The impact of natural selection on health and disease: uses of the population genetics approach in humans

Investigations of the legacy of natural selection in the human genome have proved particularly informative, pinpointing functionally important regions that have participated in our genetic adaptation to the environment. Furthermore, genetic dissection of the intensity and type of selection acting on human genes can be used to predict involvement in different forms and severities of human diseases. We review here the progress made in population genetics studies toward understanding the effects of selection, in its different forms and intensities, on human genome diversity. We discuss some outstanding, robust examples of genes and biological functions subject to strong dietary, climatic and pathogen selection pressures. We also explore the possible relationship between cancer and natural selection, a topic that has been largely neglected because cancer is generally seen as a late-onset disease. Finally, we discuss how the present-day incidence of some diseases of modern societies may represent a by-product of past adaptation to other selective forces and changes in lifestyle. This perspective thus illustrates the value of adopting a population genetics approach in delineating the biological mechanisms that have played a major evolutionary role in the way humans have genetically adapted to different environments and lifestyles over time.


Introduction
Our understanding of the patterns of human genome diversity has improved considerably over the last 10 years. The complete sequence of the human genome, which was published in 2001 (Lander et al. 2001;Venter et al. 2001), provided us with information about the location and genomic structure of genes, but did little to improve our understanding of the genetic diversity of the human genome at population level. A number of international consortia have since investigated the genetic differences both between individuals from the same population and between different populations from around the world. The International HapMap project (International HapMap Consortium 2005;Frazer et al. 2007;Altshuler et al. 2010), based on genotyping technologies, and, more recently, the 1000 genomes project (1000Genomes Project Consortium 2010, based on next-generation sequencing, have made an enormous contribution to the identification and character-ization of different types of variation in different populations worldwide. The HapMap project, for example, has cataloged both allele frequencies and levels of genetic association (assessed by measuring linkage disequilibrium, LD) across several populations, for 3.5 million single nucleotide polymorphisms (SNPs) (International HapMap Consortium 2005;Frazer et al. 2007;Altshuler et al. 2010). The 1000 Genomes project has described the location and allele frequencies of approximately 15 million SNPs, 1 million short insertions and deletions, and 20 000 structural variants. Interestingly, the data obtained suggest that each of us carries 250-300 loss-of-function variants of known genes, and that we are heterozygous for 50-100 variants known to be involved in genetic disorders (1000Genomes Project Consortium 2010Abecasis et al. 2012).
These data sets for human genetic variation in health have also been very useful for the interpretation and design of genome-wide association studies (GWAS) on many human diseases. By the end of 2012, more than 1500 GWAS had been published on more than 250 traits or diseases as varied as height, hair color, obesity, smoking behavior, schizophrenia and colorectal or bladder cancer (see, National Human Genome Research Institute catalog of published GWA studies: http://www.genome.gov/ 26525384). These GWAS have made a major contribution to our understanding of the genetic basis of human phenotypic variation in both health and disease, by identifying genes/variants associated with the variation of traits of interest, disease susceptibility/severity and differences in response to treatment.

The evolutionary and population genetics approach
Various factors shape the diversity of the human genome at the population level and may contribute to phenotypic variation. Mutation and recombination create and reshuffle, respectively, diversity within the chromosomes. Other factors, such as demographic processes and the cultural behavior of human populations, then affect allelic frequencies within populations (Salem et al. 1996;Seielstad et al. 1998;Chaix et al. 2004Chaix et al. , 2007Wilder et al. 2004). Both genetic and archeological evidence have supported a common, recent origin of all humans in Africa, followed by range expansion and dispersal out-of-Africa (i.e., the Out-of-Africa hypothesis) (Lewin 1987;Quintana-Murci et al. 1999;Cavalli-Sforza and Feldman 2003;Macaulay et al. 2005;Mellars 2006;Fagundes et al. 2007;Laval et al. 2010). However, there is increasing evidence to suggest that such dispersals of modern humans were accompanied by some degree of admixture with local populations of ancient hominids. Indeed, recent estimates suggest that Neanderthals contributed 1-4% to modern Eurasian genomes, and Denisovans contributed 4-6% to modern Melanesian genomes Reich et al. 2010) Furthermore, adaptive introgression of archaic HLA haplotypes from Neanderthal and Denisovan genomes have been documented among some modern human populations (Abi-Rached et al. 2011), highlighting overall that the Out-of-Africa hypothesis oversimplifies a much more complex scenario.
The colonization of new geographic regions led to the exposure of human populations to different, changing environments, and these differences in climate, nutritional resources or pathogens acted as selective forces, to which human populations had to adapt if they were to survive (Sabeti et al. 2006;Nielsen et al. 2007;Novembre and Di Rienzo 2009). This brings us neatly to the concept of natural selection, as genetic variants increasing fitness would have been conserved in such conditions, thereby increasing in frequency, whereas deleterious variants would have been rapidly eliminated, contributing to the process of genetic adaptation.
Investigations of the legacy of natural selection events in the past within the human genome and of the ways in which these selective events have shaped current genetic diversity have proved particularly informative, pinpointing functionally important regions of the genome (Bamshad and Wooding 2003;Akey et al. 2004;Bustamante et al. 2005;Sabeti et al. 2006Sabeti et al. , 2007Voight et al. 2006;Barreiro et al. 2008;Pickrell et al. 2009;Altshuler et al. 2010). Indeed, studies of whether and how selection has targeted particular genes in the human species as a whole, or in specific human populations, constitute a powerful approach for identifying genes that have played (and probably continue to play) an essential biological role in our survival, and distinguishing between such genes and those with a higher degree of redundancy (Sabeti et al. 2006;Quintana-Murci et al. 2007;Barreiro and Quintana-Murci 2010). Furthermore, population genetic dissection of the intensity and type of selection acting on human genes facilitates the identification of genes likely to be involved in rare, severe Mendelian diseases and makes it easier to distinguish between these genes and those most likely to be involved in complex susceptibility to disease. Overall, this approach increases our understanding of the ways in which past selection events have contributed to current differences in resistance/susceptibility to disease (Blekhman et al. 2008).

Natural selection acts in different forms
Natural selection can take many different forms and act with different intensities. The most common type of natural selection is probably purifying selection, also known as negative selection, which affects most genes, to various extents (Bamshad and Wooding 2003;Bustamante et al. 2005;Nielsen et al. 2005). Purifying selection decreases the frequency of mutations that prove to be disadvantageous to carriers in a given environment, the magnitude of this decrease depending on the degree to which the mutation is deleterious. If the effects of the variant are highly deleterious to carriers, or even lethal, the variant is purged from the population by strong purifying selection. The frequent removal of deleterious variants can also result in the occasional removal of neutral linked variation (particularly in low-recombining regions), a phenomenon known as background selection. Genes subject to this type of selection generally display a major deficit of non-synonymous mutations. Such genes are thus likely to play an essential role in the host, with few, if any, amino-acid substitutions tolerated. In this case, the occurrence of new amino-acid variants may lead to severe Mendelian diseases. Purifying selection is weaker for mutations with effects that are only mildly deleterious. This allows such mutations to accumulate at the population level and to be maintained at a low frequency. In this case, an excess of low-frequency alleles is generally observed for the gene concerned, with no additional signature. Genes evolving under weaker evolutionary constraints are generally thought to be involved in less important, or more redundant, biological functions than those evolving under strong purifying selection.
In some cases, the occurrence of new genetic variants may be advantageous in a given environment, increasing the fitness of individuals, and the frequency of such variants may therefore rise due to positive selection. This process is also known as directional or Darwinian selection (Sabeti et al. 2006;Nielsen et al. 2007). When an advantageous mutation increases in frequency in the population as a result of positive selection, linked neutral variation will be dragged along with it-a process known as genetic hitchhiking. As a consequence, variation that is not associated with the selected allele is eliminated, resulting in a selective sweep that leads to an overall reduction of genetic diversity around the selected site. In such cases, several molecular signatures may be observed, including a skew in the distribution of allele frequencies toward an excess of both lowfrequency alleles and high-frequency derived alleles (i.e., derived alleles with respect to the corresponding chimpanzee alleles), and a transitory increase in the strength of linkage disequilibrium associated with the selected allele(s).
Finally, several alleles may coexist at a given locus if they are advantageous individually or together, due to the effects of balancing selection (Charlesworth 2006;Hurst 2009). This is a general type of selective regime that favors the maintenance of diversity in a population. There are two main mechanisms by which balancing selection preserves polymorphism: heterozygote advantage and frequencydependent selection.
Studies of the effects of selection at the interspecies level (e.g., comparing the human and chimpanzee lineages) can improve our understanding of the genetic mechanisms and traits underlying speciation. In turn, studies of the effects of selection on specific human populations are particularly useful for the identification of genes/variants responsible for the phenotypic diversity observed in human populations, in health (e.g., stature, skin color, hair development) and disease (e.g., differential susceptibility to infectious diseases or differences in response to treatment or vaccination).

Neutrality tests for detecting the effects of selection
Each type of selection leaves a distinctive molecular signature (e.g., nucleotide diversity, allele frequency spectrum, haplotype length) in the genomic concerned. Such molecular signatures can be detected with various statistical tests that can be broadly subdivided into those that search for selection at the inter-species level (e.g., human versus chimpanzee) and those that focus on particular aspects of within-species data (e.g., within the human lineage or between human populations).

Detection of selection between species
Inter-species neutrality tests make use of data concerning the divergence between closely related species, such as humans and chimpanzees, to detect relatively ancient selective events, such as those occurring at the time at which the two species considered diverged. In some cases, these tests also take into account data concerning polymorphism within species, making it possible to detect both older and more recent selective events. Inter-species tests include the traditional dN/dS test (Yang 1998), the McDonald-Kreitman (MK) test (McDonald and Kreitman 1991), and its extension MKPRF (Sawyer and Hartl 1992;Bustamante et al. 2005). The dN/dS test detects selection acting on protein-coding loci by comparing the ratio of non-synonymous (d N ) to synonymous (d S ) substitutions. Under neutrality, synonymous and non-synonymous substitutions should occur at a similar rate and we would therefore expect dN/dS = 1. However, the negative selection of nonsynonymous variants would result in dN/dS < 1, whereas the positive selection of such variants would result in dN/ dS > 1. In the MK test, which also considers polymorphic data, the two classes of mutations are assumed to be evolutionarily equivalent: patterns of polymorphism and divergence should therefore be the same for both classes of mutations. Non-independence between the counts of nonsynonymous and synonymous polymorphisms and fixed differences (i.e., variants that reach fixation and are different between species) is assessed by Fisher's exact test. An excess of fixed differences for non-synonymous mutations (assumed to be subject to selection) with respect to synonymous variants is typically considered indicative of adaptive evolution. In contrast, an excess of polymorphic non-synonymous variants may result from weak negative selection or local, population-specific positive selection.

Detection of selection within and between human populations
Intra-species neutrality tests focus on the level of polymorphism within a single species, such as humans, and therefore detect selection events that have occurred more recently. They can be subdivided into distinct groups, each focusing on different aspects of the genetic data. Allele frequency spectrum-based tests determine whether the frequency spectrum of mutations conforms to the expectations of the standard neutral model, and are extensively reviewed elsewhere (Kreitman 2000;Nielsen et al. 2005). Deviations from neutrality in the distribution of allele frequencies within populations can be measured with various tests, including Tajima's D, Fu and Li's D* and F* and Fay and Wu's H tests (Tajima 1989;Fu and Li 1993;Fay and Wu 2000). Negative values of Tajima's D and Fu and Li's D* and F* generally indicate an excess of rare alleles, consistent with the occurrence of negative or positive selection, whereas positive values of these statistics typically reflect an excess of alleles of intermediate frequency, due to balancing selection. Furthermore, Fay and Wu's H can be used to detected positive selection events (i.e., selective sweep), through the demonstration of an excess of high-frequency derived alleles of the targeted genes (or linked variation).
Other statistics, such as F ST statistics, are based on the level of genetic differentiation between populations historically exposed to different selection pressures (Cavalli-Sforza 1966; Lewontin and Krakauer 1973;Excoffier et al. 1992;Weir and Hill 2002). For example, geographically restricted positive selection tends to increase the degree of differentiation between a specific human population and other human populations, resulting in an increase in F ST value at the selected locus. Conversely, balancing selection, negative selection or species-wide directional selection may result in an F ST value lower than expected under a hypothesis of neutrality (Cavalli-Sforza 1966;Bamshad and Wooding 2003).
Finally, another group of neutrality tests focus on more recent positive selection events (i.e., <30 000 years), by examining the patterns of haplotype homozygosity (haplotype length) associated with particular alleles. These tests include the long-range haplotype (LRH), integrated haplotype score (iHS), and LD decay (LDD) tests (Sabeti et al. 2002;Voight et al. 2006;Tang et al. 2007). These tests are all based on the comparison of the population frequency of a given mutation with the length of the haplotypes around it. Under neutral evolution, new alleles take a long time to reach high frequencies in the population, and haplotype lengths around these variants decrease substantially during this time, due to recombination. Thus, common alleles are typically old and associated with short haplotypes. In contrast, a variant subject to recent positive selection would be expected to have an unusually long haplotype for its population frequency, because the advantageous allele increases in frequency too rapidly for recombination to have a major effect on haplotype length.

Pitfalls due to the mimicking effects of demography
Factors other than selection, such as particular demographic processes, may also account for deviations from the neutral model. Some of the tests for detecting selection described above, including those based on the allele frequency spectrum in particular, are sensitive to the confounding effects of demography on genetic diversity patterns. For example, an excess of rare alleles, giving negative values of Tajima's D and Fu & Li's D* and F* statistics, may actually result from a sudden population expansion rather than from the effects of positive selection (Ptak and Przeworski 2002;Nielsen et al. 2005). Similarly, positive values of these tests are indicative not only of balancing selection but also of the presence of strong population structure within the study-population (i.e., the study-population is indeed subdivided into different subpopulations), which may increase the proportion of alleles of intermediate frequency in the population (Kreitman 2000;Voight et al. 2005).
However, it is possible to overcome these problems by applying a basic principle of population genetics: demographic events affect the whole genome, whereas natural selection acts more locally and is restricted to particular genomic regions. Thus, when considering the impact of demographic factors on patterns of diversity, demographic models based on multiple, non-coding regions of the genome, taking into account realistic scenarios for the demographic history of human populations (e.g., population expansion, bottlenecks) can be incorporated into neutral expectations (Schaffner et al. 2005;Voight et al. 2005;Fagundes et al. 2007;Laval et al. 2010). Similarly, empirical procedures can be used to compare the value of a given statistic for the gene of interest (e.g., Tajmas's D, F ST , etc.) with background expectations for that statistic generated from genome-wide data, which should reflect neutrality. Thus, simulation-based or empirical procedures can be used to distinguish between the effects of demographic factors and those of natural selection events targeting specific genomic regions, providing evidence of past adaptation to new climates, foods or pathogens.

Adaptive phenotypes in human populations
In the last few years, the accumulation of massive sets of genome-wide genetic variation data for diverse human populations, as in the HapMap and 1000 Genomes projects (International HapMap Consortium 2005;Frazer et al. 2007;Altshuler et al. 2010;1000Genomes Project Consortium 2010, has made it possible to test blindly for the occurrence of selection throughout the entire genome. Genome-wide scans for selection have provided long lists of genes putatively targeted by positive selection, together with information about the genomic regions and biological functions most likely to have played a role in our adaptation to the environment (reviewed by (Akey 2009)). So far, candidate-gene approaches have provided the most convincing evidence for the action of natural selection on particular genes, particularly when functional evidence is also available, despite the bias inherent to such approaches due to the need to make assumptions at the outset concerning the genes likely to be subject to selection. Furthermore, one needs to be cautious when linking the patterns of selection observed with putative phenotypes, particularly when functional analyses are performed in laboratory animals, where the genes concerned may have very different functions with respect to humans.

Adaptation to diet
Many examples of the genetic adaptation of humans to diet have been described, including milk consumption, starchrich diets, and bitter-taste perception (Bersaglieri et al. 2004;Wooding et al. 2004Wooding et al. , 2006Balaresque et al. 2007;Perry et al. 2007;Tishkoff et al. 2007;Patin and Quintana-Murci 2008). Adaptation to milk consumption, through lactase persistence, is probably one of the best-known examples of natural selection in humans. A high concentration of lactase ensures that lactose is digested effectively during the first few weeks of life. The lactase gene is then repressed, resulting in residual levels (5-10% those just after birth) of the enzyme in adults (Swallow 2003). However, some populations, particularly those that have traditionally practiced cattle husbandry, maintain the ability to digest milk into adulthood. This 'lactase persistence' trait is present at high frequency in European, Middle Eastern and pastoralist African populations (reaching frequencies of up to 90% in northern Europe) (Swallow 2003).
The genetic basis of the lactase persistence is now well known; it is inherited as a dominant Mendelian trait in Europeans, with a mutation (C/T-13910) in the promoter of the lactase gene (LCT), increasing the expression of this gene (Enattah et al. 2002). The lactase-persistence -13910T allele displays clear signatures of recent positive selection in Europeans (i.e., it lies on a long-range haplotype that largely exceeds its expected length under neutrality) (Bersaglieri et al. 2004;Nielsen et al. 2005), and there is a strong correlation between the geographic distribution of this allele and historical milk consumption (Beja-Pereira et al. 2003). Furthermore, the estimated age of expansion of this allele is between 5000 and 12 000 years ago (Beja-Pereira et al. 2003), suggesting that lactase persistence emerged in response to the cultural innovation of dairying associated with an agricultural lifestyle.
Different results have been found among East African pastoralist populations that also present the lactase-persistence trait, as these populations do not present the -13910T allele. Instead, a different mutation in the LCT promoter region (À14010C), which also increases LCT expression, has been shown to be positively selected over the last 7000 years among East African pastoralists (Tishkoff et al. 2007). Altogether, these results demonstrate that the cultural trait of milk consumption has conferred a strong selective advantage in terms of human survival in different parts of the world. The most obvious selective advantage provided by the lactase-persistence trait is the possibility for lactose-tolerant individuals to access a valuable food source in times that were characterized by food shortage. An alternative, but not mutually exclusive, hypothesis is that given that lactose-intolerant individuals present syndromes such as water loss diarrhea, individuals harboring the lactase persistence-associated alleles may have had a strong selective advantage for survival, particularly in the nutritionally poor and pathogen-rich conditions of the past. Today, lactose-intolerance can still be a disadvantageous trait, as illustrated by the considerable mortality observed in non-tolerant African children following consumption of milk products from alimentary aid.

Adaptation to climate
Another example of genetic adaptation to changing environments is provided by the exposure of ancestral populations to colder climates and lower levels of incident sunlight after early migrations out-of-Africa ). These changes in climatic conditions led to variation in the quantity, type, and distribution of melanin in the skin, resulting in various levels of skin pigmentation (Jablonski and Chaplin 2000, 2012). Darker skin was favored in regions of strong UV irradiation, such as the African continent, due to the obvious protection it provides against photodamage (e.g., sunburn, melanoma, and basal and squamous cell carcinomas) (Kollias et al. 1991). Furthermore, darker skins protect against UV-induced photolysis of folate, a metabolite essential for the normal development of the embryonic neural tube and spermatogenesis (Branda and Eaton 1978;Jablonski and Chaplin 2000). In turn, the evolution of lighter skin may reflect either a relaxation of functional constraints or a selective advantage for lighter skins, in regions of low UV radiation. There is increasing evidence supporting the selective hypothesis for lighter skin, as it allowed higher levels of vitamin D photosynthesis in regions with lower levels of UV irradiation Chaplin 2000, 2012). In addition to these biological factors, sexual selection has been proposed as a factor further contributing to selection for skin color in the human population (Harpending 2002).
A number of genes involved in skin pigmentation have been identified, with effects at various stages of the pigmentation pathway (i.e., from melanogenesis to the production and maintenance of melanosomes and the switch between eumelanin and pheomelanin production) (Shriver et al. 2003;Lamason et al. 2005;Stokowski et al. 2007;Sulem et al. 2007;Han et al. 2008;McGowan et al. 2008). There is increasing evidence to suggest that many of these pigmentation genes, such as ASIP, OCA2, SLC24A5, MATP, or TYR, have been subject to positive selection across the globe or in specific human populations (Lamason et al. 2005;Izagirre et al. 2006;Soejima et al. 2006;Myles et al. 2007;Norton et al. 2007;Pickrell et al. 2009). For example, the SLC24A5 gene, for which the A111G variant is associated with lighter skin color, presents strong signals of positive selection in Europeans, based on F ST values and the haplotype homozygosity surrounding this variant (Lamason et al. 2005;Izagirre et al. 2006;Myles et al. 2007;Norton et al. 2007; Barreiro et al. 2008;Pickrell et al. 2009). In this case, positive selection may have increased the frequency of this mutation to maximize cutaneous vitamin D synthesis in areas with lower levels of UV irradiation, such as Europe (Norton et al. 2007).
Another interesting case of selection in response to environmental conditions is adaptation to altitude to avoid hypoxia, a physiological stress of the body due to the lower levels of oxygen at altitude. Various populations that have historically lived at high altitude, such as Andeans from the Andean Altiplano and Tibetans from the Himalayan plateau, have unique sets of physiological and morphological characteristics (e.g., increased respiratory and heart rate, changes in pulmonary artery pressure, enlarged thorax) that have allowed them to adapt to high-altitude conditions, where oxygen concentrations are about 40% lower than those of sea level. The genetic basis of such adaptations to extreme environmental conditions has started to be deciphered only recently. Recent studies have provided evidence of positive selection targeting various genes or gene regions involved in oxygen metabolism and sensing in Andeans and Tibetans, the strongest signals of selection being observed in the EPAS1, EGLN1, PRKAA1, and NOS2A genes (Bigham et al. 2010;Yi et al. 2010). These studies have helped to delineate a number of functionally important loci responsible for the genetic adaptation to high altitude.

Adaptation to pathogen pressures
Another important selective pressure that has confronted humans over time is that imposed by pathogens and infectious diseases. Indeed, pathogens have been, and still are in regions in which antibiotic treatment, vaccine administration and hygiene improvements are limited, a major cause of human mortality, thus exerting strong selective pressure on the human genome (Casanova and Abel 2005;Quintana-Murci et al. 2007;Barreiro and Quintana-Murci 2010;Sironi and Clerici 2010). Pathogen-driven balancing selection has been clearly demonstrated for the human leukocyte antigen (HLA) gene, the tremendous diversity of which is strongly correlated with residence in an area in which there are large numbers of pathogen species (Prugnolle et al. 2005).
Several studies have identified multiple signatures of selection in response to the presence of Plasmodium parasites, the agent responsible for malaria (Kwiatkowski 2005). For example, the high frequency of some hemoglobinopathies has been correlated with greater resistance to Plasmodium falciparum malaria: the HbS allele or 'sickle cell trait' is a textbook example of positive selection due to an infectious disease, increasing resistance to life-threatening forms of malaria by an order of magnitude when present in the heterozygous state (Allison 1954;Ackerman et al. 2005;Kwiatkowski 2005). Another example is provided by glucose-6-phosphate dehydrogenase (G6PD) deficiency. Patients with this condition have abnormally low levels of G6PD, which is particularly important for red blood cell metabolism. More than a hundred variants can lead to this deficiency, some of which are selected because they confer greater protection against falciparum or vivax malaria (Tishkoff et al. 2001;Saunders et al. 2002;Louicharoen et al. 2009). An extreme example is provided by the DARC gene, null alleles of which result in an absence of protein, preventing Plasmodium vivax from penetrating into host cells (Tournamille et al. 1995). Positive selection for the null allele has been demonstrated in sub-Saharan Africans, and this allele is almost fixed in some Central African populations, whereas it is virtually absent from populations originating from other parts of the world (Livingstone 1984;Hamblin and Di Rienzo 2000;Hamblin et al. 2002). Plasmodium vivax, or another pathogen using the same mode of entry into host cells, has been identified as the most probable source of selective pressure in this case (Hamblin and Di Rienzo 2000;Hamblin et al. 2002). The selective pressures imposed by malaria can also act at the micro-geographic scale and be variable among closely related human populations. The case of the Fulani from West Africa, who present a specific resistance to malaria with respect to other ethnic groups living in the same area, is a prime example (Modiano et al. 1999).

Multiple mutations underlying similar adaptive phenotypes
It is interesting to note that many of the best-studied cases of genetic adaptation in humans have revealed multiple, independent mutations that confer an advantage to the same or similar selective pressures. Again, the lactase persistence trait not only constitutes one of the best documented cases of positive selection in the human genome but also it represents a clear example of convergent adaptation in response to a strong selective force related to diet. Indeed, different lactase-persistence alleles are found in Europe, the Middle East, and Africa, where they have increased in frequency independently, as a result of strong positive selection (Bersaglieri et al. 2004;Tishkoff et al. 2007;Enattah et al. 2008;Itan et al. 2010). Even within a given population (e.g., East African pastoralists), multiple mutations have been found to explain the lactase-persistence trait. Furthermore, the estimated times at which these mutations began to increase in frequency differ slightly between Europe and Africa (Bersaglieri et al. 2004;Tishkoff et al. 2007), and coincide with the cultural practice of pastoralism (and therefore milk consumption) in these populations. A possible explanation for this fascinating example of convergent adaptation within our species is that mutation rates to adaptive alleles might be high enough to allow novel adaptive mutations to arise in distinct geographic regions before any single variant spreads globally ).
Other cases of multiple mutations in a given gene, or even mutations located in different genes, underlying similar adaptive phenotypes in humans include resistance to malaria, adaptation to high altitude, or skin pigmentation. Indeed, various G6PD-deficiency alleles have been independently targeted by positive selection in African, Mediterranean, or South-East Asian populations, as they conferred higher resistance to falciparum or vivax malaria (Tishkoff et al. 2001;Saunders et al. 2002;Louicharoen et al. 2009). Likewise, different genes have been found to explain adaptation to high altitude in Tibetans (EPAS1) or Andeans (PRKAA1, NOS2A), and for the only gene targeted by positive selection in both populations, EGLN1, the selected haplotypes markedly differ between the two populations (Bigham et al. 2010;Yi et al. 2010). This clearly suggests that Tibetans and Andeans have followed different evolutionary trajectories to adapt to the same pressure, high altitude. Finally, the case of light skin in Europeans and East Asians provides with another example of convergent evolution (Norton et al. 2007). Indeed, different genes, associated with lighter skin color in these populations, have been independently targeted by positive selection in Europeans (SLC24A5, MATP, and TYR) and East Asians (ADAM17 and ATRN) (Lamason et al. 2005;Izagirre et al. 2006;Soejima et al. 2006;Myles et al. 2007;Norton et al. 2007;Pickrell et al. 2009). Although the role of sexual selection cannot be ruled out, these data support independent genetic mechanisms for evolution of skin color.

The special case of cancer selection
The relationship between selection and cancer remains blurry (Leroi et al. 2003). Some cancers affect mostly children, such as acute lymphoblastic leukemia (Ribera and Oriol 2009), and so they may select for anti-cancer adaptations that reduce the chance of death. Furthermore, although most cancers occur after the reproductive age, the variance is large, extended to much younger ages corresponding to a non-negligible fraction of individuals (Couch and Weber 1996). With this in mind, the possibility that cancers exert a direct selection pressure cannot be ruled out. In this context, theoretical considerations predict that the risk of developing cancer for large and long-lived species should be higher, in comparison with short-lived organisms, but this does not seem to be the case (Caulin and Maley 2011). This observation, known as the Peto's paradox (Peto et al. 1975), has been used to suggest that large animals might have developed some mechanisms to resist to cancer (Leroi et al. 2003;Nagy et al. 2007;Caulin and Maley 2011), supporting the notion that cancer may have been indeed evolutionary counter-selected.
The long-held view that cancers do not generally affect individuals of reproductive age has constituted a real obstacle to the use of the evolutionary genetics approach in this field. As a consequence, only a few studies have addressed directly the question of how selection has affected the evolution of cancer-related genes. Among them, the case of BRCA1 is worth mentioning, as this gene displays one of the strongest, replicated associations with breast and ovarian cancers (for a review, see (Lee and Boyer 2001)). Interestingly, BRCA1 has been found to evolve under purifying selection in humans (Pavard and Metcalf 2007). With respect to the related BRCA2 gene, which is strongly associated with the same diseases, it also appears to be evolving under purifying selection . These observations are particularly interesting, as hereditary forms of breast cancer have a high impact in young adults (Hall et al. 1990). We might therefore expect genes associated with cancers with a strong impact on young adults or children to be highly constrained by purifying selection.
Signatures of positive selection have also been reported for certain genes known to play a direct or indirect role in cancer risk. Some of these studies focused on signatures of selection between species, to highlight adaptive selection in humans, as a whole, with respect to other species. For example, at the inter-species level, BRCA1 appears to evolve adaptively, with some specific codons in its DNA repair domain being targeted by positive selection (Huttley et al. 2000;Fleming et al. 2003;Pavlicek et al. 2004). This finding and the above-mentioned signature of purifying selection in humans (Pavard and Metcalf 2007) are not incompatible but instead they suggest that BRCA1 evolved adaptively before human speciation (accumulating various amino-acid changes), subsequently becoming strongly constrained within the human lineage. At the intra-species level, some studies, which have made use of polymorphic data in humans, have found signatures of positive selection at some genes. For example, a genomic scan of 142 genes for signals of recent positive selection identified the PPP2R5E gene, which is involved in the negative regulation of cell growth and division (Voight et al. 2006;Grochola et al. 2009). Another selection scan of 132 genes found strong signatures of positive or balancing selection in a region extending over more than 115 kb in European-Americans, the strongest signals being centered on TRPV6 (Akey et al. 2004;Stajich and Hahn 2005). This gene has been shown to be up-regulated in prostate cancer and associated to an aggressive form of the disease (Wissenbach et al. 2001;Paiss et al. 2003). One of the most interesting examples is probably that of UGT2B4, which has been strongly associated with increased risk of breast cancer in Nigerians and, to a lesser extent, African Americans, and signatures of selection (balancing selection or recent positive selection) have been found in the upstream portion of the gene (Sun et al. 2011). Altogether, these examples suggest that there might be a direct or indirect relationship between cancer and selection, but the overall genome-wide proportion of cancer-related genes actually targeted by selection remains to be determined.

Maladaptation as a consequence of past selection
There is increasing evidence to suggest that some diseases of modern societies, such as obesity, hypertension, inflammatory or autoimmune diseases, allergies or even cancers, may simply be a by-product of past adaptation to other selective forces and changes in lifestyle. Genetic modifications occur more slowly than changes in lifestyle. It has therefore long been suggested that ancient selection events may have favoured variants that are no longer advantageous and may even have become detrimental in modern societies.

Changes in nutritional and pathogenic pressures over time
The 'thrifty gene hypothesis' was first introduced by James Neel (Neel 1962), who suggested that genes conferring a predisposition to diabetes (called 'thrifty genes') were historically advantageous but had become detrimental in the modern world. Some variants had been positively selected in the past because they favored the accumulation of larger amounts of fat, greatly increasing the likelihood of survival between famines. However, the change to a sedentary lifestyle and the increase in food abundance has increased the risk of developing type II diabetes in individuals carrying these variants today. Several studies have revealed particularly high risks of diabetes and high levels of obesity in populations that have recently rapidly switched to a 'Western lifestyle', such as the Native Americans of the United States (Joffe and Zimmet 1998).
Another example of maladaptation may reflect changes in the selective pressures imposed by infectious diseases over time Sironi and Clerici 2010). Pathogens are still present, but improvements in hygiene and the use of antibiotics and vaccines have greatly weakened the selection pressures they impose, particularly in Western countries. A strong, exacerbated immune response, which may have been the best way to survive in pathogen-rich environments in the past, seems to have become a burden in modern societies, as an overly vigorous response increases the risk of developing inflammatory and autoimmune diseases (Le Souef et al. 2000;Sironi and Clerici 2010) Indeed, populations with a long-term tropical ancestry introduced into temperate regions appear to have a higher risk of developing allergic inflammatory diseases and asthma, as shown for African Americans and for Asians living in the United Kingdom (Gillam et al. 1989;Gold et al. 1993;Von Behren et al. 1999). In all these cases, ethnic background was found to have a greater impact than environment on differences in the prevalence of asthma (Gilthorpe et al. 1998;Miller 2000). From a population genetics standpoint, some alleles conferring a higher risk of inflammatory and autoimmune diseases have been shown to have been under strong selective pressure in the past . This observation supports the notion that the higher risk of developing inflammatory and autoimmune disorders may be a by-product of past selection in response to infectious disease (Sironi and Clerici 2010).

Compromises during life
Another compromise that has been suggested for adaptation is antagonistic pleiotropy. Pleiotropy is defined as the situation in which one gene controls more than one phenotypic trait in an organism; 'antagonistic pleiotropy' is the term used when at least one trait is beneficial and another is detrimental. The antagonistic pleiotropy hypothesis was firstly proposed by George C. Williams in 1957 to explain senescence (Williams 1957). Indeed, variants that were advantageous early in life could become detrimental later on (Stearns et al. 2010), as genetic variation may affect several pathways at different ages. For example, high levels of testosterone in the bloodstream are associated with a greater fitness in early life, but are associated with a higher risk of developing prostate cancer at later stages (Gann et al. 1996). Another interesting case is birth weight: it has been suggested that selection for high birth weight, which is beneficial for survival early in life, is counterbalanced by a higher risk of various cancers later in life (Thomas et al. 2012).

Conclusion
Population genetic studies have greatly improved our knowledge of the way in which humans have genetically adapted to variation in environmental pressures and changes in lifestyle over time. Likewise, the investigation of how natural selection has driven the evolution of particular genes and biological functions has proven a useful tool to inform the relationship between genetic diversity, adaptive phenotypes and disease. In this context, detailed studies combining population genetic approaches making use of the massive whole-genome sequence-based data sets for various human populations, such as the 1000 Genomes Project (1000Genomes Project Consortium 2010Abecasis et al. 2012), and genome-wide association studies of multiple diseases and other traits of interest are now required. In particular, much more work is needed with respect to the increasingly recognized role of copy number variation in human adaptation and disease phenotypes, as rapid gains and losses of genomic segments can have a substantial impact on phenotypic variation (Feuk et al. 2006;Girirajan et al. 2011;Iskow et al. 2012). Similarly, we still have much to learn from the integration of data from population genetics, epigenetics and epidemiological genetics studies in populations with different lifestyles and modes of subsistence (e.g., agriculture, hunter-gathering, sedentary, nomadic) or living in different environments (e.g., urban, rural, forest). Such multidisciplinary efforts are clearly required to clarify the relationship between natural selection and disease and to improve our understanding of the evolutionary mechanisms accounting for the present-day differences in disease susceptibility, resistance or progression observed. Ultimately, these integrative approaches are likely to be essential for dissection of the contribution of genotypic, epigenetic, and environmental variables to the current risk of many diseases, facilitating improvements in their diagnosis, prevention and treatment.