Red Blood Cells Serve as a Primary Glucose Sink to Improve Glucose Tolerance at Altitude

Summary

High altitude conditions improve glucose tolerance and reduce diabetes risk, but the physiological mechanism is not well-understood. Using mouse models, we found that hypoxia alone robustly improved glucose tolerance and that the effect persisted for weeks after returning to normal oxygen levels. PET/CT imaging suggested a significant, unknown glucose sink beyond major internal organs. We hypothesized that hypoxia-induced red blood cells (RBCs) serve as this sink. Manipulating RBC numbers through phlebotomy or transfusion directly altered blood glucose, establishing RBCs as necessary and sufficient for this effect. In chronic hypoxia, RBCs showed a sustained ~3-fold increase in glucose uptake and ~2-fold increase in GLUT1 protein abundance, specifically in newly synthesized RBCs, which ultimately contributes to increased glycolytic flux towards 2,3-DPG. Mechanistically, acute hypoxia promotes displacement of GAPDH from inhibitory Band 3 binding through competitive interactions with deoxyhemoglobin, thereby boosting glycolytic flux and driving 2,3-DPG production. We also found that hypoxia or our small molecule hypoxia-mimetic, HypoxyStat, rescued hyperglycemia in mouse models of type 1 and type 2 diabetes. Our findings identify RBCs as key regulators of systemic glucose metabolism, highlighting a novel therapeutic approach for hyperglycemic disorders.

Graphical Abstract

Introduction

Numerous epidemiological studies report a decreased incidence of diabetes and improved glycemic control at high altitude^1–12 (Table 1). This phenomenon even extends beyond humans. For example, high-altitude-adapted deer mice (Peromyscus maniculatus) exhibit enhanced glucose disposal compared to their low-altitude counterparts¹³. Similarly, pigs in Tibet display improved insulin sensitivity and lower plasma glucose concentrations compared to low-altitude breeds¹⁴. Multiple high-altitude songbird species have lower glycemia and improved insulin sensitivity¹⁵. These cross-species observations suggest the existence of an evolutionarily conserved physiological mechanism that optimizes glucose disposal at altitude.

Table 1. Human studies linking high altitude with improved glycemic control.

Summary of published research articles reporting associations between chronic high-altitude exposure and glycemic control in humans. Included studies report outcomes such as blood glucose levels, glucose tolerance, diabetes risk or hyperglycemia risk, where available. The specific altitude, sample size and geographic location of each study are noted. OR = odds ratio.

Study	Altitude (m) (participants, n)	Location	Glycemia indication (basal glucose, glucose tolerance, diabetes risk, hyperglycemia risk)	Chronic effect of high altitude on glycemia levels
Xu et al., 2017 1	<3500m (control) (n=907) 3500–3999m (n=600) >4000m (n=152)	Tibet	Diabetes risk <3500m: OR = 1 3500–3999m: OR = 0.35 (0.14 to 0.93) >4000m: OR = 0.11 (0.02–0.66)	↓
Woolcott et al., 2014 2	Total sample size (n=284,945) <499m (control) (83%) 500–1499m (11%) 1500–3500m (6%)	United States	Diabetes risk <499m: OR = 1 1500–3500m: OR = 0.88 (0.81 to 0.96)	↓
Calderón et al., 1965 3	Sea level (control) (n=32) 4500m (n=23) Healthy males	Peru	Basal glycemia Sea level: 81.9 mg/dL 4500m: 69.7 mg/dL IV glucose tolerance test Improved at high altitude	↓
Calderón et al., 1966 4	Sea level (control) (n=19) 4500m (n=13) Pregnant females	Peru	Basal glycemia Sea level: 85.9 mg/dL 4500m: 71.7 mg/dL IV glucose tolerance test Improved at high altitude	↓
Picon-Reategui et al., 1970 5	150m (control) (n=12) 4358m (n=12) Healthy males + 150m (starting) (n=5) 3992m (n=5) US male athletes	Peru 6 weeks at high altitude	Basal glycemia 150m: 85.3 mg/dL 4358m: 70.7 mg/dL Basal glycemia 150m: 77.0 mg/dL 3992m: 71.6 mg/dL	↓ ↓
Lindgärde et al., 2004 6	150m (control) (n=105) 3800m (n=105) Quechua females	Peru	Basal glycemia 150m: 82.8 mg/dL 3800m: 68.5 mg/dL	↓
Castillo et al., 2007 7	150m (control) (n=10) 3250m (n=10) Healthy males	Peru	Glycemia 12h-profile 150m: 74.5 mg/dL 3250: 50.6 mg/dL	↓
Lopez-Pascual et al., 2018 8	4–6m (control) (n=152) 2758–2787m (n=108) Male and female university graduates	Ecuador	Hyperglycemia risk 4–6m: OR = 1 2,758–2787m: OR = 0.25 (0.07–0.88)	↓
Stock et al., 1978 9	580m (starting) (n=6) 3650m (n=6) Healthy subjects	Italy 3 weeks at high altitude	Basal glycemia (fed) 580m: 88mg/dL 3650m: 64 mg/dL	↓
Brooks et al., 1991 10	10m (starting) (n=7) 4300m (n=7) Healthy males	United States 3 weeks at high altitude	Basal glycemia 10m: 82.4 mg/dL 4300m: 70.76 mg/dL	↓
Lecoultre et al., 2013 11	20m (starting) (n=8) 2400m (15% O2) (n=8) Obese males	United States 10 nights in hypoxic tents	Basal glycemia 20m: 94.8 mg/dL 2400m (15% O2): 91.8 mg/dL Glucose disposal rate with insulin infusion Improved in hypoxia	↓
Horscroft et al., 2017 12	35m (starting) (n=10) 5300m (n=10) Healthy lowlanders + 1300m (starting) (n=15) 5300m (n=15) Sherpas	Nepal 2–8 weeks at high altitude	Fasting glycemia Decreased at high altitude in lowlanders Unchanged at high altitude in Sherpas Oral glucose tolerance test Improved at high altitude in lowlanders Unchanged at high altitude in Sherpas	↓ for lowlanders = for Sherpas

Many physiological variables change simultaneously at high altitude, making it challenging to determine which factor is specifically responsible for improved glucose homeostasis. High-altitude studies in human populations often lack suitable controls, leaving uncertainty about cause and effect. However, multiple lines of evidence point to decreased oxygen availability (hypoxia) as the critical variable. Some of the earliest systematic investigations into this phenomenon trace back to the pioneering work of the Harvard Fatigue Laboratory between the 1920s and 1940s¹⁶, which aimed to study human physiology under extreme environmental conditions relevant for soldiers during wartime. In a series of carefully documented experiments, researchers from the Harvard Fatigue Laboratory observed improved glucose tolerance in healthy volunteers who were transported to the Chilean Andes, at elevations reaching up to 6000 meters¹⁶. These foundational observations have since been revisited and expanded in recent research. For example, our own recent studies demonstrated dramatically reduced blood glucose concentrations in mice exposed chronically to low oxygen, without overt physiological or behavioral deficits¹⁷.

Despite extensive characterization of improved glucose tolerance at altitude, the precise physiological mechanism underlying this phenomenon remains elusive. A prevailing hypothesis has been that increased peripheral glucose consumption under hypoxic conditions is responsible for this beneficial effect. Indeed, acute hypoxia rapidly upregulates glucose transporters in peripheral tissues via the Hypoxia-Inducible Factor (HIF) transcriptional response, thereby enhancing glucose uptake^18,19. However, these acute signaling responses diminish with prolonged hypoxic exposure due to a negative feedback loop²⁰, making it unlikely that such signaling alone can fully explain the sustained improvement in glycemic control observed in chronic hypoxia. Moreover, the gradual onset of improved glycemia over weeks further suggests that slower, chronic adaptations likely contribute to the observed metabolic benefits. Thus, despite numerous studies, the mechanistic basis of long-term improved glycemia at altitude has remained an enduring mystery in the field.

Red blood cells (RBCs) are the most abundant cell type in the human body, accounting for approximately 85% of all cells and contributing to about 4% of total body mass²¹. Under conditions of chronic hypoxia, RBC numbers can nearly double, occupying upwards of 70% total blood volume²². Mature RBCs are highly specialized cells that lack organelles, such as nuclei and mitochondria. These cells are primarily composed of hemoglobin and play a central role in maintaining adequate tissue oxygenation. Due to their lack of mitochondria, RBC metabolism heavily relies on anaerobic glucose fermentation for ATP production²³. Interestingly, one of the primary hemoglobin allosteric regulators under hypoxia, 2,3-diphosphoglycerate (2,3-DPG), is continuously produced within RBCs as a glycolytic intermediate via the Rapoport-Luebering shunt²⁴. Elevated levels of 2,3-DPG during hypoxia promote enhanced oxygen release from hemoglobin to tissues²⁵, an acute metabolic switch that is observed within minutes from exposure to high-altitude hypoxia and persists for weeks²⁶. These observations suggest an intimate link between RBC glucose metabolism and hypoxic physiological adaptation, raising the intriguing possibility that RBC glucose utilization might play a central role in regulating systemic glucose homeostasis under conditions of low oxygen availability.

Therefore, we hypothesized that RBCs represent the primary glucose sink under hypoxic conditions. To test this hypothesis, we systematically evaluated the relationship between RBC abundance and blood glucose across multiple mouse models in which RBC numbers were experimentally manipulated. Our results revealed that RBC abundance and their altered metabolism in hypoxia could account for most of the observed glycemic changes during hypoxia. These findings uncover an unexpected yet fundamental role for RBCs in systemic glucose homeostasis.

Results

Improved Glucose Tolerance in High-Altitude Residents Is Recapitulated in Hypoxic Mice

High-altitude environments are characterized by alterations in several environmental conditions, including lower temperatures, increased UV radiation, decreased humidity, and reduced oxygen availability^27,28. To determine whether the improved glucose tolerance observed in high-altitude residents results specifically from hypoxia, we employed a mouse model for normobaric hypoxia exposure. We housed eight-week-old male mice under either normoxic (21% O₂) or hypoxic (8% O₂, corresponding to >5,000 meters of altitude) conditions for three weeks, regularly monitoring blood glucose and body weight. Mice are known to undergo acute stress during the first hours/days of 8% O₂ exposure (i.e. temporary immobility, food intake suppression, body weight loss), but this stress resolves within one week of acclimatization¹⁷. We observed that basal blood glucose levels significantly decreased in hypoxic mice starting from day 2 of exposure, coinciding with acute weight loss that stabilized within one week (Fig 1A, S1A). This early reduction in glucose was not explained by reduced food intake, as pair-fed normoxic mice did not show any significant changes in glycemia after three days compared to hypoxic mice (Fig S1B). The hypoglycemia effects were pronounced after the first week of hypoxia exposure (Fig 1A), when acute stress responses (behavior, food intake, etc.) are already normalized¹⁷. Furthermore, we found glucose tolerance to be substantially improved at all hypoxia timepoints tested (1, 2, and 3 weeks) (Fig 1B–C). These results indicate that hypoxia alone recapitulates the enhanced glucose tolerance previously observed in human high-altitude residents.

Figure 1. Hypoxia leads to improved glucose tolerance, which is not fully explained by increased glucose uptake by internal organs.

(A-E) Mice were exposed to hypoxia (8% O₂, n=8) or normoxia (21% O₂, n=7), followed by reoxygenation, indicated by the orange dashed line at day 23. Displayed readouts from these mice are: (A) tail vein blood glucose levels over time (mean ± SD), (B) a representative glucose tolerance test (GTT) (mean ± SD), (C) fold change of GTTs over time (mean ± SEM), (D) representative insulin tolerance test (ITT) (mean ± SD) and (E) fold change of ITTs over time (mean ± SEM). Representative GTT and ITT were performed two weeks after hypoxia. Area under the curve (AUC) values from individual curves. Unpaired two-tailed Student’s t-test. Fold change GTT and ITT was analyzed by Sidak’s multiple comparisons test. (F-G) Fasting blood glucose (F) and GTT with AUC quantification (top right) (G) after 1 week of hypoxia (11 and 8% O₂) or normoxia (21% O₂). Merging of two experiments: one including 21% and 8% O₂ exposure (n=7 and n=8, respectively) and other including 21% and 11% O₂ exposure (n=7 and n=7, respectively). Total sample sizes of n=14, n=7 and n=8 (for 21%, 11 and 8% O₂, respectively). Mean ± SD. One-way ANOVA, Dunnett’s multiple comparisons test. (H-I) Mice were exposed to intermittent hypoxia (8% O₂ for 8 hours, 21% O₂ for 16 hours, n=8), continuous hypoxia (8% O₂, n=8), or normoxia (21% O₂, n=8). Displayed readouts for these mice are: (H) tail vein blood glucose levels over time (mean ± SD) and (I) GTT after one week of oxygen treatment (Mean ± SD) with AUC quantification (top right). (J) Representative ¹⁸F-FDG PET/CT scans showing ¹⁸F-FDG signal accumulation in tissues after three weeks of hypoxia (8% O₂, n=5) or normoxia (21% O₂, n=8). Pie chart displaying the relative contributions of individual organs to the hypoxia-dependent increase in glucose uptake.

*p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001.

Next, to better understand the origin and kinetics of the improved glucose tolerance, we returned mice previously adapted to hypoxia back to normoxic conditions. We regularly monitored blood glucose and body weight to determine when glycemia normalized. Basal blood glucose returned to normoxic levels after 14 days of reoxygenation (Fig 1A), while glucose tolerance did not fully normalize until more than one month after returning to normoxia (Fig 1B–C, S1C). These results suggest that the mechanisms responsible for improved glucose tolerance under hypoxic conditions do not require continued hypoxia exposure and persist for weeks after hypoxia cessation.

To determine whether improved glucose tolerance was insulin-dependent, we performed insulin tolerance tests during hypoxia and reoxygenation. Insulin sensitivity declined during hypoxia and gradually recovered with reoxygenation, normalizing more rapidly than glucose tolerance (Fig. 1D–E, S1D). As hypoglycemia induces counter-regulatory responses, the reduced insulin effect likely reflects systemic adaptation rather than intrinsic insulin resistance. These findings suggest that glucose tolerance improvements under hypoxia occur independently of enhanced insulin signaling, in agreement with previous data showing that chronic hypoxia reduces circulating insulin levels¹⁷. Thus, increased glucose clearance in chronic hypoxia is insulin-independent.

To test whether more subtle hypoxic interventions could similarly improve glucose tolerance, we examined the effects on glucose homeostasis of (i) milder hypoxic regimes and (ii) intermittent hypoxia. We exposed mice to normoxia (21% O₂), moderate hypoxia (11% O₂) or severe hypoxia (8% O₂) for one week and measured both their fasting blood glucose and glucose tolerance. Interestingly, moderate hypoxia (11% O₂) was sufficient to sharply decrease fasting blood glucose (Fig 1F) and promote a dramatic improvement in glucose tolerance (Fig 1G). Notably, both fasting blood glucose and glucose tolerance exhibited an oxygen-dose dependent effect. The degree of hypoxic adaptation is known to be dictated by oxygen availability, as shown by the hematocrit boost induced by these two hypoxic regimes (Fig S1E). The extent of hypoglycemia was directly correlated with the hematocrit increase (Fig S1F), highlighting the translatability of our experimental model to milder hypoxic conditions.

For the intermittent hypoxia model, we housed eight-week-old WT male mice continuously at 21% O₂ (continuous normoxia), continuously at 8% O₂ (continuous hypoxia), or intermittently cycling between 21% and 8% O₂ (intermittent hypoxia) for 25 days. The intermittent hypoxia protocol consisted of daily cycles of: 8 hours of hypoxia exposure during the mice’s sleeping period and 16 hours of normoxia exposure during their awake period (Fig S1G). Intermittent hypoxia moderately reduced basal blood glucose (Fig 1H) without affecting body weight (Fig S1H), indicating that food intake was likely not impaired by sleeping in hypoxia. Similarly, body temperature did not acutely decrease under intermittent hypoxia and even showed a tendency to increase during chronic exposure (Fig S1I). Glucose tolerance tests performed at days 7, 14, and 25 revealed a moderate yet consistent improvement throughout the intervention period (Fig 1I and S1J). Taken together, these data suggest that even milder and potentially translational hypoxic interventions, such as intermittent hypoxia therapy, can moderately enhance glucose tolerance.

So far, all our data was pointing at enhanced glucose clearance as the main explanation for hypoxic hypoglycemia, but as a sanity check we assessed the contribution of glucose production pathways to this phenomenon. To assess the role of gluconeogenesis in hypoxic hypoglycemia, we conducted a pyruvate tolerance test, showing that hepatic gluconeogenesis does not significantly contribute to lowered blood glucose during hypoxia (Fig S2A). In summary, increased glucose clearance is the most plausible explanation for the improved glucose tolerance in hypoxia.

All the above glucose measurements relied on a handheld glucose meter (glucometer). Given the reported technical artifacts of some glucometers when testing blood across wide hematocrit ranges^29–31, we decided to analyze the plasma fraction alone using LC-MS as an orthogonal method. Plasma glucose levels in mice adapted to hypoxia (8% O₂) for three weeks showed a 35% decrease compared to normoxic mice, closely matching the reduction previously observed using our glucose meter on whole blood (Fig S2B). Additionally, we conducted detailed comparisons of glucose concentrations measured by LC-MS (on plasma) and glucose meters (on whole blood or plasma) using matched normoxic and hypoxic samples (Fig. S2C–D). These analyses revealed a strong correlation between LC-MS and glucose meter quantifications across oxygen tensions. Notably, plasma-based glucose meter readings demonstrated particularly high accuracy (Fig. S2D), leading us to adopt this method as our gold standard for glycemia quantification. Collectively, these results validate the accuracy of our glucose measurements and support the reliability of our experimental approach.

Improved Glucose Tolerance in Hypoxia Is Not Explained by Increased Glucose Uptake by Internal Organs, Suggesting a Missing Glucose Sink

To determine the primary site of glucose clearance under hypoxic conditions, we analyzed previously performed PET/CT scans to measure glucose uptake using the tracer 2-deoxy-2-[¹⁸F] fluoro-D-glucose (FDG)¹⁷. ¹⁸F-FDG was injected into mice housed at either normoxia (21% O₂) or hypoxia (8% O₂) for three weeks. Cellular uptake of radioactive ¹⁸F-FDG served as a surrogate marker for glucose uptake (Fig 1J). To evaluate the contribution of the main organs experiencing changes in glucose uptake upon hypoxia exposure¹⁷, we quantified the total increase in ¹⁸F-FDG accumulation in hypoxia and within these organs and the whole body. Surprisingly, we found that internal organs did not account for the majority of hypoxia-driven glucose uptake, with ~70% of the increase remaining unexplained (Fig 1J). These results indicated that the primary glucose sink responsible for enhanced glucose consumption under hypoxic conditions remained unidentified and motivated us to broaden our search space.

Erythrocytosis is Both Necessary and Sufficient to Improve Glycemic Control

RBCs are the most abundant cell type in the body, and their total mass nearly doubles after four weeks of adaptation to 8% O₂ exposure (Fig S3A), a phenomenon known as erythrocytosis. Because RBCs lack mitochondria, they primarily metabolize glucose through the Embden-Meyerhof-Parnas glycolytic pathway and the pentose phosphate pathway (PPP)²³. Previous reports indicate that hypoxic conditions promote glycolytic flux in RBCs over PPP activity^32,33, with relative fluxes of 90% and 10%, respectively³⁴. Therefore, we hypothesized that the combined effects of increased RBC numbers and enhanced glycolysis in RBCs might account for most of the glucose uptake observed under hypoxic conditions.

To directly test if RBCs are the main glucose sink during hypoxia, we reversed erythrocytosis in hypoxic mice via serial phlebotomy and investigated whether this intervention could mitigate the observed decrease in blood glucose (Fig 2A). Specifically, we extracted 15% of total blood volume (TBV) every three days from eight-week-old male mice housed under normoxic or hypoxic conditions over four weeks. Hematological and plasma glucose analyses were performed on these mice and compared to control normoxic and hypoxic mice that had never been subjected to phlebotomy. Phlebotomized hypoxic mice achieved hematocrits comparable to normoxic controls (Fig 2B, S3B). Phlebotomy did not significantly alter other blood parameters (Fig S3C). Importantly, this normalization of hematocrit significantly normalized the hypoxic hypoglycemia observed in non-bloodlet mice (Fig 2C–E), producing a stable glycemic plateau around ~170 mg/dL. Moreover, glucose tolerance tests performed after four weeks of hypoxia revealed that phlebotomy substantially diminished the improvement in glucose tolerance observed in hypoxic mice (Fig 2F–G). Glucose transporters in peripheral tissues were not significantly altered between phlebotomized and control hypoxic mice (Fig S3D–E), ruling out the contribution of other organs to this effect. Collectively, these data demonstrate that RBC depletion largely normalizes glycemia in hypoxic mice, strongly supporting erythrocytosis as necessary for the enhanced glucose tolerance observed in hypoxia.

Figure 2. Erythrocytosis is necessary and sufficient to explain hypoglycemia in hypoxia.

(A-E) Phlebotomy model, (A) experiment schematic. RBCs were depleted by serial phlebotomy in hypoxic (8% O₂, n=8) and normoxic (21% O₂, n=8) mice. A volume equivalent to 15% of total blood volume (TBV) was removed every three days. Control groups consisted of non-phlebotomized hypoxic (n=5) and normoxic (n=5) mice. Displayed readouts from these mice are: (B) RBC counts over time (mean ± SD); two-way ANOVA was used, (C) plasma glucose levels over time (mean ± SD); two-way ANOVA was used, (D) endpoint RBC counts after four weeks; Sidak’s multiple comparisons test, and (E) endpoint plasma glucose levels after four weeks; Sidak’s multiple comparisons test. (F) GTT after four weeks in hypoxic and normoxic mice with or without phlebotomy (n=5–10 per group). Mean ± SD are shown. (G) AUC values from individual GTT curves analyzed by unpaired two-tailed Student’s t-tests within each oxygen condition. (H-K) RBC transfusion model, (H) experiment schematic. RBCs were isolated from hypoxic (8% O₂) or normoxic (21% O₂) donor mice and transfused into normoxic recipients (n=10 per group). An additional control group received vehicle (saline, n=13). Packed RBCs (75%) were administered twice daily for two consecutive days to rapidly elevate RBC counts. Displayed readouts from these mice are: (I) endpoint RBC counts from two experiments; ordinary one-way ANOVA was used, (J) blood glucose over one single RBC transfusion experiment (mean ± SD); two-way ANOVA was used, and (K) endpoint plasma glucose levels from two experiments; ordinary one-way ANOVA was used. For reference, values from mice exposed to two weeks of hypoxia (8% O₂) or normoxia (21% O₂) were included in I and K.

*p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001. (comparisons between phlebotomized and control at 8% O₂).

#p < 0.05, ##p < 0.01, ###p < 0.001, ####p < 0.0001 (comparisons between phlebotomized and control at 21% O₂).

To directly assess whether increased RBC abundance is sufficient to induce hypoglycemia comparable to that observed under hypoxic conditions, we performed RBC transfusions (Fig. 2H). Male mice maintained under normoxic conditions received two retro-orbital injections per day for two consecutive days, with either vehicle or 75%-packed RBCs from donor mice housed chronically in either normoxia or hypoxia. Hematological and plasma glucose measurements were conducted one day after the final transfusion. RBC transfusion resulted in significant and comparable increases in RBC counts relative to vehicle-treated controls (Fig. 2I, Fig. S3F). These increases closely mirrored the elevation seen in mice exposed to two weeks of hypoxia. As expected, some RBC parameters, such as mean corpuscular volume (MCV), differed slightly depending on donor oxygenation status, reflecting intrinsic differences between normoxic and hypoxic RBCs (Fig. S3F), while other hematological parameters remained unchanged (Fig. S3G). Both transfusion groups exhibited marked hypoglycemia, with a slightly greater effect observed in recipients of hypoxic RBCs (Fig. 2J). Endpoint plasma glucose levels were reduced in mice receiving normoxic and hypoxic RBCs, respectively (Fig. 2K); comparable to the reduction observed after two weeks of hypoxia. Glucose transporters in peripheral tissues were not significantly altered between RBC-transfused and saline-injected mice (Fig S3H–I). Collectively, these findings demonstrate that elevated RBC numbers are sufficient to drive hypoglycemia, supporting the hypothesis that RBCs function as a major glucose sink under hypoxic conditions.

RBCs from Hypoxic Mice Have Increased Glucose Uptake

We next asked whether the observed glucose-lowering effect could be fully explained by the increased RBC count under hypoxia, or if hypoxia additionally enhances glucose uptake per RBC. To address this, we performed experiments using a non-radioactive glucose analog, 2-deoxy-D-glucose (U-¹³C) [2DG (U-¹³C)], as a tracer (Fig 3A). Like ¹⁸F-FDG, 2DG (U-¹³C) is imported into cells through glucose transporters and becomes trapped intracellularly upon phosphorylation, accumulating as 2DG-P (U-¹³C). As this molecule cannot be further metabolized through glycolysis, its accumulation serves as a surrogate marker for glucose uptake. We administered 2DG (U-¹³C) retro-orbitally to mice housed under normoxic or hypoxic conditions for three weeks, collecting blood at 2-, 10-, 30-, and 120-minutes post-injection. Plasma 2DG (U-¹³C) concentrations, measured by LC-MS, exhibited similar decay kinetics in both groups, indicating comparable tracer clearance from plasma (Fig 3B). However, the accumulation of 2DG-P (U-¹³C) within RBCs occurred much more rapidly in hypoxic mice, reflecting enhanced glucose uptake per RBC (Fig 3B, S4A). This phenomenon was even more pronounced when factoring in the increased RBC pool under hypoxic conditions. These results demonstrate that hypoxia increases glucose uptake per RBC in vivo, further contributing to the improved glucose tolerance observed during hypoxic exposure.

Figure 3. RBC glucose uptake multiplies 3-fold in hypoxia, primarily due to their newly matured RBC population.

(A-B) In vivo glucose uptake experiment, (A) experiment schematic. Mice exposed to hypoxia (8% O₂) or normoxia (21% O₂) for three weeks received a bolus of uniformly ¹³C-labeled 2-deoxy-D-glucose [2DG (U-¹³C], 1 g/kg body weight via the retro-orbital route. Blood samples were collected at 2-, 10-, 30-, and 120-minutes post-injection. Plasma and RBCs samples were analyzed by LC-MS, specifically: (B) plasma 2DG levels (left), 2DG-P per RBC over time (central) and estimated 2DG-P in all the RBC pool over time (right). (C-D) Ex vivo glucose uptake experiment: (C) experiment schematic. RBCs isolated from mice exposed to hypoxia (8% O₂) or normoxia (21% O₂) for four weeks were incubated in a solution of known glucose concentration. After incubation, remaining glucose was measured to estimate glucose uptake per RBC per sample. (D) Ex vivo glucose uptake per RBC. (E-F) Fold change of median FITC signal of GLUT1 (E) and GLUT4 (F) in RBCs from mice exposed to hypoxia (8% O₂, n=5) or normoxia (21% O2, n=5) for three weeks (left) and histograms of FITC signal from hypoxic or normoxic RBCs, including a no primary antibody control (right). Black: no primary antibody with normoxic (first) or hypoxic RBCs (second); gray: normoxia biological replicates; blue: hypoxia biological replicates. (G-J) Biotin labeling: (G) experiment schematic. Pre-existing RBCs were labeled by a biotin pulse and after four weeks of normoxia (21% O₂, n=3) or hypoxia (8% O₂, n=3) exposure RBCs were analyzed by flow cytometry. Displayed readouts from these mice are: (H) representative histograms showing FITC-A signal (corresponding to GLUT1 protein abundance) in New (APC negative) vs. Old (APC positive) RBCs per mouse, (I) median FITC-A signal and (J) working model for GLUT1 protein upregulation in newly synthesized RBCs under hypoxia.

*p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001.

As an orthogonal approach, we further evaluated RBC glucose uptake using an ex vivo system. Blood was collected from mice exposed to normoxia or hypoxia for four weeks, and RBCs were isolated by centrifugation of whole blood. Identical volumes of RBCs from normoxic or hypoxic mice were incubated ex vivo with equal volumes of plasma (1:1 RBC-to-plasma ratio), standardized to an identical initial glucose concentration (Fig 3C). After 10 minutes of incubation at 37°C under normoxic conditions, we measured residual glucose in the plasma to determine glucose uptake. Hematology analyses were performed concurrently to accurately quantify the RBC number in each sample (Fig S4B). RBCs derived from hypoxic mice exhibited a 2.5-fold increase in glucose uptake per cell compared to normoxic controls (Fig 3D). Importantly, because this experiment was conducted entirely under normoxic conditions, we conclude that the enhanced glucose uptake observed in hypoxia-derived RBCs reflects a sustained functional adaptation rather than a hyperacute response to low oxygen availability. Thus, our ex vivo data further substantiates that chronic hypoxia induces a long-lasting increase in glucose uptake capacity per individual RBC.

To assess whether the increase in glucose uptake per cell in hypoxia is mediated by the upregulation of glucose transporters in RBCs, we analyzed their protein abundance using flow cytometry (Fig S4C). In adult RBCs, GLUT1 serves as the primary glucose transporter, although GLUT4 expression can also be upregulated during periods of increased erythropoiesis^35,36. Thus, we analyzed the ‘per cell’ protein abundance of both GLUT1 and GLUT4 in RBCs from mice exposed to normoxia or hypoxia for three weeks (Fig. 3E–F). Glycolytic enzymes HK1 and GAPDH were additionally included in the analysis (Fig. S4D). Exposure to hypoxia led to a robust 60% and 48% upregulation of GLUT1 and GLUT4 abundance in RBCs (Fig. 3E–F), respectively, whose enlarged width distribution implies augmented cell heterogeneity under hypoxia. Notably, GAPDH protein abundance was also increased in hypoxic RBCs, while HK1 remained unchanged (Fig. S4D). Altogether, these results indicate that the augmented glucose uptake observed in hypoxic RBCs is likely driven by increased levels of these glucose transporters.

Possible explanations for increased GLUT1 and GLUT4 abundance include (1) a shift in distribution of RBCs towards a younger population due to enhanced erythropoiesis, (2) a transcriptional or translational upregulation of these transporters during erythroid maturation and/or (3) decreased transporter degradation in mature RBCs. Since mature RBCs are unable to synthesize new proteins, glucose transporters abundance in older RBCs must be decreased³⁷. To assess whether newly synthesized RBCs in hypoxia have increased protein abundance of glucose transporters, we performed a biotin-tracing experiment^38,39 (Fig 3G). Briefly, mice were injected with biotin for three consecutive days to label existing RBCs. After biotin treatment, they were housed in normoxia or hypoxia, resulting in the maturation of unlabeled new RBCs. After four weeks of hypoxic exposure, we analyzed the abundance of GLUT1 in their biotin-positive (“Old”) and biotin-negative (“New”) RBCs, finding that the newly synthesized RBCs accounted for GLUT1 upregulation (Fig 3H–I). Thus, we have shown that glucose uptake upregulation in hypoxic RBCs is at least partially mediated by the newly matured RBCs under chronic low oxygen conditions (Fig 3J).

Hypoxia Boosts RBC Glycolytic Flux for Increased Production of the Allosteric Hemoglobin Regulator 2,3-DPG

We next sought to determine the metabolic fate of this enhanced glucose influx. To address this, we performed an in vivo tracing experiment using uniformly labeled D-glucose (U-¹³C). We injected glucose (U-¹³C) retro-orbitally into mice exposed to normoxia or hypoxia for three weeks and collected blood at 2-, 10-, 30-, and 120-minutes post-injection (Fig 4A). Plasma glucose (U-¹³C) levels over time and RBC metabolites were analyzed by LC-MS. As expected, labeled glucose in plasma decayed promptly over time (Fig 4B), implying rapid disposal by cells. To profile glucose fates in RBCs under hypoxic conditions comprehensively, we conducted semi-targeted metabolomics (Fig 4C). This analysis revealed that the labeling of 2,3-DPG—the primary product of the LR pathway—and glycolytic intermediates increased more rapidly in RBCs from hypoxic mice (Fig 4C), in agreement with previous results^40–42. Some other metabolites whose labeling from glucose increased in hypoxia were hexose 1,6-bisphosphate, while others like hexose (glucose and hexose isomers) were decreased.

Figure 4. Glycolytic flux is increased in hypoxic RBCs to build hemoglobin allosteric regulator 2,3-DPG.

(A-F) Glucose tracer experiment, (A) experiment schematic. Mice exposed to hypoxia (8% O₂, n=5) or normoxia (21% O₂, n=5) for three weeks received a bolus of uniformly labeled glucose (U-¹³C) (1 g/kg). Blood was collected at multiple time points post-injection, and plasma and RBCs were isolated and stored at −80°C for downstream analysis. Displayed readouts from these mice include: (B) plasma fractional glucose labeling over time, (C) volcano plots of the fractional labeling delta (Δ (%Labeling8% - %Labeling21%) and their p-value after unpaired Student’s t-test analysis at 2-minutes post-injection; glycolytic (blue) and PPP (brown) intermediates are highlighted in the graph, (D) targeted metabolomics heatmap of the main metabolites from glycolysis, LR and PPP; statistical significance is depicted on the right, (E) 2,3-DPG fractional labeling in RBCs over time and (F) unlabeled and fully labeled 2,3-DPG at the earliest timepoint −2 minutes post-injection- as a qualitative indication for 2,3-DPG pool size. Val-d8 peak area was used for normalization. Uncorrected Fisher’s LSD multiple comparisons were used.

Targeted quantification of glycolytic, LR, and PPP intermediates confirmed a marked accumulation of labeled 2,3-DPG, along with other glycolytic intermediates (Fig. 4D). Importantly, 2,3-DPG serves as a key allosteric regulator of hemoglobin, facilitating enhanced oxygen release to hypoxic tissues²⁵. Fractional labeling analyses over time revealed a ~3.5-fold increase in the rate constant of 2,3-DPG labeling in hypoxic versus normoxic RBCs (Fig. 4E). Precise flux calculations would require absolute pool size quantification and a non-perturbative infusion. However, both the relative pool size and the fractional labeling of 2,3-DPG were substantially higher in RBCs from hypoxic mice (Fig. 4F), indicating a qualitatively higher metabolic flux of glucose into the LR pathway. In summary, our data demonstrate that hypoxia markedly enhances in vivo glucose utilization by RBCs, directing it toward the synthesis of 2,3-DPG, a hemoglobin modulator, as well as other glycolytic intermediates.

Hypoxia Rearranges RBC-Specific Glycolytic Metabolon

To investigate how oxygen tension regulates glucose metabolism in RBCs, we tested whether these cells undergo acute, oxygen-dependent metabolic reprogramming that could explain the rapid stimulation of glycolysis despite lacking transcriptional or translational machinery. Prior work has proposed that under normoxia, glycolytic enzymes cluster in an inactive state at the plasma membrane by binding the N-terminus of Band 3, the most abundant integral RBC membrane protein. In hypoxia, deoxyhemoglobin is thought to displace these interactions, releasing glycolytic enzymes such as GAPDH into the cytosol and thereby promoting glycolytic flux. To assess this mechanism in our system, we performed GAPDH immunofluorescence in oxygenated and deoxygenated RBCs. In both human and murine RBCs, GAPDH localized to the plasma membrane under normoxia but redistributed to the cytosol under hypoxia (Fig 5A–B and S5A–B).

Figure 5. Oxygen-dependent rearrangement of Band3 glycolytic metabolon to explain the shift in glycolytic flux.

(A-B) Z-stack immunofluorescence of GAPDH in RBCs with oxygenated (Oxy. Hb) or deoxygenated (Deoxy. Hb) hemoglobin, using (A) mouse or (B) human RBCs. (C) Proximity Ligation Assay (PLA) quantification using oxygenated or deoxygenated mouse (left) and human (right) RBCs. (D-E) STED microscopy analysis of GAPDH and Band 3 interaction using human oxygenated and deoxygenated RBCs including: (D) representative images and (E) colocalization quantification. (F) Isothermal titration calorimetry using the cytosolic N-terminus of Band 3 (residues 1–395) and recombinant GAPDH. (G) Crosslinking proteomics were performed by incubating human recombinant GAPDH with the N-terminal cytosolic domain of Band 3 (residues 1–395) in the presence of purified human hemoglobin under oxygenated or deoxygenated conditions. Quantitative analysis was enabled using TMT sixplex labeling to compare the relative abundance of crosslinks introduced by either DMTMM or DSSO. A heat map displays the top 50 crosslink pairs ranked by t-test (normoxia vs. hypoxia), highlighting SLC4A1 (Band 3)–GAPDH crosslinks enriched under normoxia and SLC4A1–HBB crosslinks enriched under hypoxia, also shown as bar graphs. A circos plot provides a global view of all identified crosslinks, color-coded by fold-change (normoxia/hypoxia) and with edge thickness proportional to crosslink score.

*p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001.

We next tested the binding dynamics of GAPDH and Band 3 using a proximity ligation assay (PLA). This revealed a significant reduction in GAPDH–Band 3 interactions under hypoxia in both species (Fig 5C and S5C–D). Moreover, to determine colocalization differences with higher spatial resolution, we performed STED microscopy in human oxygenated and deoxygenated RBCs, finding that GAPDH~Band 3 colocalization dramatically decreases under hypoxic conditions (Fig 5D–E).

To directly probe this interaction, we used isothermal titration calorimetry (ITC) with recombinantly-expressed human GAPDH and the N-terminal cytosolic domain of Band 3 (residues 1–395). GAPDH bound this domain strongly (Kd = 1.18 μM; Fig. 5F). Notably, this affinity was higher than previously reported for the shorter 56–amino acid N-terminal peptide⁴³, suggesting additional residues beyond the intrinsically disordered extreme N-terminus contribute to binding.

To further investigate the structural underpinning of this interaction, we performed quantitative TMT labeling-based crosslinking proteomics incubating recombinant GAPDH with Band 3_1–395 in the presence of oxygenated or deoxygenated hemoglobin (Fig 5G). Deoxyhemoglobin –particularly β-globin (HBB) at residue 83– showed increased interaction with Band 3 at residue 212 (2.4-fold enrichment in hypoxia). This interaction displaced GAPDH binding at Band 3 residue 218. This analysis expanded the understanding of Band 3 interactions with GAPDH. Specifically, Band 3 residues 33 and 40 crosslinked to GAPDH residues 107 and 84, respectively, while residues 122, 166, 168, 205, 218, and 254 of Band 3 interacted with GAPDH residues 106, 172, 189, 254, and 256. These regions are proximal to the GAPDH active site (C152 and H179), providing a structural explanation for how Band 3 binding inhibits GAPDH activity.

Together, these data support the oxygen-dependent rearrangement of the RBC glycolytic metabolon: under normoxia, GAPDH is sequestered at the plasma membrane by Band 3 binding, which dampens glycolytic flux. Under hypoxia, deoxyhemoglobin competes for Band 3 binding, releasing GAPDH into the cytosol and enabling glycolysis. This conserved mechanism was observed in both mouse and human RBCs (Fig S5E).

Hypoxia and RBC transfusions Rescue Hyperglycemia in Type 1 and Type 2 Diabetes Models

To investigate whether the hypoxia-induced increase in RBC number and metabolic hypoxic rewiring could serve as a therapeutic approach for hyperglycemia, we tested the efficacy of hypoxia and increased hematocrit in treating Type 1 and 2 diabetes.

First, we evaluated hypoxia’s potential as a therapy for hyperglycemia induced by Type 1 diabetes (Fig 6A). We injected 8-week-old male mice intraperitoneally with either vehicle or streptozotocin (STZ), a compound that leads to loss of insulin-producing pancreatic β-cells⁴⁴, causing insulin deficiency and hyperglycemia (Fig S6A). To better replicate the clinical timeline, hypoxia therapy was initiated in diabetic mice only after the “diagnosis of diabetes” was confirmed, two weeks after the final STZ administration. STZ- and vehicle-treated mice were then randomized into normoxia or hypoxia therapy conditions for three weeks. Hypoxic exposure induced erythrocytosis in both STZ-treated and vehicle-treated mice (Fig 6B, S6B), whereas RBC counts remained unchanged across normoxic conditions. Basal blood glucose levels and body weight were monitored every three days throughout the study period (Fig 6C and S6C). Notably, hypoxic STZ-treated mice exhibited a dramatic reduction in their initially elevated glycemia (Fig 6C, S6D). Glucose tolerance tests performed after three weeks demonstrated that hypoxia completely rescued impaired glucose tolerance caused by STZ treatment (Fig 6D and S6E), highlighting again the insulin independence of this phenomenon and the therapeutic potential of our intervention.

Figure 6. Erythrocytosis, induced by orthogonal approaches, ameliorates STZ- and HFD-induced hyperglycemia.

(A-G) Type I Diabetes Model, (A) experiment scheme. Type I diabetes was induced in male mice via five consecutive daily i.p. injections of streptozotocin (STZ) or vehicle. Two weeks post-injection, hyperglycemia was confirmed, and mice were randomized to hypoxia (8% O₂, n=8) or normoxia (21% O₂, n=8) for three weeks or to RBC transfusions (n=5) or saline injections (n=5). Displayed readouts from hypoxic vs. normoxic mice are: (B) endpoint RBC number; Sidak’s multiple comparison test, (C) tail vein blood glucose over time (mean ± SD), and (D) endpoint GTT. Displayed readouts for transfused vs. control mice include: (E) endpoint RBC number (F) blood glucose over time and (G) slope of glycemia change per individual mouse. (H-M) Type II Diabetes Model, (H) experiment scheme. Male C57BL/6J mice fed either a high-fat diet (HFD) or standard chow diet (CD) were treated daily with HypoxyStat (600 mg/kg) or vehicle for 2.5 weeks. Briefly, HypoxyStat increases hemoglobin’s oxygen affinity, limiting oxygen release and inducing tissue hypoxia. HFD-fed mice received HypoxyStat (n=5) or vehicle (n=6), and CD-fed mice received HypoxyStat (n=8) or vehicle (n=8). Displayed readouts from these mice are: (I) tail vein blood glucose, (J) endpoint RBC count; (K) endpoint GTT (mean ± SD) and (L) AUC values from individual GTT curves; Sidak’s multiple comparison test.

*p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001.

Second, to avoid the potential pleiotropic effects of hypoxia therapy, we assessed whether hypoxic RBC transfusions were sufficient to ameliorate hyperglycemia in diabetic mice (Fig 6A, S6F). Vehicle- and STZ-treated mice were transfused with hypoxic RBCs or an equivalent volume of saline for two consecutive days and their glycemia was assessed longitudinally. RBC transfusions successfully increased the RBC count of mice compared to mice injected with saline (Fig 6E and S6G). Interestingly, all mice transfused with RBCs presented a drop in their glycemia (Fig 6F–G), including the STZ-induced diabetic mice, demonstrating that the increase of RBC number is sufficient to ameliorate hyperglycemia in the context of Type 1 diabetes.

Finally, we evaluated whether HypoxyStat, a small-molecule compound we recently developed to pharmacologically cause hypoxia⁴⁵, could effectively reverse hyperglycemia induced by high-fat diet (HFD) (Fig 6H). Briefly, HypoxyStat increases hemoglobin’s oxygen affinity, limiting oxygen offloading to the tissues and inducing local tissue hypoxia⁴⁵. Male mice fed either chow diet (CD) or HFD received daily doses of vehicle or HypoxyStat via oral gavage, all while being housed under normoxic conditions. As expected, vehicle-treated HFD-fed mice developed prominent hyperglycemia compared to CD-fed controls (Fig 6I and S6H). Notably, HypoxyStat treatment completely abolished HFD-induced hyperglycemia (Fig 6I and S6H), coinciding with a robust induction of erythrocytosis (Fig 6J, S6I). Importantly, no other hematological parameters were affected (Fig S6J). Glucose tolerance was significantly impaired in vehicle-treated HFD-fed mice but fully normalized upon treatment with HypoxyStat, irrespective of dietary fat content (Fig 6K–L). Collectively, these findings demonstrate that HypoxyStat, our recently developed pharmacological hypoxia mimetic, potently rescues hyperglycemia and impaired glucose tolerance associated with diet-induced obesity.

Discussion

Reduced glycemia and improved glucose tolerance have long been observed among high-altitude populations^1–12 and across diverse organisms^13–15, yet the physiological mechanisms underlying these effects have remained unclear. Here, we identify RBCs as the primary glucose sink during hypoxia (Fig S7). Hypoxia significantly increases RBC number and these new RBCs have increased glucose uptake. We demonstrate that depletion of RBCs in hypoxic mice normalizes blood glucose, while artificially increasing RBC number under normoxia is sufficient to induce hypoglycemia. The increase in RBC number is likely the main mechanism explaining increased glucose clearance in chronic hypoxia. However, we also find that hypoxia boosts glucose uptake per individual RBC, at least in part by the upregulation of GLUT1 and GLUT4 protein abundance in newly synthesized RBCs. This increased glucose influx enables the rapid accumulation of the hemoglobin allosteric regulator 2,3-diphosphoglycerate (2,3-DPG), a critical molecule for oxygen offloading and hypoxia adaptation²⁵. These findings highlight a previously unrecognized physiological mechanism and could inspire innovative therapeutic strategies for treating hyperglycemia.

We have shown that the augmented glucose uptake capacity of younger RBCs in chronic hypoxia is coincident with increased GLUT1 and GLUT4 protein abundance per RBC. GLUT1 is the primary glucose transporter in mature RBCs^46,47, although some studies have indicated expression of alternative glucose transporters, such as GLUT4³⁶ and GLUT3⁴⁸. Given that mature RBCs are enucleated and thus unable to synthesize new proteins, their glucose uptake capacity largely reflects passive factors established at earlier developmental stages. These include GLUT1 and GLUT4 expression levels determined at the progenitor cell stage³⁷. GLUT1 expression in erythroid progenitors is known to be regulated by the hypoxia-inducible factor (HIF) pathway^49–51, which becomes activated during erythropoiesis and under chronic hypoxic conditions (erythropoietin peaks after 24–48h from initial hypoxic exposure⁵²). Because the half-life of GLUT proteins in RBCs is thought to approach the RBC lifespan itself, increased glucose uptake observed under hypoxia may be largely attributable to an expanded pool of younger RBCs, which retain higher GLUT expression established during their recent erythropoietic development³⁷. This is consistent with the notion that acclimatization involves long-term retention of RBC metabolic adaptations, as observed upon repeated ascent to high altitude. Future work will dissect the contribution of RBC age and the HIF pathway on GLUT1 and GLUT4 expression leading to improved glycemia during systemic hypoxia.

Beyond enhanced glucose uptake, we observed a marked acceleration in the conversion of glucose to 2,3-DPG and other glycolytic intermediates during hypoxia. These acute responses occur in the minute scale (rather than days/weeks) and are thus insufficiently explained by de novo erythropoiesis and transcriptional reprogramming of younger RBCs generated in hypoxia. In this context, the dynamic compartmentalization of glycolytic enzymes mediated by Band 3 in response to oxygen may be critical^53–55. Under normoxic conditions, Band 3 (an abundant RBC plasma membrane protein) sequesters glycolytic enzymes, thereby promoting flux through the PPP to support NADPH production and antioxidant defense mechanisms. In contrast, under hypoxia, deoxygenated hemoglobin displaces these enzymes from Band 3, thereby enhancing glycolytic flux and favoring the accumulation of 2,3-DPG. We have shown that this mechanism is in fact relevant in our mouse experimental set up and that it is further conserved in humans.

Our findings strongly support the idea that hypoxic hypoglycemia primarily results from accelerated and enhanced glucose uptake by RBCs. Nevertheless, changes in gluconeogenesis, glycogenolysis, or glucose absorption and secretion might also contribute to systemic hypoglycemia. Preliminary data from pyruvate tolerance tests, which serve as a proxy for hepatic gluconeogenic capacity, indicated no significant differences between oxygen conditions. However, further investigation into additional pathways, including glycogen breakdown and intestinal glucose absorption, is warranted. This will enable a comprehensive understanding of hypoxia’s influence on glucose metabolism beyond our observed effects on glucose uptake.

Our data also offer insights into broader population-level observations, clarifying previously unexplained differences in glycemic control among high-altitude residents^1–8,56. Most high-altitude populations consistently demonstrate improved glucose tolerance; however, Sherpas represent a notable exception, as they do not exhibit improved glycemic control despite chronic high-altitude exposure^12,22,57. Genetically, Sherpas carry hypoxia-adaptive variants in HIF2 (EPAS1) and related genes that blunt hypoxia-induced erythropoiesis, resulting in unusually low hematocrit levels compared to other high-altitude groups. Similarly, among Andean populations, individuals carrying the EPAS1/HIF2 missense variant rs570553380 also exhibit reduced hematocrit^58,59. Our results provide a mechanistic explanation for this discrepancy: the lack of an expanded pool of newly synthesized, glucose-avid RBCs likely prevents Sherpas from experiencing the glycemic benefits typical of other highlanders.

Our findings are also relevant beyond high altitude residents. In fact, it has been reported that patients with Chuvash polycythemia, who present marked erythrocytosis due to sustained HIF signaling in basal conditions, have decreased blood glucose levels, in agreement with our work. In a similar vein, erythropoietin (EPO) treatment has been associated with improved glucose tolerance and reduced blood glucose levels in both human subjects⁶⁰ and animal models^61–63. Similarly, testosterone replacement therapy in hypogonadic males or during gender affirming therapy induces erythrocytosis and has been linked to improved insulin sensitivity and normalization of hyperglycemia⁶⁴. Conversely, studies have shown that patients with anemia often present with hyperglycemia⁶⁵ and that diabetes is frequently associated with anemia^66–70, further supporting our proposed physiological mechanism. Collectively, these scattered yet converging observations suggest that the role of RBCs in glycemic control has been largely overlooked and may carry broader clinical implications.

A future therapeutic avenue against hyperglycemia could be the use of our small-molecule form of hypoxia, HypoxyStat⁴⁵, or even RBC transfusions after hypoxic storage. However, the risks associated with increased blood viscosity may counter the benefits for a relatively manageable condition such as diabetes. Development of future therapies based on promoting a faster turnover of RBCs that shift RBC pools towards younger, glucose-avid RBCs could be a promising avenue for novel diabetes management strategies. Such an intervention would not affect blood viscosity while potentially improving glycemic control. Nonetheless, future work is needed to translate our findings to patients in a safe and effective manner.

In summary, we have identified RBCs as key regulators of systemic glucose metabolism under hypoxic conditions. Of note, related findings were recently reported by Scherer et al.⁷¹ further corroborating our observations. Increased RBC abundance and metabolic reprogramming dramatically enhance glucose uptake, positioning RBCs as mobile glucose reservoirs. This discovery clarifies longstanding observations of improved glycemic control among high-altitude residents and explains clinical links between RBC number and blood glucose in polycythemia and anemia. Our findings thus open new therapeutic avenues, including the modulation of RBC dynamics or use of pharmacological hypoxia mimetics such as HypoxyStat, for effectively treating hyperglycemia and associated metabolic disorders.

Limitations of Study

Our study presents several limitations. First, mouse experiments in this work were solely performed using C57BL/6J mice, which tends to be more glucose intolerant than other stains^72,73. While it would be ideal to extend these findings to multiple mouse strains, the fact that this phenomenon is found in humans and other species suggests that the underlying mechanism is strain-independent and broadly conserved across mammals. Second, even if glycemic control improvement reported in epidemiological data was equivalent across sexes and ages, we acknowledge that our experiments were performed exclusively in young male mice for consistency. Given the notable decline in RBC production during aging^74–76 and the well-known differences in RBC population dynamics among sexes⁷⁷, a mechanistic analysis on the roles of age and sex in the ability of RBCs to act as regulators of glucose homeostasis would be of particular interest. Third, though we demonstrated that upregulation of glucose transporters is specific to RBCs that were “born in hypoxia”, we did not identify the molecular mechanism behind this phenomenon. Future work will dissect the signaling pathways in the erythroid precursors that lead to the upregulation of glucose transporters in hypoxia.

RESOURCE AVAILABILITY

LEAD CONTACT

Additional information or requests for resources will be fulfilled by the lead contact, Isha Jain (Isha.Jain@arcinstitute.org).

MATERIAL AVAILABILITY

No new materials were created for this study.

DATA AND CODE AVAILABILITY

Numerical values used for plots in the paper are available in: Data S1. No new code was generated for this study. Crosslinking proteomics were deposited to Pride (PXD071719, https://www.ebi.ac.uk/pride/).

STAR★Methods

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Animal model

Male C57BL/6J (#000664) mice (8–11 weeks old) from The Jackson Laboratory were used for all animal experiments. Mice were housed at 24°C on a 12h:12h light:dark cycle. Unless stated differently, mice were fed a chow diet (PicoLab 5058). Cages were randomly allocated to normoxic or hypoxic chambers within the Gladstone Institutes animal facility. Hypoxia was simulated in chambers by mixing N₂ (Airgas), O₂ (Airgas, Praxair), and room air using gas regulators. FiO₂ and CO₂ levels were continuously monitored wirelessly and checked daily. To inhibit the accumulation of CO₂, soda lime (Fisher Scientific) was placed surrounding cages in each chamber. Mouse experiments were approved by the UCSF Institutional Animal Care & Use Program (IACUC).

METHOD DETAILS

Body weight measurements

Body weight was monitored daily during the first week of hypoxia exposure and every three days during the following weeks of treatment, unless specified differently.

Body temperature measurements

Rectal body temperature was measured by a BIOSEB thermometer during their sleeping phase (9:00 AM to 10 AM, corresponding to circadian time (CT) 3 and 4, respectively). For longitudinal studies, the time for body temperature measurement was kept consistent across all time points.

Pair-feeding experiment

Food intake per mouse was calculated by measuring the daily change in food weight in single-housed cages. For the pair-fed group, food availability was restricted to match the amount consumed by hypoxic mice on the previous day, resulting in a one-day temporal lag between the hypoxic (8% O₂, n=8) group and the normoxic (21% O₂) pair-fed and ad libitum-fed (n=8 and n=8, respectively) groups. All mice were single-housed for the whole experiment. Body weight was recorded daily and blood glucose measurements were performed on whole blood using the OneTouch Ultra Plus^™ glucose meter.

Blood and plasma glucose measurements

Blood glucose levels were measured in mice during their sleeping phase (9:00 AM to 12:00 PM, corresponding to CT 3 to 6) without prior fasting interventions. Given that mice predominantly feed during their active (dark) phase (6:00 PM to 6:00 AM), we presume that these measurements were obtained after at least three hours of physiological fasting. In experiments involving multiple longitudinal blood glucose measurements, the time of sampling was kept consistent across all time points. Absolute blood and plasma glucose values can show variation between mouse strains and across fed states: our mice strain was C57BL/6J (#000664) from The Jackson Laboratory and most measures were performed without prior fasting (unless stated otherwise).

For standard blood glucose analysis, a small drop of whole blood was obtained via a gentle tail tip incision and immediately measured using the OneTouch Ultra Plus^™ glucose meter. Although we also validated the OneTouch Verio^™ glucose meter, all reported experiments utilized the Ultra Plus^™ version.

To account for potential measurement inaccuracies due to elevated hematocrit levels, we confirmed that applying the plasma fraction directly to the OneTouch Ultra Plus^™ meter yielded accurate glucose values. For plasma glucose measurements, blood was collected by submandibular vein puncture using 5 mm Goldenrod^® lancets into EDTA-coated microtubes. Samples were centrifuged at 2,000 × g for 10 minutes at 4°C, and the plasma fraction was applied to the glucose meter. These values are reported in the text as “plasma glucose,” while “blood glucose” refers to values obtained from whole-blood samples.

Glucose tolerance test

Glucose tolerance tests (GTTs) were performed by intraperitoneal injection of a glucose bolus (2 g/kg body weight), followed by serial blood glucose measurements at baseline (prior to injection) and at 15-, 30-, 60-, 90-, and 120-minutes post-injection. Each mouse’s baseline value was used as its individual reference point. The area under the curve (AUC) for blood glucose over time, relative to the baseline, was calculated using GraphPad Prism software. Individual AUCs were grouped by experimental condition to assess the impact of interventions on glucose tolerance. A reduced mean AUC compared to control mice indicated improved glucose tolerance, while an increased AUC was interpreted as impaired glucose tolerance. To visualize changes in glucose tolerance over time, AUC values were expressed as fold changes relative to the control condition within each experiment.

Insulin tolerance test

Insulin tolerance tests (ITTs) were performed by intraperitoneal injection of a human insulin bolus (1 IU/kg body weight), followed by serial blood glucose measurements at baseline (prior to injection) and at 15-, 30-, 60- and 90-minutes post-injection. Each mouse’s baseline value was used as its individual reference point. The area above the curve (referred as AUC for simplicity) for blood glucose over time, relative to the baseline, was calculated using GraphPad Prism software. Individual AUCs were grouped by experimental condition to assess the impact of interventions on insulin tolerance. A reduced mean AUC compared to control mice indicated impaired insulin tolerance, while an increased AUC was interpreted as improved insulin sensitivity. To visualize changes in insulin tolerance over time, AUC values were expressed as fold changes relative to the control condition within each experiment.

Pyruvate tolerance test

Pyruvate tolerance tests (PTTs) were performed by intraperitoneal injection of a pyruvate bolus (2 g/kg body weight), followed by serial blood glucose measurements at baseline (prior to injection) and at 15-, 30-, 60-, 90-, and 120-minutes post-injection. Each mouse’s baseline value was used as its individual reference point. The area under the curve (AUC) for blood glucose over time, relative to the baseline, was calculated using GraphPad Prism software. Individual AUCs were grouped by experimental condition to evaluate the effect of interventions on pyruvate tolerance. A lower mean AUC compared to control mice was interpreted as a reduced gluconeogenic capacity (i.e., diminished glucose production from pyruvate), while a higher AUC indicated an enhanced gluconeogenic response.

¹⁸F-FDG PET/CT scan analysis

¹⁸F-FDG PET/CT scans acquired for our previous study on hypoxic fuel rewiring¹⁷ were re-analyzed by Amide software to estimate the relative contribution of major organs to the overall increase in glucose uptake under hypoxia adaptation. Male mice were housed under normoxic (21% O₂, n=8) or hypoxic (8% O₂, n=5) conditions for 3 weeks. At the end of the exposure period, all animals received an intravenous injection of 2-deoxy-2-[¹⁸F]fluoro-D-glucose (FDG). ¹⁸F-FDG accumulation was used as a surrogate marker for tissue-specific glucose uptake.

Whole-body ¹⁸F signal was quantified by drawing same-size cylindrical regions of interest (ROIs) (“Whole-body ROI: 27, 27, 90 mm-elliptic cylinder) encompassing individual animals. To correct for urinary excretion of ¹⁸F-FDG, signal within the bladder was excluded from the total signal (ROI: 9, 9, 9 mm-ellipse). We selected organs that demonstrated a hypoxia-dependent increase in ¹⁸F-FDG uptake in previous reports¹⁷. Organ-specific ¹⁸F-FDG uptake was determined by placing same-size elliptical ROIs over the heart (ROI: 10, 8, 10 mm-ellipse), brain (ROI: 9, 16, 10 mm-ellipse), liver (ROI: 20, 10, 20 mm-ellipse) and brown adipose tissue (BAT) (ROI: 9, 9, 9 mm-ellipse), guided by anatomical landmarks visible on the corresponding CT scans. The contribution of skeletal muscle was estimated as ‘none’ as no change in FDG uptake was quantified in our previous analysis¹⁷. The increase in whole-body glucose uptake under hypoxia was calculated as the difference in the percentage of injected dose (%ID) between hypoxic mice and the normoxia average (Δ%ID = %ID_8% - %ID_21%). The relative contribution of each organ to the overall Δ%ID was then computed to identify potential drivers of increased glucose uptake and represented in a pie chart.

Hematological analysis

Complete blood counts were performed using a VetScan HM5 hematology analyzer (Abaxis, Union City, CA). Blood samples were collected via submandibular vein puncture using 5 mm Goldenrod^® lancets and immediately transferred into EDTA-coated microtubes to prevent coagulation. Samples were stored at 4°C and analyzed within 1 to 6 hours of collection.

For analysis, 50 μL of EDTA-anticoagulated blood was automatically drawn by the VetScan HM5. The sample was split into two fractions. The first fraction was used for RBC parameter analysis via impedance technology, providing measurements such as RBC count (10¹²/L) and mean corpuscular volume (MCV, fL). Hematocrit (%) was calculated as the product of RBC count and MCV divided by 10, following the formula: Hematocrit = (RBC × MCV) / 10. The second blood fraction was subjected to RBC lysis to enable quantification of non-RBC parameters. White blood cell (WBC) and platelet counts (10⁹/L) were determined by impedance in this lysed fraction. Hemoglobin concentration (g/dL) was measured photometrically at a wavelength of 540 nm using the same lysed sample.

Phlebotomy Model

To achieve sustained RBC reduction, serial phlebotomy was performed on mice housed under hypoxic (8% O₂, n=8) and normoxic (21% O₂, n=8) conditions. Every three days, 15% of the estimated total blood volume (TBV) was withdrawn via submandibular vein puncture using 5 mm Goldenrod^® lancets. To maintain circulatory volume, an equal volume of sterile saline was administered intraperitoneally following each blood draw. Phlebotomized mice were compared to control groups of non-phlebotomized mice maintained under hypoxia (n=5) or normoxia (n=5).

Blood samples were collected on days 9, 15, and 21 of exposure for complete blood count (CBC) and plasma glucose analysis. Blood was collected into EDTA-coated microtubes and stored at 4°C until processing. Samples were split into two aliquots upon collection. One aliquot was used for plasma glucose measurement. Plasma was extracted by centrifugation within 1 hour of collection, and glucose levels were subsequently measured. The second aliquot was used for hematological analysis and processed within 1 to 6 hours of collection, as described in the ‘Hematological analysis’ section above.

RBC Transfusion Model

To increase circulating RBC levels, mice housed under normoxic conditions (21% O₂) were transfused with RBCs derived from donor mice previously exposed to either hypoxia (8% O₂) or normoxia (21% O₂) for four weeks. Recipient groups included mice receiving hypoxic RBCs (n=5), normoxic RBCs (n=5), or vehicle control (saline, n=5). Blood from donor mice was collected into tubes containing citrate buffer (pH 7.0) to a final concentration of 0.3% to prevent coagulation. Samples were thoroughly mixed and centrifuged at 2,000 × g for 10 minutes at 4°C. The plasma and buffy coat layers were discarded to isolate a clean pellet of packed RBCs. This pellet was then pooled and resuspended in sterile saline to achieve a 75% hematocrit (i.e., 75% packed RBCs by volume).

Recipient mice received retro-orbital injections containing 200 μL of 75%-packed RBCs twice daily for two consecutive days. Endpoint blood analysis was conducted the following day. Blood was collected via submandibular vein puncture into EDTA-coated microtubes and analyzed immediately for hematological parameters and plasma glucose as described above.

Type I Diabetes Model

Type I diabetes was induced in 8-week-old male mice by intraperitoneal (i.p.) injection of streptozotocin (STZ; n=16) or vehicle (citrate buffer, pH 4.5; n=16) for five consecutive days. Two weeks post-injection, blood glucose levels and body weight were assessed to confirm the onset of hyperglycemia. Following verification of diabetes induction, STZ-treated hyperglycemic mice and vehicle-treated controls were randomly assigned to hypoxia or normoxia groups. Mice were exposed to either hypoxic (8% O₂; n=8 STZ-treated, n=8 vehicle-treated) or normoxic (21% O₂; n=8 STZ-treated, n=8 vehicle-treated) conditions for a period of three weeks. Throughout the hypoxia/normoxia exposure, blood glucose levels and body weight were monitored every three days to evaluate the physiological effects of hypoxia in diabetic and non-diabetic mice.

HFD-Induced Hyperglycemia Model

Male C57BL/6J mice fed either a high-fat diet (HFD; JAX^® Diet-Induced Obese, stock #380050, n=11) or standard chow diet (CD; stock #000664, n=16) were purchased from The Jackson Laboratory. Upon arrival, mice were allocated to receive either daily oral gavage with HypoxyStat (600 mg/kg) or vehicle for a duration of 2.5 weeks. Group allocation was as follows: HFD-fed mice received either HypoxyStat (n=5) or vehicle (n=6), while CD-fed mice received either HypoxyStat (n=8) or vehicle (n=8). Blood glucose levels were measured after two weeks of daily dosing. Glucose tolerance was evaluated after two and a half weeks of treatment to assess the metabolic effects of HypoxyStat in both dietary contexts.

Flow Cytometry

RBCs were isolated from mice after 3 weeks in 21% (n=5) or 8% O₂ (n=5). Blood was collected by submandibular vein puncture into EDTA-coated microtubes and stored at 4°C until processing. Samples were split into three aliquots upon collection. One aliquot was used for plasma glucose measurement to verify the difference in glycemia across oxygen tensions. The second was used for hematological analysis to obtain their CBC profile. The third aliquot was used for flow cytometry analysis.

RBCs were fixed in 4% PFA and 0.05% glutaraldehyde in PBS for 15 mins at RT on a rotating mixer. RBCs were permeabilized in 0.1% Triton X-100 in PBS for 10 mins at RT, followed by 1 hour of blocking at RT in 3% BSA and 2% donkey serum in PBS. After blocking, RBCs were incubated with primary antibodies (GLUT1 (Proteintech 21829–1-AP); GLUT4 (Invitrogen PA5–23052); GAPDH (Cell Signaling 2118S); Hexokinase I (Cell Signaling 2024S)) at 1:100 in 3% BSA in PBS on a rotating mixer overnight at 4C. RBCs were then washed 3X with PBS and stained with a secondary antibody (donkey-anti-rabbit 488-FITC (Invitrogen 21206)) at 1:600 in 3% BSA in PBS for 1 hour at RT. Stained RBCs were then washed 3X with PBS and resuspended in PBS with 25 mM HEPES and 1 mM EDTA for immunofluorescence analysis on the LSR Fortessa X-20 Cell Analyzer. A minimum of 50,000 events were acquired for each sample, and no-stain and no-primary antibody controls were included in gating.

Glucose transporters and HIF targets in peripheral tissues

The apex and half of the brain were collected at necropsy and immediately flash frozen. For RNA extraction, a stainless-steel bead (Qiagen) and 1 mL TRIzol^™ Reagent (Thermo Fisher Scientific) were added to the frozen samples, which were rapidly lysed using a tissue lyser (Qiagen). For brain samples, five 1-minute cycles of 30 Hz frequency were performed while the heart only required two. Subsequent steps for RNA isolation, reverse transcription and qPCR analysis were identical as previously described⁴⁵.

In vivo 2-deoxy-D-glucose (U-¹³C) uptake assay

2-deoxy-D-glucose (2DG) (U-¹³C) (Cambridge Isotope Laboratories) was freshly dissolved in sterile water on the day of the experiment. Mice exposed to hypoxia (8% O₂, n=5) or normoxia (21% O₂, n=5) for three weeks were administered a bolus injection of 2DG (U-¹³C) at a dose of 1 g/kg body weight via the retro-orbital route. Blood samples were collected at 2-, 10-, 30-, and 120-minutes post-injection in EDTA-coated microtubes. Collected EDTA-blood was stored at 4°C and processed at the end of the 2-hour experimental period. Blood samples collected at 2 minutes were split into two aliquots upon collection. One aliquot was used for hematological analysis, to ensure proper normalization of 2DG uptake by the total number of RBC per sample, using the formula: Volume of analyzed RBC (20 μL) × 10⁹ / MCV (fl) = Number of analyzed RBCs in 20 μL

The second aliquot was processed together with the rest of the samples. Blood samples were centrifuged at 2,000 × g for 10 minutes at 4°C to separate plasma and cellular components. A 5 μL aliquot of plasma was transferred to a clean microtube and stored at −80°C for later analysis. The remaining plasma and buffy coat were discarded to isolate a clean pellet of packed RBCs. A 5 μL RBC pellet was collected and stored at −80°C for subsequent isotopic analysis.

Ex vivo glucose uptake assay

Blood samples were collected from mice exposed to hypoxia (8% O₂, n=5) or normoxia (21% O₂, n=5) for four weeks in EDTA-coated microtubes and stored on ice until processing. Blood samples were split into two aliquots upon collection. One aliquot was used for hematological analysis, to ensure proper normalization of glucose uptake by the total number of RBC per sample, using the formula: Volume of analyzed RBC (10 μL) × 10⁹ / MCV (fl) = Number of analyzed RBCs in 10 μL

The second aliquot was centrifuged at 2,000 g for 10 minutes at 4°C and all the plasma from normoxic mice was pooled in a vial. Plasma glucose in the pooled normoxic plasma was measured for reference using the OneTouch Ultra Plus^™ glucose meter. Upon plasma and buffy coat layers removal, 10 μL of RBCs pellet from each hypoxic and normoxic sample were incubated with 10 μL of the pooled normoxic plasma. These samples were incubated at 37°C for 10 minutes in a rocker. One sample only containing plasma was included to test non-RBC specific changes in glucose levels upon incubation. After this time, samples were centrifuged to isolate the plasma that had been in contact with the experimental RBCs. Remaining plasma glucose was measured in all samples using the OneTouch Ultra Plus^™ glucose meter and the value was compared to the plasma glucose level in the incubated sample without RBCs (234 mg/dL), which did not present a significant change in glucose levels during the incubation.

In vivo biotin tracer experiment

Biotin ((+)-Biotin N-hydroxysuccinimide ester, Sigma H1759) was resuspended in DMSO to a 30 mg/mL solution and frozen at −20°C until the day of the injection. On that day, concentrated biotin was diluted to 2.5 mg/mL with PBS and was immediately injected via the retro-orbital route. Every mouse received a fixed dose of 200 μL (0.5 mg of biotin) every day for three consecutive days. The goal for this dosing strategy is to label all pre-existing RBCs with biotin and be able to separate them from the newly synthesized biotin-negative RBCs, which will have matured after this biotin pulse. Right after the last biotin injection, mice were randomized into normoxia (21% O₂, n=3) or hypoxia (21% O₂, n=3) exposure, so that biotin-negative RBCs start maturing under normoxic or hypoxic conditions, respectively.

Four weeks after the biotin pulse was done and normoxia/hypoxia exposure started, we collected blood samples from these mice into EDTA tubes and stored them on ice until processing. Blood samples were subsequently centrifuged at 2,000 g for 10 minutes at 4°C and their plasma and buffy coat carefully discarded. 200 μL of packed RBCs were placed in a new Eppendorf and processed for flow cytometry as described above (see “Flow Cytometry” from Materials & Methods section). In the step of incubation with conjugated-secondary antibody, we co-stained with Streptavidin-APC (SA1005, Thermo Fisher), to probe for biotin in parallel to GLUT1 (FITC-A) detection. 50,000 events were acquired for each sample, and no-stain and no-primary antibody controls were included for defining the biotin-negative and GLUT1-negative signal.

In vivo glucose (U-¹³C) tracer experiment

Glucose (U-¹³C) (Cambridge Isotope Laboratories) was freshly dissolved in sterile water on the day of the experiment. Mice exposed to hypoxia (8% O₂, n=5) or normoxia (21% O₂, n=5) for three weeks were administered a bolus injection of glucose (U-¹³C) at a dose of 1 g/kg body weight via the retro-orbital route. Blood samples were collected at 2-, 10-, 30-, and 120-minutes post-injection in EDTA-coated microtubes. Collected EDTA-blood was stored at 4°C and processed at the end of the 2-hour experimental period. Samples were centrifuged at 2,000 × g for 10 minutes at 4°C to separate plasma and cellular components. A 5 μL aliquot of plasma was transferred to a clean microtube and stored at −80°C for later analysis. The remaining plasma and buffy coat were discarded to isolate a clean pellet of packed RBCs. A 5 μL RBC pellet was collected and stored at −80°C for subsequent isotopic analysis.

Polar metabolites extraction

Plasma was extracted from blood by centrifuging at 2,000 g for 10 minutes at 4°C and collecting supernatant. For tracer experiments, plasma metabolites were extracted by adding 45 μL of 80% methanol to every 5 μL of plasma. Samples were vortexed for 10s and incubated at −80°C for 20 minutes. Next, samples were centrifuged at 20,000g for 10 minutes at 4°C and supernatant was collected. Supernatants were subsequently lyophilized at 4°C and resuspended in 40 μL 60% ACN. Samples were sonicated, mixed in a thermomixer and incubated for 20 minutes at 4°C. Samples were centrifuged at 20,000 g for 20 minutes at 4°C and supernatants were transferred to suitable vials for LC-MS injection.

For RBC polar metabolites extraction, 400 μL of 40:40:20 (acetonitrile:methanol:water) solvent was added to 20 μL of isolated RBCs. Samples were mixed in a thermomixer, sonicated and incubated for 1 minute in liquid nitrogen twice. After the second cycle, samples were incubated in ice for 20 minutes and centrifuged at 20,000 g for 20 minutes at 4°C. Supernatants were subsequently lyophilized at 4°C and resuspended in 100 μL 60% ACN. Samples were sonicated, mixed in a thermomixer and incubated for 20 minutes at 4°C. Samples were centrifuged at 20,000 g for 20 minutes at 4°C and supernatants were transferred to suitable vials for LC-MS injection.

Peak area quantification of targeted metabolites was performed using TraceFinder. Metabolite peak areas were normalized to the internal standard peak area included in each experiment. For absolute quantification of plasma glucose by LC-MS, 5 mM uniformly labeled glucose (U-¹³C) was used as a standard, enabling quantification of unlabeled glucose in plasma samples. In the glucose uptake experiment using labeled 2-deoxy-D-glucose (U-¹³C) [2DG (U-¹³C)], 20 μM unlabeled 2-deoxy-D-glucose phosphate (2DG-P) served as a standard, allowing for absolute quantification of fully labeled 2DG-P in RBC samples. In the glucose tracer experiment using glucose (U-¹³C), 100 μM L-Valine-d8 was used as a standard to enable relative quantification of all detected metabolites.

LC-MS

Samples were run on an Orbitrap Exploris 240 (OE240) high resolution mass spectrometer (Thermo Fisher Scientific) using electrospray ionization. The OE240 was coupled to hydrophilic interaction chromatography on a Vanquish Horizon ultra-high performance liquid chromatography (UHPLC, Thermo Fisher Scientific). 2 μL of polar metabolite samples were injected into the LC-MS system and separated on a iHILIC-(P) Classic column (2.1×150 mm, 5 μm; part # 160.152.0520; HILICON AB). The autosampler was maintained at 4°C and the column was maintained at 40°C during runs. Mobile phase A was 20 mM ammonium bicarbonate in water, with ammonium hydroxide added to reach a pH of 9.6. Mobile phase B was acetonitrile. Flow rate of 200 μL/min was used for separation, using the following gradient: from 0–18 minutes 85 to 20% B, from 18–20 minutes hold at 20% B, from 20–20.5 minutes a linear increase from 20 to 85% B, and from 20.5–28 minutes hold at 85% B. A 10-min equilibration round was used before every run. The OE240 ran in full-scan, polarity-switching mode with an ion voltage of 3.5 kV in positive mode and 3.25 kV in negative mode, and the scan range was 70–1000 m/z. The orbitrap resolution was 60000, RF Lens was 60%, AGC target was 1e7, and the maximum injection time was 200 ms. The sheath gas was set to 35 units, auxiliary gas was 10 units, and sweep gas was 0.5 units. The ion transfer tube temperature was 300°C, and the vaporizer temperature was 35°C.

Semi-targeted metabolomics

For semi-targeted metabolomics, the Mass Spectrometry Metabolite Library of Standards (MSMLS, IROA Technologies) was run on the LC-MS system. A library of fragmentation spectra was generated by running standards in negative and positive mode using data-dependent MS2. HCD collision energies were 15, 45, and 90%, and the Orbitrap resolution was 30,000. For compound identification, quality control samples containing equal volumes from each experimental sample were run using the same ddMS2 parameters using a targeted mass list of compounds from the standard library that were detectable. In addition, untargeted compound identification was conducted using AcquireX with the same ddMS2 parameters.

Compound identification and peak area determination were conducted using Compound Discoverer (Thermo Fisher Scientific). A local database containing fragmentation spectra of MSMLS compounds and an online database through mzCloud were used to identify compounds.

Immunofluorescence and Proximity Ligation Assay (PLA)

For human blood samples, healthy volunteers provided blood upon hospital admission, which was drawn into heparin-containing tubes. For mice blood samples, cardiac blood collection was performed under appropriate anesthesia to ensure minimal discomfort to the animals. Briefly, C57BL/6 mice were anesthetized until full loss of reflexes was confirmed. The thoracic cavity was opened, and blood was collected directly from the heart using a sterile syringe and needle. Collected blood was immediately transferred to the heparin tubes for downstream analysis. RBCs were separated from plasma and the buffy coat, then washed three times (2,000 × g, 10 min, 4 °C) with HBSS. Oxyhemoglobin (Oxy) RBC samples were exposed to ambient room air, while deoxyhemoglobin (Deoxy) samples were deoxygenated using a telemetry system until hemoglobin oxygen saturation was confirmed to be below 20%. Following deoxygenation, Deoxy samples were fixed under hypoxic conditions within a glove box.

Packed RBCs were immediately mixed with 30 μL of fixing buffer (a 1:99 ratio of 8% glutaraldehyde to 4% paraformaldehyde) at room temperature (RT), gently suspended, and rotated for 20 minutes. Afterward, cells were washed three times with rinsing buffer (0.1 M glycine and 0.05% sodium azide), then permeabilized with permeabilization buffer (0.1% Triton^® X-100 in rinsing buffer) for 10 minutes at RT, followed by three additional rinses.

For immunofluorescence staining, we used a mouse monoclonal anti-GAPDH antibody (sc-166545) and a rabbit monoclonal anti-Band 3 antibody [EPR1425(2)] (Abcam ab175214). As secondary antibodies, we used Alexa Fluor 488 donkey anti-mouse IgG (Thermo 2563848) and Alexa Fluor 488 donkey anti-rabbit IgG (Thermo 2622406). Proximity Ligation Assays (PLAs) were performed using the Sigma Duolink kit. Imaging was conducted with a Nikon Eclipse Ti2 microscope at 100× magnification, and subsequent image analysis was performed using NIS-Elements software.

STED microscopy

Super-resolution imaging was performed using an Olympus IX-series inverted microscope equipped with an Abberior Facility Line STED system (Abberior Instruments GmbH, Göttingen, Germany), with a 60× oil immersion objective (NA 1.4). STAR Orange- and STAR Red-conjugated secondary antibodies (Abberior) were used for detection. Excitation was performed at 561 nm (STAR Orange) and 640 nm (STAR Red), with depletion at 775 nm. Image analysis for localization studies was performed using Imaris software, version 10.0.2 (Oxford Instruments).

Recombinant expression of Band 3 N-term domain (1–395) and GAPDH

Unlabeled proteins were grown in Luria broth (LB) while labeled proteins were grown in M9 minimal media (6 g/L Na₂HPO₄, 3 g/L KH₂PO₄, 0.5 g/L NaCl, 1 g/L ¹⁵NH₄Cl, 2 g/L glucose (¹³C₆-glucose if preparing double labeled protein), 2 mL of 1M MgSO4, 100 μL of 1 M CaCl2, 10 mg/L thiamine) unless stated otherwise.

All Band 3 and GAPDH protein constructs contained an N- or C-terminal 6xHis-SUMO. Constructs were cloned into the bacterial expression vector, pET21b, using the NdeI restriction site. Growth media with proper antibiotics was inoculated with a colony from a freshly transformed LB agar plate and shaken at 37 °C for 16 hours. Fresh media was inoculated with 2.5% of the overnight growth and shaken at 37 °C until an OD of 0.6 was reached. Protein expression was induced by the addition of 1 mM isopropyl-beta-D-thiogalactoside (IPTG). Induced cultures were shaken for 4 hours at 37 °C before being harvested by centrifugation at 4,500 rpm for 10 min.

The purification strategy of all protein constructs was similar. Bacterial pellets were resuspended in Lysis Buffer (50 mM Tris pH 7.5, 300 mM NaCl, 3% glycerol, 10 mM imidazole, 10 mM DTT) and lysed by sonication with 7 cycles of 30 sec on, 30 sec off. Cellular debris was removed by centrifugation at 13,000 rpm for 30 min and 4 °C. Lysate was applied to a column packed with Ni Sepharose Excel resin (Cytiva) and washed with 7 column volumes of Lysis Buffer. Bound protein was eluted with 4 column volumes of Elution Buffer (50 mM Tris pH 7.5, 300 mM NaCl, 400 mM imidazole, 5% glycerol). The elution fractions were pooled, and SUMO protease (ulp1, in-house) was added. The sample was dialyzed overnight into Lysis Buffer at 4 °C using dialysis tubing with a 3.5K molecular weight cut-off (MWCO). The sample was applied to a column packed with Ni-NTA His•Bind Resin (Sigma) and the flow-through material was collected. The flow-through fraction was concentrated and applied to a Superose 6 Increase column (Cytiva) equilibrated with Storage Buffer (20 mM Bis-Tris pH 6.5, 150 mM NaCl). Fractions containing target protein were pooled, concentrated, and stored at −80 °C.

Purification of hemoglobin

Human Hemoglobin protein was purified from RBC lysate using size exclusion chromatography and polished by anion exchange chromatography. Briefly, human RBCs were lysed through overnight dialysis against nanopure water with 1X HALT Protease Inhibitor Mix (ThermoFisher, MA, USA) at 4 °C. RBC lysate was applied to a Superdex 200pg column and the fractions corresponding to the size and coloration of hemoglobin were collected and pooled. The running buffer for this experiment was 20 mM Bis-Tris, 50 mM NaCl pH 6.5. Pooled fractions were then diluted with 10 equivalents of 50 mM Tris pH 7 and Anion exchange chromatography utilizing a Source 15Q Anion Exchange Column, 12 mL column volume, (Cytiva, New Jersey, USA) was used for polishing. The A and B buffers for the run were 50 mM Tris pH 7 and 50 mM Tris, 1 M NaCl pH 7, respectively, and the elution gradient for the experiment was 0–45% in 9 column volumes. Chromatography was performed at a flow rate of 1.5 mL/min and at 4 °C. The resulting fractions were checked by SDS-PAGE gel for purity.

Isothermal Titration Calorimetry (ITC)

All isothermal titration calorimetry binding experiments were performed with a MicroCal iTC200 (Cytiva) set at 25°C. GAPDH and Band 3 (1–395) recombinantly expressed human proteins were dialyzed into matching buffer (20 mM Bis-Tris pH 6.5, 50 mM NaCl, 2 mM DTT). The cell contained GAPDH at 0.695 mM while the syringe contained Band 3 at 30 μM. Reference power was set to 10 μcal/s with a constant stirring speed of 1,000 rpm. In total, 20 injections were performed. The first injection was excluded from the data analysis. Experiments were performed in duplicate and the results were analyzed with the Origin ITC module.

Crosslinking proteomics

Purified hemoglobin (Hb), GAPDH and Band 3 N-terminus Domain (1–395) (B3) proteins were dialyzed into 50mM Bis-Tris, 50mM NaCl pH 6.5 buffer and concentrated using spin concentrators. Proteins were then mixed together at a 5:5:1 Hb:GAPDH:B3 ratio and placed inside a hypoxia hood overnight to degas. After overnight degassing, the sample mix was aliquoted into six 75 uL sample replicates, and three of the replicates were brought out of the hood into the oxygenated atmosphere. These three replicates outside the hood were allowed to equilibrate prior to further analysis for 20 min.

Samples both inside and outside the hood were crosslinked with either 4 mM disuccinimidyl sulfoxide (DSSO, Thermo Fisher) or 20 mM 4-(4,6-Dimethoxy-1,3,5-triazin-2-yl)-4-methylmorpholinium chloride (DMTMM, Sigma Aldrich) for 1 hour at 25 °C. Residual crosslinker was quenched by the addition of 1 M Tris pH 8 for 30 minutes. Crosslinked samples were diluted with 8 M urea to a final concentration of 2 M urea. DTT and IAA were added to a final concentration of 5 mM and 10 mM, respectively, and incubated at room temperature for 20 minutes in the dark. 5 equivalents of acetone was then added to the sample and acetone precipitation was allowed to proceed overnight at −80°C to remove urea and Tris. The protein pellets from acetone precipitation were resuspended in 2X S-Trap Lysis Buffer (10% SDS, 100 mM TEAB pH 8.5) and applied to an S-Trap (Protifi, NY, USA) for digestion. Lys-C (FUJIFILM Wako) and Glu-C (Promega) were added to the samples at a ratio of 1:50 and incubated overnight at 37 °C with constant shaking. After the standard S-Trap elution, samples were dried down and TMT labeled with TMTsixplex (Thermo Fisher Scientific) following the manufacturer’s instructions. Approximately 10 μg of peptide sample was desalted using Spin Tips (Pierce) for subsequent LC-MS/MS analysis.

TMT labeled peptides were analyzed by nano-UHPLC-MS/MS (EASY-nLC 1200, Fusion Lumos Tribrid, Thermo Fisher Scientific). Sample was loaded directly onto an in-house packed 100 μm i.d. × 250 mm fused silica column packed with CORTECS C18 resin (2.7 μm, spherical solid core). Samples were run at 400 nL/min over a 120-min linear gradient from 4 to 40% acetonitrile with 0.1% formic acid. The mass spectrometer was operated in positive ion data-dependent mode. MS1 scans were run in the orbitrap from 375 to 1,700 m/z at a resolving power of 120,000. Full scan automatic gain control (AGC) and maximum injection time were set to defaults. Fragmentation was performed by stepped higher-energy collisional dissociation (HCD) of 25, 35, and 55%. Filtering was performed by the quadrupole with an isolation window of 0.7 m/z. MS2 AGC was set to 200 % and maximum injection time was set to 200 ms. MS2 fragment ions were measured by the orbitrap operating at a resolution of 50,000 with the first mass set to 120 m/z. The scan frequency was determined by a 3 s total duty cycle.

Raw files were loaded into Proteome Discoverer 3.2 (Thermo Fisher Scientific) and searched against the three proteins used in the crosslinking experiment using the MS Annika 3.0 plugin⁷⁹. Search parameters for the search for DSSO crosslinks included carbamidomethylation-C as a fixed modification, oxidation-M, DSSO-n-term,S,T,Y,K, DSSO/amidated-S,T,Y,K, and DSSO/hydrolysed-S,T,Y,K as variable modifications, allowing for 4 missed cleavages. Search parameters for the search for DMTMM crosslinks included carbamidomethylation-C as a fixed modification, oxidation-M, DMTMM-[n-term,K-c-term,D,E], allowing for 4 missed cleavages. Precursor mass tolerance was set to 15 ppm, with MS/MS mass tolerance set to 20 ppm. Crosslink FDR cutoffs (peptides and crosslinks) were set to 0.5% and 0.1%, respectively. Results were visualized using xiVIEW⁸⁰. The data was deposited to the ProteomeXchange Consortium via the PRIDE⁸¹ partner repository with the dataset identifier PXD071719 and 10.6019/PXD071719.

Statistical analysis

Sample sizes for each measurement were based on previously reported results and varied depending on the data type and study design. Data are presented as mean ± standard deviation (SD) in dot plots unless otherwise specified. For comparisons between two groups, either paired or unpaired t-tests were used depending on the experimental design. When data met assumptions of normality, a t-test was applied; otherwise, a non-parametric Mann–Whitney test was used. For comparisons involving more than two groups, one-way analysis of variance (ANOVA) followed by Dunnett’s multiple comparisons test was performed. When data involved two categorical independent variables (e.g., time and genotype, or genotype and treatment), a two-way ANOVA or a mixed-effects model (with Geisser–Greenhouse correction) was applied. Specifically, mixed-effects models were used when repeated measurements were taken from the same group of individuals over time, while two-way ANOVA was used when each measurement came from distinct individuals. Statistical outliers were identified using Grubb’s test (extreme studentized deviate method). All statistical analyses were performed using Prism 10 (GraphPad Software, California, USA), unless otherwise noted. A p-value < 0.05 was considered statistically significant.

Supplementary Material

2

Data S1. Unprocessed source data underlying all blots and graphs. Related to Figures 1–6 and Supplemental Figures 1–6

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-GLUT1	Proteintech	21829-1-AP
Anti-GLUT4	Invitrogen	PA5-23052
Anti-GAPDH	Cell Signaling	2118S
Anti-Hexokinase I	Cell Signaling	2024S
Donkey anti-rabbit 488-FITC	Invitrogen	21206
Anti-GAPDH	Santa Cruz Biotechnology	sc-166545
Anti-Band 3	Abcam	ab175214
Donkey anti-mouse Alexa Fluor 488	Thermo	2563848
Donkey anti-rabbit Alexa Fluor 488	Thermo	2622406
STAR Orange-conjugated antibody	Abberior	STORANGE
STAR Red-conjugated antibody	Abberior	STRED
Chemicals, peptides, recombinant proteins
D-(+)-Glucose	Sigma Aldrich	G7021
Insulin solution human	Sigma Aldrich	I9278
Sodium Pyruvate	Gibco	11360070
TRIzol reagent	Thermo Fisher	15596018
2-Deoxy-D-glucose 6-phosphate sodium salt	Sigma Aldrich	SMB00932
2-Deoxy-D-glucose (U-¹³C₆, 99%)	Cambridge Isotope Laboratories	CLM-10466-PK
L-VALINE (D8, 98%)	Cambridge Isotope Laboratories	DLM-488-0.25
D-Glucose (U-13C6, 99%)	Cambridge Isotope Laboratories	CLM-1396-2
(+)-Biotin N-hydroxysuccinimide ester	Sigma Aldrich	H1759
Streptavidin-APC	Thermo Fisher	SA1005
Streptozocin	Sigma Aldrich	S0130
HypoxyStat	WuXi Apptec	N/A
Critical commercial assays
QuantiTect Reverse Transcription Kit	Qiagen	205313
Maxima SYBR Green/ROX qPCR Master Mix	Thermo Fisher	K0222
Deposited data
Values used to create all graphs in the paper	This paper. Data S1
Crosslinking proteomics	Pride	PXD071719
Experimental models: organisms/strains
C57BL/6J mice	Jackson Laboratory	#000664
Diet-Induced Obese mice	Jackson Laboratory	#380050
Oligonucleotides
Slc2a1 Forward: GCTTCTCCAACTGGACCTCAAAC	Midha et al. 202317	N/A
Slc2a1 Reverse: ACGAGGAGCACCGTGAAGATGA	Midha et al. 202317	N/A
Slc2a4 Forward: GGTGTGGTCAATACGGTCTTCAC	Midha et al. 202317	N/A
Slc2a4 Reverse: AGCAGAGCCACGGTCATCAAGA	Midha et al. 202317	N/A
Acer2 Forward: GTGCGAGGACAACTACACTATC	Blume et al. 202545	N/A
Acer2 Reverse: CACATGCAGATGGGAGGTAAA	Blume et al. 202545	N/A
Adm Forward: ATGTCTCAGCAAGGTGTAAGG	Blume et al. 202545	N/A
Adm Reverse: TTCTTCATCCACAGGCGATAAT	Blume et al. 202545	N/A
Angpt2 Forward: ACAGCTGTGATGATAGAGATTGG	Blume et al. 202545	N/A
Angpt2 Reverse: CGAGTCTTGTCGTCTGGTTTAG	Blume et al. 202545	N/A
Vegfa Forward: TGGTTCTTCACTCCCTCAAATC	Blume et al. 202545	N/A
Vegfa Reverse: GGTCTCTCTCTCTCTTCCTTGA	Blume et al. 202545	N/A
Software and algorithms
AMIDE software	Loening and Gambhir78	https://amide.sourceforge.net/
TraceFinder 5.1 General Quan	Thermo Fisher	https://www.thermofisher.com/us/en/home/industrial/mass-spectrometry/liquid-chromatography-mass-spectrometry-lc-ms/lc-ms-software/lc-ms-data-acquisition-software/tracefinder-software.html
QuantStudio 5	Applied Biosystems	https://www.thermofisher.com/order/catalog/product/A34322
GraphPad Prism	GraphPad	https://www.graphpad.com/
Other
Ultra Plus Flex™ glucose meter	OneTouch	OTSUS05_0021
Ultra Plus® test strips	OneTouch	OTSUS05_0024
Thermometer	BIOSEB
5mm-blood lancets	Goldenrod®	NC9416572
VetScan HM5 Hematology Analyzer	Abaxis
Reagent Pack for VetScan HM5, Abaxis	ZOETIS	10023312
Tissue lyser	Qiagen

Highlights.

• High altitude residents and hypoxic mice display improved glucose clearance

• Increased RBC number in hypoxia is necessary and sufficient for hypoxic hypoglycemia

• Glycolytic metabolon disruption in hypoxic RBCs supports increased 2,3-DPG production

• Inhaled and small molecule forms of hypoxia rescue type 1 and 2 diabetes models