Human-specific selection has probably acted on CREs of TR genes whose HI causes NDDs/NCPs

Research into the evolution of human-specific traits has benefited from the increasing availability of genome sequences of other great apes, such as the chimpanzee, and of archaic humans, such as Neanderthals and Denisovans (Pääbo, Reference Pääbo2014). These advances have increased our understanding of the genetic peculiarities that distinguish anatomically modern humans (AMHs) from their closest relatives, both living and extinct. For example, comparative genomic analyses have identified hundreds of so-called human accelerated regions (HARs) – genomic loci that are highly conserved among vertebrates yet show accelerated sequence divergence in the human lineage. Importantly, most HARs appear to be driven by positive selection, rather than by non-adaptive processes such as GC-biased gene conversion (Kostka et al., Reference Kostka, Hubisz, Siepel and Pollard2012; Hubisz & Pollard, Reference Hubisz and Pollard2014; Levchenko et al., Reference Levchenko, Kanapin, Samsonova and Gainetdinov2018). Moreover, almost all HARs are non-coding CREs (Pollard et al., Reference Pollard, Salama, King, Kern, Dreszer, Katzman and Haussler2006; Prabhakar et al., Reference Prabhakar, Noonan, Pääbo and Rubin2006; Bird et al., Reference Bird, Stranger, Liu, Thomas, Ingle, Beazley and Dermitzakis2007; Lindblad-Toh et al., Reference Lindblad-Toh, Garber, Zuk, Lin, Parker, Washietl and Kellis2011; Capra et al., Reference Capra, Erwin, McKinsey, Rubenstein and Pollard2013; Gittelman et al., Reference Gittelman, Hun, Ay, Madeoy, Pennacchio, Noble and Akey2015), in line with the hypothesis that adaptive divergence in human evolution, particularly with regard to cognitive traits, is primarily driven by regulatory changes (King & Wilson, Reference King and Wilson1975; Haygood et al., Reference Haygood, Babbitt, Fedrigo and Wray2010; Enard et al., Reference Enard, Messer and Petrov2014; Peyrégne et al., Reference Peyrégne, Boyle, Dannemann and Prüfer2017; Liu et al., Reference Liu, Mosti and Silver2021). Strikingly, many HARs have been found to regulate genes that are involved in the TR of brain and face development and for which HI leads to NDDs/NCPs (Table 1; Doan et al., Reference Doan, Shin and Walsh2018; Levchenko et al., Reference Levchenko, Kanapin, Samsonova and Gainetdinov2018; Kozlenkov et al., Reference Kozlenkov, Vermunt, Apontes, Li, Hao, Sherwood and Dracheva2020; Girskis et al., Reference Girskis, Stergachis, DeGennaro, Doan, Qian, Johnson and Walsh2021). Indeed, the vast majority of haploinsufficient TR genes implicated in the evolution of human brain/face development show changes in CREs, rather than coding changes (Table 1). Finally, among 29 human tissues or cell types scanned for signals of positive selection on TF binding sites in CREs (on the basis of predicted binding affinity changes), brain-related cell types show the highest proportion of positive selection (Liu & Robinson-Rechavi, Reference Liu and Robinson-Rechavi2020; see also Babbitt et al., Reference Babbitt, Haygood, Nielsen and Wray2017). Together, these findings underscore the importance of positive selection on the regulatory architecture of haploinsufficient brain- and face-related TR genes in the evolution of human-specific traits.

Table 1. Haploinsufficient transcriptional regulation (TR) genes implicated both in neurodevelopmental disorders (NDDs)/neurocristopathies (NCPs) and in the evolution of human brain and/or face development

Notes: 1, If available, the OMIM phenotype number is given in parentheses. Listed clinical features do not necessarily appear in all (or even most) patients. ADHD, attention deficit hyperactivity disorder; ASD, autism spectrum disorder; BD, bipolar disorder; ID, intellectual disability; NCP, neurocristopathy; NDD, neurodevelopmental disorder; SCZ, schizophrenia. 2, Changes occurred in the evolutionary lineage leading to anatomically modern humans (AMHs), either compared with archaic humans or with non-human primates. CRE, cis-regulatory element; HAR, human accelerated region. 3, References for human-specific evolutionary changes of the particular gene. For references for the NDD/NCP caused by haploinsufficiency of the gene, see Table S1 in the Supplementary Material.

Evidence of reduced TR gene expression during human evolution

That selection appears to have acted predominantly on CREs, and that the TR genes regulated by those CREs are exquisitely dosage sensitive (as evidenced by their HI), suggests that the associated human-specific traits evolved, at least in part, owing to changes in the genes’ expression level. Indeed, with respect to craniofacial traits, there is good support for this idea, in particular, for a reduction in gene expression (i.e. partial LOF). As described in detail below, several studies indicate (1) that the phenotypic changes in craniofacial morphology that occurred from archaic to modern humans are due to reduced TR gene expression and (2) that these phenotypic changes seen during human evolution resemble the phenotypic changes seen in HI-associated disorders, which are likewise due to reduced TR gene expression. In other words, there seems to be a parallelism between the evolutionary and the pathological consequences of reduced TR gene expression for craniofacial features (Figure 2). For example, reduced expression of the haploinsufficient TF gene NFIX causes Malan syndrome, whose clinical features include a high forehead and a prominent chin (Priolo et al., Reference Priolo, Schanze, Tatton-Brown, Mulder, Tenorio, Kooblall and Hennekam2018). Intriguingly, the same genetic and phenotypic changes (reduced NFIX expression, resulting in a more prominent forehead and chin) are likely to have also occurred in the evolution from archaic to modern humans (Gokhman et al., Reference Gokhman, Nissim-Rafinia, Agranat-Tamir, Housman, García-Pérez, Lizano and Carmel2020). A similar correspondence between the evolutionary and the pathological changes in craniofacial traits following TR gene downregulation has been suggested for BAZ1B (reduced facial bones; Williams–Beuren syndrome; Zanella et al., Reference Zanella, Vitriolo, Andirko, Martins, Sturm, O'Rourke and Testa2019), and SATB2 (smaller jaws, flatter face, higher forehead; SATB2-associated syndrome; Weiss et al., Reference Weiss, Harshman, Inoue, Fraser, Petrov, Ahituv and Gokhman2021); both cases are discussed in more detail in the next section. Taken together, these findings suggest that downregulation of haploinsufficient TR genes might have played an important role in the evolution of human craniofacial traits.

Figure 2. The parallelism between the evolutionary and the pathological consequences of reduced TR gene expression for craniofacial development. The figure shows several craniofacial features of archaic humans, healthy anatomically modern humans (AMHs), and AMHs with a haploinsufficiency-associated disorder. Both the evolutionary changes from archaic humans to AMHs and the pathological changes from healthy AMHs to individuals with the disorder are due to reduced TR gene expression. Adapted from Gokhman et al. (Reference Gokhman, Nissim-Rafinia, Agranat-Tamir, Housman, García-Pérez, Lizano and Carmel2020).

What do we know about TR gene expression changes in the evolution of the human brain? A promising approach to address this question is to compare gene expression patterns in brain samples from humans and other primates. In the past few years, it has become possible to compare brain transcriptomes at single-cell resolution, which enables the detection of cell type-specific gene expression differences. Interestingly, comparisons between primates show that, in the human lineage, gene expression divergence is higher in non-neuronal cell types such as oligodendrocytes than in neurons (Berto et al., Reference Berto, Mendizabal, Usui, Toriumi, Chatterjee, Douglas and Konopka2019; Hu et al., Reference Hu, Li and Wang2020; Khrameeva et al., Reference Khrameeva, Kurochkin, Han, Guijarro, Kanton, Santel and Khaitovich2020; Pembroke et al., Reference Pembroke, Hartl and Geschwind2021). Strikingly, genes that showed human-specific downregulation in oligodendrocytes were enriched for functions related to TR; among those downregulated TR genes were ARID1B, ARID2 and TCF20 (Berto et al., Reference Berto, Mendizabal, Usui, Toriumi, Chatterjee, Douglas and Konopka2019). Other brain-related TR genes that probably experienced reduced expression during human evolution are FOXP2 and CUX1. Maricic et al. (Reference Maricic, Günther, Georgiev, Gehre, Ćurlin, Schreiweis and Pääbo2013) found a regulatory change in FOXP2 specific to modern humans that diminishes binding of the TF POU3F2, thus driving reduced reporter gene expression compared with the ancestral allele. Finally, using allele-specific expression data from human–chimpanzee hybrid cortical spheroids, Starr et al. (Reference Starr, Gokhman and Fraser2022) recently found that expression of CUX1 from the human allele is lower than expression from the chimpanzee allele. These results suggest that reduced TR gene expression has played a role in human brain evolution. We do not know the specific effects that downregulation of these TR genes had on human brain development. However, this knowledge is crucial to understand how reduced expression affected the evolution of human cognition and social behaviour. Therefore, we cannot yet determine the relationship between the evolutionary and the pathological consequences of reduced TR gene expression for brain-related traits (remember that, for craniofacial traits, there is tentative evidence of a parallel relationship). The fact that TR HI in NDDs generally leads to ID and other impairments of brain function (Table 1) argues against a simple parallel relationship between the pathological and the evolutionary consequences of reduced TR gene expression with respect to brain-related traits, because cognitive function has increased, rather than decreased, during human evolution.

While few TR genes have so far been studied in detail, the findings described above suggest that reduced TR gene expression was involved in the evolution of both human craniofacial and brain-related traits. Thus, the results also add to the long-standing discussion about the developmental mechanisms that contributed to the evolution of human-specific traits. While many researchers have argued for a major role of heterochrony (or more specifically, delayed gene expression) in human evolution (Gould, Reference Gould1977; Somel et al., Reference Somel, Franz, Yan, Lorenc, Guo, Giger and Khaitovich2009), the findings presented here emphasize the role of reduced gene expression (i.e. heterometry; see Arthur, Reference Arthur2000). Nevertheless, the conclusion that reduced TR gene expression was important during human evolution does not preclude the possibility that, in other developmental contexts, it was increased TR gene expression that fuelled the evolution of human-specific traits. Moreover, the effects of TR gene expression changes on development and evolution of craniofacial traits are not necessarily the same as those on development and evolution of brain-related traits (see Box 3). Also, our conclusion is currently based only on a handful of cases and hence somewhat speculative, while several crucial questions remain open. Nevertheless, we think that HI of TR genes predisposes these genes to LOF mutations that do have evolutionary implications (see Box 2). Therefore, we believe that future research will unearth further evidence supporting the significance of TR gene downregulation in the evolution of human-specific traits.

Selection for prosociality and against fear and aggression: the human self-domestication hypothesis

In light of the evidence of reduced TR gene expression during human evolution, an important question is: what selective forces were responsible for this downregulation and the associated emergence of human-specific traits? One hypothesis is that these traits evolved in part through human self-domestication (HSD) (Hare, Reference Hare2017; strictly speaking, the term is a misnomer – in contrast to animal domestication, HSD did not involve any deliberate selection or goal-directedness; Sánchez-Villagra & van Schaik, Reference Sánchez-Villagra and van Schaik2019). According to the HSD hypothesis, selection during human evolution favoured in-group prosociality over aggression and fear. In the wake of this selection, humans evolved traits similar to those characteristic of other domestic animals. In domesticated mammals, this suite of behavioural, physiological and morphological traits that emerged via selection for tameness is referred to as the ‘domestication syndrome’ (Wilkins et al., Reference Wilkins, Wrangham and Fitch2014). Several craniofacial traits that distinguish modern humans from their archaic ancestors, such as a short, retracted face and a reduced brow ridge (Figure 2), fit the domestication syndrome, which can be seen as support for the HSD hypothesis (Hare, Reference Hare2017). In their seminal paper, Wilkins et al. (Reference Wilkins, Wrangham and Fitch2014) propose that the domestication syndrome ultimately results from multiple partial LOF mutations affecting dosage-sensitive NC genes (the ‘neural crest/domestication syndrome’ hypothesis; for an up-to-date account, see Wilkins et al., Reference Wilkins, Wrangham and Fitch2021). Indeed, many haploinsufficient TR genes involved in NC development, including BAZ1B, CHD7, MITF, PAX3, PAX9, RAI1, SATB2, SOX9, SOX10 and ZEB2, are not only known to cause human NCPs (Table S1), but also are candidate genes for the domestication syndrome (Wilkins et al., Reference Wilkins, Wrangham and Fitch2014; Pendleton et al., Reference Pendleton, Shen, Taravella, Emery, Veeramah, Boyko and Kidd2018). Hence, the neural crest/domestication syndrome hypothesis provides an appealing mechanism underlying HSD. Yet what is the evidence for this hypothesis with respect to human evolution?

In order to test the neural crest/domestication syndrome hypothesis with regard to HSD, Zanella et al. (Reference Zanella, Vitriolo, Andirko, Martins, Sturm, O'Rourke and Testa2019) focused on Williams–Beuren syndrome, whose characteristic features include a reduced, retracted face and pronounced friendliness and sociality – traits that fit the domestication syndrome. HI of the chromatin regulator BAZ1B has long been known as the causal factor for most of the characteristic features of Williams–Beuren syndrome (Lu et al., Reference Lu, Meng, Morris and Keating1998; Peoples et al., Reference Peoples, Cisco, Kaplan and Francke1998; Lalli et al., Reference Lalli, Jang, Park, Wang, Guzman, Zhou and Kosik2016). Zanella et al. (Reference Zanella, Vitriolo, Andirko, Martins, Sturm, O'Rourke and Testa2019) showed that BAZ1B is a master regulator of NC genes involved in craniofacial development, and Kuhlwilm and Boeckx (Reference Kuhlwilm and Boeckx2019) found nearly fixed mutations in the CREs of BAZ1B in the genomes of modern, but not archaic, humans. These mutations are probably partial LOF mutations causing slight reductions in gene expression and, ultimately, in craniofacial morphology, corresponding to the facial retraction from archaic to modern humans (Wilkins, Reference Wilkins2020). Strikingly, very similar findings were recently described for another NC gene, the TF SATB2. HI of SATB2 causes SATB2-associated syndrome (also termed Glass syndrome), whose features include small jaws, flat face and a friendly personality – again, traits that match the domestication syndrome. Weiss et al. (Reference Weiss, Harshman, Inoue, Fraser, Petrov, Ahituv and Gokhman2021) found a fixed mutation in the CRE of SATB2 in modern, but not archaic, humans, which probably reduced SATB2 expression and contributed to the retracted face characteristic of modern humans. In light of the correspondence between the pathological and the evolutionary changes in craniofacial traits following reduced TR gene expression (see above), these findings provide tentative support for the neural crest/domestication syndrome hypothesis as a mechanism underlying HSD: selection for prosociality led to reduced expression of cranial NC-associated genes (e.g. BAZ1B, SATB2), and this NC hypofunction caused, pleiotropically, the craniofacial changes that distinguish modern humans (and which are paralleled both in the associated HI syndromes and in the domestication syndrome of domesticated mammals). Moreover, the neural crest/domestication syndrome hypothesis as a possible mechanism for HSD has also been linked to the evolution of human cognition and language, based on a set of genes (many of them encoding TRs) that are involved both in NC development and in brain globularization (Benítez-Burraco et al., Reference Benítez-Burraco, Theofanopoulou and Boeckx2018; Thomas & Kirby, Reference Thomas and Kirby2018).

Given the appeal of the HSD hypothesis as an explanation for the emergence of human cognitive and social traits, it is not surprising that researchers have used it to understand the origin and persistence of psychiatric disorders. SCZ, for example, is viewed as a hyper-domestication syndrome, resulting from NC hypofunction and hence from the same selective pressures that triggered HSD (Benítez-Burraco et al., Reference Benítez-Burraco, Di Pietro, Barba and Lattanzi2017; Šimić et al., Reference Šimić, Vukić, Kopić, Krsnik and Hof2021). In fact, this idea is only one of several evolutionary hypotheses that trace the existence of psychiatric disorders to the action of natural selection. In contrast, in the next section, we argue that the important role of LOF of TR genes in the etiology of psychiatric disorders is more compatible with the view that these disorders persist despite, not because of, natural selection.

Evolutionary models for the persistence of psychiatric disorders

NDDs such as ASD and SCZ seem to present an evolutionary paradox: they have a strong genetic component and are associated with considerable fitness costs, and yet they are surprisingly prevalent. Given the high heritability and high fitness costs of these disorders, why has natural selection not eliminated the genetic variants that predispose to them? This ‘paradox of common, harmful, heritable mental disorders’ (Keller & Miller, Reference Keller and Miller2006) has long puzzled evolutionary geneticists and psychiatrists alike (e.g. Huxley et al., Reference Huxley, Mayr, Osmond and Hoffer1964). In essence, it boils down to a fundamental question in evolutionary genetics: what causes genetic variation in fitness-related traits to persist, given that selection is expected to minimize it?

Simply put, there are two classes of evolutionary models to explain genetic variation in fitness-related traits and, by extension, in the liability to psychiatric disorders: mutation-selection balance (or more precisely, mutation-selection-drift) and balancing selection (Keller, Reference Keller2018). The mutation-selection-drift model explains genetic variation as a balance between the loss of deleterious alleles through purifying selection and the emergence of new ones through mutation; in addition, random events cause some deleterious alleles with small effects to drift to high frequencies. Under the alternative model, balancing selection, allelic variation at a given locus is maintained by the balance between positive and negative selective forces. In this scenario, risk alleles associated with psychiatric disorders confer some benefit at least under some conditions; hence, selection for the beneficial trait can maintain disorder risk as a by-product. For example, Crow (Reference Crow1997) famously argued that SCZ was ‘the price that Homo sapiens pays for language’. Indeed, the idea that SCZ represents a maladaptive by-product of positive selection during human evolution has been expressed repeatedly (Burns, Reference Burns2004; Crespi et al., Reference Crespi, Summers and Dorus2007; Sikela & Searles Quick, Reference Sikela and Searles Quick2018), and the HSD approach to SCZ is one version of this idea (see above). Similar hypotheses have been suggested for ASD (Baron-Cohen, Reference Baron-Cohen2003; Ploeger & Galis, Reference Ploeger and Galis2011) and bipolar disorder (Wilson, Reference Wilson1998; Greenwood, Reference Greenwood2020). A major difference between mutation-selection balance and balancing selection concerns the role of selection: under mutation-selection balance, mental disorders exist and persist despite natural selection, whereas under balancing selection, they do so because of natural selection (Durisko et al., Reference Durisko, Mulsant, McKenzie and Andrews2016). Which model, then, best explains the paradox of harmful yet common psychiatric disorders?

As our understanding of the genetic architecture of ASD, SCZ and other highly heritable psychiatric disorders is increasing, much research is consistent with a simple mutation-selection-drift model (Keller & Miller, Reference Keller and Miller2006; Keller, Reference Keller2018). First, there is broad support for the important role that rare deleterious mutations with large effect size play in the etiology of NDDs: as described above, a considerable part of ASD and SCZ risk is due to rare heterozygous LOF variants in TR genes that do not tolerate such LOF because of their HI. In addition, disorder risk also depends on a large number of deleterious variants with small effect sizes (background polygenic risk). In fact, the effects of many deleterious mutations will be so small that they may drift to high frequencies. Since the extremely polygenic nature of complex traits (Boyle et al., Reference Boyle, Li and Pritchard2017) implies a large mutational target size of psychiatric disorders, this helps to explain their high prevalence (Keller, Reference Keller2018). Second, there is compelling evidence that, on average, risk alleles associated with these disorders are under weak to strong purifying selection (Rees et al., Reference Rees, Moskvina, Owen, O'Donovan and Kirov2011; Mullins et al., Reference Mullins, Ingason, Porter, Euesden, Gillett, Ólafsson and Stefansson2017; Keller, Reference Keller2018; Pardiñas et al., Reference Pardiñas, Holmans, Pocklington, Escott-Price, Ripke, Carrera and Walters2018; Huang & Siepel, Reference Huang and Siepel2019; Esteller-Cucala et al., Reference Esteller-Cucala, Maceda, Børglum, Demontis, Faraone, Cormand and Lao2020; Rapaport et al., Reference Rapaport, Boisson, Gregor, Béziat, Boisson-Dupuis, Bustamante and Patin2021; Wendt et al., Reference Wendt, Pathak, Overstreet, Tylee, Gelernter, Atkinson and Polimanti2021). These lines of evidence converge in recent population genetic analyses which apply a simple mutation-selection balance model (either as a deterministic approximation or explicitly incorporating drift) to provide direct estimates of the strength of selection against heterozygous LOF in humans (Cassa et al., Reference Cassa, Weghorn, Balick, Jordan, Nusinow, Samocha and Sunyaev2017; Weghorn et al., Reference Weghorn, Balick, Cassa, Kosmicki, Daly, Beier and Sunyaev2019; Agarwal et al., Reference Agarwal I., Myers and & Przeworski2022). Strikingly, negative selection was strongest for LOF-intolerant TR genes (Cassa et al., Reference Cassa, Weghorn, Balick, Jordan, Nusinow, Samocha and Sunyaev2017).

Recent studies provide further interesting insights into the nature of negative selection at LOF-intolerant genes. First, purifying selection not only acts on coding sequences, but also on the promoters of LOF-intolerant genes, presumably to safeguard promoter hypomethylation and to curb coding mutation rates (Boukas et al., Reference Boukas, Razi, Bjornsson and Hansen2022). Second, negative selection at LOF-intolerant genes comes not only in the form of natural selection (against variants increasing pre-reproductive mortality or decreasing fertility), but also in the form of sexual selection. Gardner et al. (Reference Gardner, Neville, Samocha, Barclay, Kolk, Niemi and Hurles2022) found that LOF variants in these genes reduce reproductive success much more in males than in females, and that this reduction is mediated primarily through cognitive and behavioural traits, rendering male carriers of such variants less likely to find mating partners (see also Liu et al., Reference Liu, Akimova, Ding, Jukarainen, Vartiainen, Kiiskinen and Ganna2022).

Lastly, we would like to add two further points. First, while mutation-selection-drift offers a convincing explanation for the persistence of highly heritable psychiatric disorders such as ASD and SCZ, other NDDs, and especially those with lower heritability, might require different types of evolutionary explanation. For example, many symptoms of depression, which has much lower heritability than other psychiatric disorders, can be seen as adaptive defences to minimize fitness loss in the face of adverse life situations (Keller, Reference Keller2018; Syme & Hagen, Reference Syme and Hagen2020). Second, acceptance of mutation-selection-drift as the major explanation of the persistence of psychiatric disorders does not preclude a role of positive selection in providing the conditions for the evolutionary origin of these disorders. One can easily imagine a scenario in which mutations that altered brain development of early H. sapiens were selected for, became fixed, and thus helped to establish human-specific cognitive and behavioural abilities (e.g. mutations that gave rise to HARs). At the same time, these mutations could have, if only indirectly, made human brain development more susceptible to disruption and allowed disorders such as ASD and SCZ to arise (we thank an anonymous reviewer for drawing our attention to such a scenario). However, since these mutations became fixed early on in human evolution, such a scenario does not explain individual differences in genetic risk to psychiatric disorders in modern human populations (Keller, Reference Keller2018).

Taken together, positive selection might have been involved in the emergence of psychiatric disorders, and we also cannot completely rule out balancing selection as a possible explanation for the persistence of some risk alleles associated with these disorders. Nevertheless, mutation-selection-drift is likely to be a more general explanation for the observed genetic variation in disorder risk (Keller, Reference Keller2018). Both the evidence for disease-causing LOF alleles in LOF-intolerant TR genes and for strong purifying selection against such alleles are fully in line with this conclusion.