Genomic Evidence for Natural Selection Underlying High-Altitude Adaptive Hemoglobin Levels Among Peruvian Andeans

Abstract

High-altitude hypoxia presents an extreme environmental pressure that challenges human survival, growth, and reproduction. Despite this challenge, humans have thrived on the Andean Altiplano for millennia, displaying several unique physiological responses to hypoxia such as elevated hemoglobin concentration ([Hb]). This trait closely resembles the acclimatization response observed among high-altitude sojourners but is distinct from the sea-level normative [Hb] that characterizes the Tibetan adaptive response. As recent candidate-gene efforts to understand the role of natural selection in shaping Andean [Hb] have produced conflicting results, it remains unclear what role natural selection may have played in shaping this unique hematopoietic response. Using genome-wide array data from Peruvian Andeans, we identified two genomic regions containing three genes, PDE1B, PPP1R1A, and RASGEF1B, that show evidence of recent positive selection and are associated with [Hb]. Importantly, Andean alleles within these regions are associated with lowered [Hb], suggesting that recent polygenic selection may be acting to reduce [Hb] within this population. We observe the greatest divergence of Andean allele frequencies from other global populations within the PDE1B/PPP1R1A region and use WGS data and publicly available expression and Hi-C data to more closely identify how natural selection may be acting within this region to impact [Hb]. We identify a selective sweep that favors 11 PDE1B expression-decreasing alleles and is located at the boundary of a topologically associating domain spanning several hemoglobin-linked genes. In sum, this study provides novel evidence that polygenic natural selection may be acting to lower Andean [Hb] in a manner phenotypically convergent with Tibetan populations.

Significance.

Long-resident human populations of the Andes display elevated hemoglobin concentration compared to sea-level values. This is in contrast to high-altitude Tibetan populations who exhibit a blunted hematopoietic response to hypoxia. Much is known concerning the evolutionary genetic history of Tibetan hemoglobin concentration, but among Andeans, there has been limited genomic investigation. Here, we show that polygenic natural selection may be operating among Andeans to lower hemoglobin concentration. One gene in particular, PDE1B, contains eQTLs that affect PDE1B expression as well as variation that may impact the expression of nearby hematopoiesis-related genes. These findings suggest that ongoing polygenic natural selection is operating on Andean hemoglobin concentration in a manner phenotypically convergent with Tibetans.

Introduction

Humans evolved under the atmospheric conditions found at sea level with limited exposure to high-altitude hypoxia, or the low oxygen availability found at high elevations (>2,500 m). This reduced atmospheric oxygen limits aerobic metabolism, thereby challenging human growth, development, and reproduction (Beall 2006; Storz and Scott 2019). Nonetheless, humans have inhabited the Andean Altiplano of South America (average elevation of 3,750 m) for over 11,000 years (Rademaker et al. 2014). At this altitude, the partial pressure of oxygen is approximately 35% lower than at sea level. Andean populations have thrived at this elevation, displaying a unique suite of physiologically adaptive traits that help them overcome the challenges of high-altitude hypoxia. This suite includes traits such as increased lung volume, high exercise capacity (VO₂Max), and elevated hemoglobin concentration ([Hb]) (Frisancho et al. 1995; Brutsaert et al. 1999, 2003).

One trait of particular interest for high-altitude adaptation is hemoglobin concentration ([Hb]). Hemoglobin is the oxygen-carrying metalloprotein found in erythrocytes. It is responsible for the transport and delivery of oxygen to all bodily tissues. At low elevation, “normal” concentrations of hemoglobin are generally estimated to range from 14 to 18 g/dL in healthy male adults and 12 to 16 g/dL in healthy female adults (Blanc 1968; Pagana et al. 2019). Variation in [Hb] is attributable to multiple factors including, but not limited to, sex, age, genetic background, and environment (Whitfield et al. 1985; Evans et al. 1999; Yang et al. 2007; Timoteo et al. 2023). Lowland sojourners to high altitude show elevated [Hb] with acclimatization to high-altitude hypoxia (Lundby et al. 2004; Windsor and Rodway 2007; Cheung et al. 2014). At high elevation, reductions in atmospheric oxygen result in hypoxemia, a decreased partial pressure of blood O₂, and reduced tissue oxygenation. Acclimatization to this hypoxic stress induces several biological processes, including a hematopoietic response that results in increased [Hb]. This allows for increased blood O₂ carrying capacity and the maintenance of arterial oxygen content (Dijkhuizen et al. 1977; Rizvi et al. 2017). Andeans also display similarly elevated [Hb] at high altitudes, averaging ∼17 g/dL in females and 18 g/dL in males (León-Velarde et al. 2000; Beall 2007). However, increases in [Hb] can lead to polycythemia and chronic mountain sickness (CMS), a progressive condition characterized by excessive erythrocytosis, pulmonary hypertension, and blood hyperviscosity that occurs at highest frequency in the Andes (Monge et al. 1989; Danesh et al. 2000). Severe cases of CMS can result in cor pulmonale and congestive heart failure (Villafuerte and Corante 2016; Corante et al. 2018). In contrast, Tibetans, a second high-altitude indigenous human population, exhibit sea-level normative [Hb] at high altitude that likely has been the result of increased plasma volume rather than a blunted erythropoietic response (Beall and Reichsman 1984; Beall and Goldstein 1987; Beall et al. 1998; Stembridge et al. 2019). This lowered [Hb] observed among Tibetans prevents increases in blood viscosity, which, in conjunction with the increased ventilatory response observed among this population, allows for sufficient tissue oxygenation and improved exercise capacity, with a reduced risk of excessive erythrocytosis and related cardiovascular disease (Beall et al. 2010; Simonson et al. 2015; Stembridge et al. 2019).

Two key genes, EGLN1 and EPAS1, have been found to be under positive selection among Tibetans and involved in these adaptive physiological mechanisms (Beall et al. 2010; Bigham et al. 2010; Simonson et al. 2010). These genes are part of the hypoxia inducible factor (HIF) pathway, an evolutionary ancient signaling network that controls cellular responses to hypoxia by regulating erythropoiesis, angiogenesis, and anaerobic metabolism. Genetic variation within EPAS1 has been associated with the low [Hb] observed among Tibetans (Beall et al. 2010; Yi et al. 2010). Variation in EGLN1 also has been associated with the low Tibetan [Hb] (Simonson et al. 2010; Xiang et al. 2013), and functional investigation has identified a Tibetan EGLN1 (PHD2) loss-of-function allele involved with altered oxygen sensing (Simonson et al. 2010; Xiang et al. 2013; Song et al. 2014, 2020). However, several other studies have been unable to report similar associations between EGLN1 variation and [Hb] (Wuren et al. 2014; Tashi et al. 2017). Taken together, these studies indicate that positive selection within the HIF pathway has resulted in protective genetic adaptations that limit the erythropoietic response found in Tibetans, while maintaining adequate oxygen delivery to tissues.

Among Andeans, recent investigations of EGLN1 and EPAS1 have identified genetic associations with [Hb], with variation in EPAS1 associating with lowered [Hb] and variation in EGLN1 associating with elevated [Hb] (Yasukochi et al. 2020; Lawrence et al. 2024). Moreover, the haplotypes associated with lowered [Hb] among Tibetans are not found among Andeans (Bigham et al. 2013; Heinrich et al. 2019). The limited scope of candidate-gene studies such as these provide incomplete portraits of how natural selection may be acting on [Hb], as it remains unclear if natural selection is acting to lower or raise [Hb] concentration among Andeans. Furthermore, [Hb] is a polygenic trait with contributing variation from multiple genomic regions (Garner et al. 2000; Yang et al. 2007; Li et al. 2013). To date, there has been little exploration of the underlying genetic control of [Hb] among high-altitude populations outside of these two candidate regions, EGLN1 and EPAS1. Here, we performed a genome-wide investigation to examine how ongoing positive natural selection may be acting to shape [Hb] among Peruvian Andeans. Our results broaden our understanding of how natural selection has acted across the Andean genome to influence [Hb], highlighting the importance of examining complex, polygenic phenotypes in a genome-wide context.

Results

Peruvian Quechua Population Characteristics

We recruited 603 Peruvian study participants of high-altitude Andean ancestry who self-identified as Quechua using a migrant study design. This design included three groups who differed in their developmental and lifetime exposure to hypoxia: (i) high-altitude Quechua (HAQ) participants who were born, raised, and resided at high altitude (n = 301); (ii) migrant Quechua (MQ) participants who were born at high altitude (above 3,000 m), but down migrated to low altitude within their lifetime (n = 150); and (iii) low-altitude Quechua (LAQ) participants who were born, raised, and resided at low altitude (n = 152). LAQ are second-generation down-migrants to low altitude whose parents were born and raised at high altitude, but then migrated to low altitude Lima, Peru, during their lifetime. HAQ study participants were recruited in Cerro de Pasco, Peru (4,338 m). MQ and LAQ study participants were recruited in Lima, Peru (154 m). All Peruvian Quechua participants were healthy adults between the ages of 18 and 35 (mean age 24.6 ± 5.1; Table 1). [Hb] was significantly elevated among HAQ compared to sea-level values (LAQ + MQ) with an average of 19.2 ± 1.7 g/dL among HAQ males and 16.1 ± 1.4 g/dL among HAQ females compared to average concentrations of 14.9 ± 0.9 g/dL among low-altitude males and 12.6 ± 0.8 among low-altitude females (LAQ + MQ, P = 8.27e-13 among males, 9.16e-11 among females; Table 1).

Table 1.

Peruvian Quechua participant characteristics

	All (n = 603)	HAQ (n = 301)	LAQ (n = 152)	MQ (n = 150)	LAQ + MQ (n = 302)
Proportion female	0.50	0.48	0.50	0.53	0.52
Age (years)	24.59 ± 5.12	24.29 ± 4.99	24.45 ± 4.58b	25.48 ± 5.45	24.96 ± 5.05
Height (cm)	158.60 ± 10.56	158.17 ± 8.14	161.02 ± 8.96	158.09 ± 7.81	159.56 ± 8.52
Weight (kg)	61.67 ± 10.23	59.50 ± 8.20a,b,c	66.34 ± 11.77b	61.71 ± 9.49	64.04 ± 10.93
Hemoglobin (g/dL)
Males	17.09 ± 2.54 (16.81,17.38)	19.16 ± 1.70a,b,c (18.89,19.43)	14.91 ± 0.92 (14.70,15.12)	14.89 ± 0.96 (14.66,15.12)	14.90 ± 0.94 (14.75,15.05)
Females	14.32 ± 2.10 (14.08,14.56)	16.14 ± 1.43a,b,c (15.90,16.37)	12.61 ± 0.74 (12.44,12.77)	12.68 ± 0.92 (12.47,12.88)	12.64 ± 0.84 (12.51,12.77)

95% confidence intervals (CI) are in parenthesis. Sample sizes (n) are provided in the table header. Data are means ± SD.

HAQ, high-altitude Quechua; LAQ: low-altitude Quechua; MQ, migrant Quechua.

^a P ≤ 0.05 versus LAQ. ^bP ≤ 0.05 versus MQ. ^cP ≤ 0.05 versus LAQ + MQ.

Ancestry Estimates Among Peruvian Quechua

Genome-wide variant data were generated using the Affymetrix (Santa Clara, CA) Axiom Biobank Genotyping Array for all 603 study participants as well for 100 Mexican Maya speakers (Tzeltal, Tzotzil, and Ch’ol) recruited from Palenque, Chiapas, Mexico (197 m), representing a closely related low-altitude Indigenous American population. Related individuals at the first, second, and third degrees were removed, yielding a total of 573 unrelated Peruvian Quechua and 71 unrelated Mexican Maya participants (Manichaikul et al. 2010). Principal component analysis (PCA) (Purcell et al. 2007; Chang et al. 2015) revealed that Peruvian Quechua formed a cluster with Mexican Maya and overlapped closely with other Indigenous American populations, including Peruvian and Mexican Ancestry populations from the 1,000 Genomes Project (1KG) (1000 Genomes Project Consortium et al. 2015) (Fig. 1a). Admixed Peruvian and Mexican individuals showed increased dispersion toward European populations. ADMIXTURE analysis (Alexander and Lange 2011) revealed Indigenous American ancestry estimates that ranged from 99.99% to 53.29% among Peruvian Quechua and 99.99% to 52.04% among Mexican Maya. On average, Peruvian Quechua study participants were estimated to have 88.79% Indigenous American ancestry, 8.01% European ancestry, 1.03% African ancestry, and 1.29% Asian ancestry. Estimates among Mexican Maya individuals averaged 94.34% Indigenous American ancestry, 3.58% European ancestry, 1.31% African ancestry, and 0.42% East Asian ancestry (Fig. 1b). To limit potential confounding due to European admixture, a total of 334 Peruvian Quechua individuals (196 HAQ, 79 MQ, and 59 LAQ) and 57 Mexican Maya individuals with more than 90% Indigenous American ancestry were included in the selection scan.

Fig. 1.

Principal component and admixture analysis of Peruvian Quechua participants. a) PCA for all recruited Peruvian Quechua (PER) and lowland Mexican Maya (MEX) participants, including publicly available data from 1000 Genomes populations. PC1 and 2 account for 5.2% and 2.9% of total observed variance, respectively. The recruited PER and MEX cohorts (black squares and triangles) overlap with the Peruvian in Lima, Peru, and Mexican Maya Ancestry from Los Angeles populations (orange squares and triangles). b) ADMIXTURE individual bar plots (K = 5) results for all recruited participants as well as European (CEPH), East Asian (CHB), and African (YRI) reference population panels. Peruvian Quechua and Mexican Maya cohorts show individuals with inferred levels of European admixture ranging from 0% to 50%, averaging 8.01% of European admixture among Peruvian Quechua and 3.58% among Mexican Maya.

Peruvian Quechua Show Signatures of Natural Selection Throughout the Genome

We applied two statistics to test for evidence of positive directional selection in the Andean genome: Cross-Population Number of Segregating Sites per Length (XP-nSL) (Szpiech et al. 2021) and Population Branch Statistic (PBS). XP-nSL was calculated by comparing Peruvian Andeans to Mexican Maya across 440,343 individual variants that passed QC and had no missing data. XP-nSL scores ranged from −5.65 to 7.17. Positive XP-nSL values indicated positive selection among Peruvian Andeans. We identified 22,018 variants with XP-nSL values above 1.69 in the top 5% of the empirical distribution (Fig. 2a; Table S1). A region on chromosome 3 between 69,285,366 and 69,531,012 (hg38) contained the top seven significant XP-nSL results with the FRMD4B intronic variant rs13063380 (XP-nSL = 7.17) as the most significant SNP (Fig. 2a; Table S1). PBS was calculated for 122,842 variants with minor allele frequencies above 5% (MAF > 0.05) in each of the three populations used for comparison: Peruvian Andeans, Mexican Maya, and 1KG Han Chinese (CHB). We did not enforce a minor allele cutoff for the calculation of XP-nSL. Therefore, more variants passed to the calculation of XP-nSL than PBS. We identified 6,142 variants with PBS values above 0.08 in the top 5% of the empirical distribution (Fig. 2b; Table S2). The most extreme PBS value was for the variant rs4147578 (PBS = 0.50) located on chromosome 14 within the first intron of GSTZ1.

Fig. 2.

Genome-wide selection scan results from a) XP-nSL and b) PBS statistical tests. The genome-wide significance threshold (95th percentile) for each statistic is marked by the dashed lines. Significant SNPs in HIF-pathway genes previously identified in selection scans are shown in red and identified by gene name.

We mapped the significant variants onto genes including up and downstream windows of 5 and 2 kb, respectively. Our selection scan identified significant SNPs in several HIF target genes. Several of these genes previously have been found to be under selection among high-altitude populations, including EGLN1, HIF1A, NOS2, EGFR, ARID1B, and GAPDH. For EGLN1, we identified one intronic variant, rs2491403 (PBS = 0.14, rank = 1,162), and two downstream variants, rs2437150 (PBS = 0.12, rank = 2,007) and rs2064766 (PBS = 0.15, rank = 1,023), within the top 5% of our PBS empirical distribution. For HIF1A, we identified one exonic variant, rs149348765, significant for XP-nSL (XP-nSL = 3.60, rank = 1,243). NOS2 contained eight SNPs nominated by XP-nSL, of which one SNP was also nominated by PBS (rs6505483, PBS = 0.12, rank = 2,003, XP-nSL = 2.45, rank = 7,330). The intronic variant rs6965663 (PBS = 0.16, rank = 780) was significant for EGFR. ARID1B contained seven SNPs that were nominated by XP-nSL (XP-nSL = 1.75 to 1.97). Four variants annotated to GAPDH were identified within the top 5% of XP-nSL results (XP-nSL = 1.76 to 1.84). To mobilize these allele-based and haplotype-based statistics, we merged the significant results from each test to obtain a set of 24,823 unique variants nominated by either of the selection scan statistics. Of these selection-scan nominated loci, a total of 976 were identified by both statistics (Table S3).

Genomic Association Testing Reveals Three Genes With Evidence of Selection Influencing Hemoglobin Concentration

To determine whether natural selection has acted on the Andean genome to influence [Hb], we performed a genome-wide association study (GWAS) to test for the effects of SNP genotype on this phenotype. We limited our GWAS to selection-nominated variants (n = 24,823) identified in our selection scan and performed QC filtering and excluded variants within the HLA region for association testing. In total, we tested 11,456 variants for association with [Hb] among 507 Peruvian Quechua (273 HAQ, 110 LAQ, and 124 MQ) participants with estimated Indigenous American ancestry of 80% or more. This relaxed admixture threshold was implemented to improve statistical power for association testing. We performed a linear association test using an additive inheritance model that included sex and the first five genetic PCs as covariates. We included the additional covariate of recruitment altitude to control for the effect of acclimatization. Of the 11,456 variants tested, none passed FDR multiple testing correction. The top three variants included RASGEF1B intronic SNP rs17005141 (β = 0.29 [0.16 to 0.42], P = 0.08), PDE1B 5′ variant rs884510 (β = 0.14 [0.07 to 0.21], P = 0.08), and the intergenic SNP rs16829653 located between MFSD1 and IQCJ (β = −1.7 [−0.25 to −0.09], P = 0.08; Fig. 3a and Table 2; Table S4).

Fig. 3.

Association testing results for a) SNP-based and b) gene-based tests, with top results from each test shown in red. No SNPs were significant after multiple testing correction for the SNP-based association testing. The FDR significance threshold for the gene-based test is marked by the dashed line. Gene-based association testing revealed two genomic regions containing three genes significantly associated with hemoglobin concentration.

Table 2.

Gene associations with hemoglobin concentration [Hb]

Gene	FDR	rsID	Chr:Pos (hg38)	DA/AA	PER DAF	MAY DAF	AMR DAF	EUR DAF	EAS DAF	AFR DAF
RASGEF1B	0.047	rs2159657	4:81434439	G/A	0.94	0.92	0.80	0.71	0.78	0.37
		rs17005141	4:81458127	C/T	0.05	0.04	0.08	0.16	0.21	0.09
		rs12649837	4:81469773	T/C	0.08	0.08	0.11	0.17	0.18	0.15
PDE1B/PPP1R1A	0.047	rs884510	12:54578515	C/T	0.25	0.42	0.42	0.37	0.38	0.51

AA, ancestral allele; AFR, African; AMR, American; Chr, chromosome; DA, derived allele; DAF, derived allele frequency; EAS, East Asian; EUR, European; MAY, Mexican Maya; PER, Peruvian; Pos, position.

Standard GWAS is known to overlook weak but legitimate genetic associations with small individual effects (Manolio et al. 2009), particularly with investigations of polygenic traits where cohort sizes are limited. In these cases, gene-based analyses of association can be leveraged to examine the joint association of variants within a gene while accounting for local linkage disequilibrium (LD). Here, we performed gene-based association testing in MAGMA using three models including principal components regression, a SNP-wise mean model, and SNP-wise Top 1 model (De Leeuw et al. 2016). With this multi-model approach, P-values from all three models are internally combined to form an aggregate P-value. The 11,456 variants that passed association testing QC filters were mapped onto gene bodies, including 5 kb upstream and 2 kb downstream transcriptional windows, resulting in a total of 2,732 genes tested.

Three genes had aggregate P-values that passed FDR multiple testing correction: RASGEF1B, PDE1B, and PPP1R1A (FDR-adjusted P = 4.68e-2 for all three; Fig. 3b and Table 2; Table S5). RASGEF1B included three intronic SNPs, rs2159657 (derived allele frequency (DAF) (G) = 0.94), rs17005141 (DAF (C) = 0.05), and rs12649837 (DAF (T) = 0.08). The derived alleles for rs17005141 and rs12649837 were associated with higher [Hb] and showed lower frequencies in Peruvian Quechua compared to 1KG populations (AMR, EUR, EAS, and AFR) with only rs2159657 showing a higher frequency of the [Hb] decreasing derived allele in Peruvian Quechua compared to these populations (Table 2). For PDE1B and PPP1R1A, a single SNP, rs884510 (DAF (C) = 0.25), drove the significant association for both genes, mapping to the 3′ UTR of PDE1B and the downstream transcriptional window of PPP1R1A. The rs884510 ancestral allele (T) was associated with decreased [Hb] and found at the highest global frequency among Peruvians (1KG Peruvians from Lima [PEL] = 0.71, Peruvian Quechua with 80% or more Indigenous American ancestry [PER] = 0.75; Table 2). Based on these results, we chose to more robustly characterize the signature of positive selection within the PDE1B and PPP1R1A region using whole-genome sequencing (WGS) data.

WGS Data Reveal Several PDE1B eQTLs Under Selection

We explored PDE1B and PPP1R1A WGS data generated from a subset of participants including 46 Peruvian Quechua and 48 Mexican Maya individuals (Fig. S1) (Jorgensen et al. 2023). These data included 103 variants located within the gene of PDE1B or PPP1R1A and 49 variants located in their upstream transcriptional windows. We repeated the XP-nSL and PBS selection scan analyses, identifying 33 variants within PDE1B, including rs884510, that showed significant signals of selection for one or both of these statistics (Fig. 4; Tables S6 and S7). No variants within PPP1R1A were found to pass genome-wide significance thresholds. We examined the 33 PDE1B significant variants for associations with changes in both PDE1B and PPP1R1A gene expression using the GTEx eQTL Calculator (GTEx Consortium 2020). Data for 25 of the 33 variants were available for eQTL testing, which included expression profiles from blood vessel, heart, kidney, lung, spleen, and whole blood tissue types. No expression data were available for any bone tissues. Eleven variants were identified as eQTLs for PDE1B in at least one tissue, while none of the 25 variants were found to be eQTLs for PPP1R1A. Ten of these loci were found to act as eQTLs in whole blood, of which eight loci showed suppression of alleles associated with decreased PDE1B expression among Peruvians. The variant rs884510 was not found to act as an eQTL for either PDE1B or PPP1R1A in any tissues. The top three eQTLs in whole blood were the intronic variants rs10876566, rs7954532, and rs2669406 (Table S7).

Fig. 4.

Results for a) XP-nSL and b) PBS selection scan statistics using WGS data for the genomic region surrounding PDE1B and nearby hematopoiesis-related genes. Genome-wide selection significance thresholds are marked by the dashed lines. eQTLs for PDE1B that pass genome-wide thresholds for at least one statistic are highlighted in orange, and the three most significant eQTLs, rs10876566, rs7954532, and rs2669406, are highlighted in red. Genes within this region are shown in the schematic below.

Variation in PDE1B May Impact Expression Profiles of Nearby Hematopoiesis-Related Genes

We examined the location of PDE1B in the context of the gene's surrounding 3D genomic architecture. Topologically associating domains (TADs) are self-interacting, loop-like domains, whose boundaries act as partitions between TADs, preventing interaction between cis-regulatory elements and their target genes. On a genome-wide level, TAD boundaries are enriched for complex-trait heritability and are more evolutionarily constrained than the regions within TADs (McArthur and Capra 2021; Rajderkar et al. 2023). To examine the 3D genomic architecture of the region surrounding PDE1B, we used publicly available Hi-C data and corresponding TAD maps from the 3D Genome Browser (Dixon et al. 2012; Wang et al. 2018). We found that among various cell types, the genomic region containing PDE1B was located at a fairly conserved TAD boundary containing several hemoglobin and erythropoiesis associated genes including NCKAP1L, HZF (ZNF385A), ITGA5, and NFE2. This boundary was preserved among 12 cell lines including primary, stem, and cancer lines. Following evidence that nearby genes and genes located within a shared TAD show similar expression patterns (Gonzalez-Sandoval and Gasser 2016), we investigated whether PDE1B is co-expressed with NCKAP1L, HZF, ITGA5, and NFE2. We queried co-expression networks for bone marrow, heart, and lung tissues in GeneFriends (Raina et al. 2023). In all three tissues, PDE1B was found to be co-expressed with NCKAP1L (Pearson correlation coefficient = 0.65, 0.64, and 0.88 for bone marrow, heart, and lung tissues).

Discussion

High altitude has imposed severe environmental pressures on Andean highlanders for over 11,000 years (Rademaker et al. 2014), resulting in a unique suite of hypoxia-responsive phenotypes. Among these traits, hemoglobin concentration [Hb] is elevated compared to sea-level values, ranging from an average of ∼17 g/dL in Andean females to ∼18 g/dL in Andean males at high altitude (León-Velarde et al. 2000; Beall 2007). This response mirrors the hematopoietic response observed among lowlanders acclimatized to high altitude. Increased [Hb] at high elevations helps offset the low ambient oxygen availability by increasing arterial oxygen content, but can lead to polycythemia and eventually result in CMS. In contrast, Tibetan highlanders show sea-level normative [Hb] values, demonstrating that lowered [Hb] may be adaptive through lowered blood viscosity and improved hemodynamics (Simonson et al. 2015; Stembridge et al. 2019). We hypothesized that natural selection has acted in response to high-altitude hypoxia to shape [Hb] among Andeans. We detected novel associations with lowered [Hb] for two genomic regions containing three genes, PDE1B, PPP1R1A, and RASGEF1B, that show signals of positive selection. These results suggest that polygenic selection may be acting to select genetic variation that blunts the Andean hematopoietic response despite the documented elevated [Hb] that has long been considered part of the Andean adaptive suite to hypoxia and point to phenotypic convergence between Andeans and Tibetans.

At first, our evidence of natural selection acting to lower [Hb] among Andeans may appear paradoxical. The elevated [Hb] observed among Andeans has been considered part of the Andean adaptive suite for decades. So why, then, would our data support natural selection operating on genetic variation to blunt the hematopoietic response? One explanation is that early-stage polygenic directional selection, or a selective sweep across multiple loci that is in its beginning stages, may be acting in favor of this blunted response to counteract elevated hemoglobin concentrations induced by acclimatization responses. While moderately elevated [Hb] can improve blood oxygen capacity and tissue oxygenation, excessively elevated [Hb] and the associated increases to blood viscosity can result in reduced cardiac output and microcirculatory blood flow, theoretically limiting the passive diffusion of oxygen across the alveolar-capillary membrane (Wagner 1996; Storz and Bautista 2022). Furthermore, excessive erythrocytosis is a hallmark of CMS, a progressively debilitating condition that can lead to myocardial infarction, and is found at frequencies of approximately 6% to 33% in the Andean highlanders, compared to 0.91% among Tibetan highlanders (Spielvogel et al. 1981; Monge et al. 1989; De Ferrari et al. 2014). Selection acting to dampen this hematological response to hypoxia may increase evolutionary fitness by limiting the development of excessive erythrocytosis and CMS. This explanation is in line with recent findings demonstrating selection at an Andean-specific EPAS1 variant to lower hematocrit (Lawrence et al. 2024).

A second explanation is that polygenic stabilizing selection may be operating on multiple loci to shape optimal hemoglobin concentration among Andean highlanders. Optimal hemoglobin concentration is [Hb] raised enough to increase blood oxygen saturation, but not elevated to the point of inducing pathologically increased blood viscosity. Previous work has suggested that optimal hemoglobin at high altitude may in fact be very close to sea-level values (Reeves and Leon-Velarde 2004). There can be numerous combinations of positively and negatively associated loci that jointly result in the optimal [Hb] phenotype (De Vladar and Barton 2014). In this way, directional selection operates at the level of the genotype, but stabilizing selection operates at the level of the phenotype. Previous research showing genetic variation in EPAS1 associating with lowered hematocrit and genetic variation in EGLN1 associating with elevated [Hb] supports a model of polygenic stabilizing selection (Yasukochi et al. 2020; Lawrence et al. 2024). Here, we only detected gene associations with lowered [Hb], but these findings do not preclude the absence of gene associations with elevated [Hb].

Polygenic adaptation plays a powerful role in shaping human evolution and is likely more prevalent and impactful in facilitating rapid adaptation to environmental pressures than hard selective sweeps (Pritchard et al. 2010; Hernandez et al. 2011; Gnecchi-Ruscone et al. 2018; Barghi et al. 2020). Hemoglobin concentration is a complex trait with variants in over 800 genes likely contributing to its spectrum of variation (Chen et al. 2020; Vuckovic et al. 2020; Barton et al. 2021). It therefore follows that genetic adaptations underpinning this trait within the context of high-altitude hypoxia would involve multiple loci that together shape [Hb]. Our results further suggest that polygenic selection may be influencing this complex phenotype by operating within a more diverse set of biological pathways than previously thought. Our findings emphasize the value of using genome-wide approaches that are equipped to detect polygenic signatures of selection and association when investigating the genetic basis of complex phenotypes. When inclusive to populations that are not well-represented in genomic studies, these approaches are especially well equipped to identify functional genomic variation underlying complex human phenotypes and can contribute to current understandings of these phenotypes among broader global populations.

Our most compelling association with [Hb] was for the gene PDE1B. PDE1B encodes for a subunit of the calmodulin-stimulated cyclic nucleotide phosphodiesterase that participates in the NO-sGC-cGMP signaling pathway (Hanson et al. 2008; Adderley et al. 2009). Activation of this pathway stimulates hematopoiesis by inducing erythropoiesis in erythroid and myeloid cell lines and upregulates γ-globin gene expression in primary erythroblasts (Ikuta et al. 2001, 2010). A single SNP, the 3′ UTR variant rs884510, was responsible for the PDE1B gene association wherein the ancestral Peruvian allele (T) was associated with a 1.39 standard deviation decrease in [Hb]. Analysis of WGS data for this region revealed signals of recent, ongoing selection, including for several eQTLs that may act to increase PDE1B expression levels. Previous genomic investigation has identified this gene to be under positive selection among Andean Calchaquíes, another high-altitude population (Eichstaedt et al. 2015). This upregulation would result in the increased hydrolyzation of cGMP, and therefore the suppression of the NO-sGC-cGMP signaling pathway and a reduction of erythropoiesis. Further, augmentation of PDE1B expression may alter the expression of numerous hematopoiesis-related genes located within the bordering TAD, including NCKAP1L. NCKAP1L encodes the cytoskeleton scaffold protein HEM-1, which is involved in erythrocyte membrane stability and highly expressed within hematopoietic cells (Hromas et al. 1991; Park et al. 2008). Together, this evidence provides a mechanism by which recent and ongoing positive selection may have impacted PDE1B gene expression to facilitate functional adaptive change by lowering [Hb] among Andean highlanders.

In addition to PDE1B, we found evidence that positive selection may be acting on PPP1R1A and RASGEF1B to lower Andean [Hb]. PPP1R1A encodes for a protein phosphatase 1 (PP1) regulatory inhibitor subunit. Modulation of PP1 expression has been previously associated with ischemic injury, heart failure progression, cardiomyopathy, and ventricular tachycardia (Pathak et al. 2005; Nicolaou et al. 2009; Wittköpper et al. 2010). Given the potential shared functionality of rs884510, it is possible that natural selection may be influencing [Hb] by targeting PPP1R1A. However, our analysis of WGS data suggests that positive selection in this area is more likely acting to influence PDE1B expression as variants with publicly available expression data were all found to function as eQTLs for PDE1B, while none were eQTLs for PPP1R1A. RASGEF1B encodes for a subunit of RAS guanine nucleotide exchange factor (GEF) 1, and genetic variation in this gene has been associated with coronary heart disease and with white-coat hypertension (Yaman et al. 2009; Van Der Harst and Verweij 2018; Rimpelä et al. 2019). Of the three intronic RASGEF1B SNPs highlighted by our analysis, none have previously been significantly associated with any phenotype.

Genetic variation in PDE1B, PPP1R1A, and RASGEF1B has not been associated previously with lowered [Hb] among Tibetans. Rather, associations with lowered [Hb] have been identified for variants within the genes EPAS1, EGLN1, PPARA, and HMOX2 (Beall et al. 2010; Simonson et al. 2010, Yi et al. 2010; Xiang et al. 2013; Yang et al. 2016; Tashi et al. 2017). Among Andeans, previous work has identified associations for EGLN1 and EPAS1 variation with raised [Hb] and lowered hematocrit, respectively (Yasukochi et al. 2020; Lawrence et al. 2024), but our genome-wide analysis did not identify significant SNP or gene associations with Andean [Hb] for genes previously associated with this phenotype in either Andeans or Tibetans. Rather, our study builds on these previous investigations to show genome-wide evidence for selection on genetic variation to lower [Hb] among Andeans. On the whole, this evidence suggests that natural selection may be acting to lower [Hb] among Peruvian Quechua in a manner phenotypically convergent with Tibetans and highlights how natural selection may act on different gene sets across distinct populations to produce similar phenotypes. The lack of association with [Hb] for EGLN1 and EPAS1 may reflect an absence of a true association for these two genes or may be a function of the combined effects of our relatively small sample size with array-based genotyping data. To the latter point, it is important to note that data for the Andean-specific EPAS1 variant previously associated with lowered [Hb] was not included in the genotyping array and therefore unavailable for our analysis, highlighting the potential importance of novel variation contributing to Andean adaptive phenotypes.

High altitude presents a variety of environmental pressures beyond hypoxia including increased UV exposure, extreme temperature fluctuations, cold exposure, decreased availability of arable land, dietary shifts, and pathogen exposure. Furthermore, high-altitude hypoxia has been shown to select for genetic variation contributing to high-altitude phenotypes other than [Hb] including VO2Max, pulmonary artery pressure, nitric oxide biosynthesis, among others (Peng et al. 2017; Brutsaert et al. 2019; Jorgensen et al. 2023). Therefore, we expect that our results have identified signals of selection in genes involved in a variety of other adaptive traits beyond [Hb] and are worthy of future exploration. This includes the genes FRMD4B and GSTZ1. FRMD4B harbored the most significant SNP for XP-nSL, the intronic variant, rs13063380. FRMD4B is a member of the GRP1 signaling complex that helps regulate cellular polarity and is involved in insulin sensing (Garner et al. 2014). The GSTZ1 intronic variant rs4147578 showed the most extreme PBS value. GSTZ1 encodes a metabolic enzyme involved in amino acid catabolism, particularly phenylalanine and tyrosine.

Andeans are an Indigenous American population for whom many novel variants may have yet to be identified and consequently would not be included on standard genotyping arrays. Therefore, the signals of positive selection and gene associations with lowered [Hb] may have been driven by causal variation in LD with variants in our dataset. Analysis of WGS data allows for the identification of novel and/or high frequency Andean variants driving the associations with Andean [Hb]. Similarly, population-specific differences in underlying genetic architecture that may be unique to this population also complicate the interpretation of publicly available gene expression and Hi-C (TAD map) data, which largely originate from European or Han Chinese populations. Our findings would benefit from future work investigating the functional consequences of the [Hb] associated variants identified by this study.

Conclusion

We observed widespread signals of natural selection and identified gene associations with lowered [Hb] for two genomic regions containing three genes, PDE1B, PPP1R1A, and RASGEF1B. eQTL and TAD analysis suggested that variation in PDE1B may be acting to increase PDE1B expression and may affect the co-expression of nearby hematopoietic genes. Together, these findings suggest that recent, ongoing selection may be acting across multiple genomic regions to drive lowered hemoglobin concentrations among Andeans, thus providing novel evidence for the role of polygenic selection in shaping convergent phenotypic adaptations among Andean and Himalayan populations.

Materials and Methods

Participant Recruitment, Data Collection, and Phenotype Comparison

All study participants provided written informed consent in Spanish at the time of enrollment, and 3 mL of whole blood was collected from each individual for [Hb] measurement and DNA extraction. [Hb] was measured using a HemoCue Hb201+ analyzer (AngelHolm, Sweden). Blood was field stabilized in Cell Lysis buffer (Qiagen, Valencia, CA), then hand-carried to the University of Michigan where DNA extraction was performed in the Bigham Lab for Anthropological Genomics using the Puregene Protocol (Qiagen, Valencia, CA). All study participants provided basic anthropometric data including weight, height, sex, and age. This study was approved by the Institutional Review Boards of the University of Michigan, Syracuse University, the Universidad Peruana Cayetano Heredia (UCPH, Lima, Peru), and the Centro de Investigación y Docencia Económicas (CIDE, Mexico City, Mexico). Student's t-test was used to detect difference between the means of phenotypic characteristics for study groups using unequal or equal variances, depending on the phenotype.

Quality Control Filtering

Genotype data for 603 Peruvian Quechua and 100 Mexican Maya participants were generated using the Affymetrix (Santa Clara, CA) Axiom Biobank Genotyping Array that included 628,679 variants including haplotype tagging SNPs, exonic SNPs, and eQTL variants. Quality control filtering was performed on the full set of variants. Variants that passed Affy Best Practices, had call rates > 95%, and passed the Hardy-Weinberg exact test with a threshold of 1e-6 were retained for analysis (Purcell et al. 2007; Chang et al. 2015). After initial QC filtering, 611,852 variants remained. Peruvian Quechua, Mexican Maya, and data for all 1KG populations were merged. Strand-flipped variants and variants that did not have an identifiable strand were identified and removed using conform-gt (Browning and Browning 2016). In total, 536,061 variants passed QC filtering. Participant relatedness was estimated using KING (Manichaikul et al. 2010) and individuals exhibiting 3rd degree relationships or higher were removed, retaining the individual with the higher genotyping rate for each dyad. This resulted in a dataset of 573 Peruvian and 71 Mexican participants. Affymetrix array data were mapped to hg19 at the time of data delivery. All analyses were performed using hg19 coordinates with the results lifted over to hg38 using UCSC Genome Browser Liftover tool (Hinrichs et al. 2006).

PCA and Admixture Analysis

Principal component and individual ancestry analyses were conducted on 96,532 SNVs using Plink 2.0 and ADMIXTURE (Purcell et al. 2007; Alexander and Lange 2011; Chang et al. 2015) with additional data from the 1KG Project. Variants were filtered to exclude insertion or deletion polymorphisms, low frequency variants (MAF < 0.025), and variants in LD (r² > 0.2). A global ancestry estimate for each individual in the dataset was calculated using an unsupervised model (K = 5), and Indigenous ancestry was calculated as the sum of Peruvian Quechua and Mexican Maya ancestral components.

Phase Estimation, Selection Scan, and Overrepresentation Analysis

Haplotype phase estimation was performed using SHAPEIT4.4.2 (Delaneau et al. 2019). As inclusion of related individuals can improve phase estimation, we included all Peruvian Quechua and Mexican Maya individuals, as well as 1000 Genomes Project data from CEU, YRI, and EAS individuals. Two test statistics that detect evidence of positive selection, XP-nSL and PBS, were computed using a panel of 334 Peruvian Quechua individuals and 57 Mexican Maya individuals. The individuals included in this analysis showed over 90% Indigenous ancestry as estimated by our ADMIXTURE analysis. XP-nSL was computed in Selscan 1.3.0 for variants with complete observations (no missing data) and no minor allele frequency filtering (Szpiech and Hernandez 2014) to identify haplotypes that increased in frequency after Peruvian Quechua split from Mexican Maya. XP-nSL scores were then normalized across the genome using the norm function (Szpiech and Hernandez 2014). F_ST values for each SNP with a minor allele frequency above 5% (MAF > 0.05) were calculated using Hudson's F_ST estimator (Hudson et al. 1992; Bhatia et al. 2013). PBS was calculated by comparing pairwise F_ST values between Peruvians, Mexican Maya, and 1KG Han Chinese (CHB). Both Hudson's F_ST and PBS were calculated in R using KRIS (Chaichoompu et al. 2018). Genetic variants or windows falling in the top 5% of each of the empirical distributions were determined to be statistically significant (Akey et al. 2002).

Genome-Wide Association Tests

A total of 26,723 variants were found to be significant with either XP-nSL or PBS, which were passed forward for genome-wide association testing. We performed HLA masking to remove a total of 653 variants from the HLA region (Chr 6: 28,477,797 to 33,448,354). We enforced a minor allele frequency threshold of 0.05 that resulted in 11,456 passing variants. SNP-based and gene-based association tests were performed in Plink 2.0 and MAGMA using these QC passing variants (De Leeuw et al. 2016). To assess the effect of LD, we conducted LD pruning of the GWAS results using PLINK with the command –indep-pairwise 50 5 0.8, which retains largely independent SNPs for downstream analysis. No significant associations were detected after pruning, so the primary results presented here are based on the full, non-pruned dataset.

For the SNP-based test, we performed linear association using an additive inheritance model. For the gene-based test, the annotation file was built using NCBI gene locations, and the 11,456 variants were mapped to gene coordinates as well as up (5 kb) and down (2 kb) stream transcriptional regions. These variants mapped to a total of 2,735 genes. Three complementary gene analysis models were run in MAGMA, including a principal components regression that provides improved detection of association in low LD areas, a SNP-wise mean model, which is more sensitive to areas of higher LD, and SNP-wise Top 1 model, which has improved power when a small proportion of variants within each gene show association. These models were run using the “multi-all” model in MAGMA, which internally combines resulting P-values into an aggregated P-value for facilitated interpretation. Sex, recruitment altitude, and the first five genetic PCs were again included as covariates. To account for the selection-nominated variants that were excluded due to low minor allele frequencies (MAF < 0.05), we performed an additional gene-based association test that only included these low-frequency variants. This analysis did not reveal any additional genes showing significant associations with [Hb].

WGS Data: Selection Scan and eQTL Analysis

WGS data for 46 Peruvian Quechua and 48 Mexican Maya individuals were generated on an Illumina NovaSeq (single S4 flow cell) by the University of Michigan's DNA Sequencing Core. These data were filtered using quality control and variant calling procedures outlined by Jorgensen (Jorgensen et al. 2023) and were aligned to hg38. Variants surrounding PDE1B and PPP1R1A were extracted. We performed a selection scan in this region using identical procedures performed in the previous section for XP-nSL and PBS statistics. XP-nSL was calculated by comparing Peruvian Andeans to Mexican Maya. PBS was calculated between Peruvian Andeans, Mexican Maya, and 1KG CHB. The CHB data were obtained from the 1KG 30× high coverage WGS release that was aligned to hg38. Variants were annotated using the Ensembl Variant Effect Predictor (McLaren et al. 2016). We used the eQTL calculator database to identify which of these variants have been identified as eQTLs for either PDE1B or PPP1R1A in a range of phenotype-relevant tissues in the GTEx Project (Lonsdale et al. 2013, François et al. 2020). Variants that were found to have P-values below the pre-computed P-value threshold for that gene-tissue pairing were reported as significant eQTLs.

TAD Boundary and Co-expression Analysis

We obtained publicly available TAD maps from the 3D Genome Browser. These TAD maps were predicted from Hi-C data using a pipeline created by Dixon (Dixon et al. 2012). We calculated conservative 40 kb windows surrounding PDE1B to determine TAD boundaries that fall within these windows. Gene co-expression analyses were performed in GeneFriends (Raina et al. 2023) using SRA data (Katz et al. 2022). Gene expression data from whole blood was not available in this database.

Supplementary Material

Supplementary material is available at Genome Biology and Evolution online.

Author Contributions

K.Z.: data analysis (lead), data curation (lead), funding acquisition (supporting), investigation (lead), methodology (lead), visualization (lead), writing—original draft and reviewing (lead), and conceptualization (supporting). K.C.J.: data analysis (supporting), methodology (supporting), and writing—reviewing (supporting). J.W.: data collection (supporting). M.K.: data collection (supporting). G.E.: data collection (supporting). J.L.I.: data collection (supporting). A.M.B.: data collection (supporting). O.A.G.: data analysis (supporting), data collection (supporting), and writing—reviewing (supporting). G.G.: data analysis (supporting). L.N.P.: data collection (supporting). F.S.L.: data collection (supporting) and funding acquisition (supporting). M.R.-C.: data collection (supporting). F.L.-V.: data collection (supporting) and funding acquisition (supporting). T.D.B.: data collection (supporting), funding acquisition (supporting), and writing—reviewing (supporting). A.W.B.: funding acquisition (lead), data collection (lead), data curation (supporting), methodology (supporting), investigation (lead), supervision (lead), project administration (lead), conceptualization (lead), and writing—original draft and reviewing (lead).

Funding

This work was supported by NSF grants BCS-1132310 (to A.W.B., T.D.B., and F.L.-V.) and BCS-1638642 (to A.W.B., T.D.B., and F.S.L.), Leakey Foundation grant (to A.W.B.), and an Elizabeth Caroline Crosby Grant from the University of Michigan (to A.W.B.). The University of California, Los Angeles also supported this work through Graduate Research Mentorship, Graduate Summer Research, and Dissertation Year Fellowship Awards (to K.T.Z.).

Data Availability

Genotyping data are available through the Dryad digital repository (https://doi.org/10.5068/D1XH3R).