Genomic Data Reveal a Complex Making of Humans

There is great diversity in primates with regard to hair appearance, including its distribution, shape and colour1. Hair plays a range of important functions, including thermal regulation, camouflage, sensory and social signalling, and the evolution of hair has been proposed to be influenced by both natural and sexual selection1. Humans differ from other primates in having lost most terminal body hair (via hair follicle miniaturization), possibly in connection to the adaptive development of more efficient sweating, linked to bipedalism2, 3. However, considerable head hair has been retained in modern humans and its appearance shows extensive variation between individuals4, 5, 6. Head hair appearance is highly heritable7, 8, 9, 10 and certain traits show high differentiation between continental native populations. For instance, variation in hair colour is essentially restricted to West Eurasia11, whereas straight hair is virtually absent from sub-Saharan Africa4. It has been proposed that variation in head hair appearance has been influenced by selection during the evolution of modern humans2, 3, 11. Although genetic association studies for human hair traits are few to date, loci influencing male pattern baldness, scalp hair colour and shape (curliness) have been identified in samples of European and East Asian ancestry6. Consistent with the key role of androgens in balding, the androgen receptor (AR) gene region has been highlighted as a major determinant among several loci associated with male-pattern baldness12. The genes associated with hair colour, involved in various aspects of melanocyte biology and melanin synthesis, also often impact on skin and eye pigmentation13. Interestingly, different genes have been associated with straight hair in Europeans and East Asians, suggesting that this trait evolved independently at least twice. The most robust associations for straight hair have implicated Trichohyalin (TCHH, a structural hair protein) in Europeans14, 15, and EDAR (a cell signalling receptor) in East Asians16, illustrating the range of cellular mechanisms that can impact on hair shape.


At the top are shown drawings illustrating the seven hair features examined in the CANDELA study sample. Thick lines connect these features with the candidate genes identified in regions with SNPs reaching genome-wide significant association (Table 1).

Several of the regions showing genome-wide significant association include strong candidate genes. Scalp hair shape and beard thickness are strongly associated with SNPs in 2q12 (Fig 3a,b) and two other traits (eyebrow thickness and monobrow) show genome-wide suggestive P values for SNPs in this same region (Supplementary Fig. 6). Associated SNPs overlap the EDAR (ectodysplasin A receptor) gene. EDAR acts as part of the EDA-EDAR-EDARADD signalling pathway23 during prenatal development to specify the location, size and shape of ectodermal appendages, such as hair follicles, teeth and glands23. Strongest association with hair shape was observed for SNP rs3827760 (Fig. 3a and Table 1), coding for a V370A substitution in EDAR. This variant has been robustly associated with hair shape in East Asians16, 24, 25. In a previous study of the CANDELA sample, which examined 30 ancestry informative markers, we found association between hair shape and a SNP in EDAR that is in strong linkage disequilibrium (LD) with rs3827760 in the 1000 genomes data17. The EDAR SNP showing strongest association with beard density is not the coding SNP rs3827760 associated with hair shape but rather rs365060 located ~62 kb upstream of rs3827760 in the first intron of EDAR (Fig. 3b). Several SNPs in this intron have smaller association P values than rs3827760 and analyses conditioning on rs3827760 suggest that the association signal of SNPs in this intron is independent from that observed at rs3827760 (Supplementary Fig. 7). These intronic SNPs are located in a different LD block and there is a recombination hotspot between them and rs3827760 (Fig 3a,b and Supplementary Fig. 7A). Interestingly, the first intron of EDAR is rich in regulatory elements and SNPs in this intron show evidence of recent selection in Europeans (discussed further below). Hypohidrotic ectodermal dysplasia is a Mendelian disorder caused by mutations in the EDA-EDAR-EDARADD pathway and is characterized by sparse scalp hair, eyebrows and eye lashes23. Transgenic mice with increased Edar function have been shown to have thickened and straightened hair fibres26, 27. We therefore examined chin hair follicle density in wild-type mice and in an Edar gain-of-function transgenic strain (EdarTg951/Tg951)26. Consistent with the effect of EDAR on beard thickness, we found that the EdarTg951/Tg951 strain has significantly lower chin hair follicle density compared with wild-type mice (Fig. 4a,b).


Figure 3: Association plots for six regions with SNPs showing genome-wide significant association to hair traits.

Eyebrow thickness shows genome-wide significant association to SNPs on 3q23 overlapping the forkhead box L2 (FOXL2) gene (Fig. 3d). Rare mutations in the FOXL2 gene region (including coding variants as well as upstream and downstream intergenic rearrangements) cause blepharophimosis syndrome (BPES)34, an autosomal dominant eyelid malformation often accompanied by thick eyebrows. Mouse experiments have shown that Foxl2 is expressed around the eyes up to the time that hair is formed (E13.5)35 and a mutant with altered Foxl2 expression (and BPES features) typically shows hair loss around the eyes36.

Monobrow shows genome-wide significant association with SNPs in 2q36 with strongest association being observed for marker rs2395845 located ~70 kb downstream of the paired box gene 3 (PAX3) gene, an interesting candidate in the region (Fig. 3e). Rare mutations of PAX3 have been shown to cause Waardenburg syndrome type 1 (WS1). WS is a clinically and genetically heterogeneous Mendelian disorder of neural crest derivatives whose manifestations include deafness, a range of pigmentation abnormalities, broad nasal bridge and monobrow (seen in ~85% of WS1 patients37). Intronic SNPs within PAX3 have been implicated in recent GWASs of facial morphology, particularly in relation with nasion position (the point just above the nasal bridge)38, 39. PAX3 is a key transcription factor during embryogenesis and analysis of mouse mutants have confirmed that it is essential to guide normal development of neural crest derivatives40.


Figure 4: EDAR effects on mouse facial hair follicle density and expression of PRSS53 in anagen (growing) human hair follicles.

Hair greying shows genome-wide significant association to SNP rs12203592 in intron 4 of the interferon regulatory factor 4 gene (IRF4; Fig. 3f). This SNP also shows association with hair colour in our sample (Table 1) and in previous studies this SNP has been associated with skin, hair and eye pigmentation13. Recent in-vitro studies have shown that rs12203592 impacts on the function of an enhancer element regulating IRF4 expression and the induction of tyrosinase (TYR), a key enzyme in melanin synthesis41. In addition to IRF4, hair colour (but not hair greying) shows genome-wide significant association to four other well-established pigmentation gene regions (SLC45A2, TYR, OCA2/HERC2 and SLC24A5; Table 1 and Supplementary Fig. 8)13. Finally, for balding, we replicate association to the well-established Androgen Receptor/Ectodysplasin A2 Receptor (AR/EDA2R) locus on Xq12 (ref. 12; Supplementary Fig. 8).

The other four genomic regions showing genome-wide association include no strong candidate genes with established roles in hair biology (Table 1 and Supplementary Fig. 8). Beard thickness is associated with SNPs in 7q31, 4q12 and 6q21 where the nearest genes are, respectively: forkhead box P2 (FOXP2), ligand of numb-protein X 1 (LNX1) and prolyl endopeptidase (PREP; Supplementary Fig. 8). None of these genes have known functions specifically related to hair. Similarly, balding is associated with SNPs in the second intron of the Glutamate receptor delta-1 subunit gene (GRID1) on 10q22 but this gene has no documented role in hair biology.


Figure 5: Processing of PRSS53 and signals of selection in the PRSS53 gene region.

The colour of hair results from melanin pigments transferred to hair fibre keratinocytes from hair follicle melanocytes. These melanocytes differentiate from melanoblasts that migrate from the neural crest into hair follicles early in development6. Some hair follicle melanoblasts remain undifferentiated and serve as stem cells for the periodic replenishment of mature melanocytes, melanogenesis occurring only in the anagen phase of the hair growth cycle. Among the several hundred gene products known to participate in melanogenesis, recent association studies have identified a handful that influence hair colour variation in Europeans13. Most of these associations have been replicated here (Table 1), including that of the derived T allele at SNP rs12203592 in IRF4 with lighter hair colour. It has been shown experimentally that IRF4 interacts with the microphthalmia-associated transcription factor (MITF, a key regulator of the expression of many pigment enzymes and differentiation factors), to activate the expression of TYR (a rate-limiting essential enzyme in melanin synthesis). The derived T allele at rs12203592 leads to reduced TYR expression and melanin synthesis, consistent with the association of this allele with lighter hair colour41. In line with the geographic distribution of light hair colour, the T allele at rs12203592 is essentially absent outside Europe (Supplementary Table 7). Interestingly, we find that the T allele at SNP rs12203592 is also associated with increased hair greying. Experimental evidence suggests that the mechanism of hair greying involves incomplete maintenance of melanocyte stem cells in the hair follicle6. Importantly, MITF is known to affect melanocyte survival via its regulation of anti-apoptotic Bcl2 expression, a key factor in protection of the hair follicle against oxidative stress6. To probe the mechanism by which IRF4 might impact on hair greying, it will therefore be important to evaluate whether the T allele at rs12203592 influences MITF in terms of melanoblast stem cell maintenance and survival or via melanocyte loss post differentiation.

The skin on different parts of the body has different hair characteristics, with the final hair distribution dependent upon the spacing pattern laid down during development, the extent of skin growth that occurs subsequent to pattern establishment, and to hormonal and ageing effects. The development of skin at different body sites is known to be controlled by an underlying transcription factor code52 to which FOXL2, FOXP2 and PAX3 may contribute in defining hair distribution on specific areas of the face. As hair of the beard is produced through a two-stage process of embryonic hair follicle patterning followed by a post-pubertal androgen-driven transformation into terminal hair, beard thickness could be modulated by genes acting either prenatally or at puberty. Our finding of reduced hair placode density in embryonic mice with increased Edar expression suggests that the basis for this variation lies in the recognized role of EDAR in developmental hair patterning53. It is likely that EDAR function affects hair follicle density on most or all of the human body, as described in the mouse26, but that on the head this effect is most readily apparent as variation in beard thickness. Analyses focusing on the mechanism by which associated genetic variants affect regional facial hair density should provide insights into developmental patterning in humans, and perhaps yield clues into the genetic basis for the striking modification of hair distribution that has occurred in human evolution54.

Adhikari, K. et al. A genome-wide association scan in admixed Latin Americans identifies loci influencing facial and scalp hair features. Nat. Commun. 7:10815 doi: 10.1038/ncomms10815 (2016).

The Yiddish language is over one thousand years old and incorporates German, Slavic, and Hebrew elements. The prevalent view claims Yiddish has a German origin, whereas the opposing view posits a Slavic origin with strong Iranian and weak Turkic substrata. One of the major difficulties in deciding between these hypotheses is the unknown geographical origin of Yiddish speaking Ashkenazic Jews (AJs). An analysis of 393 Ashkenazic, Iranian, and mountain Jews and over 600 non-Jewish genomes demonstrated that Greeks, Romans, Iranians, and Turks exhibit the highest genetic similarity with AJs. The Geographic Population Structure (GPS) analysis localized most AJs along major primeval trade routes in northeastern Turkey adjacent to primeval villages with names that may be derived from "Ashkenaz." Iranian and mountain Jews were localized along trade routes on the Turkey's eastern border. Loss of maternal haplogroups was evident in non-Yiddish speaking AJs. Our results suggest that AJs originated from a Slavo-Iranian confederation, which the Jews call "Ashkenazic" (i.e., "Scythian"), though these Jews probably spoke Persian and/or Ossete. This is compatible with linguistic evidence suggesting that Yiddish is a Slavic language created by Irano-Turko-Slavic Jewish merchants along the Silk Roads as a cryptic trade language, spoken only by its originators to gain an advantage in trade. Later, in the 9th century, Yiddish underwent relexification by adopting a new vocabulary that consists of a minority of German and Hebrew and a majority of newly coined Germanoid and Hebroid elements that replaced most of the original Eastern Slavic and Sorbian vocabularies, while keeping the original grammars intact.



There have been other papers which show that Ashkenazi Jews form a separate cluster from gentile whites in the United States. This is important again in the context of biomedical studies attempting to ascertain the genetic roots of particular diseases; population substructure (e.g., Jew vs. non-Jew) may result in confounded associations. Also, one of the authors of the paper is David Goldstein, author of the fascinating Jacob’s Legacy. In any case, on to the PC charts where the real action is. Do note that I’ve resized and added explanatory labels here & there for clarity.



As you can see, there is almost perfect separation between the Jewish and non-Jewish clusters here. PC 1 = first principle component of variation, and PC 2 second principle component of variation. These are the two largest independent dimensions of genetic variance extractable out of the data set. The authors note that there is almost perfect separation along PC 1, and, they note that most of the gentile whites who are closest to the Jews on this PC are of Italian or Eastern Mediterranean origin. This is important later.



As you can see here, the interesting point is that Jewish ancestral quanta is roughly predictive of genetic position. This shouldn’t be that surprising once we know that Jews and non-Jews separate so cleanly (e.g., someone who is biracial would be located between their two parent racial clusters on any plot), but it is striking nonetheless in reaffirming the genetic reality of Jewishness. The United States has an easy genealogical history for Jews, as the ancestors of all self-identified American Jews today arrived on the order of 4-6 generations ago. These individuals were likely Jewish relatively far back in their lineages, and admixture is easy to recall because it is of recent vintage.

In conclusion, we show that, at least in the context of the studied sample, it is possible to predict full Ashkenazi Jewish ancestry with 100% sensitivity and 100% specificity, although it should be noted that the exact dividing line between a Jewish and non-Jewish cluster will vary across sample sets which in practice would reduce the accuracy of the prediction. While the full historical demographic explanations for this distinction remain to be resolved, it is clear that the genomes of individuals with full Ashkenazi Jewish ancestry carry an unambiguous signature of their Jewish heritage, and this seems more likely to be due to their specific Middle Eastern ancestry than to inbreeding.