Generation and Statistical Analysis of fat are of interest given their importance in Knockin Mouse eproduction(Neville et aL, 1998; Lefevre et aL, 2010). A role for Ectodysplasin signaling in mammary gland development is 370A Increases Eccrine Gland Number in Mice suggested by loss-of-function mutants in which glands are Eccrine sweat glands in humans are widespread throughout the present and functional, but gland branching and size of the skin, reflecting their critical role in heat dissipation, but in mice mammary tree are reduced(Chang et aL., 2009; Mikkola, 2011; and most other mammals, they are restricted to the plantar Voutilainen et al., 2012). In contrast, overexpression of Edar surfaces where they serve in traction. In spite of this difference. nd its ligand Eda-A1 lead to the converse phenotypes(Chang loss-of-function mutants have demonstrated the conserved et aL., 2009; Voutilainen et al. 2012 role of Ectodysplasin signaling in eccrine gland formation across We assessed five aspects of the 4th and 9n mammary glands mammals(Gruneberg, 1971; Mikkola, 2011). Because eccrine in pre-estrus mice: branch number, branch density, gland length, glands are absent in Edar loss-of-function mutants, we evaluated gland area, and mammary fat pad area. Only branch density and the effect of 370A on eccrine gland number in our mouse model mammary fat pad size were affected by the 370A genotype We scored eccrine gland number in four of the six hindlimb MANOVA, univariate main effects: p= 0.044 and p= 0.018, footpads(Figures 6A-6F. Edar genotype was significantly respectively, Figure 5 and Table S5). 370A homozygotes had associated with eccrine gland number in all footpads(MANOV higher branch density than either 370V or 370V/370A mice(p= p=4.3 x 10-7, see Generation and Statistical Analysis of the 0.018 and p=0.047, respectively, Figure 5E) and smaller fat 370A Knockin Mouse, Table S5). 370A homozygous animals pads(p=0.007 and p=0.030, respectively, Figure 5F). Although, had more sweat glands per footpad than wild-type 370Vhomo- body weight was not affected by Edar genotype(ANOVA, p= zygotes(p< 0.01 for all footpads, Figure 6G), and in most 0. 459), linear regression revealed a small effect of body footpads 370V/370A heterozygotes showed an intermediate on gland and fat pad size(Generation and Statistical increase(p <0.01 for footpads FP-3, FP-4, FP-5, Figure 6G) of the 370A Knockin Mouse). To control for this effect, Because a single copy of 370A was sufficient to increase eo alyzed the effect of Edar genotype on these traits using body crine gland number in our model, we directly tested whether weight as a covariate. In this analysis, fat pad area was still signif- 370A is a gain-of-function allele by analyzing its ability to rescue 694ce152,691-702, February14,2013e2013Es

Mammary Gland Branching and Fat Pad Size Are Altered in 370A Mice The mammary gland and surrounding stromal tissue, the mammary fat pad, are of interest given their importance in reproduction (Neville et al., 1998; Lefe` vre et al., 2010). A role for Ectodysplasin signaling in mammary gland development is suggested by loss-of-function mutants in which glands are present and functional, but gland branching and size of the mammary tree are reduced (Chang et al., 2009; Mikkola, 2011; Voutilainen et al., 2012). In contrast, overexpression of Edar and its ligand Eda-A1 lead to the converse phenotypes (Chang et al., 2009; Voutilainen et al., 2012). We assessed five aspects of the 4th and 9th mammary glands in pre-estrus mice: branch number, branch density, gland length, gland area, and mammary fat pad area. Only branch density and mammary fat pad size were affected by the 370A genotype (MANOVA, univariate main effects: p = 0.044 and p = 0.018, respectively, Figure 5 and Table S5). 370A homozygotes had higher branch density than either 370V or 370V/370A mice (p = 0.018 and p = 0.047, respectively, Figure 5E) and smaller fat pads (p = 0.007 and p = 0.030, respectively, Figure 5F). Although, body weight was not affected by Edar genotype (ANOVA, p = 0.459), linear regression revealed a small effect of body weight on gland and fat pad size (Generation and Statistical Analysis of the 370A Knockin Mouse). To control for this effect, we rean￾alyzed the effect of Edar genotype on these traits using body weight as a covariate. In this analysis, fat pad area was still signif￾icantly affected by genotype (ANCOVA, p = 0.045), whereas gland area was significantly affected by body weight but not by genotype (Generation and Statistical Analysis of the 370A Knockin Mouse). 370A Increases Eccrine Gland Number in Mice Eccrine sweat glands in humans are widespread throughout the skin, reflecting their critical role in heat dissipation, but in mice and most other mammals, they are restricted to the plantar surfaces where they serve in traction. In spite of this difference, loss-of-function mutants have demonstrated the conserved role of Ectodysplasin signaling in eccrine gland formation across mammals (Gru¨ neberg, 1971; Mikkola, 2011). Because eccrine glands are absent in Edar loss-of-function mutants, we evaluated the effect of 370A on eccrine gland number in our mouse model. We scored eccrine gland number in four of the six hindlimb footpads (Figures 6A–6F). Edar genotype was significantly associated with eccrine gland number in all footpads (MANOVA, p = 4.3 3 107 , see Generation and Statistical Analysis of the 370A Knockin Mouse, Table S5). 370A homozygous animals had more sweat glands per footpad than wild-type 370V homo￾zygotes (p < 0.01 for all footpads, Figure 6G), and in most footpads 370V/370A heterozygotes showed an intermediate increase (p < 0.01 for footpads FP-3, FP-4, FP-5, Figure 6G). Because a single copy of 370A was sufficient to increase ec￾crine gland number in our model, we directly tested whether 370A is a gain-of-function allele by analyzing its ability to rescue Figure 2. Generation of the 370A Mouse (A) Conservation of human and mouse EDAR DD. 370V is in bold with superscript asterisk. (B) Targeting strategy for the introduction of the 370A mutation into the mouse Edar locus. The construct spans part of the Edar genomic sequence including exons 11 and 12 and contains the T1326C mutation (red line). The targeting vector also contained a neomycin (Neo) resistance cassette flanked by LoxP sites (purple arrows) inserted into the EDAR 30 untranslated region (UTR). Neo was excised by breeding to a ubiquitously expressing b-Actin Cre line. The final genomic structure of the knockin Edar locus is shown with exons as boxes, coding sequence in dark gray, UTR in black and all other genomic sequence as a black line. Diphtheria toxin A selection cassette (DTA) is shown in blue and vector sequence in light gray. (C) Appearance of 370V, 370V/370A and 370A animals and confirmation of the T1326C (underlined) substitution. See also Figure S6. 694 Cell 152, 691–702, February 14, 2013 ª2013 Elsevier Inc.

Cell zygotes(MANOVA post hoc tests, p=0.002, Figure 6H). In agreement with a gain-of-function model, 370A/379K heterozy (p < 0.05 for all footpads, Figure 6H and Table S5) 370A Is Associated with More Eccrine Glands and other Pleiotropic Effects in Humans 4 cell 3 cell 2 cell The change in eccrine gland number we observed in the 370A mouse has important implications for the distribution of variation in this trait in human populations. However, association studies of sweat gland density with nonpathological variation at the EDAR locus have not been reported in humans. To examine whether 370A is associated with altered eccrine gland number in humans, we carried out an association study 82 in individuals of han descent from an established cohort in taiz- 370V hou, China Wang et al., 2009). To sample a sufficient number of 370v/370A the rare 370V alleles, we first genotyped the 370A SNP and found 2, 226 370A homozygotes, 340 370V/370A heterozygotes, and 6 370V homozygotes. We then contacted all individuals with at least one copy of the 370V allele and enrolled 187 of them(184 370V/370A and 3 3701, along with 436 370A individuals and collected phenotypes related to ectodermal appendages ( Table S6 and Association Study of 370A in a Han Chinese Population) Because only three individuals were homozygous for the 370V Number of cells for 370A in statistical analysis of the collected data Consistent with previous reports(Kimura et al., 2009: Park et aL., 2012), 370A was associated with single and doub shoveling of the upper incisors (Wald test, p= 0.0077 and p=0.0004, respectively; Table S6). Additionally, 370A was significantly associated with the presence of a protostylid cusp and the absence of lower third molars (Wald test, p=0.0079 ■370V and p=0.0123, respectively; Association Study of 370A in a 370V/370A Han Chinese Population and Table S6) We tested for an association between 370A and eccrine sweat gland number using the starch-iodine method to measure the number of activated glands in digit pads of the thumb and inde finger(Juniper et aL., 1964: Randall, 1946 ). In agreement with our mouse findings, 370A homozygous individuals had significantly more active eccrine glands than 370V/370A individuals(two- ailed t test, p=0.011, Figure 7). Testing all three genotypes Number of cells sing linear-regression in an additive model revealed a strong Figure 3. 370A Allele Increases Hair Size in the Mouse Coat association between 370A and eccrine gland density ( Wald (A) Representative images of hairs show medulla cell number serves test, p=0.0047; Table S6). This association remained significant a proxy for hair shaft thicknes when we controlled for age, sex, and potential population (B and C)370A mice have a larger proportion of thicker hairs than mice substructure(Association Study of 370A in a Han Chinese pressing the ancestral allele the average frequency (+SEM)of awl( B)and Population and Table S6). DISCUSSION post hoc tests: p<0.05(),p<0.01("). See also Table $5. study integrated population genetic analyses, a humanized the eccrine gland phenotype of mice heterozygous for the mouse model, and human association study to characterize downless(dl,E379K@Edarloss-of-functionmutation(Headonanaturalhumangenevariant.Combiningtheseapproache and Overbeek, 1999). 379K is classically considered a recessive allowed us to determine the direct biological effects of 370A mutation, and animals heterozygous for the 379K allele are and cast new light on their evolutionary uences Extend- described as wild-type(Headon and Overbeek, 1999). However, ing this strategy to other candidate adaptive alleles stands to our quantitative method for scoring eccrine glands revealed advance our understanding of the effects of recent selection a subtle decrease in eccrine gland number in 379E/379K hetero- on the diversification of modern humans. Cell 152, 691-702, February 14, 2013(2013 Elsevier Inc. 695

the eccrine gland phenotype of mice heterozygous for the downless (dlj , E379K) Edar loss-of-function mutation (Headon and Overbeek, 1999). 379K is classically considered a recessive mutation, and animals heterozygous for the 379K allele are described as wild-type (Headon and Overbeek, 1999). However, our quantitative method for scoring eccrine glands revealed a subtle decrease in eccrine gland number in 379E/379K hetero￾zygotes (MANOVA post hoc tests, p = 0.002, Figure 6H). In agreement with a gain-of-function model, 370A/379K heterozy￾gous animals had more eccrine glands than 370V/379K animals (p < 0.05 for all footpads, Figure 6H and Table S5). 370A Is Associated with More Eccrine Glands and Other Pleiotropic Effects in Humans The change in eccrine gland number we observed in the 370A mouse has important implications for the distribution of variation in this trait in human populations. However, association studies of sweat gland density with nonpathological variation at the EDAR locus have not been reported in humans. To examine whether 370A is associated with altered eccrine gland number in humans, we carried out an association study in individuals of Han descent from an established cohort in Taiz￾hou, China (Wang et al., 2009). To sample a sufficient number of the rare 370V alleles, we first genotyped the 370A SNP and found 2,226 370A homozygotes, 340 370V/370A heterozygotes, and 6 370V homozygotes. We then contacted all individuals with at least one copy of the 370V allele and enrolled 187 of them (184 370V/370A and 3 370V), along with 436 370A individuals and collected phenotypes related to ectodermal appendages (Table S6 and Association Study of 370A in a Han Chinese Population). Because only three individuals were homozygous for the 370V allele, we focused on individuals homozygous and heterozygous for 370A in statistical analysis of the collected data. Consistent with previous reports (Kimura et al., 2009; Park et al., 2012), 370A was associated with single and double shoveling of the upper incisors (Wald test, p = 0.0077 and p = 0.0004, respectively; Table S6). Additionally, 370A was significantly associated with the presence of a protostylid cusp and the absence of lower third molars (Wald test, p = 0.0079 and p = 0.0123, respectively; Association Study of 370A in a Han Chinese Population and Table S6). We tested for an association between 370A and eccrine sweat gland number using the starch-iodine method to measure the number of activated glands in digit pads of the thumb and index finger (Juniper et al., 1964; Randall, 1946). In agreement with our mouse findings, 370A homozygous individuals had significantly more active eccrine glands than 370V/370A individuals (two￾tailed t test, p = 0.011, Figure 7). Testing all three genotypes using linear-regression in an additive model revealed a strong association between 370A and eccrine gland density (Wald test, p = 0.0047; Table S6). This association remained significant when we controlled for age, sex, and potential population substructure (Association Study of 370A in a Han Chinese Population and Table S6). DISCUSSION This study integrated population genetic analyses, a humanized mouse model, and human association study to characterize a natural human gene variant. Combining these approaches allowed us to determine the direct biological effects of 370A and cast new light on their evolutionary consequences. Extend￾ing this strategy to other candidate adaptive alleles stands to advance our understanding of the effects of recent selection on the diversification of modern humans. Figure 3. 370A Allele Increases Hair Size in the Mouse Coat (A) Representative images of mouse hairs show medulla cell number serves as a proxy for hair shaft thickness. (B and C) 370A mice have a larger proportion of thicker hairs than mice expressing the ancestral allele. The average frequency (±SEM) of awl (B) and auchene (C) hairs of each size is shown. Significance levels of differences between 370V and 370A animals by ANOVA post hoc tests: p < 0.05 (*), p < 0.01 (**). See also Table S5. Cell 152, 691–702, February 14, 2013 ª2013 Elsevier Inc. 695

Cel Figure 4. 370A Does Not Increase Meibo- mian Gland Size rbital (white arrow) and suborbital (red arrow) D Differences between genotypes did not reach statistical significance. See also Table s5 12 370v/370A 370A level that requi priori knowledge of biological t his is especially useful for stu Melic variants of 6 n Ancient Asian Origin for 370A patially explicit simulation, haplotype Superorbital and maximum likelihood analyses sug Gland Type gest that 370A originated once in central China more than 30,000 years BP with a selective coefficient that is one of the Modeling Human Adaptive Variation in Mice highest measured in human populations. Our results are consis- The laboratory mouse is an established tool for understanding tent with previous inferences that 370A must have arisen prior to pathological variants in the human genome(Cox and Brown, 15,000 BP(Bryk et aL., 2008; Peter et al., 2012) and the first 2003). Recent work on an adaptive variant of the FoxP2 gene peopling of the Americas(Goebel et al., 2008: O'Rourke and showed that a mouse model can also enable functional examina- Raff, 2010)but also suggest that the allele likely emerged in tion of nonpathological, hominin-specific alleles(Enard et al., East Asia even earlier. It should be noted that haplotype-based 2009). To our knowledge, our study is the first to demonstrate methods, such as that used by Bryk and colleagues(Bryk that mice can be used to model the phenotypic effects of adap- et al., 2008)assume recombination occurs between distinct tive variation within H. sapiens, rather than between humans and haplotypes. However, in a case of rapid local fixation, as is likely other species. There are several advantages to this approach. for a strongly selected and semidominant allele like 370A, First, the existence of inbred mouse strains allows phenotypes recombination of the selected haplotype with itself would be to be evaluated on a genetically homogeneous background, masked, reducing the observed number of recombinations and making it possible to isolate the effects of a variant and draw leading to underestimation of the time of origin( Figure S7) conclusions about the causal effect of a genetic change. In the Thus, our findings shift the context in which to consider the current case, the 370a knockin mouse shows that the derived selective forces that could have acted on 370A mutation is sufficient to alter multiple traits in vivo. Second, the tractability of an animal model allows us to easily explore novel Phenotypic Consequences of 370A traits, e.g, eccrine gland number, and ones that are not readily A comparison of mice harboring the 370V and 370A alleles on assayed in humans, e. g, mammary gland structure. The results the same genetic background revealed multiple differences from mouse models can thus serve to inform human association including increased hair thickness, increased eccrine gland studies pursuing the identification of traits sensitive to candidate number, reduced mammary fat pad size, and increased mammary gland branch density in mice carrying 370A. With Successfully modeling human adaptive alleles in mice relies the exception of mammary fat pad size, which has not been on the conservation of target organ form and function between analyzed in gain-of-function models, these phenotypes are the two species. There are clear limitations to this approach. expected if 370A confers modestly enhanced signaling activity For example, in this study, the absence of murine dental features on EDAR. This mode of action was previously proposed based homologous to the phenotypes observed in humans, such as on the observation that 370A can potentiate NFKB signaling incisor shoveling or the presence of protostylid cusps, makes in vitro(Bryk et al., 2008; Mou et al., 2008) and a clinical case it difficult to equate any changes in 370A mouse dental report in which 370A was associated with reduced severity of morphology to a specific human dental trait -and in fact we hypohidrotic ectodermal dysplasia caused by an EDA missense observed no gross phenotypic changes in the dentition of mutation(Cluzeau et al., 2012). The hair phenotype of 370A 70A animals. Despite such caveats, when direct homology animals is consistent with this model as is the increase in eccrine exists, modeling an allele's effects in vivo has the advantage of gland number and mammary gland branching. We demon- enabling assessment of phenotypic impact on a whole-organism strated this in vivo for the eccrine gland trait by showing 370A cel152,691-702, February14,2013@2013E

Modeling Human Adaptive Variation in Mice The laboratory mouse is an established tool for understanding pathological variants in the human genome (Cox and Brown, 2003). Recent work on an adaptive variant of the FoxP2 gene showed that a mouse model can also enable functional examina￾tion of nonpathological, hominin-specific alleles (Enard et al., 2009). To our knowledge, our study is the first to demonstrate that mice can be used to model the phenotypic effects of adap￾tive variation within H. sapiens, rather than between humans and other species. There are several advantages to this approach. First, the existence of inbred mouse strains allows phenotypes to be evaluated on a genetically homogeneous background, making it possible to isolate the effects of a variant and draw conclusions about the causal effect of a genetic change. In the current case, the 370A knockin mouse shows that the derived mutation is sufficient to alter multiple traits in vivo. Second, the tractability of an animal model allows us to easily explore novel traits, e.g., eccrine gland number, and ones that are not readily assayed in humans, e.g., mammary gland structure. The results from mouse models can thus serve to inform human association studies pursuing the identification of traits sensitive to candidate adaptive alleles. Successfully modeling human adaptive alleles in mice relies on the conservation of target organ form and function between the two species. There are clear limitations to this approach. For example, in this study, the absence of murine dental features homologous to the phenotypes observed in humans, such as incisor shoveling or the presence of protostylid cusps, makes it difficult to equate any changes in 370A mouse dental morphology to a specific human dental trait—and in fact we observed no gross phenotypic changes in the dentition of 370A animals. Despite such caveats, when direct homology exists, modeling an allele’s effects in vivo has the advantage of enabling assessment of phenotypic impact on a whole-organism Figure 4. 370A Does Not Increase Meibo￾mian Gland Size (A–C) Representative images from 370V (A), 370V/ 370A (B), and 370A (C) mouse eyelids. Super￾orbital (white arrow) and suborbital (red arrow) Meibomian glands are visible through the connective tissue. (D) Average glandular area is shown (±SEM). Differences between genotypes did not reach statistical significance. See also Table S5. level that requires no a priori knowledge of biological targets. This is especially useful for studying allelic variants of genes with unknown or pleiotropic functions. An Ancient Asian Origin for 370A Spatially explicit simulation, haplotype, and maximum likelihood analyses sug￾gest that 370A originated once in central China more than 30,000 years BP with a selective coefficient that is one of the highest measured in human populations. Our results are consis￾tent with previous inferences that 370A must have arisen prior to 15,000 BP (Bryk et al., 2008; Peter et al., 2012) and the first peopling of the Americas (Goebel et al., 2008; O’Rourke and Raff, 2010) but also suggest that the allele likely emerged in East Asia even earlier. It should be noted that haplotype-based methods, such as that used by Bryk and colleagues (Bryk et al., 2008) assume recombination occurs between distinct haplotypes. However, in a case of rapid local fixation, as is likely for a strongly selected and semidominant allele like 370A, recombination of the selected haplotype with itself would be masked, reducing the observed number of recombinations and leading to underestimation of the time of origin (Figure S7). Thus, our findings shift the context in which to consider the selective forces that could have acted on 370A. Phenotypic Consequences of 370A A comparison of mice harboring the 370V and 370A alleles on the same genetic background revealed multiple differences, including increased hair thickness, increased eccrine gland number, reduced mammary fat pad size, and increased mammary gland branch density in mice carrying 370A. With the exception of mammary fat pad size, which has not been analyzed in gain-of-function models, these phenotypes are expected if 370A confers modestly enhanced signaling activity on EDAR. This mode of action was previously proposed based on the observation that 370A can potentiate NFkB signaling in vitro (Bryk et al., 2008; Mou et al., 2008) and a clinical case report in which 370A was associated with reduced severity of hypohidrotic ectodermal dysplasia caused by an EDA missense mutation (Cluzeau et al., 2012). The hair phenotype of 370A animals is consistent with this model, as is the increase in eccrine gland number and mammary gland branching. We demon￾strated this in vivo for the eccrine gland trait by showing 370A 696 Cell 152, 691–702, February 14, 2013 ª2013 Elsevier Inc

Cell B E品 370v37 04 Figure 5. 370A Reduces the Size of the Mammary Fat Pad and Increases Mammary Gland Branch Density iD) Whole mount preparations of stained mammary glands. (A)Gland area(dotted line)and fat pad area(dashed line) are quantified from the main lactiferous duct(arrow head). Representative images are shown of 370v(B), 370v/370A (C), and 370A (D). (E and F) Average branch density (+sEM)(E and mean fat pad area(SEM)()are shown. Significance levels by ANOVA post hoc tests: p<0.05(),p<0.01(), p<0.001("). See also Table S5. escued the reduction in eccrine gland number of d heterozy- 370A is not sufficient to cause a significant change in either gotes. The finding that 370A mice have reduced fat pads reveals Meibomian or mammary gland size. Our results suggest the a hereto unappreciated role for Ectodysplasin signaling in regu- magnitude of the effects exerted by the 370A allele are more Dtem2:的 vel of change in Ectodysplasin signaling. This inference shifts The differences between the 370A knockin mouse phenotypes the discussion of potential adaptive consequences of 370 and those of loss- and gain-of-function models emphasize the toward other ectodermal appendages, which were affected in advantage of a more accurate mouse model. a dramatic change our model. This highlights the importance of employing as close hair size and shape and a disordered hair coat are observed a genetic mouse model as possible as a proxy for studying hen Ectodysplasin signaling is strongly augmented in Edar human genetic variation. transgenic mice carrying multiple copies of the wild-type Edar gene(Mou et aL, 2008)or in K14-Eda-A1 transgenic mice(Cui Pleiotropic Effects and Potential Selective Forces et al., 2003; Mustonen et aL., 2003). In contrast, the changes Favoring the 370A Allele directly attributable to the 370A allele are in the same direction, Our study provides evidence that 370A was selected in East but a smooth hair coat with subtle changes in hair size is observed. Asia, but the question of which of its observed pleiotropic pheno- everal of the other phenotypes observed in stronger gain-of- types were adaptations and which were exaptations remains. function models are not detected in 370A mice. In particular, One possibility is that selection favored individuals with an Cell 152, 691-702, February 14, 2013(2013 Elsevier Inc. 697

rescued the reduction in eccrine gland number of dlj heterozy￾gotes. The finding that 370A mice have reduced fat pads reveals a hereto unappreciated role for Ectodysplasin signaling in regu￾lating the formation not only of the mammary gland, in which EDAR is expressed (Pispa et al., 2003), but also of the surrounding mesenchymal support tissue. The differences between the 370A knockin mouse phenotypes and those of loss- and gain-of-function models emphasize the advantage of a more accurate mouse model. A dramatic change in hair size and shape and a disordered hair coat are observed when Ectodysplasin signaling is strongly augmented in Edar transgenic mice carrying multiple copies of the wild-type Edar gene (Mou et al., 2008) or in K14-Eda-A1 transgenic mice (Cui et al., 2003; Mustonen et al., 2003). In contrast, the changes directly attributable to the 370A allele are in the same direction, but a smooth hair coat with subtle changes in hair size is observed. Several of the other phenotypes observed in stronger gain-of￾function models are not detected in 370A mice. In particular, 370A is not sufficient to cause a significant change in either Meibomian or mammary gland size. Our results suggest the magnitude of the effects exerted by the 370A allele are more modest than those modeled to date, and that a different subset of ectodermal appendages may be preferentially sensitive to this level of change in Ectodysplasin signaling. This inference shifts the discussion of potential adaptive consequences of 370A toward other ectodermal appendages, which were affected in our model. This highlights the importance of employing as close a genetic mouse model as possible as a proxy for studying human genetic variation. Pleiotropic Effects and Potential Selective Forces Favoring the 370A Allele Our study provides evidence that 370A was selected in East Asia, but the question of which of its observed pleiotropic pheno￾types were adaptations and which were exaptations remains. One possibility is that selection favored individuals with an Figure 5. 370A Reduces the Size of the Mammary Fat Pad and Increases Mammary Gland Branch Density (A–D) Whole mount preparations of stained mammary glands. (A) Gland area (dotted line) and fat pad area (dashed line) are quantified from the main lactiferous duct (arrow head). Representative images are shown of 370V (B), 370V/370A (C), and 370A (D). (E and F) Average branch density (±SEM) (E) and mean fat pad area (±SEM) (F) are shown. Significance levels by ANOVA post hoc tests: p < 0.05 (*), p < 0.01 (**), p < 0.001 (***). See also Table S5. Cell 152, 691–702, February 14, 2013 ª2013 Elsevier Inc. 697

Cel G E150 a370v/370A 370V370A 370A Figure 7. 370A Is Associated with Increased Eccrine Sweat Gland Density in Humans of a 370v/370A heterozygous (A) and a 370A homozygous( B) individual ed size is 1. 30cm Average sweat gland density is shown for each genotype(+SEM) Significance of difference by two tailed t test: p<0.05 0). See also Table S6. increased number of eccrine glands. a high density of eccrine glands is a key hominin adaptation that enables efficient evapo- traspiration during vigorous activities such as long-distance walking and running(Carrier et al., 1984; Bramble and Lieber- 370v379K man, 2004). An increased density of eccrine glands in 370A 370A/379K carriers might have been advantageous for East Asian hunter- gatherers during warm and humid seasons, which hinder evapotranspiration. Geological records indicate that China was relatively warm and humid between 40,000 and 32, 000 years ago, but between 32,000 and 15,000 years ago the climate became cooler FP.5 nd drier before warming again at the onset of the Holocene ng 1; Yuan et al., 2004). Throughout this time period, however, China may have remained relatively humid Figure 6. 370A Increases the Number of Eccrine Sweat Glands in Mice lue to varying contribution from summer and winter monsoons -C) Representative whole-mount preparations of the volar hindfoot skin of 70V (A, 370V/370A (B), and 370A(C)mice. Gland ducts appear as thin blue iew of FP-3 from 370V(D), 370V/370A(E, and 370A(F) ("). Quantification of average gland number per FP (SEM)across the three All experiments were carried out on fifth on FvB backcross animals but those shown in (H) were a FVB by C3HeB/FeJ outcross. The difference in 370A rescues the decrease in eccrine gland number in 379K heterozy- genetic background accounts for the difference in eccrine gland number gous mutant mice. Average gland number per FP is shown (+SEM). Signifi- between wild-type animals in(G and H). See also Table s5. 698 Cell 152, 691-702, February 14, 2013@2013 Elsevier Inc

increased number of eccrine glands. A high density of eccrine glands is a key hominin adaptation that enables efficient evapo￾traspiration during vigorous activities such as long-distance walking and running (Carrier et al., 1984; Bramble and Lieber￾man, 2004). An increased density of eccrine glands in 370A carriers might have been advantageous for East Asian hunter￾gatherers during warm and humid seasons, which hinder evapotranspiration. Geological records indicate that China was relatively warm and humid between 40,000 and 32,000 years ago, but between 32,000 and 15,000 years ago the climate became cooler and drier before warming again at the onset of the Holocene (Wang et al., 2001; Yuan et al., 2004). Throughout this time period, however, China may have remained relatively humid due to varying contribution from summer and winter monsoons Figure 6. 370AIncreases the Number of Eccrine Sweat Glands inMice (A–C) Representative whole-mount preparations of the volar hindfoot skin of 370V (A), 370V/370A (B), and 370A (C) mice. Gland ducts appear as thin blue tubes emerging from inside the footpads (FP). (D–F) Detail view of FP-3 from 370V (D), 370V/370A (E), and 370A (F) mice. (G) Quantification of average gland number per FP (±SEM) across the three genotypes. (H) 370A rescues the decrease in eccrine gland number in 379K heterozy￾gous mutant mice. Average gland number per FP is shown (±SEM). Signifi- cance of differences by ANOVA post hoc tests: p < 0.05 (*), p < 0.01 (**), p < 0.001 (***). All experiments were carried out on fifth generation FVB backcross animals, but those shown in (H) were a FVB by C3HeB/FeJ outcross. The difference in genetic background accounts for the difference in eccrine gland number between wild-type animals in (G and H). See also Table S5. Figure 7. 370A Is Associated with Increased Eccrine Sweat Gland Density in Humans (A and B) Representative cropped active sweat gland images of the digit tips of a 370V/370A heterozygous (A) and a 370A homozygous (B) individual. Cropped size is
1.30cm2 . (C) Active sweat gland density is significantly increased in 370A individuals. Average sweat gland density is shown for each genotype (±SEM). Significance of difference by two tailed t test: p < 0.05 (*). See also Table S6. 698 Cell 152, 691–702, February 14, 2013 ª2013 Elsevier Inc

Cell (Sun et al., 2012). High humidity, especially in the summers, may traits such as hair phenotypes and breast size. This creates have provided a seasonally selective advantage for individuals conditions in which biases in mate preference could rapidly better able to functionally activate more eccrine glands and evolve and reinforce more direct competitive advantages. Is sweat more effectively(Kuno, 1956). To explore this hypoth- Consequently, the cumulative selective force acting over time esis, greater precision on when and where the allele was under on diverse traits caused by a single pleiotropic mutation could selection-perhaps using ancient DNA sources-in conjunction have driven the rise and spread of 370A with more detailed archaeological and climatic data are needed er E branching or fat pad size could have been adaptive. The increased branching of 370A mouse mammary glands and th Haplotype Analysis importance of mammary tissue in evolutionary fitness(Anderson DNA from1, 064 individuals from 52 global populations was acquired from the et al., 1983; Oftedal, 2002)make this organ an interesting candi- uman Genome Diversity Panel(HGDP-CEPH)(Cann et aL, 2002). We used MassARRAY iPLEX Gold to pe 48 SNPs in a 0.83cM date. Alterations in gland structure have been reported to disrupt (1, 400 kb) region surrounding the 370A allele and combined our data with lactation in mice(Ramanathan et al., 2007), suggesting a func- published HGDP-CEPH data(Jakobsson et al., 2008; Li et al, 2008). After tional consequence for this change Unfortunately it is not excluding monomorphic SNPs and SNPs with inconsistent genotypes, we highlighting the importance of animal models. Reports of smaller from 51 populations. we interred haplotype data by phasing with fastPHASE breast size in East Asian women(Maskarinec et al., 2001; Chen in the region revealed a 139 kb block surrounding 370A(Figure S1).We al.,2004)are notable in light of the effects of 370A on fat pad counted the number of chromosomes of each haplotype in each population size and the importance of breast morphology in human mate and plotted the frequencies on a world map. preference(Furnham et al., 1998, 2006: Dixson et aL, 2011). Further analysis of the functional implications of 370A in the Forward simulation mouse and development of methods to assay these phenotypes The spatially explicit model takes into account evolutionary processes such as in humans are critical to evaluate such hypotheses and also to population structure, drift, and natural selection(tan et al,2009),imple- demographic processes, including population growth, sporadic long-range gated, such as those linked to differential susceptibility to respi- migration, cultural diffusion of farming technology gene flow between demes ratory disease(Clarke et aL., 1987; Mauldin et aL., 2009) and between cultural groups, and the effects of the spread of farming on In light of 370A's pleiotropy it is possible that selection acted carrying capacities (see Modeling the origins and Sprea on multiple traits. The tendency to seek a single driving character Approximate Bayesian Computation Framework and Figure S2 for details) is underlain by the perception that pleiotropic changes are inherently disadvantageous. Evolution is believed to proceed We applied an ABC inference framework to estimate p of interest primarily through mutations in gene regulatory regions rather (Bertorelle et al, 2010: Csillery et al, 2010). We compared summary statistics han exons because this reduces pleiotropic effects(King and (370A allele frequency in 29 populations) reco Wilson, 1975; Stern, 2000; Carroll, 2008 ) From the perspective observed of this model, a specific effect of 370A,s pleiotropic conse- the differences were sufficiently small. We calculated the Euclidean distance quences was favored under the conditions present in East Asia (8)between the simulated and observed statistics for each simulated data and conferred an advantage with other neutral or deleterious set and retained those with the smallest values. Parameter sets were obtained according to the associated 8. See Modeling the origins and spread of 370A in traits hitchhiking along with the selected trait. However, the large an Approximate Bayesian Computation Framework for details on posterior coefficient of selection on 370A contrasts with the relatively density and choice of simulation cut-off modest magnitude of structural changes on any one affected trait and suggests alternative interpretations. One possibility is Mouse strains that the effects of 370A were magnified by coselection on o construct mice carrying the 3/A allele, a targeting vector conta another variant. For example, a coding variant of the related 326C point mutation was injected into J1 ES cells(Brigham and Women's EDA2R gene affects human hair and has swept to fixation Hospital Transgenic Mouse Facility, Boston USA). Chimeric mice that trans- hitted the knockin allele to the germline were recovered from a correctly East Asia ( Sabeti et al., 2007; Prodi et al., 2008) rgeted clone after injection into C57BL/6 blastocysts. The chimera was Alternatively, it could be precisely the pleiotropic nature of bred to a ubiquitously expressing B-Actin Cre line(gift from Susan Dymecki 370A that allowed multiple distinct selective forces to act on Harvard Medical School, Boston USA. Mice subsequently bred onto this variant over its long history, when many of the postulated an PVB (Charles River Laboratories)background for five generations, by which selective pressures such as temperature and humidity changed 370A Knockin Mouse for details of knockin construction). We obtained df mice dramatically. The fact that EDAR acts mostly on ectodermal on a C3HeB/FeJ background from Jackson Laboratories and crossed them ppendages and that the phenotypic effects of the 370A allele onto an FVB background for one generation. Mouse work was performed in are not extreme reduces the costs of pleiotropy and would facil- accordance with protocols approved by the Harvard M ate this process. Thus, what were initially neutral changes in Committee on Anima some appendages driven by 370A would gain adaptive signifi- cance in the face of new selective pressures. It is worth noting Mouse Hair Size lair from the back of pl9-P21 that largely invisible structural changes resulting from the 370A nted on slides in Gelvatol and allele that might confer functional advantage, such as increased analyzed on a Nikon eclipse cope to score medulla cells. A minimum of 700 hairs were sc mouse. 370V (n= 12), 370V/ eccrine gland number, are directly linked to visually obvious 370A(n=11), and 370A(n Cell 152, 691-702, February 14, 2013(2013 Elsevier Inc. 699

(Sun et al., 2012). High humidity, especially in the summers, may have provided a seasonally selective advantage for individuals better able to functionally activate more eccrine glands and thus sweat more effectively (Kuno, 1956). To explore this hypoth￾esis, greater precision on when and where the allele was under selection—perhaps using ancient DNA sources—in conjunction with more detailed archaeological and climatic data are needed. Alternatively, another phenotype, such as mammary gland branching or fat pad size could have been adaptive. The increased branching of 370A mouse mammary glands and the importance of mammary tissue in evolutionary fitness (Anderson et al., 1983; Oftedal, 2002) make this organ an interesting candi￾date. Alterations in gland structure have been reported to disrupt lactation in mice (Ramanathan et al., 2007), suggesting a func￾tional consequence for this change. Unfortunately, it is not possible to assess mammary gland branching in living humans, highlighting the importance of animal models. Reports of smaller breast size in East Asian women (Maskarinec et al., 2001; Chen et al., 2004) are notable in light of the effects of 370A on fat pad size and the importance of breast morphology in human mate preference (Furnham et al., 1998, 2006; Dixson et al., 2011). Further analysis of the functional implications of 370A in the mouse and development of methods to assay these phenotypes in humans are critical to evaluate such hypotheses and also to analyze additional potential 370A phenotypes yet to be investi￾gated, such as those linked to differential susceptibility to respi￾ratory disease (Clarke et al., 1987; Mauldin et al., 2009). In light of 370A’s pleiotropy, it is possible that selection acted on multiple traits. The tendency to seek a single driving character is underlain by the perception that pleiotropic changes are inherently disadvantageous. Evolution is believed to proceed primarily through mutations in gene regulatory regions rather than exons because this reduces pleiotropic effects (King and Wilson, 1975; Stern, 2000; Carroll, 2008). From the perspective of this model, a specific effect of 370A’s pleiotropic conse￾quences was favored under the conditions present in East Asia and conferred an advantage with other neutral or deleterious traits hitchhiking along with the selected trait. However, the large coefficient of selection on 370A contrasts with the relatively modest magnitude of structural changes on any one affected trait and suggests alternative interpretations. One possibility is that the effects of 370A were magnified by coselection on another variant. For example, a coding variant of the related EDA2R gene affects human hair and has swept to fixation in East Asia (Sabeti et al., 2007; Prodi et al., 2008). Alternatively, it could be precisely the pleiotropic nature of 370A that allowed multiple distinct selective forces to act on this variant over its long history, when many of the postulated selective pressures such as temperature and humidity changed dramatically. The fact that EDAR acts mostly on ectodermal appendages and that the phenotypic effects of the 370A allele are not extreme reduces the costs of pleiotropy and would facil￾itate this process. Thus, what were initially neutral changes in some appendages driven by 370A would gain adaptive signifi- cance in the face of new selective pressures. It is worth noting that largely invisible structural changes resulting from the 370A allele that might confer functional advantage, such as increased eccrine gland number, are directly linked to visually obvious traits such as hair phenotypes and breast size. This creates conditions in which biases in mate preference could rapidly evolve and reinforce more direct competitive advantages. Consequently, the cumulative selective force acting over time on diverse traits caused by a single pleiotropic mutation could have driven the rise and spread of 370A. EXPERIMENTAL PROCEDURES Haplotype Analysis DNA from1,064 individuals from 52 global populations was acquired from the Human Genome Diversity Panel (HGDP-CEPH) (Cann et al., 2002). We used Sequenom MassARRAY iPLEX Gold to genotype 48 SNPs in a 0.83cM (
1,400 kb) region surrounding the 370A allele and combined our data with published HGDP-CEPH data (Jakobsson et al., 2008; Li et al., 2008). After excluding monomorphic SNPs and SNPs with inconsistent genotypes, we obtained a final data set of 280 SNPs in 984 samples (1,968 chromosomes) from 51 populations. We inferred haplotype data by phasing with fastPHASE (Scheet and Stephens, 2006). Examination of linkage disequilibrium patterns in the region revealed a
139 kb block surrounding 370A (Figure S1). We counted the number of chromosomes of each haplotype in each population and plotted the frequencies on a world map. Forward Simulation The spatially explicit model takes into account evolutionary processes such as population structure, drift, and natural selection (Itan et al., 2009), imple￾mented here for a semidominant allele. The model also considers various demographic processes, including population growth, sporadic long-range migration, cultural diffusion of farming technology, gene flow between demes and between cultural groups, and the effects of the spread of farming on carrying capacities (see Modeling the Origins and Spread of 370A in an Approximate Bayesian Computation Framework and Figure S2 for details). Approximate Bayesian Computation We applied an ABC inference framework to estimate parameters of interest (Bertorelle et al., 2010; Csille´ry et al., 2010). We compared summary statistics (370A allele frequency in 29 populations) recorded after each simulation to observed frequencies (Table S1) and accepted only those simulations in which the differences were sufficiently small. We calculated the Euclidean distance (d) between the simulated and observed statistics for each simulated data set and retained those with the smallest values. Parameter sets were obtained according to the associated d. See Modeling the Origins and Spread of 370A in an Approximate Bayesian Computation Framework for details on posterior density and choice of simulation cut-off. Mouse Strains To construct mice carrying the 370A allele, a targeting vector containing the T1326C point mutation was injected into J1 ES cells (Brigham and Women’s Hospital Transgenic Mouse Facility, Boston USA). Chimeric mice that trans￾mitted the knockin allele to the germline were recovered from a correctly targeted clone after injection into C57BL/6 blastocysts. The chimera was bred to a ubiquitously expressing b-Actin Cre line (gift from Susan Dymecki, Harvard Medical School, Boston USA). Mice were subsequently bred onto an FVB (Charles River Laboratories) background for five generations, by which the Cre transgene was also removed (Generation and Statistical Analysis of the 370A Knockin Mouse for details of knockin construction). We obtained dlj mice on a C3HeB/FeJ background from Jackson Laboratories and crossed them onto an FVB background for one generation. Mouse work was performed in accordance with protocols approved by the Harvard Medical Area Standing Committee on Animals. Mouse Hair Size Hair from the back of P19–P21 pups was mounted on slides in Gelvatol and analyzed on a Nikon eclipse E1000 microscope to score medulla cells. A minimum of 700 hairs were scored from each mouse. 370V (n = 12), 370V/ 370A (n = 11), and 370A (n = 13) animals were analyzed. Cell 152, 691–702, February 14, 2013 ª2013 Elsevier Inc. 699

Cel Meibomian Glands selected in this scheme can be seen as a random sampling for the 370A gland preparations were made from the left eyelids were fixed flat on Whatman paper in 4% Paraf Detailed phenotype collection and calling procedure as well as statistical er), and photographed on a Leica MZFLIll ster methods are provided in Association Study of 370A in a Han Chinese Popula quipped with a Nikon DXM1200F camera Total glandular area w tion. We performed all human subjects work in accordance with approved om images using ImageJ (v 1.46, (Schneider et al., 2012)-3 protocols by Fudan and Harvard Universities. 370V/370A (n= 25), and 370A (n= 20)animals were analyzed SUPPLEMENTAL INFORMATION The 4th and gh inguinal mammary glands and associated fat pads were supplement si t ental Proce issected from 6-week-old, virgin female mice. Whole mount preparations figuresandsixtablesandcanbefoundwiththisarticleonlineathttp:/dx.doi mammary glands and staining of the ductal tree were performed as org/10.1016cel201301016 /ndex html Mammary glands were fixed flat in Carnoy's fix then stained with carmine alum(Sigma Aldrich). Stained glands were dehydrated into Xylenes (Thermo ACKNOWLEDGMENTS Fisher), mounted flat, and photographed on a Leica MZFLIll s quipped with a Nikon DXM1200F camera. Several high-resolution images of This project was funded by a grant from the Harvard University Science and each gland were merged for analysis in Photoshop (Adobe Systems ). Engineering Committee Seed Fund for Interdisciplinary Science to CJT, ce and Engi- were measured from the main lactiferous duct to the dorsolateral edge of the neenng and an NIH Innovator Award 1DP20D006514-01 to PCS: a BIRT gland Gland length was measured between the distal-most ductal termini at Award AR055256-04S1 from NIAMS to BAM; an NIH grant R37 HD032443 her end of the gland. Total branch number was assessed by counting all CJT; and funding from the american School of Prehistoric Research to uctal termini per gland using the Image Cell Counter plug- DEL Human association study work was additionally supported by NSFC was calculated by dividing total branch number by gland length. Left and 30890034: MOST 2011BAI09B00: MOH 201002007 to LJ. YI was funded by right glands of each animal were averaged together. 370v (n 10), 370v/ an AXA Research Fund postdoctoral fellowship and PG by the LeCHE Marie 70A (n= 19), and 370A(n= 11)animals were assessed R37 054364 grant. We acknowledge the UCL Legion High Performance Mouse Eccrine Glands omputing Facility and support services and thank O. Bar Yosef, C. Zhao E. Rohling, S Schaffner, L. Gaffney, C. Edwards, J. Vitti, S Tabrizi and A Tariel skin from both hindfeet and ating in Dispase ll(Roche)as described for feedback on the manuscript and A Carpenter, C walby, and M Moro previously(Okada et al, 1983).Next, the epidermis was peeled away from for data analysis help. The authors declare no conflict of interest the underlying demis. Eccrine gland ducts remained associated with the 21,2012 ns Revised: November 22.2012 also stained with 0.5% Oil Red O(Sigma), which stains sebaceous Accepted: January 4, 2013 glands. Footpads 1 and 2 were not analyzed because their high eccrine gland Published: February 14, 2013 density prevents accurate scoring. The number of eccrine glands per footpad indfeet. 370V(n= 17), 370V/370A(n= 18), and REFERENCES 70A(n= 16)animals were assessed to evaluate the effect of 370A on eccrine gland number and 370V/379E(n= 12), 370V/379K (n= 11), and 370A/379K(n= Akey, JM.(2009). humans: where do we go from here? Genome Res. 19, 711-722. Details of all statistical tests of mouse data are reported in Generation and Philippe, P, Raphael, D, Stini, W.A., and Esterik, P v (1983 ). The Reproduc Beaumont, M.A, Zhang W, and Balding, D.J.(2002). Approximate Bayesian Association Study computation in population genetics. Genetics 162, 2025-2035 Population le studied a Han chinese Bertorelle, G, Benazzo, A, and Mona, S(2010. ABC as a flexible framework te demography over space and time some cons, many pros. MoL. from five closely located villages in Taizhou(ages 35-65) and local students of C Taizhou Professional Technology College(ages 18-21). All participants spent Bramble, D M, and Lieberman, D.E. (2004). Endurance running and the evolu- the majority or entirety of their lives in Taizhou and are expected to be tion of Homo. Nature 432, 345-352 Hardouin, E, A., DNA Extraction, Genotyping, and Sample and Myles, S.(2008). Positive selection in East As EDAR allele DNA extraction and at Fudan ur ity. Upon that enhances NF-kappaB activation. PLOS ONE 3, e2209. enrollment in the cohort, each participant's blood samples collected el, V, Piouffre, L, and stored in the cohort database. DNA was isolated using standard phenol Bodmer. J. Bodmer W.F. Bonne-Tamir. B. Cambon-Thomsen. A. et al hloroform extraction. The EDAR SNP,rs3827760, was genotyped using the (2002 ). A human genome diversity cell line panel. Science 296, 261-262 NaPshot Multiplex System which included seven other SNPs that showed signatures of positive selection in East Asia. Genotype calling was performe Carrier, D R, Kapoor, A K, Kimura, T, Nickels, MK, Satwanti, Scott, E.C. y GeneMapper v2.0. We compiled a priority list of potential study participan So, J K, and Trinkaus, E(1984). The Energetic Paradox of Human Running ased on genotype results. 370V allele cariers had top priority, followed by the and Hominid Evolution. Curr. Anthropol. 25, 483-495. allele carriers of the other seven SNPs. From the 2, 572 samples geno. Carroll, SB(2008). EVO-devo and an expanding typed, we contacted the top 1, 000 individuals, and enrolled 623 in this study n Cell 13 (427 from the villages and 196 from the college). Among the other SNPs Chang, S.H., Jobling, S, Brennan, K, and Headon, Enhanced genotyped, none are on the same chromosome as 370A, and none were asso- Edar signalling has pleiotropic effects on craniofacial ciated with 370A. Therefore. we concluded that the 437 370A individuals PLos oNE 4. 07591 700cel152,691-702, February14,2013e2013Es

Meibomian Glands Meibomian gland preparations were made from the left eyelids of 6-week-old mice. Eyelids were fixed flat on Whatman paper in 4% Paraformaldehyde (Thermo Fisher), and photographed on a Leica MZFLIII stereomicroscope equipped with a Nikon DXM1200F camera. Total glandular area was measured from images using ImageJ (v.1.46, (Schneider et al., 2012)). 370V (n = 13), 370V/370A (n = 25), and 370A (n = 20) animals were analyzed. Mammary Glands The 4th and 9th inguinal mammary glands and associated fat pads were dissected from 6-week-old, virgin female mice. Whole mount preparations of mammary glands and staining of the ductal tree were performed as described (http://mammary.nih.gov/tools/histological/Histology/index.html#a1). Mammary glands were fixed flat in Carnoy’s fix then stained with carmine alum (Sigma Aldrich). Stained glands were dehydrated into Xylenes (Thermo Fisher), mounted flat, and photographed on a Leica MZFLIII stereomicroscope equipped with a Nikon DXM1200F camera. Several high-resolution images of each gland were merged for analysis in Photoshop (Adobe Systems). Image analysis was performed with ImageJ. Fat pad area and glandular area were measured from the main lactiferous duct to the dorsolateral edge of the gland. Gland length was measured between the distal-most ductal termini at either end of the gland. Total branch number was assessed by counting all ductal termini per gland using the ImageJ Cell Counter plug-in. Branch density was calculated by dividing total branch number by gland length. Left and right glands of each animal were averaged together. 370V (n = 10), 370V/ 370A (n = 19), and 370A (n = 11) animals were assessed. Mouse Eccrine Glands Epidermal whole-mount preparations were prepared by dissecting the volar skin from both hindfeet and incubating in Dispase II (Roche) as described previously (Okada et al., 1983). Next, the epidermis was peeled away from the underlying dermis. Eccrine gland ducts remained associated with the epidermis and were stained with a 0.1% solution of Nile Blue A (Sigma Aldrich) and observed on a Leica MZFLIII stereomicroscope. Epidermal preparations were also stained with 0.5% Oil Red O (Sigma), which stains sebaceous glands. Footpads 1 and 2 were not analyzed because their high eccrine gland density prevents accurate scoring. The number of eccrine glands per footpad was averaged across both hindfeet. 370V (n = 17), 370V/370A (n = 18), and 370A (n = 16) animals were assessed to evaluate the effect of 370A on eccrine gland number and 370V/379E (n = 12), 370V/379K (n = 11), and 370A/379K (n = 13) animals were analyzed to evaluate the effect of 370A on the 379K mutation in separate crosses. Glands were analyzed from mice aged 3 to 6 weeks. Details of all statistical tests of mouse data are reported in Generation and Statistical Analysis of the 370A Knockin Mouse and Table S5. Association Study Population We studied a Han Chinese population from an established Taizhou longitudinal cohort in Jiangsu Province, China (Wang et al., 2009), that recruited individuals from five closely located villages in Taizhou (ages 35–65) and local students of Taizhou Professional Technology College (ages 18–21). All participants spent the majority or entirety of their lives in Taizhou and are expected to be homogeneous. DNA Extraction, Genotyping, and Sample Selection DNA extraction and genotyping were performed at Fudan University. Upon enrollment in the cohort, each participant’s blood samples was collected and stored in the cohort database. DNA was isolated using standard phenol/ chloroform extraction. The EDAR SNP, rs3827760, was genotyped using the SNaPshot Multiplex System which included seven other SNPs that showed signatures of positive selection in East Asia. Genotype calling was performed by GeneMapper v2.0. We compiled a priority list of potential study participants based on genotype results. 370V allele carriers had top priority, followed by the rare allele carriers of the other seven SNPs. From the 2,572 samples geno￾typed, we contacted the top 1,000 individuals, and enrolled 623 in this study (427 from the villages and 196 from the college). Among the other SNPs genotyped, none are on the same chromosome as 370A, and none were asso￾ciated with 370A. Therefore, we concluded that the 437 370A individuals selected in this scheme can be seen as a random sampling for the 370A association study. Detailed phenotype collection and calling procedure as well as statistical methods are provided in Association Study of 370A in a Han Chinese Popula￾tion. We performed all human subjects work in accordance with approved protocols by Fudan and Harvard Universities. SUPPLEMENTAL INFORMATION Supplemental Information includes Extended Experimental Procedures, seven figures, and six tables and can be found with this article online at http://dx.doi. org/10.1016/j.cell.2013.01.016. ACKNOWLEDGMENTS This project was funded by a grant from the Harvard University Science and Engineering Committee Seed Fund for Interdisciplinary Science to CJT, PCS, BAM and DEL; a Packard Foundation Fellowship in Science and Engi￾neering and an NIH Innovator Award 1DP2OD006514-01 to PCS; a BIRT Award AR055256-04S1 from NIAMS to BAM; an NIH grant R37 HD032443 to CJT; and funding from the American School of Prehistoric Research to DEL. Human association study work was additionally supported by NSFC 30890034; MOST 2011BAI09B00; MOH 201002007 to LJ. YI was funded by an AXA Research Fund postdoctoral fellowship and PG by the LeCHE Marie Curie FP7 framework. Work in APM’s laboratory was supported by an NIH R37 054364 grant. We acknowledge the UCL Legion High Performance Computing Facility and support services and thank O. Bar Yosef, C. Zhao, E. Rohling, S. Schaffner, L.Gaffney, C. Edwards, J. Vitti, S.Tabrizi and A.Tariela for feedback on the manuscript and A.Carpenter, C. Wa¨ hlby, and M. Morgan for data analysis help. The authors declare no conflict of interest. Received: September 21, 2012 Revised: November 22, 2012 Accepted: January 4, 2013 Published: February 14, 2013 REFERENCES Akey, J.M. (2009). Constructing genomic maps of positive selection in humans: where do we go from here? Genome Res. 19, 711–722. Anderson, P., Frisch, R.E., Graham, C.E., Manderson, L., Orubuloye, I.O., Philippe, P., Raphael, D., Stini, W.A., and Esterik, P.V. (1983). The Reproduc￾tive Role of the Human Breast. Curr. Anthropol. 24, 25–45. Beaumont, M.A., Zhang, W., and Balding, D.J. (2002). Approximate Bayesian computation in population genetics. Genetics 162, 2025–2035. Bertorelle, G., Benazzo, A., and Mona, S. (2010). ABC as a flexible framework to estimate demography over space and time: some cons, many pros. Mol. Ecol. 19, 2609–2625. Bramble, D.M., and Lieberman, D.E. (2004). Endurance running and the evolu￾tion of Homo. Nature 432, 345–352. Bryk, J., Hardouin, E., Pugach, I., Hughes, D., Strotmann, R., Stoneking, M., and Myles, S. (2008). Positive selection in East Asians for an EDAR allele that enhances NF-kappaB activation. PLoS ONE 3, e2209. Cann, H.M., de Toma, C., Cazes, L., Legrand, M.-F., Morel, V., Piouffre, L., Bodmer, J., Bodmer, W.F., Bonne-Tamir, B., Cambon-Thomsen, A., et al. (2002). A human genome diversity cell line panel. Science 296, 261–262. Carrier, D.R., Kapoor, A.K., Kimura, T., Nickels, M.K., Satwanti, Scott, E.C., So, J.K., and Trinkaus, E. (1984). The Energetic Paradox of Human Running and Hominid Evolution. Curr. Anthropol. 25, 483–495. Carroll, S.B. (2008). Evo-devo and an expanding evolutionary synthesis: a genetic theory of morphological evolution. Cell 134, 25–36. Chang, S.H., Jobling, S., Brennan, K., and Headon, D.J. (2009). Enhanced Edar signalling has pleiotropic effects on craniofacial and cutaneous glands. PLoS ONE 4, e7591. 700 Cell 152, 691–702, February 14, 2013 ª2013 Elsevier Inc