What kind of selection will favor an intermediate phenotype?

2.2 Single binding site detection with sequence conservation

TF binding sites are often under purifying selection as shown by cross-species comparison of ChIP-derived or curated TF binding sites (Birney et al., 2007; Eisen, 2007; He et al., 2011b; Roy et al., 2010; Sandmann et al., 2006; Zeitlinger et al., 2007). Consequently, PWM predictions can be filtered by phylogenetic footprinting (Tagle et al., 1988; Wasserman et al., 2000), retaining a PWM match only if the TF binding site is conserved (Fig. 5.1B). Note that conserved binding sites are often the strongest sites (Håndstad et al., 2011) and weak sites may be missed (although a PWM approach will favor strong sites anyway, when stringent thresholds are applied). Also, species-specific sites or sites under positive selection (He et al., 2011a) will also be missed. Deciding whether a site is conserved or not can be done arbitrarily, for example requiring that the orthologous sequences of the site, as determined by pairwise or multiple alignments, need to be the same across a number of species, or that the corresponding position in the alignment is also a positive hit for the PWM. Examples of available software tools that perform simple PWM scanning across species are ConSite (Sandelin et al., 2004) and rVista (Loots and Ovcharenko, 2004), which are mostly used to predict conserved PWM instances in a promoter of a particular gene under study. A similar approach is used by the program TFLOC (Transcription Factor binding site LOCater) that identifies PWM matches on multiz multiple alignments (Fujita et al., 2010). Candidate-binding sites are scored by the sum of normalized log-odds scores of each species, and a threshold is determined empirically from the distribution of maximal scores across all 5-kb upstream regions. Hence, genome-wide scores and conservation are used to determine the highest confidence sites. The predictions of TFLOC for all TRANSFAC PWMs are available as a track in the UCSC Genome Browser (Fujita et al., 2010). A reverse strategy can also be used, starting with the detection of sequence conservation, potentially followed by matching conserved DNA words with PWMs or consensus sequences, as done, for example, in EvoPrinter (Odenwald et al., 2005) using multiple alignments; in FootPrinter (Blanchette and Tompa, 2003) using parsimony scores in the phylogenetic tree; and in rMonkey (Moses et al., 2006) using an evolutionary model. Additional confidence in conserved motifs can be gained by comparing the conservation of a particular motif or k-mer across the genome with random expectations. This strategy is used in combination with the branch length score (BLS) of the motif along the phylogenetic tree throughout several applications from the Kellis laboratory (Kellis et al., 2003; Stark et al., 2007; Xie et al., 2005). In these studies, the BLS score is combined with statistics based on genome-wide occurrences of conserved sites compared to random motif shuffles. When a particular consensus site or PWM has significantly more conserved matches in the genome than expected by chance (e.g., using shuffled motifs), genome-wide motif prediction for those motifs is feasible, although the search space is usually still limited to proximal regions to achieve acceptable false positive rates (Kheradpour et al., 2007). The target gene predictions for highly conserved motifs for Drosophila, detected using the BLS and confidence scores and then matched with known consensus sites or PWMs, can be found online (Kheradpour and Stark). When a new PWM is determined for a TF, it can be matched with the k-mers of this motif collection to check whether it has favorable conservation properties; or alternatively the BLS scoring procedure, including PWM shuffles, has to be applied to the PWM under study.

Importantly, the turnover of TFBS can be high, hence even a “functionally conserved” site may actually not be found at the same position in the alignment (Ludwig et al., 2000). Yet, they often reside within conserved regions (see further below). To also capture “moved” sites within the same CRM, some approaches allow for motif movement, such as the BLS implementation and—by definition—the network-level conservation approach. Tavazoie and colleagues used such an approach to identify conserved motifs on a genomic scale, requiring a motif to be present in orthologous promoters, but not necessarily aligned (Elemento and Tavazoie, 2005; Pritsker et al., 2004). This kind of binding site conservation, being present in the same region but not necessarily aligned, is also called binding site “preservation” (Berman et al., 2004). For a further overview of conservation and divergence of CRMs, we refer to a recent review (Meireles-Filho and Stark, 2009). In conclusion, the use of binding site conservation or preservation may be useful for proximal promoters, yet additional cues are often necessary to further increase the specificity of TF binding site and target gene predictions.

8.3.2 What Are the Forces That Reduce Genetic Variation?

There are several types of natural selection. Purifying selection is the most prevalent form of selection as it constantly sweeps away deleterious mutations that are produced in each generation. It is what keeps us fit. However, once in a while a mutation may arise that increases the fitness of the individual. These selectively advantageous alleles can replace other alleles and become “fixed” in the population (i.e., reach the frequency of 1) through directional selection. Strong directional selection, such as frequent pesticide application, may result in recurrent bottlenecks so that the population contains only the variation present in a small, surviving subpopulation. Therefore directional selection, in theory, has the effect of reducing the diversity of alleles and therefore the genetic variation in a population.

But genetic variation can also be decreased because of chance alone, through a process known as genetic drift. Genetic drift is the change in allele frequencies from generation to generation that occurs in finite-sized populations due to the random sampling of gametes (containing particular alleles) that will constitute the zygotes of the next generation. Its effect is more pronounced in smaller populations and inevitably leads to the fixation of a particular allele (and the loss of others). Inbreeding also increases as populations get smaller, further decreasing population variability. Genetic drift can be distinguished from selection because the whole genome is affected, not just a particular locus. Migration is a counteracting force to genetic drift. By mixing alleles among populations, migration distributes and homogenizes genetic variation among populations.

IV THE ORGANISM AND THE CELL

Long open reading frames are statistically improbable and will not persist without purifying selection. However, the protein-binding motifs seen in enhancers and promoters of genes are not complex, and new patterns of expression may arise through random mutations finding sequences capable of binding cell-type-specific transcription factors. Jenkins et al. (1995) and Ludwig and Kreitman (1995), studying evolutionary changes in enhancers of the fushi tarazu and even-skipped pair rule genes of Drosophila, found protein-binding sites in these enhancers to be surprisingly labile evolutionarily, suggesting either that the genes are evolving subtly different new expression patterns or that they have true redundancy in their control.

Since selection operates at the level of the organism, the effect of a mutation on fitness is the result of its combined effects in all tissues. It may be that the only way to express a protein in all cells where it is needed requires its expression in some other cells as well. Redundant gene expression is likely. Gene duplications may allow more precise control of the expression of functions. The myogenic genes myf-5 and myogenin are not redundant but are expressed in different tissues. Does the lack of genetic redundancy imply a lack of functional redundancy in the proteins? The effect on the rib cage of a myf-5‒ mutation was examined using a gene disruption in which the coding sequence for myogenin was integrated into the myf-5 locus in place of the coding sequence, thereby causing expression of Myogenin in cells that would normally express Myf-5 (Yang et al., 1996). The resulting mice were wild type, indicating functional redundancy of Myogenin and Myf-5 in the rib cage.

4.2 Initiating trans change with detrimental fitness effects

If network co-option is deleterious, the initiating trans change should be lost due to purifying selection (Fig. 3B, ii) unless it is fixed by drift, which is more likely in small populations and when the fitness consequence is mild (Fisher, 1930). It is also possible that the phenotypic consequences of a given co-option event on the novel tissue were initially beneficial or neutral, and only later became detrimental (e.g., accompanying a change in environment that alters selective regime or developmental plasticity, epistatic changes that reveal larger effects on phenotype, etc.). In such cases, the upstream mutation may have already been fixed in the population. In either of the above cases, the detrimental effects of network co-option could either be eliminated by another change at the upstream trans factor that reverts the co-option, or the effects could be reduced over time by evolving tissue specific repression of downstream network nodes individually. Any given case of this latter process would be indistinguishable from an initial state of partial or aphenotypic co-option, although in some cases a comparison across species or populations that diverged after the initiating trans change might reveal a history of modifications deactivating the co-opted network.

20.5.2 Innate immune genes

Unlike the MHC, most avian innate immune genes appear to be single copy genes. Despite predominant purifying selection and general structural conservatism, in animals, several innate immune genes exhibit significant levels of interspecific as well as intraspecific variation and appear to be under positive selection [288]. This is especially true for the pattern recognition receptors (PRRs) which trigger inflammatory responses, broadly effective immune mechanisms that carry high costs in terms of resource allocation and host tissue damage [52]. For a range of avian pathogens it has been shown that innate immune cells from pathogen-resistant birds respond more readily to infectious challenge than those from susceptible individuals, which suggests differences in properties of their molecular repertoires [289].

Toll-like receptors (TLRs) represent the most well studied PRRs in birds. In its ancestral state the avian TLR family consists of ten receptors differentially adapted to distinct ligand binding [290]. Interspecific variation is observed in some avian taxa in duplication of TLR7 (several clades within Passeriformes, Charadriiformes, Cuculiformes and Mesiornithiformes [290–293]) and pseudogenisation of TLR5 (several clades within Passeriformes, Psittaciformes, Cariamiformes, Trogoniformes, Phaethontiformes, Eurypygiformes, and Apodiformes [290,294]). Despite gene conversion homogenizing certain regions between some TLR paralogues (e.g., in TLR1 or TLR2 [290]), strong positive selection has been repeatedly detected on interspecific level in avian TLRs. This site-specific diversifying selection appears to be stronger in the TLRs, that face the extracellular space and bind complex ligands (TLR10 [TLR1A], TLR1 [TLR1B], TLR2A, TLR2B, TLR4 and TLR5), than in TLR15 and endosomal TLRs (TLR3, TLR7, TLR21) [290].

While generally conserved across specific pathogen types, the immunogenic microbial structures (named microbe-associated molecular patterns, MAMPs) recognized by TLRs show significant structural variability [295,296]. This may create a strong selective pressure on TLRs to adapt to specific ligand variants. Indeed, despite their much conserved general functionality [297], differences linked to disease resistance have been shown even between TLR alleles originating from the same species. In birds, this has been documented for chickens [298], where the existing genetic variation between breeds much exceeds human variation [299]. Similar levels of intraspecific polymorphism were also observed for TLRs in some free-living birds [300], including the lesser kestrel (Falco naumanni: Falconiformes) and the house finch, but not in all, as evidence mainly from inbred populations [292,301–303]. In Attwater’s prairie-chicken it has been shown that several parameters of immune function (lysozyme activity, hemolysis) and survival are related to certain TLR alleles [271]. High TLR allelic variation can result from balancing selection [304]. Thus the pattern of adaptation through high genetic variation resembles the above-mentioned variation in MHC involved in the adaptive immunity. However, unlike the adaptive immune response which can take as long as 1 to 2 weeks to clear an infection [74], innate immunity is a rapid, ready-to-act defense mechanism. Thus balancing selection in these genes could provide advantage in early immunological decisions, directing subsequent responses and preventing later immunopathology.

The finite number of MAMP structural types seemingly corresponds with the frequent pattern of convergent evolution observed in avian TLRs [290]. Patterns of positive selection were mainly reported in extracellular domains being located either in or topologically close to ligand-binding sites reported earlier for model species [305,306]. These adaptations do not appear to affect the molecular shape of the ligand-binding region, but they alter its surface properties, namely the charge distribution [306,307]. In birds, only about 4.5% of codons experience episodic positive selection across all TLR loci [216], yet this variation may significantly affect the molecular phenotype. Evolutionary analysis of the TLR4 ligand-binding region in a sample of 55 passerine species from two different geographical regions indicated phylogenetic patterns in the charge clustering, but its association to any of the tested ecological characteristics remains unsupported [307]. This is consistent with Fahrenholz’s rule, in which cospeciation between hosts and microbes [308] results in their mirrored phylogenies. If this is the case, parsing signals of phylogeny from relevant environmental pressures (namely pathogen communities) can prove particularly difficult.

PRR other than TLRs, such as Nucleotide-binding oligomerization domain (NOD)-like receptors (NLRs) and RIG-I-like receptors (RLRs), remain poorly studied in birds other than poultry. In one study, melanoma differentiation-associated gene 5 (MDA5), a nonself RNA-sensing RLR, showed low variability in the helicase domain across 62 avian species [309]. Although the pattern observed in the retinoic acid-inducible gene I (RIG-I) was different from MDA5, the sites under positive selection were located far from the RNA-binding sites. This suggests general ligand-recognition conservatism in avian RLRs, accompanied by selection on only minor adjustments in pathogen responsiveness. Interestingly, increased positive selection acting on the galliform laboratory of genetics and physiology 2 (LGP2) gene, a third RLR that otherwise exhibits few signs of positive selection in birds, might correspond to the loss of RIG-I in Galliformes [309]. Further evidence then suggests conservation of the downstream molecular machinery in birds including the Mx genes ([310], but see [311]). Genetic diversity of avian tetherins that are also effector molecules involved in antiviral defense remains unknown [312].

Compared to PRRs, immune signaling molecules, such as cytokines appear to evolve mostly under negative selection [212], with little evidence for directional selection [305]. This confers that these signal mediators form the structural conservatism required for their universal functionality. Despite this, some genes of pleiotropic cytokines, such as IL1B [313], and some cytokine receptors, such as interleukin 4 receptor alpha-chain (IL4R alpha) [314], have been reported to show signals of balancing selection in galliform species.

Finally, avian defense peptides and proteins (known also as antimicrobial proteins) exhibit mostly negative selection, with occasional evidence for balanced diversity, namely among β-defensins (AvBDs) and cathelicidins [315,316]. Despite this, site- and branch-specific episodic diversifying selection has been detected in such genes, mainly in AvBD3, AvBD7, and AvBD10 [315]. This selection appears to affect mainly residues close to conserved cysteines forming the characteristic β sheet-rich fold. Among 44 waterfowl species, Chapman and colleagues [317] described remarkably low putative functional diversity for five AvBD genes. Here radical changes associated with balancing selection were observable only in the duplicated AvBD3. Balanced polymorphisms have also been revealed among passerine AvBDs, despite their otherwise conservative nature [318]. Functionally-relevant polymorphism has been confirmed in AvBD7 in great tits, as in vitro testing revealed differential inhibitory effects of two AvBD7 variants on gram-positive bacterium Staphylococcus aureus [319]. This pattern probably reflects differences in the molecular charge and hydrophilicity of the AvBD variants [214]. Several sites under positive selection were found also in lysozyme studies across chicken populations [320].

Despite recent progress in examining the diversity of avian innate immune genes, the list of such genes with potential ecological and evolutionary implications could certainly be expanded much further. Since virtually nothing is known about the functional diversity of most of these genes in birds, wildlife immunogenetics clearly needs to expand its scope to arrive at a more comprehensive understanding of the evolution of avian disease susceptibility and resistance. Future investigation should also target noncoding regions involved in gene expression regulation, such as promotor regions and untranslated regions in exons.

12.23 Human Gene Variants Associated with Mesoderm Specification Defects

What kinds of genetic mutations cause these defects? Theoretical considerations lead to the following two predictions.

Due to evolutionary purifying selection of essential genes, the frequency and structural severity of heritable core mesoderm gene (CMG) variants should be diminished relative to other viable genes.

Given the roles of such CMGs as “master switches” in early development, large effects from hypomorphic variants are also expected.

These two predictions receive statistical support from a bioinformatic comparison of the mutational spectrum and human disease association of human genes orthologous to essential mouse genes vs. human genes orthologous to non-essential mouse genes (Park et al., 2008). Continuing with our Nodal/T-centric perspective on mesoderm formation, some disease associations will now be considered for these and related genes. In the course of this, some general issues relevant to the study of CMG variants in humans will be discussed.

2.4.3 Natural Selection

Natural variations exist among the individuals in any population. Many of these differences do not affect survival or reproductive fitness (e.g. the eye color variations in humans), but some differences may improve the chances of survival of a particular group of individuals. Natural selection results in the fixation of these advantageous variations in the population, leading to greater adaptability to and reproductive success in the environment. Thus, natural selection drives the evolutionary engine.

Natural selection can be of two types, based on its effect on the fate of genetic variations: purifying (negative) selection and positive (Darwinian) selection. Purifying selection removes deleterious variations, whereas positive selection fixes beneficial variations in the population and promotes the emergence of new phenotypes. As a result, natural selection acts on populations to determine the allele frequency and distribution of quantitative traitsn over generations. The principal types of selection determining the distribution of traits across a population are directional, stabilizing, disruptive, and balancing selection.

Directional selection favors the advantageous allele so that its proportion (and the associated phenotype) increases in the population. As a result, both the allele frequency and the phenotype are skewed in one direction and away from the average phenotype (Figure 2.5A). A popular example is the phenomenon of industrial melanism in the peppered moth (Biston betularia). This species has both light- and dark-colored phenotypes. Before the industrial revolution in England, the light-colored phenotype was predominant. During the industrial revolution, the trees on which the peppered moths rested were blackened by soot. The darker background gave the dark-colored moths an advantage in hiding from predatory birds and at the same time made the light-colored moth more visible and prone to predation. As a result, over time the dark-colored moths proliferated and became the predominant phenotype while the light-colored moth population was significantly reduced. Through regulation and legislation, the environment started clearing up. As a result, the balance between light-colored and dark-colored varieties was reversed and the light-colored variety proliferated again.

Stabilizing selection is known to be the most prevalent type of natural selection; it favors the intermediate (average) phenotype of the trait, and in doing so it removes the extreme phenotypes of the trait from the population (Figure 2.5B). Thus, stabilizing selection reduces genetic variability in the population. It is generally accepted that stabilizing selection maintains the DNA and protein sequences over evolutionary time. However, Kimura64 demonstrated that under stabilizing selection, extensive neutral evolution can occur through random genetic drift. In other words, many cryptic neutral genetic changes may occur in natural populations while maintaining the phenotype unchanged. A common example of stabilizing selection is the mortality and birth weight in human babies. It is well known that both very large and very small human babies suffer high mortality rates; hence, the intermediate weight is the most favored phenotype for survival.

Disruptive selection (diversifying selection) favors the two extreme phenotypes of the trait and minimizes the average phenotype. Thus, disruptive selection creates a bimodal distribution of a trait in the population; consequently, it is the opposite of stabilizing selection in the outcome (Figure 2.5C). Disruptive selection is an important driving force behind sympatric speciationo. An example of disruptive selection is provided by the mimicry and survival of the African butterfly Pseudacraea eurytus. In this species, the coloration ranges from reddish yellow to blue, with some intermediate colors. The extreme colors mimic other butterflies that are not normally preyed upon by the local predatory birds. In contrast, butterflies with intermediate coloration are devoured by the predators in greater numbers. Therefore, butterflies with extreme coloration survive in greater proportion compared to those with intermediate coloration. Another example of disruptive selection is the selection of the two extreme trophic phenotypes in the spadefoot toad (Spea multiplicata). Using a mark-recapture experiment in a natural pond, Martin and Pfennig65 showed that the spadefoot toad can have different trophic phenotypes depending on the resource availability. However, disruptive selection favors the two extreme phenotypes, the small-headed “omnivore phenotype,” which feeds mostly on detritus, and a large-headed “carnivorous” phenotype, which feeds on and whose phenotype is induced by the fairy shrimp. By foraging more effectively on the two alternative resource types, these extreme phenotypes avoid competition for food resources and are favored by disruptive selection, whereas the intermediate phenotypes are reduced in number.

Balancing selection (balanced polymorphism) maintains polymorphism in the population with respect to an allele of a trait. Therefore, balancing selection maintains genetic diversity in the population. A classic example of balancing selection is the heterozygote advantage in areas in Africa with high incidence of malaria. Sickle cell anemia reduces life expectancy and is caused if an individual is homozygous for a variant of hemoglobin (HbS/HbS). A red blood cell (RBC) containing HbS becomes sickle-shaped and is extremely sensitive to oxygen deprivation. However, the malarial parasite Plasmodium cannot survive in such sickle-shaped RBCs. Thus, heterozygous individuals, containing one normal copy and one variant copy of the hemoglobin gene (HbA/HbS), are at a survival advantage in areas with high incidence of malaria. In contrast, individuals homozygous for normal hemoglobin (HbA/HbA) are at an increased risk of death by malaria. Thus, selection maintains the apparently deleterious HbS allelic variant in the population, and balances between strong selection against both HbA/HbA and HbS/HbS genotypes by providing a selective advantage to the HbA/HbS genotype.

Based on the scale of changes, selection can lead to microevolution and macroevolution. Microevolution means small changes in the genome and is also associated with changes in gene frequency in a population. Over time, the accumulated small changes collectively can be significant enough to create certain new traits so that the group possessing those traits could be assigned an infra-species category, such as a subspecies or variety under the original species. In contrast, macroevolution means evolutionary changes leading up to the formation of species or higher taxa. The mechanisms for both micro- and macroevolutionary processes are generally the same.

Detecting Selection With Samples Over Time

Natural selection is often expected to cause changes over time in the frequency of an allele, particularly when dealing with positive selection on a newly arisen beneficial mutation. For balancing and negative selection, natural selection may cause allele frequencies to be unusually stable over time and resistant to changes induced by other evolutionary forces, such as genetic drift and mutation. In both situations, natural selection leads to outliers with respect to allele frequency changes over time. Statistical techniques for detecting and estimating selection from data gathered over multiple generations through time have long been used in experimental population genetics, particularly with organisms with short generation times such as Drosophila (e.g., Templeton, 1974). Such multigeneration approaches traditionally have had limited applicability in humans. One method was to identify populations with deep pedigrees and to use that information coupled with genetic surveys on current individuals to infer the fate of genetic variants over time and the role of selection on shaping those fates. A recent example of this approach is the work of Peischl et al. (2018) on French Canadians, a population that colonized Quebec in the 17th century followed by a large expansion in population size and occupied territory. They discovered a pattern of “relaxed selection” on the expansion front with more deleterious variation in the front than in the core population that represented the original settlement. This pattern arose in just 6–9 generations and was consistent with greater genetic drift effects on the expanding front. Recall that Eq. (4.9) shows that drift interacts with a growing population size to enhance the survival of selectively deleterious mutants, and Fig. 10.1 shows that deleterious alleles have an increased chance of fixation when genetic drift is strong. Although Peischl et al. (2018) called this pattern “relaxed selection,” there is no evidence in this case of the selection coefficients becoming smaller in magnitude. This situation is better described as nonrelaxed selection interacting with enhanced genetic drift coupled with the effects of population growth at the wave front of the expansion. As emphasized throughout this book, evolutionary properties emerge from the relative strengths of several evolutionary forces, and by increasing the role of drift and population growth at the wave front, the evolutionary impact of natural selection is diminished even though selection itself is not relaxed. This pattern of genetic variation being more influenced by genetic drift and population growth at the population's wave front has also been called genetic surfing (Peischl et al., 2016), and the term “relaxed selection” should be avoided unless there is actual evidence that the selection coefficients have diminished in magnitude over time.

Few human populations have extensive and deep pedigrees, making the approach discussed above of limited applicability. However, the advances in studying ancient DNA (Chapter 7) have made samples across time more and more feasible, thereby allowing the use of methods for time serial samples in humans (Malaspinas, 2016). For example, Mathieson et al. (2015) assembled genome-wide data on 230 ancient individuals from western Eurasia dated between 6500 and 300 BCE. This was a very dynamic period in European history characterized by the transition to agriculture and the admixture of ancient populations. By contrasting the ancient DNA with modern European genome data, they inferred that most present-day Europeans can be regarded as a mixture of three ancient populations: western hunter-gatherers, early European farmers, and steppe pastoralists. Hence for a neutral allele, we would expect that the current allele frequency should be close to a linear combination of the allele frequencies of these three ancient populations (Eq. 6.22). This linear predicted allele frequency constituted their null hypothesis under selective neutrality. They then scanned the modern European genome database to identify allele frequencies that were significant outliers from the predicted allele frequency under the null hypothesis of neutrality. They only looked for large outliers, which means they could only detect positive selection in their scan. They found 12 genes with multiple SNP outliers above the genome-wide significance threshold (Fig. 10.12). Their strongest signal was for the allele causing lactase persistence in the gene LCT in Europe, which shows a large increase in frequency in the last 4000 years in Europe. As will be discussed in more detail in Chapter 12, this allele has been under strong selection when humans began to use cattle for milk as a food resource. Three other genes under significant selection in Fig. 10.12 are also related to diet: FADS1-2, associated with plasma lipid and fatty acid concentration; DHCR7, associated with vitamin D levels that is a strong selective force in higher latitudes (Chapter 12); and SLC22A4, hypothesized to have experienced a selective sweep to protect against ergothioneine deficiency in agricultural diets. Two other signals of allele frequency change driven by selection are in the pigment genes SLC45A2 and HERC2, which are also associated with vitamin D selection in higher latitudes (Chapter 12). Two other regions are known targets of selection: TLR1-6-10, that appears related to resistance to leprosy, tuberculosis, or other mycobacteria; and MHC, that is also related to pathogen resistance as noted earlier.

Mathieson et al. (2015) also examined selection on a quantitative trait, height, over time in Europeans. Recall that Turchin et al. (2012) found significant selection on height between two contemporary European populations by contrasting the frequencies of alleles between the two populations, with the alleles assigned phenotypic effects on height (Chapter 8). Mathieson et al. (2015) performed a similar contrast between modern and ancient allele frequencies. However, the phenotypic effects of alleles on height could only be estimated in the modern population. Recall that the average excess and average effects are functions of allele frequencies and that changes in allele frequencies can change both their magnitude and sign, as demonstrated in Chapter 9. Hence, the analysis with ancient DNA depends on the assumption that the allele frequencies have not changed much over time such that the current average effects or excesses have for the most part not changed their sign over the time period between the ancients and the moderns. Given this assumption, they detected significant selection for both increased and decreased height in several comparisons between ancient and modern European populations, as well as between pairs of ancient populations.

Pharmacogenetics is in its Infancy

A general observation in the field of complex diseases is that most common genetic variants only confer a modest effect on risk. This stands to reason, in that strongly deleterious mutations would have been subject to purifying selection and kept at very low frequencies in the population. It is therefore hoped that extending the bounds of allele frequencies to uncommon variants, through next-generation sequencing techniques applied to extreme cases or varied populations, will lead to the identification of rarer variants that have stronger effects. However, while selection pressure may have been acting for a long time (in evolutionary terms) on certain phenotypes, modern pharmacology is relatively recent. Thus, there may not have been enough time for genetic determinants of drug response to undergo purifying selection, and therefore genetic effects on drug response may be stronger than those seen for disease pathogenesis (Link et al., 2008; Mega et al., 2008; Shuldiner et al., 2009).

In T2D, pharmacogenetic investigation remains at a very early stage. In retrospective studies conducted in the GoDARTS cohort, for which clinical outcomes are electronically available and DNA samples have been collected, carriers of the risk variant at TCF7L2 showed diminished response to sulfonylurea therapy (Pearson et al., 2007) and were more likely to require insulin therapy (Kimber et al., 2007). This finding suggests that the impairment on insulin secretion conferred by TCF7L2 cannot be overcome with an insulin secretagogue, and stands in contrast to that seen for the β-cell sulfonylurea receptor/potassium channel, where carriers of the risk alleles at KCNJ11/ABCC8 displayed improved response to the specific sulfonylurea drug (gliclazide) that has shown an in vitro allelic effect (Feng et al., 2008; Hamming et al., 2009).

An early report that a missense polymorphism in the OCT1 metformin transporter (which is responsible for the absorption of the drug into hepatocytes) alters response to metformin (Shu et al., 2007) has not been substantiated in the GoDARTS study (Zhou et al., 2009). On the other hand, a polymorphism in a different metformin transporter (MATE1), which catalyzes the disposition of drug into bile and urine, seemed to affect metformin response in a different small retrospective study (Becker et al., 2009); this has been confirmed in the Diabetes Prevention Program, where ~1000 participants at high risk of diabetes who received metformin for approximately three years for diabetes prevention showed little benefit from metformin treatment only if they were homozygous for the risk allele (Jablonski et al., 2010).

It is hoped that the integration of clinical datasets in which pharmacotherapeutic information is available will allow for the deployment of the same genome-wide methods that have enabled the discovery of disease genes, but this time in a search for genetic determinants of drug response. Such an undertaking would move the field from the current “trial and error” or “one size fits all” mentality in T2D therapeutics to a setting where drug choices will be predicated on individual characteristics. A GWAS for metformin response has been recently completed (cite PMID 21186350).

Pharmacogenetics is in Its Infancy

A general observation in the field of complex diseases is that most common genetic variants only confer a modest effect on risk. This stands to reason, in that strongly deleterious mutations would have been subject to purifying selection and kept at very low frequencies in the population. However, while selection pressure may have been acting for a long time (in evolutionary terms) on certain phenotypes, modern pharmacology is relatively recent. Thus, there may not have been enough time for genetic determinants of drug response to undergo purifying selection, and therefore, genetic effects on drug response may be stronger than those seen for disease pathogenesis [172–174].

In T2D, pharmacogenetic investigation remains at a very early stage. In retrospective studies conducted in the GoDARTS cohort, for which clinical outcomes are electronically available and DNA samples have been collected, carriers of the risk variant at TCF7L2 showed diminished response to sulfonylurea therapy [175] and were more likely to require insulin therapy [176]. This finding suggests that the impairment on insulin secretion conferred by TCF7L2 cannot be overcome with an insulin secretagogue, and stands in contrast to that seen for the β-cell sulfonylurea receptor/potassium channel, where carriers of the risk alleles at KCNJ11/ABCC8 displayed improved response to the specific sulfonylurea drug (gliclazide) that has shown an in vitro allelic effect [123,124].

An early report that a missense polymorphism in the OCT1 metformin transporter (which is responsible for the absorption of the drug into hepatocytes) alters response to metformin [177] has not been substantiated in GoDARTS [178]. On the other hand, a polymorphism in a different metformin transporter (MATE1), which catalyzes the disposition of drug into bile and urine, seemed to affect metformin response in a different small retrospective study [179]; this has been confirmed in the Diabetes Prevention Program, where ~1000 partici­pants at high risk of diabetes who received metformin for approximately 3 years for diabetes prevention showed little benefit from metformin treatment only if they were homozygous for the risk allele [180]. A recent syste­matic review of the literature by Maruthur et al. has highlighted that additional high-quality controlled studies are needed to identify robust pharmaco­genetic results. Nevertheless, the following drugs and loci were noted worthy of additional follow-up: (1) metformin with SLC22A1, SLC22A2, SLC47A1, PRKAB2, PRKAA2, PRKAA1, and STK11; (2) sulfonylureas with CYP2C9 and TCF7L2; (3) repaglinide with KCNJ11, SLC30A8, NEUROD1/BETA2, UCP2, and PAX4; (4) pioglitazone with PPARG2 and PTPRD; (5) rosiglitazone with KCNQ1 and RBP4; and (5) acarbose with PPARA, HNF4A, LIPC, and PPARGC1A [181]. It is hoped that the integration of clinical datasets where pharmacotherapeutic information is available will allow for the deployment of the same genome-wide methods that have enabled the discovery of disease genes, but this time in a search for genetic determinants of drug response. Such an undertaking would move the field from the current “trial and error” or “one size fits all” mentality in T2D therapeutics to a setting where drug choices will be predicated on individual characteristics.