Cultural Genomics: Promises and Challenges

Prior Discussions About the Relations Between Genes and Culture

Although earlier evolutionists had to face the issue of the relative roles of culture and biology in human evolution (e.g., the invention of clothing might have stopped necessary physical changes to fight against cold climate; Wallace, 1864), serious discussions about how genes and culture(s) coevolve did not emerge until the middle of the last century. In the beginning, the principles of evolutionary theory such as variations and natural selection were borrowed to explain cultural evolution (Campbell, 1965). This idea was then systematically examined by a number of evolutionary biologists and social scientists. Dawkins (1976) coined and popularized the term “meme” as culture’s equivalent of gene, which led to the field of memetics. Cavalli-Sforza and Feldman (Cavalli-Sforza & Feldman, 1981; Feldman & Cavalli-Sforza, 1976) and Lumsden and Wilson (1981) used population genetics models to help formulate mathematical models for cultural evolution as well as gene–culture coevolution. Together with Wilson’s sociobiology (Wilson, 1975), however, these models were viewed as representing biological reductionism. Coming from the social sciences, Boyd and Richerson (1985) expanded on the previous work by focusing on the importance of environmental conditions, social learning, cultural rules, and potential conflicts between the dual inheritances of genes and culture, the last of which was recently further expanded by Paul (2015). According to these coevolution models or the dual inheritance theory (DIT), culture evolves partly through a Darwinian selection process, with analogous mechanisms such as transmission through social learning, random variations due to errors in social learning, selection pressure (cultural sanctions, norms), and cultural drift (gradual shift in fads and fashions, analogous to genetic drift). These ideas also led to efforts to contemplate an eventual unification or “consilience” of biological and behavioral sciences and even humanities (Gintis, 2007; Wilson, 1998).

Despite the significant strides in theoretical models of gene–culture coevolution, empirical evidence has been at best uneven for the different aspects of the theory. Memetics, behavioral economics, as well as psychology have collected data that are consistent with the idea that cultural evolution follows some of the same principles of biological evolution (e.g., Heylighen & Chielens, 2009; Richerson & Boyd, 2005). In terms of the interactions between genes and culture, discussions have focused on potential conflicts between the two forms of inheritance (see a recent discussion by Paul, 2015) or the mismatches between modern environments and the environments of evolutionary adaptiveness (EEA; Cosmides & Tooby, 1997). Most of such discussions rely on a static model of human biological evolution, as typified by the notion that “Modern skulls house a stone-age mind” (Cosmides & Tooby, 1997). Indeed, psychologists have been very comfortable with the widely accepted belief that all commonalities across groups of humans are the product of the same biological/genetic evolution (evolutionary psychology, which uses universality as proof of evolutionary adaptation) and any group differences are cultural in nature (cross-cultural psychology). Paul (2015) recently summarized that view for the DIT: “These differences between societies and their cultural systems are thus the result of differences in the cultural symbol systems themselves, not of genetic information” (p. 7). Indeed, this idea can be traced back to Franz Boas, one of the founders of cultural anthropology, who emphasized the importance of culture in the differences between civilized and primitive minds (Boas, 1901). When it comes to dynamic interactions between genes and culture, theoretical models exist but the only often-mentioned example of adaptations is the selection of the LCT gene for lactase persistence due to the cultural practice of dairy farming (Tishkoff et al., 2007).

Ideological and political issues aside (to which we will return in a later section), the static model of human molecular evolution (i.e., humans have stopped to evolve once we invented culture) prevailed for so long mainly because of a lack of genomic information needed to examine recent human evolutionary history and its group differences, as well as because of the prevailing idea of neutral theory (i.e., mutations are neutral; Kimura, 1983). With the abundance of genomic information accumulated over the past decade, group variations in genomes have been systematically documented, via the HapMap Project and now the 1000 Genomes Project. Evidence of strong recent selection (within the timescale of cultural history; C. Chen et al., 2016; Fu & Akey, 2013; Hawks et al., 2007; Scheinfeldt & Tishkoff, 2013; E. T. Wang et al., 2006) has finally shifted theoretical debate away from neutral theory and allowed us to view the human genome as malleable, making it possible to examine the dynamic relations between culture and genes. The time has come for cultural genomics.

What Is Cultural Genomics?

Cultural genomics aims to understand functional genomic differences across human cultural groups and the origins of such group differences, especially the role of culture as a driver of genetic evolution (C. Chen et al., 2016). Cultural groups can be defined as any groups that show differences in their beliefs, norms, customs, or ancestry. Broadly defined (Cohen, 2009), they can be racial/ethnic groups, religious groups, socioeconomic status (SES) groups, within-country regional groups, schooled versus nonschooled groups, and so forth. If ecological differences are considered as a driving force behind cultural variations, the term “eco-cultural genomics” would also be appropriate.

Cultural genomics builds on the progress of the DIT, but unlike DIT’s tilt toward understanding cultural evolution, cultural genomics focuses more on the interactions between culture and genes in recent evolutionary history, with a tilt toward the culture’s role in genetic evolution, analogous to cultural psychology’s emphasis on the role of culture in human behavior. Cultural genomics is an interdisciplinary field, at the intersection of (cross-)cultural psychology, physical and cultural anthropology, evolutionary biology, human behavioral ecology, evolutionary psychology or sociobiology, population genetics (especially molecular population genetics), and genomics. Because it is an emerging field, there is still no standard terminology to describe this field. Richerson and Boyd (2005) use the traditional term of gene–culture coevolution. Laland, Odling-Smee, and Myles (2010) use gene–culture coevolution as well as their own niche-construction theories. Henrich (2011) uses gene–culture coevolution as well as encultured brains (Henrich, 2015) or “the cultural brain hypothesis” (Jaquet, 2012). Kim and her colleagues (Kim & Sasaki, 2014; Sasaki, LeClair, West, & Kim, 2016) have preferred the term gene–culture interactions. Minkov and Bond (2015) have used “cross-cultural sociogenetics.” We (C. Chen et al., 2016) have used the terms “the encultured genome” and “(cross-)cultural genomics.” Rossi (2016) has also used the term “psychosocial and cultural genomics” for an online journal, which focuses on psychotherapy. In addition, Creagan (2012) also introduced the idea of “cross-cultural genomics” in a piece that used malaria and sickle cell disease as an example to illustrate that

cross-cultural genomics can study how the genes of the individuals in a culture affect that culture and how that culture interacts with its environment . . . and how all those aspects affect the interaction of that culture with other cultures. (p. 87)

It is worth mentioning that Rossi, Creagan, and Kim tend to emphasize the role of genes in behavior and how it may differ in different groups, whereas Richerson, Laland, Henrich, and Chen and their colleagues have also emphasized culture as a driving force behind recent human evolution.

Four Conceptual Issues in Cultural Genomics: Level or Unit of Analysis, Timescale, Mechanism, and Direction of Influence

Theoretically, any cultural element (indeed any environmental element) can interact with genes. Cultural elements can be grouped in many different ways. Most often, cultural psychologists have focused on cultural groups (e.g., Eastern vs. Western cultures) and cultural dimensions (e.g., collectivism vs. individualism, Triandis, 1995; tightness vs. looseness, Gelfand et al., 2011; or multidimensional models such as those of Hofstede, 2001, and Schwartz, 2012). Depending on the unit/level of cultural groups, cultural genomics can address different research questions. The most commonly used unit (and hence the most intuitive to understand) is racial or ethnic groups in both psychology and genetics (e.g., the HapMap Project and the 1000 Genomes Project). However, the unit can be enlarged or reduced. On one hand, we can treat the human species as one unit and study species-wide cultural genomics. Indeed, almost all discussions on human evolution and gene–culture coevolution have stayed at this level of analysis. Only a small amount of discussion and research has moved beyond this level to the commonly defined racial/ethnic group level. Even less research has touched smaller units of cultural groups such as within-ethnicity or within-country subgroups or within individuals whose mind is encultured. In sum, the unit of analysis for gene–culture interactions can go from one human to the total number of homo sapiens that have ever lived.

Once a level of analysis is determined, the timescale for gene–culture interactions is constrained. On one end, the study of species-wide gene–culture interaction covers the beginning of the species to the present; on the other end, at the individual level, the timescale for gene–culture interaction is just the individuals’ lifetime. In between the two extremes are the evolutionary histories of individual racial/ethnic groups or social subgroups. It is obvious, but nevertheless important, to note that all timescales of gene–culture interactions are also subject to the timescale constraints of given cultural elements, or more precisely, the timespan of their relevance (e.g., the agrarian period of a particular group).

The timescale is important because it determines the mechanisms of gene–culture interactions. At the species-wide level and some group levels, gene–culture interactions have been modeled by previous DIT mathematical models. The mechanisms involved include genetic mutation, gene flow, genetic drift, natural selection, sexual selection, and so forth. Most recently, introgression from Archaic species such as Neanderthals and Denisovans to modern humans has been added as a new mechanism (e.g., Dannemann, Andrés, & Kelso, 2016; Simonti et al., 2016; Vernot & Akey, 2014). Moving toward smaller units of groups and to individuals, many of these mechanisms are irrelevant (too few mutations, no new introgression); instead, epigenetic mechanisms, gene–gene interactions, and social selection (to be discussed later) are the potential mechanisms involved.

Finally, the nature of the gene–culture interaction needs to be specified. First of all, in psychology, the word “interaction” has been closely associated with statistical interaction. What we have discussed so far has mostly used this word in its original meaning (i.e., interact, or reciprocal influence). In other words, gene–culture interaction is about how genes influence culture and how culture influences genes. Therefore, the nature of gene–culture interaction can be decomposed into genes influence culture (i.e., traditional, biological determinism), culture influences genes (i.e., culture as a driver of human evolution as mentioned earlier), and a statistical gene–culture interaction (a.k.a. culture as a superorganic moderator of genes’ effect on behavior or genes as moderators of culture’s effect on behavior). The statistical interaction can be due to at least two possibilities. The first possibility is that culture is just a proxy of particular genetic variation that happens to covary with culture, so that cultural differences in genetic effects are due to a given gene’s interaction with the gene that covaries systematically with culture. In other words, this type of gene–culture interaction is really a gene–gene interaction—one gene variation affects the expression of another gene. The second possibility is that cultural factors (e.g., environmental stress) actually affect gene expression through epigenetic processes. Little work has been done to differentiate the different types of interactions.

As is clear from the above discussion, cultural genomics needs to specify the level or unit of analysis, timescale, mechanisms, and directions of the interactions. In the following sections, we describe examples of gene–culture interactions at various levels and the mechanisms involved.

Species-Wide Adaptation

Any culture-driven species-wide adaptation is by definition shared by all human beings today and in the past, but not shared with other animals. Although debates abound, most likely candidates for human-specific species-wide adaptations would include language, consciousness, morality, technology, self-domestication, and so forth. Of these “big” or ultimate questions about the origins of human-specific capacities, only language and self-domestication have been linked to several specific genetic bases. We will briefly discuss these two.

In the case of language evolution, the FOXP2 gene is the most celebrated case. Its role in the development of normal speech and language (particularly grammar) has been well documented. Two mutations in FOXP2, at positions 911 and 977 in exon 7, separated humans as well as Neanderthals from other primates (Krause et al., 2007). These mutations are believed to set the stage for the emergence of human language, suggesting that genetic mutations led to arguably the most important cultural element of all—language. Of course, FOXP2 is likely to be a necessary but by no means sufficient condition for language development. A number of other genes have been implicated in language processing (e.g., Mozzi et al., 2016; Wong, Morgan-Short, Ettlinger, & Zheng, 2012). Some of them have been found to show signs of recent selection (e.g., FOXP2, ROBO1, ROBO2, CNTNAP2, and KIAA0319; Mozzi et al., 2016), perhaps due to a strong selection pressure presented by human culture’s reliance on language for group living. Human language seems to be a result of the mutual influence between genes and culture.

Another species-wide adaptation may fit the broad concept of human “self-domestication.” Just as humans have tamed (or domesticated) wild animals, we have created systems (parenting, schooling, laws, and punishment) to shape human behaviors or to educate or civilize the individuals. As Steve Jones (2000) succinctly put it, “Man has tamed many creatures, but has himself become the most domestic animal of all” (p. 310). In one form or another, this idea has been around for more than a century (see Brune, 2007, for a summary; see Henrich, 2015, for a recent discussion). Some (necessarily selective) evidence deduced from human behavioral and physical traits seems to support this idea. Like domesticated animals, human aggression has seen dramatic decreases over time (e.g., Pinker, 2011), which is also consistent with physical traits associated with domestication such as the shrinking of the human brain, thinning of the skull, and a reduction in sexual dimorphism (Brune, 2007; Jones, 2000). Similarly, the secular trend of earlier puberty for human females (Tanner, 1962) is consistent with the effect of animal domestication on early sexual maturity in females (B. R. Clark & Price, 1981). Also consistent is evidence of recent slowing down of reaction times (Woodley, Nijenhuis, & Murphy, 2013). In terms of personality traits, the classic domestication experiment of silver foxes showed that the wild ancestors evolved into well-behaved animals, after generations of domestication over 20 years, with significant changes in levels of neurotransmitters (Jones, 2000). For humans, self-domestication is shown by the trends toward larger cooperative groups, sedentary environment, an emphasis on conformity, morality, as well as the coresidence of unrelated individuals (Hill et al., 2011).

Domestication seems to have cascading effects on many behavioral and physical traits as mentioned above. Therefore, it is certain that at least more than a few genes are involved. Researchers have documented genomic differences between domesticated plants and animals and their wild counterparts and attempted to link them to the archaeological data of the history of domestication (e.g., Larson & Burger, 2013; Zeder, Emshwiller, Smith, & Bradley, 2006). Recently, G. D. Wang et al. (G. D. Wang, Zhai, et al., 2013; G. D. Wang, Xie, et al., 2014) systematically analyzed genomic differences between domesticated animals and discussed their implications for humans. They found extensive overlap between the list of genes under positive selection during domestication of dogs with the list of positively selected genes in humans, with parallel evolution most apparent in genes for digestion and metabolism, neurological process, and cancer. We (C. Chen et al., 2016) have recently examined behavioral correlates of neurotransmitter genes that showed evidence of recent positive selection. Results showed that a number of these genetic variants were linked to behavioral and personality traits that are consistent with the effect of self-domestication. For example, the recently selected alleles of a set of neurotransmitter genes (e.g., HTR5B, GRIA1-4, TPH1, GRM1) were linked to lower IQ (performance), longer discriminate reaction times, and poorer memory; and, the recently selected alleles of another set of genes (e.g., HTR1B, HTR1E, CHRM2, NLN) were linked to personality traits such as lower self-directedness, lower intrinsic motivation, higher avoidance, and higher anxiety. It is worth noting that in contrast to lower IQ based on the performance tests, the selected alleles were also associated with higher performance on school-related skills such as reading and number processing. It is unclear at the moment whether these differentiated effects may have been responsible for the two contrasting trends—lowered general intelligence due to self-domestication over the past 10,000 years and increased IQ scores during the past century known as the Flynn effect (Trahan et al., 2014). Indeed, evidence is accumulating that “[d]omestication put pressure on man’s genes, as it did on those of his plants and animals” (Jones, 2000, p. 131).

In addition to language and self-domestication, many other human-specific traits can be studied from the perspective of cultural genomics. Researchers need to identify potential sources of species-wide selection pressure. Group living, which had probably been important for both self-domestication and the origin of language, might have been important for other species-wide adaptations such as conformity (as well as morality) because of the increased prevalence of pathogens with group living (Fincher, Thornhill, Murray, & Schaller, 2008). Indeed, immune-related genes represent a large category of genes that have shown recent selection (e.g., Fu & Akey, 2013; E. T. Wang et al., 2006). Other species-wide selection pressures might have included cultural elements such as widespread tool use (Taylor, 2010; Wrangham, 2009) including the use of fire (see a recent article on a mutation in the aryl hydrocarbon receptor [AHR] gene linked to attenuated reactions to environmental pollutants including toxic chemicals in smoke; Hubbard et al., 2016). Much is to be discovered about how the human genome has adapted under these selection pressures. It should be mentioned that when studying species-wide adaptations, we need to be aware of three subtypes: species-wide adaptations before the out-of-Africa split (which are in general shared among all modern human beings), species-wide adaptations since the split but relying on the same biological/genetic mechanisms, and species-wide adaptations since the split but relying on different genetic mechanisms. Further complicating the issue is that it is not straightforward to define what constitutes the “same” genetic mechanism, which can range from the same single-nucleotide mutation in genes or widely removed regulatory regions (Kellis et al., 2014), to different mutations in the same gene or even different mutations in different genes in related biochemical pathways. Convergent evolution, where the same phenotype is selected for despite differences in the underlying genetic mechanism, greatly complicates genomic analyses of gene–culture interactions. At the genetic level, such changes may be difficult to distinguish from group-specific adaptations (see the next section). Unfortunately (in terms of research difficulty), convergent evolution also seems to be widespread, from plants to animals including human beings (McGhee, 2011).

Group-Specific Adaptations

Strictly speaking, all adaptations are local adaptations and, thus, group specific. Only when the group is (was, in the case of human beings) together in one locality or when convergent evolution occurs, do we see species-wide adaptations. Not surprisingly, all the widely cited examples of recent human adaptations have come from group-specific adaptations: malaria-driven sickle cell anemia, skin color, lactase persistence, altitude adaptation, bitter taste adaptation, and so forth. All of them also show convergent evolution (i.e., different mutations produce the same phenotype).

As mentioned at the start of this article, group-specific adaptations have been considered since ancient times. Echoing Hippocrates’ eco-cultural view, Georges-Louis Leclerc Comte de Buffon (1707-1788), a French pioneer evolutionist, discussed the gradual process of group-specific adaptations for humans:

After multiplying and spreading over the whole surface of the earth, they have undergone various changes by the influence of climate, food, mode of living, epidemic diseases, and the mixture of dissimilar individuals. At first these changes were not so conspicuous, and produced only individual varieties; these varieties became afterwards specific, because they were rendered more general, more strongly marked, and more permanent by the continual action of the same causes; they are transmitted from generation to generation, as deformities or diseases pass from parents to children. (Quoted from Cavalli-Sforza, Menozzi, & Piazza, 1994, p. 17)

Buffon’s idea of “permanent” group-specific adaptations foreshadowed the discoveries of functional genes that reached fixation for some human groups. One well-known example is EDAR, a gene related to ectodermal development (e.g., hair, teeth, and sweat glands). The G allele of SNP rs3827760 in this gene is linked to thicker hair, has an allele age of a little more than 10,000 years, and has reached near fixation for Chinese and other East Asian and Native American groups, but is almost nonexistent among Europeans, Africans, and South Asians (Bryk et al., 2008). Such extreme group-specific genetic adaptations are rare. Most group-specific adaptations show partial selective sweeps (strong recent selection but not reaching fixation yet; Fu & Akey, 2013; Hernandez et al., 2011).

One of the best examples of partial replacement is selection for malaria resistance. In regions of the world where malaria is endemic, dozens of gene variants have been selected for that confer resistance, often in the same geographic location (Karlsson, Kwiatkowski, & Sabeti, 2014; Kwiatkowski, 2005). Often, as in the case of sickle cell anemia, these variants, when present in two copies, lead to disease. In these cases, having one copy of the variant gene (HBB^sc) confers malaria resistance, whereas two copies results in sickle cell disease. Obviously, the selected HBB^sc variant cannot rise to 100% frequency in the population, because the balance of positive and negative selection pressure (i.e., HBB/HBB individuals die from malaria, HBB/HBB^sc heterozygotes resist malaria, and HBB^sc/HBB^sc individuals die of sickle cell disease) keeps both alleles in the population. Furthermore, when multiple alleles for malaria resistance arise and are selected for in the same population (such as HBB, G6PD, and ABO variants; Hedrick, 2011), none will likely reach fixation. Inheriting any one of these alleles reduces the positive selective pressure on a second inherited allele, because the first allele already confers resistance. Given that the majority of malaria resistance genes/alleles are recent adaptations (Karlsson et al., 2014), that many independent variants have been selected for, and that some of the variants have deleterious consequences if inherited in two copies, it is not surprising that these group-specific variants have not reached fixation.

What do the many documented examples of group/region-specific adaptation (such as malaria resistance) tell us about cultural genomics? The malaria example indicates that convergent evolution at the phenotypic level (malaria resistance) can be accomplished by divergent evolution at the genotypic level (different variants arise in different genes, often in the same broadly defined population, i.e., Africa). Such genetic divergence is widespread. For example, within neurotransmitter genes, we (C. Chen et al., 2016) have identified 55 sites of natural selection, but only 11 (e.g., MAO-A, SLC6A4, and six HTR genes) were common across the four groups studied (Yoruba, Chinese and Japanese, Europeans, and Australian aborigines). Given this group diversity, lack of reproducibility of genetic association with behavioral traits between groups should be expected, even though phenotypic adaptation might be similar. Furthermore, pooling data from individuals with diverse geographic populations for genome-wide association studies (GWAS) is likely to dilute region-specific variants, reducing, rather than improving, power to detect these variants. Determining, then, if a group-specific genetic adaptation is associated with cultural differences is the continuing challenge of cultural genomics.

In the face of these challenges, several research groups have examined or at least speculatively discussed relations between group-level genetic variations and group variations in phenotypes of interest to psychologists. For example, we (C. Chen, Burton, Greenberger, & Dmitrieva, 1999) first proposed the hypothesis that the global distribution of DRD4 7R, a variant dopamine receptor, can be explained by natural selection related to different groups’ migratory experience. After we proposed this controversial hypothesis, both molecular evidence (Ding et al., 2002; E. T. Wang et al., 2004; E. T. Wang et al., 2006) and behavioral mechanistic evidence (e.g., Eisenberg, Campbell, Gray, & Sorenson, 2008) have supported the idea of recent selection of DRD4 7R. Other groups also confirmed the migration hypothesis with either refined local data (e.g., Tovo-Rodrigues et al., 2010) or expanded global data (Matthews & Butler, 2011).

Another commonly studied genetic polymorphism is 5-HTTLPR, a variant in the promoter region of the serotonin transporter (SLC6A4). The SLC6A4 gene has undergone recent adaptive selection, as determined by a linkage disequilibrium decay test (C. Chen et al., 2016; Hawks et al., 2007; E. T. Wang et al., 2006). Chiao and Blizinsky (2010) discussed how global variations of the frequency of 5-HTTLPR variants may be related to cultural orientations of collectivism/individualism (also see Kitayama et al., 2014, for a moderating role of DRD4 in cultural orientation of independence vs. independence). Way and Lieberman (2010) also discussed potential connections between cross-cultural variations in socioemotional adjustment and variations in 5-HTTLPR as well as variations in two other socioemotional genes (MAO, OPRM1). Fischer (2013) reported that at the cultural group level, 5-HTTLPR interacted with the presence of threats to affect cultural groups’ endorsement of social hierarchies. Similarly, Kong (2014) reported that national wealth interacted with 5-HTTLPR to affect corruption and Kong (2015) reported that democracy interacted with 5-HTTLPR to influence social trust. Recently, Minkov and Bond (2015) examined group variations in 5-HTTLPR, DRD4, and AR and correlated them with long-/short-term time orientation. They (Minkov & Bond, 2017) also found that group variations in the FAAH gene were associated with happiness.

Finally, although most researchers used human language as the example of universal biology but culture-specific behavior (e.g., Paul, 2015), it is plausible that some aspects of language differences might have had genetic correlates. As mentioned earlier, specific genes have been implicated in various aspects of language and the allele frequencies of these genetic variants vary across groups. Recently, linguists have found evidence that ecological factors play important roles in the evolution of the sounds of different languages (Everett, 2013; Maddieson & Coupé, 2015). As Levinson and Dediu (2013) recently speculated, “The distribution of genetic variants can bias the direction of cultural evolution, and cultural evolution can, in principle, channel biological evolution” (p. 23). To acknowledge the mutual influence or interaction between genes and culture, they also added in a footnote: “Of course, it is entirely possible that the reverse process, whereby speaking a tone language generates weak selective pressure on the processing and acquisition of specific cues, is active as well” (p. 23).

The factors and processes involved in group-specific or local adaptations have been well studied with plants and to some extent animals. Savolainen, Lascoux, and Merita (2013) summarized recent advances in ecological genomics of local adaptations based on reciprocal transplant experiments (field experiments in which individuals from two or more populations are reared in their respective native and nonnative environments). Such experiments showed the well-documented antagonistic pleiotropy (alternative alleles at a given locus are favored in different environments, i.e., evidence for divergent evolution), convergent evolution (one locus affects fitness/phenotype in one environment, but a different locus affects fitness/phenotype in a different environment) or parallel evolution (the same mutation at the same locus leads to the same trait in different species), and polygenicity (different loci affect the same trait).

True reciprocal transplant experiments are not permissible for humans, but quasi-experimental versions can be found due to reciprocal migrations. Indeed, migration is one of the main sources of group-specific selection pressures. Many of the existing examples of local adaptations are a direct result of some human groups moving into new territories (early humans’ move to northern Europe led to lighter skin color; moving to Tibetan and Andean plateaus led to altitude adaptations), or the process of migration itself may have accounted for the selection of DRD4 7R, as mentioned earlier (C. Chen et al., 1999; Matthews & Butler, 2011). After all, migration comes with many potential changes in local environments: latitude, altitude, climate, fauna and flora, lifestyles (sedentary vs. migratory, large vs. small group living), and pathogens (e.g., malaria parasites’ deadly effects on early Europeans who went to colonial African countries, where the local African populations had numerous resistance variants, as discussed above). Migration has also led to contacts with other human groups, which would introduce new pathogens (e.g., Europeans’ spreading of pathogens for smallpox, influenza, bubonic plague, and pneumonic plagues to Native Americans and aboriginal Australians; Karlsson et al., 2014), new genes (i.e., gene flow among groups), and the special case of gene flow—introgression from Archaic humans (e.g., Neanderthals and Denisovans) to modern humans (Hawks, 2013). In addition to these major sources, under the umbrella of migration-related factors, other factors include the advent of agriculture, domestication and herding, or sudden processes such as the Black Death. As is obvious, the variety of group-specific processes can have a wide-ranging selection on all relevant biological systems (the immune, sensory, digestive, and neuronal systems, etc.; Hawks et al., 2007; E. T. Wang et al., 2006).

The mechanisms mentioned above emphasized natural selection, but group-specific adaptations can also result from sexual selection. Indeed, the selection of EDAR in East Asians has been commonly attributed to sexual selection (Bryk et al., 2008) because hair thickness (or teeth shape and sweat gland function) does not seem likely to carry a significant effect on survival. In fact, Darwin believed that the origin of human racial variations was not in the conditions of life, but rather through sexual selection. He wrote,

the external characteristic differences between the races of man cannot be accounted for in a satisfactory manner by the direct action of the conditions of life; the differences between the races of man, as in color, hairiness, form of features, etc., are of a kind which might have been expected to come under the influence of sexual selection.

As we discussed earlier, modern research has proven him wrong in some aspects of his above claim (e.g., color), but the jury is still out on others (e.g., hairiness).

Understanding group-specific adaptations is also crucial to future research of gene–behavior associations. Up to now, population stratification has been treated as a nuisance for gene–behavior association analysis, as discussed above. Group-specific variants can produce both false positive results (when they reside fortuitously in one of the two compared groups) and false negative results (when the actual subpopulation causative variant is diluted out with subpopulations without the variant, a common problem with GWAS, as discussed below). With cultural genomics, it is a nuisance no more, but rather a central focus of research. Informed by cultural genomics, a good understanding of convergent and divergent evolution of different groups—or local adaptations—can guide future gene–behavior association studies.

Subgroup Adaptations or Within-Group Gene–Culture Interactions

Although culture has usually been broadly defined (e.g., human-made part of the environment), its operationalization has typically stayed at the level of race or country or ethnicity. Cohen (2009, 2016) has argued convincingly that at least a few other areas of research should be integrated into (cross-)cultural psychology, including SES, religion, and regional differences. Gene–culture interactions have been rarely studied within a racial or ethnic group because such groups are assumed to be genetically homogeneous. However, there is both indirect and direct evidence that such an assumption may not be true. First, much research has documented assortative mating. Although education is still the dominant force in assortative mating, there is some evidence for genetically assortative mating (Domingue, Fletcher, Conley, & Boardman, 2014). Such mating strategy’s long-term consequences are yet to be studied. Second, although it varies greatly around the world (from so-called egalitarian societies to caste systems), social hierarchy is a fact of life in all cultures. Social stratification leads to uneven distribution of resources, which theoretically create differential selection pressures and adaptive strategies. Back in 1991, Belsky, Steinberg, and Draper (1991) proposed a controversial theory of how early cultural and environmental stress can lead to differential life strategies mediated by biological (including genetic) processes. Over the decades, though still controversial (Rickard, Frankenhuis, & Nettle, 2014), this theory has obtained growing support from empirical research (Belsky, 2014). As G. Clark’s (2014) controversial book shows, there might be much stability in socioeconomic hierarchy across many generations. Third, the most important factor in social hierarchy in modern societies is education level. Recent GWAS studies (Martin et al., 2011; Okbay et al., 2016; Rietveld et al., 2013) identified genome-wide significant SNPs associated with educational attainment (see Hayden, 2016, for a debate on the significance of this line of research). Potential behavioral mediators have also been identified (Domingue, Belsky, Conley, Harris, & Boardman, 2015; Rietveld et al., 2013; Ward et al., 2014; Zhu, Chen, Moyzis, Dong, & Lin, 2015). We (C. Chen et al., 2013) also examined college enrollment as a means of social stratification that may involve underlying genetic stratification—social selection. We found that several genetic variants related to neurotransmitters were disproportionally represented among college students, and these variants were linked to cognitive and personality traits that were conducive for school achievement.

Individual-Level Gene–Culture Interactions and Gene–Culture Compatibility

In the classic debate regarding cross-cultural versus cultural psychology (Greenfield, 2000; Triandis, 2000), one of the key issues is whether culture is superorganic (Kroeber, 1917) as an antecedent variable (Lonner & Adamopoulos, 1997) or embedded within an individual (Jahoda, 1992; Shweder, 1990). What we have discussed so far parallels cross-cultural psychology and treats culture as superorganic. When examining gene–culture interactions at the individual level, culture cannot be treated as superorganic (Richerson & Boyd, 2001). In this case, culture is an interpretation by individuals of their experiences. On the genetic side, for the previous levels of analysis, genetic variations covary with cultural variations (either across time or across groups) and the task was to determine the extent of their covariation, causal direction, and potential mechanisms involved. At the individual level, genetic variations are an individual difference (within a group), and one obvious mechanism involved in gene–culture interaction is epigenetic (i.e., cultural/environmental factors affecting gene expression via mechanisms such as methylation and histone modification), rather than natural, sexual, or social selection as discussed above.

Research that fits this level analysis includes most of the gene–environment interaction (G×E) research. Most prominent is the line of research guided by the differential susceptibility theory (Belsky, 1997; Belsky & Pluess, 2009) or related ideas such as biological sensitivity (Boyce & Ellis, 2005). To the extent that environmental factors considered in this research are cultural in nature, it would belong to gene–culture interactions at the individual level. However, conventionally (cross-)cultural psychology has a narrower focus on groups as mentioned earlier, so we will not make further comments on this line of research.

Nevertheless, if one extends G×E research into two or more cultural groups or into cultural dimensions within a culture, it becomes cultural genomics. For example, Kim and her group have extended G×E to gene–culture interaction and reported differential gene–behavior relations across different cultural groups such as Americans versus Koreans or Japanese involving 5-HTTLPR, HTR1A, and OXTR (e.g., Kim, Sherman, Mojaverian, et al., 2011; Kim, Sherman, Sasaki, et al., 2010; LeClair et al., 2014; see Sasaki et al., 2016, for a recent summary). It should be noted that such findings have been described as gene–culture or culture–gene interactions in the literature because they showed statistical interactions. To differentiate the statistical interactions from interactions as in mutual influence, we describe them either as statistical interactions or as differential relations across groups. Hastie et al. (2012) reported statistical interaction between ethnicity and OPRM1 in their effect on pain. Kitayama et al. (2014) reported that DRD4 moderated cultural orientation differences. G. D. Wang, Xie, Peng, Irwin, and Zhang (2014) reported that in different cultural groups, different alleles of COMT were linked to better working memory as well as larger hippocampal size. In terms of differential gene–behavior associations by cultural dimensions within a culture, Dressler, Balieiro, Ribeiro, and Santos (2009) found that the relations between the prototypicality of family life (cultural consonance) and depressive symptoms varied by 5HT2A genotype. Recently, Luo et al. (2015) also reported a statistical interaction between OXTR and independent value orientation in their effect on empathy. There are many such statistical interactions, although some (an unknown proportion) of them may simply be statistical artifacts. As in any field of science, only replication will help us identify the true effects. It is important to emphasize again that to truly understand the statistical interactions, we need to consider convergent evolution and to determine whether these statistical interactions are due to epigenetic processes or gene–gene interactions, as mentioned earlier.

Finally, evolution is all about gene–environment (culture) fit or compatibility, which takes place at the individual level. Because neither genes nor the environment remains constant, their relations are dynamic and, thus, they coevolve. When individuals who carry a gene or a set of genes evolved for a given environment face a new environment, evolutionary mismatch takes place. Classic examples of evolutionary mismatch include human physical problems such as back problems, obesity, diabetes, and so forth. Researchers have speculated about evolutionary mismatch in mental illnesses such as attention-deficit/hyperactivity disorder (ADHD; Jensen et al., 1997) and postpartum depression (Gluckman & Hanson, 2006; Hahn-Holdbrook & Haselton, 2014; Nesse & Williams, 1996). Finally, evolutionary psychologists have discussed mismatches in social cognition and various cognitive biases (Barkow, Cosmides, & Tooby, 1995; Buss, 2005). Evolutionary mismatch is an important factor in our understanding of gene–culture interactions, but it needs to go beyond a theoretical construct to a method of hypothesis testing with empirical evidence.

Methodological Challenges and Possible Solutions

As an emerging discipline, cultural genomics faces a number of methodological challenges. The first major challenge is the sample size for a given unit (or level) of analysis. When studying species-wide adaptations, we have a sample size of one—homo sapiens, despite comparisons with a growing list of other species whose genomes have been analyzed. In this case, any initial findings can only be considered as very tentative, especially given the likelihood of spurious relationships, because many human-specific genomic changes may just be genetic “noise.” This area of research particularly requires a strong theory and detailed mechanistic evidence (including abnormal or exceptional cases). We see the emergence of human language and the trend toward human self-domestication as most promising in meeting these requirements in the near future.

The sample size improves when we move down the level of analysis toward group-specific adaptations. Multiple contrasts have been used to examine the classic examples of skin pigmentation (variations along the latitude across different longitudes), malaria resistance (different tropical regions), lactase persistence (different regions with dairy farming), altitude adaptation (different plateaus), and migration (different routes). One strength of this level of analysis is that conclusions in this area can be falsified if a new contrast that should have confirmed the results fails to do so. Nevertheless, group-level analysis is still limited in the number of possible groups available for analysis, which is likely to lead to false positives when a large number of group-level behavioral measures are used to search for significant correlations with a given genetic polymorphism (e.g., DRD4 VNTR, 5-HTTLPR of SLC6A4, or MAO-A VNTR). This situation may be further compounded by the large number of genetic polymorphisms within the same gene (e.g., see a discussion of various polymorphisms of SLC6A4 by Murdoch, Speed, Pakstis, Heffelfinger, & Kidd, 2013). To reduce false positives, researchers should follow the best research practices such as Simmons et al.’s (2011) recommendation for greater transparency/disclosure; Finkel, Eastwick, and Reis’s (2015) recommendations for preregistration, material and data sharing, replication, avoiding piecemeal publication, and increasing sample size; and Dick et al.’s (2015) specific recommendations for candidate genes research, emphasizing well-justified selections of genes and their relevant variants, environmental variables, analysis (e.g., models of genetic effects from dominant, recessive, dosage, heterozygotic advantage, polygenic), and sample size (for statistical power). In addition, Galton’s problem (i.e., spatial autoregression) is a specific threat in research on group-specific adaptations when using simple correlational or regression analyses with groups that are too similar to one another (for a recent example, see Eisenberg and Hayes’, 2011, methodological challenge to the analysis of Chiao and Blizinsky, 2010). Researchers have proposed two-stage least square regression (e.g., Dow, 2007; Kelejian & Prucha, 1998) or network analysis (e.g., Dow, Burton, White, & Reitz, 1984) to deal with such a threat. Such methods should be adopted more widely.

Moving further down the level of analysis, subgroup-specific adaptations and individual-level gene–culture interactions should theoretically have fewer obstacles in getting larger sample sizes and more replication samples. These areas of research are most promising, and should make up the bulk of future research in this field. Promising as they may be, psychologists (and others) are, nevertheless, likely to feel somewhat apprehensive at the moment. Such an apprehension is understandable because there was high expectation that once the human genome was sequenced, understanding human behaviors and diseases would follow rapidly. Indeed, the Human Genome Project promised that “It will help us to understand and eventually treat many of the more than 4000 genetic diseases that afflict mankind, as well as the many multifactorial diseases in which genetic predisposition plays an important role” (Human Genome Project, 1991). As it turns out, there are more secrets to the human genome and its connections to human behaviors and diseases than anticipated, and progress has been slow. Researchers have asked questions such as the following: Are most of the findings false positives? Where is the missing heritability and can we find it? Is the classical infinitesimal model of quantitative genetics correct? Is GWAS replacing the candidate gene approach? Where do we go from here? These are important questions that are subject to much debate, and each of them deserves extensive treatment. In the following paragraphs, we focus on the ways in which these questions are relevant to cultural genomics and how cultural genomics can contribute to a discussion of these issues.

First, much ink has been spilled on the “replication crisis” covering several disciplines, ranging from social psychology (Open Science Collaboration, 2015, but see Gilbert, King, Pettigrew, & Wilson, 2016) to biomedical research (Ioannidis, 2005; Ioannidis & Khoury, 2011; but see Jager & Leek, 2013), and from earth sciences (Santer, Wigley, & Taylor, 2011) to computer science (Peng, 2011). Many factors have contributed to this problem, including publication bias, p hacking, and underpowered sample size. Further complicating this issue in the field of behavior genetics are statistical interactions (e.g., gene-by-environment, gene-by-culture, gene-by-gender, gene-by-gene) that have been found in many studies (some of which are mentioned earlier). Such interactions are two-edged swords. On one hand, they necessarily compound the issue of multiple comparisons by increasing the number of statistical tests, and hence lead to more false positives. On the other hand, cultural genomics would argue that local adaptations of a given trait would likely involve different genetic variants in different groups, and thus statistical interactions (i.e., differential genetic effects by cultural groups) are to be expected. As discussed earlier, such statistical interactions may involve subgroups within a racial/ethnic group, an issue that has not received wide attention. Such likely genetic heterogeneity may have confounded or reduced the effects of genes in previous findings, leading to both false positives and false negatives. The only way to separate true from false positives is through replications with evolutionarily equivalent samples.

Second, the gap between the heritability estimates based on twin studies and the proportion of variance explained by specific genes has been called “missing heritability” (e.g., Manolio et al., 2009). Many factors may have contributed to this gap. For example, heritability might have been overestimated by twin studies because of issues about the assumptions involving gene–environment correlation and measurements. The low proportions of variance explained by specific genes may have been because many genes make small contributions—that is, the infinitesimal model of quantitative genetics (Fisher, 1918) or recently termed “fourth law” of behavior genetics (Chabris, Lee, Cesarini, Benjamin, & Laibson, 2015). The use of the genome-wide complex trait analysis (GCTA; Yang, Lee, Goddard, & Visscher, 2011) has attempted to solve this problem and with some success. There are many other possible explanations for “missing heritability,” however, including gene–gene interactions (Zuk, Hechter, Sunyaev, & Lander, 2012), which can be studied for a few genes, but remain computationally too intensive at the genome-wide level. From the perspective of cultural genomics, both polygenicity and gene–gene interactions should be considered because of the likely differences in the evolutionary history of groups and subgroups.

Third, when the infinitesimal hypothesis is believed to be a “law,” some researchers have argued against the traditional candidate gene approach and recommended using only GWAS with very large sample sizes (~100,000; Benjamin et al., 2012; Chabris, Hebert, et al., 2012; Chabris, Lee, et al., 2015). GWAS is certainly a powerful tool. As Price, Spencer, and Donnelly (2016) recently summarized, GWAS has identified more than 10,000 associations, with many of them replicated across studies. In addition to confirming earlier results from candidate genes, large-scale GWAS have also helped identify novel variants implicated in diseases.

GWAS’ limitations, however, have to be noted. First of all, GWAS has varying levels of coverage of the genome from low to high density and even high density scans miss many hard-to-type regions of the genome (e.g., Manolio et al., 2009; Saccone et al., 2009). A number of important genetic variations such as variable number tandem repeats (VNTRs, a short nucleotide sequence organized as a tandem repeat) and indels (insertions and deletions of bases) are not included in whole-genome chips. Large regions of the genome remain unanalyzed with current technology. Unfortunately, this is not improved with second-generation whole genome sequencing (WGS), because the short read lengths used in these approaches result in accurate analysis of only 70% to 80% of the genome (The 1000 Genomes Project Consortium, 2015; Goldfeder et al., 2016), and sequencing errors are common (L. Chen, Liu, Evans, & Ettwiller, 2017). The second limitation concerns replicability of findings. Although GWAS replication is becoming more promising (Price et al., 2015), studies based on candidate gene approaches have an approximately 50% reproducibility, consistently better than GWAS (Kaboh et al., 2012; Lohmueller, Pearce, Pike, Lander, & Hirschhorn, 2003; Manolio et al., 2009; Zuk et al., 2012). Furthermore, genes found in GWAS often identified the same genes previously reported by the candidate gene approach, with well-known examples such as the relationship between APOE and Alzheimer’s (The International Consortium for Blood Pressure Genome-Wide Association Studies, 2011; Kamboh et al., 2012; Neale et al., 2010). Third, as discussed earlier, large-scale GWAS has to pool large numbers of diverse samples, which ignores the diverse evolutionary histories of subgroups and their potentially unique causative alleles. Furthermore, for cultural genomics, measurement equivalence of behavioral traits will be a significant challenge across groups. Fourth, although it may be feasible in fields with large amounts of funding and large consortia to pursue GWAS for specific diseases (e.g., schizophrenia, ADHD, cancers) or easy-to-obtain phenotypes (e.g., educational attainment, height), large-scale GWAS is not realistic for most other behavioral measures of interest to psychologists and small-scale GWAS has insufficient statistical power. Fifth, unlike candidate gene research, which is hypothesis driven with at least some prior knowledge of the gene’s functions, GWAS is hypothesis free and data intensive and currently still has difficulty incorporating models of gene–gene and gene–environment interactions (Polderman et al., 2015). Given that all the phenotypes of interest to psychologists have a strong environmental component, it is unclear how future GWAS will be improved with regard to incorporating environmental factors if smaller hypothesis-driven studies are not used first to discover the relevant factors. The interaction between the DRD4 7R allele and activity level associated with longevity is one such study (Grady et al., 2013). Finally, whereas GWAS is one approach to finding genes associated with culture and behaviors, research then has to proceed to analyzing individual genes. As Geschwind and Flint (2015) recently suggested, given the number of genes already implicated with behavior, we need to consider the shift from “looking for genes” to “understanding their functions” (p. 1493).

Therefore, the apparent “conflict” between candidate genes approaches and GWAS is only an illusion. GWAS and the candidate gene approach should complement each other: GWAS and WGS should continue to discover new candidate genes (especially by increasing their error-free coverage of the genome), whereas the candidate gene approach explores the functions of candidate genes (both known and to be discovered) in evolutionary and cultural contexts. Given that there are numerous unanalyzed reasons for negative results in GWAS, we propose that it is premature to suggest that the only path forward is to focus solely on a GWAS approach.

Finally, in addition to the unanalyzed genetic variations as mentioned above by the current technology, another reason to delay the canonization of the infinitesimal model as a “law” is our limited understanding of how most genes work. Before the modern synthesis, genes were defined as biological units responsible for specific traits. Such a one-to-one correspondence has driven much of our expectations until today. Along the way, however, geneticists redefined genes as stretches of DNA that code for proteins, now broadened to include extensive regulatory networks, noncoding RNAs, and numerous other features imbedded in our genomes (Kellis et al., 2014). This change in definition adds multiple steps to the “long chains” from genotype to phenotype. One promising way of putting the humpty-dumpty of “genes” back together is to study gene(s) at their functional level. Efforts have been made in using pathway or system-level or gene-set analysis (e.g., L. Wang, Jia, Wolfinger, Chen, & Zhao, 2011) or gene (regulation) networks (e.g., Sansom, 2011; Varshney et al., 2017). This approach holds much promise as more and more expression data become available to allow for a better redefinition of gene sets or networks (Kellis et al., 2014).

To meet the genotype redefinition effort halfway, phenotype refinement is also needed. Mendel did not pick peas’ traits randomly, but chose traits that appeared to systematically vary. Refinement of phenotypes was responsible for the success of identifying the BRCA genes for breast cancer and the APOE gene for Alzheimer’s. Another approach to refining phenotypes is to use “endophenotypes” or intermediate phenotypes. For diseases, researchers have used cognitive correlates and neural correlates as endophenotypes with the assumption that these correlates would mediate the genetic effects on diseases. Closely related to phenotype refinement is sample refinement (phenotype refinement for BRCA and APOE was based on early age of onset). Sample refinement can define the relevant subgroups for a given disorder/trait that might have been missed otherwise. Indeed, many genetic defects only affect a small number of people. For example, phenylketonuria (PKU) afflicts only 0.01% of the newborns in the United States. PKU is an autosomal recessive genetic error of metabolism, which involves the inability to metabolize the amino acid phenylalanine. It is genetically heterogeneous, with more than 400 different mutations in the PAH gene, many on different genetic backgrounds from different subgroups. If “cognitive deficits” had been used as the phenotype in a study of a representative sample of infants, the PAH gene would not have been discovered in a GWAS analysis. It took an understanding of the biochemical mechanism underlying this one type of “cognitive deficit” to identify the underlying gene. “Cognitive deficit” was too broad a phenotype.

PKU is also illustrative of another common myth that if the variance explained by most known genetic variants is small, there must be a large number of genes working together to produce the phenotype. It would be certainly true that a large number of genes are related to the phenotype, but they need not work together to produce the phenotype. For the population, PAH variants account for only 0.01% of “cognitive deficits,” but it does not mean that many small-effect variants must work together to produce the phenotype in a single individual, the commonly stated assumption. For PAH mutation carriers, the effect on the variance is 100%: They will have PKU. There are now known to be many forms of “cognitive deficits” (PKU, fragile-X, etc.), each one of which results in the deficit independently. This confusion between “small” effects on the population variance and the significance of the same gene variant for the individual needs to be emphasized. It is likely many genes relevant to cultural genomics will exhibit similar distributions to the known genetic variants for other phenotypes, as discussed throughout this article. Small effects on population variance does not mean unimportant, as often assumed. Given that PKU is the only “cognitive deficit” disorder that has a successful environmental intervention (elimination of phenylalanine from the diet), it is certainly very important for those individuals, even though the treatment is useless for the other 99.9% of individuals with possible “cognitive deficits.” We must accept the fact that gene variants that underlie cultural genomics will likely be as diverse as the cultures themselves; yet, ultimately practical applications and interventions must account for this variability. There is no “one size fits all.” As discussed above, if we are truly to aspire to an equitable society, ignoring our differences will, in fact, lead to inequality.

In sum, the field of cultural genomics is moving toward refining (or redefining) genotype, phenotype, sample (particularly its evolutionary histories), and environment. Such refinements should help us better understand polygenicity and pleiotropy, which may be a result of misspecification of genotype, phenotype, sample, and/or environmental factors. Social scientists are adept at building multilevel models to consider the complex relations among variables. Just like polygenicity and pleiotropy, behavioral studies typically deal with many predictors of the same outcome and many outcomes from the same predictor. For example, cognitive abilities, motivation, learning style, school environment, family SES, structure, parenting, and peer factors have all been linked to school achievement; Family SES has been linked to not only school achievement but also physical and mental health as well as social cognition and relationships. Multilevel models can help move the field of behavior genetics forward toward understanding the complex gene–behavior relations and gene–culture interactions. As a starting point, several researchers have proposed high-level integration or meta-dimensional and multistage analysis (Ritchie, Holzinger, Li, Pendergrass, & Kim, 2015). Such analysis uses powerful multistage analytical approaches such as least absolute shrinkge and selection operator (LASSO) regression, kernel-based or graph-based analyses, Bayesian, and machine learning approaches to analyze multiomics data, from genomic, epigenomic, transcriptomic, and proteomic, as well as environment variables.

Cultural Genomics’ Answer to Ideological and Political Challenges

For a long time, discussions of the interactions between genes and culture have often ended up entangled with biological determinism or reductionism because of the common but incorrect assumption that any gene–culture interaction goes in only one direction—from genes to behaviors and, thus, to culture. Early attempts at bypassing this conundrum have been mostly evasive, as mentioned at the beginning of this article, and thus have proved ineffective in resolving the issue. From the perspective of cultural genomics, the ideological and political challenges need to be resolved by emphasizing three points.

First, we need to emphasize that for any given timescale and any level of analysis, there are truly interactive processes between culture and genes. If anything, cultural processes are perhaps the strongest driving force behind recent human evolution (C. Chen et al., 2016). With that corrected perspective, biology is no longer to be feared but something to work with. We need to understand the extent to which local eco-cultural contexts have shaped the genomes of different groups and how they interact with cultural changes. By understanding individual and group differences (including their evolutionary histories), we can strive for equality in opportunities and outcomes. Uncovering group differences can appear risky initially, but a correct perspective can help minimize the risk and indeed help us appreciate the differences. Perhaps, one analogy of this paradigmatic shift is that from early imperialist anthropology and cross-cultural psychology to cultural relativism since the turn of last century (Boas, 1901). In other words, if it is possible to talk about cultural differences without using the lens of human group hierarchy (or cultural imperialism), it should be possible to talk about biological differences carefully and safely without leading to racism. Biological diversity is just as necessary as cultural diversity, both of which should be treasured. For any species, genetic diversity is preferable, because species survival depends on adaptation to an ever-changing environment. Indeed, what we have proposed here is the possibility that cultural diversity itself can drive biological diversity. If we indeed value cultural diversity, we should value it even more when it has turned into biological diversity.

Second, most of human diversity (either cultural or biological) can be viewed as different groups using different means to attain the same end of group survival and well-being. This notion of convergent evolution is emphasized by cultural genomics as discussed earlier. As a result of convergent evolution, different genetic variations can be linked to the same or similar phenotypes. In other words, a given trait may be linked to one genetic variation in one group but to another genetic variation in another group. If one focuses on just one of the two genetic variations, but not on both, any conclusions about group differences is likely to be incomplete and may even be misleading (Murdoch et al., 2013). With increasing genomic information, system-level or pathway approaches and polygenic scores should help balance the picture. Cultural genomics ultimately aims to integrate genes’ history (molecular evolution, including introgression), biochemical functions, epigenetic mechanisms, behavioral correlates (especially refined phenotypes, as mentioned earlier), and selection pressure posed by cultural elements as well as its mechanisms (natural, sexual, or social selection).

Finally, cultural genomics can help us to work with biological diversity. Rapidly changing cultural and physical environments (either due to migration or cultural change) necessarily present challenges to the original cultural or biological (and necessarily local) adaptations. Such challenges can be met by cultural means. Classic examples are sunscreens for pale-skinned individuals in the sun-drenched areas such as Southern California and vitamin D supplements for the dark-skinned individuals in ultraviolet (UV)-scarce northern Europe. Similarly, compulsory schooling is perhaps the most important cultural invention that can ideally provide equality in opportunities, and hopefully in outcomes, regardless of the individuals’ ontogenetic history or groups’ phylogenetic history. Encoded in the human genome are the instructions to build a brain for learning.

Concluding Remarks

Given the strong theoretical arguments for gene–culture interaction and the encouraging preliminary evidence, the emerging field of cultural genomics has great promise in helping us understand how genes and culture mutually influence each other at different levels (from species wide to within each individual). At the same time, this field is confronted by many challenges that need to be overcome. In this article, we have outlined the major framework of this field and addressed the challenges and offered some tentative solutions. In this concluding paragraph, it is worth keeping in mind that science is a gradual cumulative endeavor. Indeed, when Darwin proposed his theory of natural selection, he did not know what a gene was. We may or may not know all the specific mechanisms of how culture and genes interact at the moment, but that day will come, albeit gradually. The genome is large (>3 × 10⁹ base pairs), but it is not infinitely large. Along the way, there will be setbacks because science is about exploration and discovery, which necessarily comes with false starts. Indeed, one can repeat the quote often attributed to Albert Einstein, “If we knew what it was we were doing, it would not be called research, would it?” An optimistic view is that the recent pessimism regarding the large number of presumed false positives is, in fact, a sign of an accelerated pace of research. A field is better off with positives to separate into true and false ones than one with only negatives, which would spark no new research. Two things are certain: Cultural genomics is an idea worth exploring, and the tools for such exploration are now available, and becoming more sophisticated, to move beyond the theoretical modeling of gene–culture interactions to empirical testing.