Higher oxygen content and transport characterize high-altitude ethnic Tibetan women with the highest lifetime reproductive success

Four hundred and seventeen women 46 to 86 y of age reported 2,193 pregnancies, resulting in 2,076 livebirths, 71 miscarriages, and 46 stillbirths. They averaged 5.3 pregnancies and 5.2 livebirths with a range of variation from zero to 14 (Fig. 1 and Table 1). Table 1 describes all the women, a large subset with complete information, including echocardiograms and HVR tests, and a small subset without echocardiograms or HVR tests. Our analyses focus on livebirths because women may not have recognized early, lost pregnancies; parallel analyses of pregnancies appear in Supporting Information. The two measures of lifetime reproductive success (number of pregnancies and number of livebirths) correlate highly (r = 0.971, P < 0.001).

We first considered natural selection on oxygen transport phenotypes. 364 women provided complete information on all variables. The best parametric analysis had a mixture of linear and nonlinear components and showed that women 63 y and older had the most livebirths (Table 2); those women had their children before family planning technology became accessible. An earlier age at first birth and a longer marriage predicted more livebirths, reflecting a longer exposure to the possibility of pregnancy. A high percent of oxygen saturation of hemoglobin, an intermediate hemoglobin concentration of approximately 13.4 gm/dL, and a low HHRR predicted more livebirths. Wide left ventricular outflow tract relative to body size (LVOTDiamIndex) or having twins interacted with a later age at first birth to predict more livebirths. We speculate that lower HHRRs allow a longer time to fill the ventricle with oxygenated blood. The hemoglobin and oxygen saturation results align with previous reports on Tibetan women (16, 36, 37). Our findings link high oxygen content (hemoglobin concentration and oxygen saturation) and oxygen transport (HHRR and LVOTDiamIndex) to more livebirths.

Fig. 2, Left shows the relationship between the predicted number of livebirths and hemoglobin concentration and the distribution of hemoglobin concentration values. The predicted livebirths increase sharply from the lowest hemoglobin concentrations to the mode, decrease gradually at higher values, and suggest stabilizing selection. Fig. 2, Right shows that ever-higher oxygen saturation predicts more livebirths and suggests directional selection.

Next, a nonparametric, data-driven, and machine-based approach [classification and regression tree analysis (38)] identified homogenous groups of women based on specific combinations of traits and values most associated with a particular number of livebirths. Fig. 3 shows the resulting tree. The variable in the box at the top labeled “length of the first marriage, yrs” links a sequence of nodes/boxes to an oval at the bottom. That oval shows the average number of livebirths of a maximally homogeneous group of women with the traits described. The tree in Fig. 3 shows the length of the first marriage as the primary factor distinguishing women with few or many livebirths, followed by the age at first pregnancy/birth. At the bottom left, women with 2.4 livebirths were married for fewer than 24 y and had a first pregnancy at 27 or later. At the bottom right, five traits identified two groups of women with eight or more livebirths among those with longer marriage, earlier first birth, and a right ventricle stroke distance (RVOTvti, the distance blood travels toward the lungs per heartbeat) above the sixth percentile (>11 cm). Women averaging 9.4 livebirths also had left ventricle outflow tract diameter normalized for body surface area (LVOTDiamIndex) above the 65th percentile (>1.5 cm/m²). Women with 8.5 livebirths had an LVOTDiamIndex below the 65th percentile, a pulse below the 5th percentile (68 f/min), and a percent oxygen saturation above the 32nd percentile. One successful combination of traits included oxygen transport traits reflecting deoxygenated blood flow into the lungs and oxygenated blood out of the left ventricle of the heart into the aorta and systemic circulation. A second successful combination of traits reflected a combination of high blood flow into the lungs and a smaller left ventricle coupled with a low pulse and high oxygen saturation.

A good parametric analysis provides the best-fitted model within an assumed form, such as linear, quadratic, or specific nonlinear models, and is efficient and easy to interpret if the model form is a reasonable approximation to the data. Nonparametric analysis, which does not assume a specific form, has minimal assumptions and offers robustness and flexibility in real-data analysis when the model form is unknown. If parametric and nonparametric fits yield similar results, the assumed parametric form is justified. Hence, both analyses are crucial for responsible data science.

Indeed, the parametric and nonparametric analyses reinforce and refine one another. Both identified the mother’s age at her first pregnancy/birth and the length of her first marriage as the most influential variables. The parametric analysis showed a linear correlation between higher saturation and more births; the tree-based analysis located the favorable range above 90%–the average was 91% (Table 1). Both analyses identified lower heart rates as beneficial, whether under added hypoxia or in room air. Both linked a large left ventricle outflow diameter, particularly above 1.5 cm/m², to many livebirths. Two lines of evidence suggest the larger diameter supported reproduction rather than resulted from it. A larger ventricle outflow diameter can offset somewhat the effects of a late first birth (Table 2) and a smaller one can support many births when combined with high saturation (Fig. 3).

We hypothesized that pulmonary, respiratory, hematological, and cardiovascular elements of a suite of traits involved in oxygen delivery and the response to high-altitude hypoxia would predict lifetime reproductive success and show genomic associations. We found influential and significant hematological and cardiovascular predictors. High-altitude native resident Tibetan women with higher oxygen content (modal hemoglobin concentration and progressively higher oxygen saturation) and oxygen transport (lower HHRR and wider left ventricles) tend to have more livebirths. Generally, these results matched expectations. However, respiratory and pulmonary traits did not predict the number of livebirths. Previous work with younger samples of Tibetans across the altitude range of 3,658 to 4,200 m reported distinctly high resting ventilation, HVR (1, 3), and minimal elevation of pulmonary artery pressure (7, 10). Therefore, we hypothesized that these traits would improve reproductive success. This study did not support that hypothesis, perhaps because of the older age range, different measurement protocols, or sensitivity to other environmental factors.

Our results refined the understanding of hemoglobin concentration values by showing that the optimum hemoglobin concentration of 13.4 gm/dL for livebirths is close to the mode of 13.5 gm/dL. Sharply lower values below the optimum show fewer and fewer predicted livebirths, likely related to arterial oxygen content too low to support pregnancy or fetal development (Fig. 2). The slow decline in predicted livebirths above the optimum may reflect a change from past selection against very high hemoglobin concentrations that raise blood viscosity. High blood viscosity burdens the cardiovascular system; ventricles work harder, cardiac output falls, impaired blood flow, and oxygen diffusion ensues. Resulting costs to the individual include poorer exercise capacity and a higher risk of Chronic Mountain Sickness (44, 45). None of the women in this study had excessively high hemoglobin concentrations, consistent with past directional selection favoring lower values. These findings are consistent with studies and meta-analyses showing that intermediate maternal hemoglobin concentrations before and during pregnancy are optimal for pregnancy outcomes in many populations (46–51). Finding this association among Andean women living as high as 4,000 m seems paradoxical; however, robust expansion of plasma volume during pregnancy may be particularly relevant in the context of the generally high hemoglobin concentrations among Andean highlanders (52). The available evidence on blood volume during high-altitude pregnancy is equivocal. Comparing women of Andean and European descent at 3,600 m, Vargas and colleagues reported that absolute blood volume was lower among Andean women, although the two samples maintained the same volume per kilogram body weight through pregnancy (53). Available evidence finds that the EGLN1/EPAS1 oxygen homeostasis pathway does not exert a large influence on hemoglobin concentration in the Andean populations as it does in the Tibetan (54–57).

An expanded plasma volume (not a lower hemoglobin mass) contributes to Tibetans’ unelevated hemoglobin concentrations (52). An expanded baseline plasma volume, further expansion during pregnancy, and cardiovascular responses to hypoxia (58) could tax the cardiovascular system. This study provides evidence of cardiovascular associations with lifetime reproductive success. The tree-based analyses showed that the full range of LVOTDiamIndex values could support high reproductive success, depending on the values of other traits. Relatively large left ventricles (LVOTDiamIndex) pumping oxygenated blood to the systemic circulation predicted more livebirths if blood flow to the lungs was above a low minimum (6 to 8th percentile of RVOTvti). Interestingly, a relatively small left ventricle also supported many births if combined with an oxygen saturation at or above 90% (the mean was 91%). A second combination of traits supporting many pregnancies was the rapid flow of deoxygenated blood to the lungs (RVOTvti) and significant volumes of oxygenated blood from the heart (LVOTDiamIndex and LADIndex) to the systemic circulation. These combinations imply large caliber arteries or effective flow-mediated vasodilation to accept large blood volumes from the heart without raising blood pressure. Some women do not accomplish this. The tree-based analyses of the 53 women without echocardiograms or HVR tests show high blood pressure among some women.

The values of the traits involved in natural selection, correlating with or influencing lifetime reproductive success in this sample, lie close to or within the usual sea-level range, such as unelevated hemoglobin concentration and higher oxygen saturation. This is consistent with the hypothesis “in which …directional selection... offsets environmentally induced changes.” (6 page 4125).

Regarding genetic selection, we conducted hypothesis-free GWAS and hypothesis-driven candidate gene analyses. Genome-wide analyses did not detect genome-wide significant associations with the candidate variables influencing reproductive success; none showed selection signals, similar to a recent study of new Tibetan mothers and neonates at high altitudes (37). This report of a genetic association with oxygen saturation remains to be replicated, as do other reports (39–42, 58). Candidate gene analyses replicated previous reports of an association with hemoglobin concentration. This study identified a set of SNP sites just below the cutoff for genome-wide significance whose most frequent alleles increased saturation by 2.8%. This result supports our earlier report that ethnic Tibetan women with higher estimated oxygen saturation genotypes had higher reproductive success (36). Furthermore, we reported a genome-wide significant association of EPAS1 with low deoxygenated hemoglobin (Hemoglobin concentration × (100-percent of oxygen saturation)/100) and polygenes lowering deoxygenated hemoglobin (27). Selection for modal hemoglobin concentration and against deoxygenated hemoglobin likely reinforces selection for higher oxygen saturation and may represent another instance of the pleiotropic effects of EPAS1 (59).

This study design combines retrospective reproductive history and current physiology and reasons that traits measured after the reproductive period inform about maternal biology during the childbearing years (for example, maintaining a relatively high percent of oxygen saturation or an optimal hemoglobin concentration). However, traits measured after reproduction may also reflect selective survival, biological aging, social and cultural factors, or the costs of reproduction. We consider these potential confounders next.

This study focused on the primary candidate physiological systems for adaptations to high-altitude hypoxia: respiratory, hematological, and cardiovascular systems comprising the distinctive Tibetan biology. Future work could address whether the expanded plasma volume reported for male Tibetans (52) occurs among women and the consequences of further expansion during pregnancy (71). Pairing measurements of cardiopulmonary with peripheral vascular traits could be informative, as would serum traits reflecting hemoglobin synthesis and iron homeostasis (e.g., erythropoietin or hepcidin). We do not have serum measurements of relevant biomarkers because community engagement before mounting the fieldwork revealed that the women emphatically did not want to provide blood samples. We respected that wish.

The modest sample size, by GWAS standards, afforded statistical power sufficient to detect only large phenotypic effects. Simultaneously, the large sample by physiology standards gave necessary statistical power to identify traits relevant to reproductive success. Balancing the need for careful measures of physiology with the need for sufficient sample sizes for GWAS will continue to challenge field studies.

Strengths include managing these limitations using several statistical approaches to increase confidence in our findings. Multiple analyses converged on a few outcome variables and increased our confidence in their associations with lifetime reproductive success, although we cannot infer causality. Nonlinear transformation of age, BMI, and hemoglobin concentration reflected the data and biologically realistic relationships. Tree-based analyses of these complex data provided flexible modeling with minimal assumptions. Cross-validation analyses confirmed the robustness of these models and the likelihood of similar performance in a replication sample. A follow-up study with a larger sample size is desirable. Good to excellent intraclass correlations and small differences between test–retest measurement of the physiological traits reflect our effort to reduce measurement noise (SI Appendix, Table S2C). Formally considering sociocultural factors resulting in prolonged exposure to the possibility of pregnancy enabled the detection of influential oxygen delivery traits. The large sample of women who had completed their families enabled assessing lifetime reproductive success in the context of their social and physical environments, further strengthening this study.

The two measurements of the percent of oxygen saturation of hemoglobin strengthened the study. Fingertip measurements obtained by transmittance plethysmography enabled longitudinal analysis of variables collected by reflectance plethysmography of the fingertip. Forehead measurements obtained by reflectance plethysmography correlated with reproductive success and had genomic associations (16, 36).

Linking multiple oxygen delivery traits with reproductive success and genomes in a sample exposed to lifelong hypoxic stress and the added physiologic stress of pregnancy advances the understanding of human evolution and adaptation. Lifetime reproductive success is a complex trait. Including sociocultural information, linear and nonlinear analyses provide a fresh approach to studying this essential outcome measure of natural selection. These findings show that overlapping physiological homeostatic systems support a favorable internal environment for reproductive success in the extreme external environment of high-altitude hypoxia.

In summary, we detected traits relating to high lifetime reproductive success in a sample of more than 400 ethnic Tibetan women, lifetime residents at ≥3,500 m. Stabilizing selection favoring a modal hemoglobin concentration and directional selection favoring high oxygen saturation of hemoglobin together increased oxygen content and links to high lifetime reproductive success. Combining optimal levels of hemoglobin concentration near the sample mode with oxygen saturation ranging up to 99% raised oxygenated hemoglobin concentration without increasing blood viscosity and correlated with more livebirths. Indicators of blood flow into the lung and out of the heart related nonlinearly to high lifetime reproductive success. Two beneficial combinations defend convective oxygen transport from the heart into the systemic circulation. High flow of deoxygenated blood to the lungs and oxygenated blood into the systemic circulation benefited reproductive success as did relatively low flow to the lungs combined with oxygen saturation. These findings suggest multiple ways to maintain homeostasis and achieve many livebirths.

Linkages among hematological and cardiovascular phenotypes, genomics, and reproductive success among ethnic Tibetan women residing at 3,500 m or higher in Nepal demonstrate natural selection favoring oxygen delivery under the severe stress of high-altitude hypoxia.

We thank the women who participated in this study and the Upper Mustang residents who graciously hosted our research team. We are grateful to our field director, Kunzom Thakuri, nurses Tashi Yangchen Bista and Tsering Wangmo, and research assistants Dhiki Dhoka, Dolma Tsering, and Nhima Lhamo, who worked tirelessly and effectively. The fieldwork benefitted substantially from the advice of Amchi Gyatso Bista of Lo Monthang. The Center for Molecular Dynamics Nepal provided much appreciated logistical support. We thank the Nepal Health Research Council for their review and approval of the study. Sarah Emch performed the echocardiograms. Kaylee Sarna and Jillian Schulte managed the data entry and verification. Hannah Boehringer provided three figures. NSF Award 1153911 to CMB supported this research. NIH award HL119577 to A.D.R. and HL145470 to T.S.S. partially supported this work.