The Genetic Basis of Chronic Mountain Sickness

Abstract

Chronic mountain sickness (CMS) is a disease that affects many high-altitude dwellers, particularly in the Andean Mountains in South America. The hallmark symptom of CMS is polycythemia, which causes increased risk of pulmonary hypertension and stroke (among other symptoms). A prevailing hypothesis in high-altitude medicine is that CMS results from a population-specific “maladaptation” to the hypoxic conditions at high altitude. In contrast, the prevalence of CMS is very low in other high-altitude populations (e.g., Tibetans and Ethiopians), which are seemingly well adapted to hypoxia. In recent years, concurrent with the advent of genomic technologies, several studies have investigated the genetic basis of adaptation to altitude. These studies have identified several candidate genes that may underlie the adaptation, or maladaptation. Interestingly, some of these genes are targeted by known drugs, raising the possibility of new treatments for CMS and other ischemic diseases. We review recent discoveries, alongside the methodologies used to obtain them, and outline some of the challenges remaining in the field.

More than 140 million people around the globe reside at high altitude (≥3,000 m) in locations such as the Ethiopian Highlands in Africa, the Himalaya Mountains in Asia, and the Andes Mountain Range in South America. Although there is no doubt that the elevation in these regions represents stressful environmental conditions, chiefly due to low environmental O₂, adaptation of high-altitude dwellers has varied qualitatively and quantitatively. For example, a higher hemoglobin concentration and lower oxygen saturation have been observed in Andean highlanders compared with Tibetans or Ethiopians at similar altitude (3, 5). Furthermore, infants of Tibetans have a higher birth weight and arterial O₂ saturation compared with infants of other populations (Han Chinese) at similar altitudes (3). In fact, a sizeable percentage of individuals in these populations (as much as 16% in certain regions, especially males) are maladapted and are thus threatened by the low levels of inspired O₂ to this day (31). The primary manifestation of this maladaptation to high altitudes is chronic mountain sickness (CMS) or Monge's disease, first described by Carlos Monge in the Andes in 1925 (29). It is characterized by polycythemia (hematocrit > 65%) and hypoxemia (O₂ < 85%), both of which improve upon descent from altitude. The most frequent symptoms and signs of CMS are headache, dizziness, breathlessness, palpitations, sleep disturbance, mental fatigue, and confusion (30). People affected by CMS often suffer from stroke and myocardial infarction in early adulthood, mostly due to increased blood viscosity and tissue hypoxia. Why some are affected by this disease and not others has so far been a mystery. Differences in adaptation patterns between the high-altitude populations suggest that there are, at least in part, distinct genetic mechanisms underlying the adaptations.

Despite this genetic basis having been proposed for many years, it is only in the past few years that our understanding of human adaptation to high altitude has accelerated (1, 6, 22, 34, 37, 39, 42, 46, 47, 51). These rapid discoveries have mostly resulted from the advent of genomic technologies, particularly deep sequencing, as well as the concurrent developments in computational genetics. Understanding the molecular basis of high-altitude adaptation and maladaptation will also provide us with a unique handle on genes that are important for human health and disease. This is particularly true of conditions where oxygen deprivation plays a major etiological role, such as arterial pulmonary hypertension, myocardial ischemia and injury, stroke, and polycythemia (to name a few). It is possible that learning about regions of the human genome that evolved over many thousands of years to allow adaptation to hypoxia will lead to a better understanding of adaptation in humans subjected to other stresses.

In this review, we highlight several recent studies (including ours) dealing with the genetic underpinnings of high-altitude adaptation or mal-adaptation. CMS is a maladaptation to high altitude, and any understanding of CMS will likely shed light on relevant genetic and physiological mechanisms. We evaluate the methods used in previous studies and the results obtained. In addition, we highlight some of the remaining open questions in the field.

A Genomic Approach to Understanding Chronic Mountain Sickness

Adaptation to hypoxia (as well as the diagnosis of CMS) is often measured using related phenotypes, such as blood oxygen saturation, Hb, or hematocrit levels (5, 30). From available data, we can surmise that the different populations all adapted separately: their phenotypic values show distinct inherited traits. It is possible that a founder effect may have been at work when the populations first migrated, and standing genetic variation in the founders was subjected to selective constraints leading to this variable adaptation. Although most high-altitude populations in the world are well-adapted, in some regions (e.g., the Andes), 10–20% of the male population is threatened by the CMS syndrome (31). Such populations provide us with a unique opportunity to study the nature of adaptation by contrasting the genetics of well adapted individuals and those with CMS (51). CMS is believed to arise, at least partly, due to an excessive production of red blood cells (RBC). By increasing RBC production, humans attempt to mitigate the effect of low environmental oxygen via increased oxygen carrying capacity in the blood. Hence, CMS is considered an adaptation to life under chronic hypoxia at altitude. Indeed, sea-level-dwelling humans visiting high altitudes show a similar, albeit more modest, response. However, a long-term consequence of this increase in RBC production is increased blood viscosity, leading to vascular sludging and increasing the likelihood of vascular occlusion, stroke, myocardial ischemia, and infarcts in early adulthood. It also results in uneven blood flow through the lungs, increasing the ventilation-perfusion mismatch and leading to hypoxemia, contrary effects to what is desired. In trying to adapt, the organism has effectively responded in a way that is adverse to its survival and well being (mal-adaptation) at high altitude (51). Therefore, the genetic basis for CMS can be investigated in the context of adaptation to hypoxic environments at high altitudes, and our survey keeps that perspective. FIGURE 1A describes a generic strategy that is implicit in many highlander studies. The first steps of this strategy, correlating genotypes and phenotypes and/or scanning the genome for signatures of natural selection, reveal candidate genes. Due to many factors that may confound this step, candidate genes must then be validated using various approaches. Our review focuses on these two steps, where current studies suggest a complex, multigenic adaptation.

FIGURE 1.

Proposed workflow for hypoxia-related therapeutics and schematic genealogical tree

A: a proposed workflow for hypoxia related therapeutics, starting with genetic samples and ending with candidate therapeutic targets. B: schematic genealogical tree illustrating the evolution of a non-recombining genomic fragment across three populations, one of which migrates to high altitude (HA population) and undergoes genetic adaptation, whereas the others remain at low altitudes (LA and Outgroup populations). The bottom of the tree (leaves) represents individuals sampled from the current generation, whereas the upper sections reflect the past genealogy. In the HA population, hypoxia imposes positive natural selection on the beneficial allele (blue star), increasing its frequency (in the non-CMS group) at the expense of individuals carrying the maladapted allele (CMS). As long as phenotypic variation persists in the adaptive trait (e.g., Hb levels are still variable in the HA population, meaning the selective sweep is ongoing), genetic association may find variants associated with the trait. However, after the trait reaches fixation or given small effect sizes and/or smaller cohorts, genome-wide association (GWA) is unlikely to reveal the adaptive genes. Neutrality tests can be used to pinpoint genomic regions under selection in both settings (i.e., pre- and postfixation, and given a smaller sample). These tests utilize properties of the genealogical tree. The LSBL/PBS tests approximate the branch length leading to the MRCA of the HA population, which is unusually high in regions under selection (see long branch with blue SNPs). Tajima's π uses the mean allelic heterogeneity, which is unusually low in regions under selection (since HA individuals are genetically similar given their relatively recent MRCA). The iHS/EHH tests use haplotype homozygosity, which is unusually high and spans longer regions under selection (most variation in HA individuals, shown as SNPs on the path from MRCA to the present HA individuals, is common to the entire HA sample). Common practice is to genotype a population sample, followed by imputation from a nearby, and densely sequenced, reference population (e.g., the LA population). Because imputation relies on conserved linkage disequilibrium (LD) between target and reference populations, and LD is strongly altered by selective sweeps, imputation will be inaccurate in regions evolving under strong selection. This further illustrates the importance of WGS. MCRA, most recent common ancestor.

Genome-Wide Association Studies and Disease Traits

Association tests measure correlation between genotype segregation and a phenotype, and can be applied to any of the many hypoxia-related phenotypes (e.g., Hb levels). Different studies have explored this with different designs involving the choice of phenotype, population (mainly Tibetan, Ethiopian, and Andean highlanders), genotyping technology, and statistical methods. It was recognized early on that there are population-specific differences in highlander phenotypes. For instance, Tibetan highlanders have lower Hb levels but also lower O₂ saturation levels compared with Andeans (3). Hanaoka et al. (18) showed that serum erythropoietin (EPO) levels in Sherpas at 3,440 m was equal to that in non-Sherpas at a much lower altitude, indicating that Sherpas have a resistance response to EPO levels. Moore (30) and Beall (4, 5, 7) review many relevant traits across highlander populations, including hematocrit and hemoglobin levels, O₂ saturation, arterial O₂ content, ventilatory response to hypoxia, exhaled NO, and pulmonary vasoconstriction. In our context, individuals with and without CMS can be used in a case-control association study, with the caveat that there may be significant population substructure [see Figure S3 of Zhou et al. (51)]. Most association studies to date have focused on Hb levels due to ease of measurement and its correlation with other traits of interest. In a typical genome-wide association study (GWAS), the association between sampled genotypes and a specific trait is measured in case-control cohorts using statistical tests (see Table 1). The merits of different statistical tests have been previously reviewed (23, 45) and will not be discussed here.

Table 1.

Genes identified via association with a hypoxia-tolerance phenotype

Study	Phenotype	Population(s)	Assay	Gene Candidates
Simonson et al. (39)	Hb	Tibet (n = 31)	Genotyping array (Affymetrix 6.0)	EPAS1, PPARA
Yi et al. (47)	Hb, erythrocyte	Tibet (n = 50)	Exome sequence (NimbleGen)	EPAS1
Beall et al. (6)	Hb	Tibet (n = 70)Tibet (n = 91)	Genotyping array(Illumina 0.5M)	EPAS1
Scheinfeldt et al. (37)	Hb	Ethiopia (n = 42)	Genotyping array(Illumina 1M)	ARNT2, THRB
Alkorta-Aranburu et al. (1)	Hb, O2 saturation	Ethiopia (n = 260)Amhara (102H, 60L)Oromo (63H, 35L)	Genotyping array (Illumina)	RORA*COL6A1*SLC30A9* HGF*

^*Not genome-wide significant after multiple testing correction. H and L, high- and low-altitude individuals, respectively.

In the domain of genomic technologies, we now have a multitude of options, including candidate gene sequencing, SNP genotyping arrays, whole exome sequencing (WES), and whole genome sequencing (WGS). Each of these options has difficult trade-offs, arguably making this the single most important decision in the study-design process. For example, genotyping arrays sample a large corpus (up to several million) of genomic loci but suffer from a serious problem of ascertainment bias [recently summarized by Lachance and Tishkoff (24) and illustrated schematically in FIGURE 1B]. Normally, conserved haplotype structure between populations implies that, even if only a small collection of SNPs is sampled (e.g., by array), most alleles can be inferred from a previously sequenced reference population (10). Although generally accurate, this strategy (genotype imputation) may fail in genomic regions affected by strong selection (blue dots in FIGURE 1B), since the haplotype structure is highly sensitive to selective sweeps and is expected to diverge in such regions. Although some [e.g., Yi et al. (47)] have used whole exome sequencing to overcome such issues, this technology does not sample the vast noncoding portions of the genome, including regulatory and many noncoding RNA regions. Indeed, one of the most important sites discovered to date was in an intron of the EPAS1 gene not specifically targeted by the exon array.

A second problem, unrelated to the technology domain, is the reduced power stemming from the multigenic response to hypoxia. As with many other complex traits, most published GWA studies to date have failed to achieve genome-wide significance after multiple testing correction. Finally, adaptive alleles fix in the population, reducing phenotypic variation in their respective traits, and consequently reducing the power of association testing. Therefore, researchers have focused on candidate genes with known physiology (mainly HIF pathway genes; see Table 1). Simonson et al. (39) used a short list of five regions and zoomed in on EPAS1. Yi et al. (47) focused on genes annotated by the ontology term “response to hypoxia.” Beall et al. (6) looked carefully at sites around the EPAS1 gene and validated them through independent cohorts. Scheinfeldt et al. (37) and Alkorta-Aranburu et al. (1) studied Ethiopian highlanders and reported many interesting candidate genes; however, their results do not achieve genome-wide significance after Bonferroni correction for any SNPs located near genes. Yet, they validated through secondary means and identified a number of novel genes, suggesting a very different adaptive response in Ethiopians compared with Tibetan highlanders. In summary, association tests (particularly with ascertained SNPs) must be applied to larger populations, validated on independent cohorts, or focused on a reduced set of candidate regions. A viable approach to identifying an unbiased list of candidate genes is through searching for genomic signatures of natural selection, discussed next.

Natural Selection and Disease Pathogenesis

Given the strong selective constraints stemming from low environmental oxygen, identifying genetic signatures of natural selection in highlander populations provides us with an alternative approach to genotype/phenotype association for candidate gene detection. FIGURE 1B provides a schematic of the evolutionary history of a short (non-recombining) chromosomal segment under positive selection. The bottom (“leaves”) of the evolutionary tree corresponds to the genomic region in extant individuals of different subpopulations, whereas the top (“root”) of the tree represents the region in the most recent common ancestor (MRCA). Mutations (green or blue circles) on a specific lineage are inherited by all its descendants. The highlander (HA) migration exerts a selective constraint. Consequently, individuals carrying a beneficial allele (blue star) rise rapidly in frequency in the population, outcompeting other individuals. These individuals 1) present the non-CMS phenotype; 2) have a recent common ancestor, with all other non-CMS individuals sharing a longer than average branch of common mutations (i.e., blue circles); and 3) have had limited time to differentiate (i.e., few mutation and recombination events), leading to a lack of allelic diversity and long homogenous haplotypes in the region. Statistical tests capturing these characteristics have been used to identify regions under selection (Table 2). For example, LSBL/PBS (38, 47) and F_ST (21) were used by Yi et al. (47), Bigham et al. (9), Alkorta-Aranburu et al. (1), Zhou et al. (51), and Udpa et al. (42) to approximate the branch length leading to MRCA. Zhou et al. (51) and Udpa et al. (42) used multiple tests to measure allelic diversity (or lack thereof). Simonson et al. (39) and Scheinfeldt et al. (37) used the iHS test (36, 44) to compute the decay of haplotype similarity. These tests may be confounded by recent admixture of populations, which can also lead to lack of allelic diversity in some cases. Recently, an admixture-correction was proposed by Huerta-Sanchez et al. (22) to refine the analysis.

Table 2.

Genes identified via tests of selection

Study	Population(s)	Assay	Gene Candidates
Bigham et al.* (9)	Tibet (n = 49) Andes (n = 49)	Genotyping array (Affymetrix 6.0)	Tibet: EGLN1, EPAS1 Andes: EGLN1, TH, NOS2A, PRKAA1
Simonson et al. (39)	Tibet (n = 31)	Genotyping array (Affymetrix 6.0)	EPAS1, EGLN1, CYP2E1, EDNRA, ANGPTL4, CAMK2D, HMOX2, CYP17A1, PPARA, PTEN
Yi et al.* (47)	Tibet (n = 50)	Exome sequence (NimbleGen)	EPAS1, HBB, HBG2, FANCA, PKLR, HFE
Beall et al. (6)	Tibet (n = 35)	Illumina Quad (0.5M)	EPAS1
Xu et al. (46)	Tibet (n = 46)	Genotyping array (Affymetrix 6.0)	EPAS1, EGLN1
Peng et al.* (34)	Tibet (n = 1334)	Genotyping array (Affymetrix 6.0)	EPAS1, EGLN1
Scheinfeldt et al.* (37)	Ethiopia Amhara HA (n = 28) Aari/Hamer LA (n = 19)	Genotyping array (Illumina 1M)	CBARA1, ARHGAP15, RNF216, SYNJ2, NAT2, AIMP1, VAV3, ARNT2, THRB
Alkorta-Aranburu et al.* (1)	Ethiopia (n = 260) Amhara (102H, 60L) Oromo (63H, 35L)	Genotyping array (Illumina)	CUL3, ADRBK1, CORO1B, ASF1B, MAPKAPK2, ADH6, SLC30A9, TMEM33
Huerta-Sanchez et al.* (22)	Ethiopia Tigray-Amhara (n = 47) Oromo (n =21)	Genotyping array (Illumina Omni 1M)	BHLHE41, CASP1, SMURF2
Zhou et al. (51)	Andes CMS (n = 10) non-CMS (n = 10)	Whole genome sequencing (Illumina)	SENP1, ANP32D
Udpa et al. (42)	Ethiopia Amhara (n = 7) Oromo (n = 6)	Whole genome sequencing (Illumina)	CIC, LIPE, PAFAH1B3, EDNRB

^*Complete gene set not listed due to space limitations.

Taken together, these studies reveal the complexity of understanding hypoxia adaptation. On one hand, small sample sizes, low effect, and reduced phenotypic diversity (after fixation of a beneficial allele) make it difficult to achieve genome-wide significance in association studies. Indeed, with the exception of the EPAS1 genes in Tibetans, few genes have been identified with genome-wide significance. On the other hand, recent population admixture and complex demographic histories may confound tests of selection. We believe that, with further refinement, both methodologies will likely yield additional insights.

Another major issue is the choice of genetic assaying technology. Genotype arrays are designed to exploit haplotype blocks by directly assaying only select SNPs from the existing variation (so-called “tag SNPs”), and then using imputation (10) to fill in missing genotypes. Yet, Udpa et al. (42) and Zhou et al. (51) demonstrated that highlander populations may have a different haplotype structure specifically in areas under selection, and that many regions with strong signature of selection would be missed by genotype arrays or exome sequencing (FIGURE 2). Instead, these studies used WGS from small population samples of highlanders. In a study of Andean highlanders with CMS and non-CMS phenotypes, Zhou et al. (51) identified 11 regions genome-wide with significant haplotype frequency differentials between the CMS and non-CMS individuals, which are consistent with selective sweeps. Two distinct regions contained genes that had fly orthologs and could be validated in a model organism system (an erythropoiesis regulator, SENP1, and an oncogene, ANP32D). These studies illustrate the potential of whole genome sequencing in identifying the genetic basis of CMS and long-term hypoxia adaptation in general. Although the studies by Zhou et al. (51) and Udpa et al. (42) illustrate the advantages of WGS, it is currently too expensive for use on large cohorts. Therefore, a cost-effective design may be created using tests of selection on WGS of a smaller cohort, followed by genotyping and association tests on a larger cohort. The identified genes would be candidates for secondary validation, followed by functional testing.

FIGURE 2.

The effects of sequence assay on genome-wide scans for selection

A and B: test statistic values on chromosome 19, when taking into account all variants discovered by WGS (A) or only the subset found in a common ∼1M SNP genotyping array (B) (1% FDR computed separately based on the genome-wide distribution of test statistic values). Highlighted in green is 1 of the 11 significant peaks reported in our laboratory's study (51), which does not exceed the 1% FDR using only genotype data. C and D: SNP frequency profiles of the highlighted (green) region in non-CMS (blue) compared with MXL (brown, inverted) showing all variants from WGS (C) or only the subset present in genotyping (D). WGS reveals many variants in the region, allowing a robust estimate of the allele frequency distribution, whereas genotyping detects only a handful of alleles, making inference of adaptive evolution difficult. However, genotyping studies in large populations can be used to validate the frequency differences obtained via WGS of smaller cohorts. Figure adapted from Zhou et al. (51), with permission from Elsevier.

From Association to Causative Genes and Pathways

Independent Cohorts

Most highlander studies to date are not powered to achieve genome-wide significance for association, even when large population cohorts have been used [e.g., Peng et al. (34)]. Therefore, many studies seek to replicate their findings on independent cohorts from the same population. Good examples of this are the EPAS1 and EGLN1 genes, which have been shown repeatedly as important for hypoxia adaptation in Tibetans, pointing to the essential role of the HIF pathway. Interestingly, these genes appear to play a less substantial role in Andeans and have not been observed in Ethiopians (see Tables 1 and 2). In general, even genes that appear repeatedly in multiple highlander cohorts require further investigation to elucidate their specific biological role.

Model Organisms

To validate the effects of genes found via genomic scans for association or selection, animal models may be of great value (42, 51). In the context of CMS, the idea is to determine the functional impact of genetic variants identified as potentially significant in both affected (CMS) and unaffected (non-CMS) individuals. Drosophila melanogaster provides a powerful in vivo model to dissect the genetic mechanisms that contribute to human disease (8, 15, 33), including aging (17, 28), neurological and cardiac disease (13, 25, 27), cancer (35, 43), and the mechanisms underlying hypoxia tolerance or susceptibility (2, 50, 52). Several genes obtained from human studies of high-altitude adaptation (and maladaptation) have orthologs in the Drosophila genome and could thus be tested for effects on hypoxia tolerance. For example, Zhou et al. (51) observed a dramatic increase in survival when the expression of candidate genes SENP1 and ANP32D (obtained from genomic tests of selection) was reduced in flies under hypoxic conditions, indicating a likely role for these genes in human adaptation to high altitude (see FIGURE 3). Indeed, SENP1 is known to regulate erythropoiesis, and Senp1^−/− mice die of anemia in early life. If CMS pathogenesis is even partially caused by abnormal polycythemia, then SENP1 may be of prime importance, potentially linking erythropoiesis to the pathogenesis of CMS.

FIGURE 3.

Experimental validation in a model system of candidate genes for human high-altitude adaptation

Downregulation of human SENP1 and ANP32D orthologs in Drosophila enhances survival under hypoxia. The da-Gal4 driver was used to ubiquitously knock down the individual candidate genes by crossing with respective UAS-RNAi lines. Eclosion rates were then measured at 21% and 5% O₂. A: significant increase in eclosion rate under 5% O₂ in three RNAi lines targeting the same human SENP1 ortholog (CG32110). B: the differences in eclosion rates were also significant in the two lines targeting the human ANP32D ortholog (Mapmodulin). Each bar represents mean 5 SE of eclosion rate. The w1118 and da-Gal4 stocks were tested and used as background controls. Figure adapted from Zhou et al. (51), with permission from Elsevier.

Although to the best of our knowledge the model organism work done by Zhou et al. (51) and Udpa et al. (42) represents the only such work to validate results from studies of genetic variation in high-altitude human populations, other models have also been used in studying acute or chronic hypoxic stress (20, 40, 49). Zebrafish and C. elegans have been used, mostly to dissect the genetic basis of response to hypoxic stress and genetic predisposition. These model systems could be very useful to corroborate findings from human studies where cellular and molecular studies are difficult to perform.

In Vitro Models

Another possible approach for validation of candidate genes obtained from statistical tests of association or selection is to study the relevant phenotypic effects in vitro. Zhou et al. (51) determined the expression levels of candidate genes in fibroblasts (obtained from skin biopsies) placed under decreasing levels of O₂. Furthermore, a similar approach can be applied to reprogrammed iPS cells. This can be very useful toward replicating the disease in a dish, as has been done before in other cases (11). As shown by Zhou et al. (51), such in vitro models are able to capture aspects of CMS. In this way, we may better understand the effect of genetic variants on phenotype in the condition when the disease is manifested.

Functional Networks

Naively, one might expect that similar adaptive genes should surface among populations experiencing similar selective stresses. To some extent this is true, as is the case for EGLN1 that exhibits a strong signature of selection in both the Andean and the Tibetan populations (see Table 2). Yet, a striking observation from Tables 1 and 2 is the relative lack of overlap in candidate adaptive genes among the different populations. Partially, this may result from technical limitations in the respective studies (meaning the true overlap may be greater than currently observed). Nevertheless, we believe the small overlap stems chiefly from the structure and connectivity of the underlying genetic networks. Differently put, although different genes are involved across different populations, it is plausible that similar mechanisms and/or pathways are at play. For instance, different loss of function (LOF) mutations in critical genes in a pathway, or different mutations disrupting regulatory sites of such genes, may suffice to mediate the chronic hypoxia response.

In FIGURE 4, we show a genetic network constructed by GeneMania (32) from the genes reported in the studies appearing in Tables 1 and 2. The network includes connections representing physical (protein-protein) interactions as well as known pathways. GeneMania reports a statistically significant enrichment of several relevant biological processes, including response to hypoxia (FDR = 2.28 × 10⁻⁶), blood circulation (FDR = 2.06 × 10⁻³), endothelial cell proliferation (FDR = 4.76 × 10⁻³), and response to oxidative stress (FDR = 9.98 × 10⁻³). In addition, we note that to a certain extent the network segregates into components that correspond to underlying physiological processes. Importantly, genes observed across different populations are often present in the same component, further supporting the hypothesis of similarity at the process level.

FIGURE 4.

GeneMania (32) network constructed from candidate genes for adaptation to hypoxia

The network contains two types of edges: physical interaction (red) and known pathways (blue), and includes genes from Tables 1 and 2 (green, blue, and yellow circles), and additional genes with direct connections (gray circles). Genes with no connecting edges are not shown. Genes are shaded according to the geographical region in which they were identified. Note that many genes from the hypoxia response pathway are directly implicated in multiple populations. The hypoxia response directly affects metabolism. Specifically, the transcription factor HIF1A also upregulates Angiopoietin-like protein 4 (26), which in turn regulates the PPAR-dependent expression of LIPE. Other genes impacted by hypoxia involve the vascular system, such as the vasoconstrictor EDNRB, and MAP kinase 2, which influences pulmonary vascular permeability (12). The Fanconi anemia complex genes [which also complex with Spectrin (SPTA)] are key members of a DNA repair pathway that are regulated by hypoxic stress. In addition, the FANCG gene interacts with cytochrome P450 protein CYP2E1 (14). Together, the studies demonstrate the complex, multi-locus adaptation to hypoxia achieved by different populations.

Future Directions

Although there is little doubt that genetic factors underlie human adaptation to high altitude, there is a paucity of investigations into the role of epigenetics in these adaptations. There is reason to suspect that the harsh environmental stress at high altitudes may cause changes in DNA methylation or histone modification. Indeed, a study by Hartley et al. (19) showed experimentally that major epigenetic changes occur in a culture system using primary neurons. Moreover, preliminary experiments from our laboratory have recently demonstrated that even a short period of hypoxia induces many changes in DNA methylation that last for weeks after exposure (Zhou D, Haddad GG, unpublished observations).

Another important area is proving causation when genes and pathways are identified as candidates. Zhou et al. (42, 51) recently used fibroblasts from skin biopsies in conjunction with reprogramming strategies to study the role of candidate genes in hypoxia tolerance. In addition, Zhau et al. (48) studied signaling pathways that may underlie causation in human (in vitro) models of adaptation and maladaptation to hypoxia.

The accelerated rate of discovery using newer technologies and the potential for rapid functional validation using reprogrammed iPSC underscore the opportunities in genetics-based approaches. We stress that the methodology is not limited to hypoxia tolerance and can be applied to any condition caused by evolutionary constraints, including adaptation to parasitic infections, and (through a study of somatic variation in individual cells) to cancer progression. In this domain, we may therefore be reaching the stage for studies that aim at understanding the cell physiology and biology of adaptation.

Finally, it is important to highlight here that a number of available drugs target some of the gene products reported by the studies on high-altitude adaptation. For example, PPAR(alpha), EDNRB, EPAS1, and EGLN1, among others, are targeted by existing drugs or compounds (16). It remains an intriguing possibility that some of these drugs could direct new treatments for CMS and other ischemic diseases.