Inosine promotes erythrocyte metabolic reprogramming and restores oxygen release for rejuvenation via 2,3-BPG-PNP axis

Liu, Wuping; Yang, Zhaoyu; Chen, Changhan; Yu, Fang; Wu, Mengzhi; Yao, Zhouzhou; Fan, Yuhua; Xie, Tingting; Wan, Linlin; Chou, Tiansheng; Feng, Xianjing; Qi, Hao; Chou, Yuyu; Zhao, Juan; Liu, Juan; Yang, Zhiyu; Zhang, Yujin; Kellems, Rodney E.; Xia, Yang

doi:10.1038/s41421-026-00877-6

Download PDF

Article
Open access
Published: 17 March 2026

Inosine promotes erythrocyte metabolic reprogramming and restores oxygen release for rejuvenation via 2,3-BPG-PNP axis

Wuping Liu ORCID: orcid.org/0000-0001-7723-4161^1,2^na1,
Zhaoyu Yang^1,3^na1,
Changhan Chen ORCID: orcid.org/0009-0009-8342-5680^1,2^na1,
Fang Yu^1,4^na1,
Mengzhi Wu^1,2,
Zhouzhou Yao^1,2,
Yuhua Fan^1,5,
Tingting Xie^1,6,
Linlin Wan^1,7,
Tiansheng Chou^1,2,
Xianjing Feng^1,4,
Hao Qi^1,2,
Yuyu Chou ORCID: orcid.org/0000-0003-3050-0746^1,2,
Juan Zhao^2,8,
Juan Liu^1,2,
Zhiyu Yang^1,2,
Yujin Zhang^1,2,
Rodney E. Kellems^1,9 &
…
Yang Xia^1,4,5,10

Cell Discovery volume 12, Article number: 19 (2026) Cite this article

4140 Accesses
10 Altmetric
Metrics details

Subjects

Abstract

Aging-related diseases are aggravated by tissue hypoxia; however, the underlying mechanism remains unknown. Here, we report that the oxygen (O₂) release capacity of red blood cells (RBCs) gradually decreases with age and is closely associated with aging-related tissue dysfunction. Metabolomic profiling of human and mouse RBCs and genetic studies in mice revealed that the reduction in 2,3-bisphosphoglyceric acid (2,3-BPG) content mediated by a decrease in bisphosphoglycerate mutase (BPGM) activity is a metabolic checkpoint underlying decreased RBC O₂ release capability and dysfunction with advancing age. When glucose metabolism is impaired, erythroid inosine, transported by equilibrative nucleoside transporter 1 and converted to ribose 1-phosphate by increased purine nucleoside phosphorylase (PNP) activity, is an important compensatory fuel for RBCs during aging. In a preclinical study, inosine supplementation successfully alleviated the age-dependent reduction in BPGM activity that mediates glucose metabolic impairment, decreased O₂ delivery, and tissue dysfunction. Finally, we unexpectedly discovered that 2,3-BPG acts as an inhibitor of PNP in RBCs by competing with the phosphate (Pi)-binding domain and interacting with residues serine 33 and alanine 116. Our studies revealed that impaired glucose metabolic reprogramming resulting from decreased BPGM activity underlies RBC bioenergetic decline and is a novel hallmark of aging. As 2,3-BPG levels decrease during aging, its inhibitory effect on PNP is reduced, resulting in increased PNP activity and inosine catabolism as an alternative fuel, suggesting that inosine is a potential rejuvenating therapy.

Erythropoiesis–inosine metabolic axis failure underlying retinal neurodegeneration in glaucoma: novel diagnoses and therapies

Article Open access 13 February 2026

Comprehensive metabolic profiling across five lifespan stages in murine hippocampus and cortex reveals sex-related variation in age-related cognitive decline

Article Open access 19 January 2026

Sphingosine-1-phosphate suppresses GLUT activity through PP2A and counteracts hyperglycemia in diabetic red blood cells

Article Open access 14 December 2023

Introduction

The aging population is increasing rapidly, overwhelming health care systems and creating an economic burden worldwide¹. Elderly individuals are widely considered to have increased susceptibility to multiple diseases, including cardiovascular disease, obesity, metabolic disorders, and neurodegenerative diseases^2,3,4. Given that aging-related diseases are frequently driven by tissue microenvironment hypoxia⁵, extensive research has been devoted to understanding the cellular and molecular mechanisms underlying tissue response to hypoxia during aging^6,7,8. Hypoxia-induced transcription factors (HIFs) orchestrate changes in the expression of genes involved in vascular angiogenesis, metabolism, and immune function to combat hypoxia-mediated tissue damage and the progression of aging^9,10,11,12. However, the specific cellular and molecular mechanisms initiating and driving age-dependent tissue hypoxia remain unknown. Additionally, how to mitigate age-related tissue microenvironment hypoxia, slow the onset of functional decline, and promote healthy aging are extremely challenging in both basic and clinical research.

Mature red blood cells (RBCs), the most abundant and only cell type that delivers oxygen (O₂) in the body, are extremely sensitive to hypoxia and can release more O₂ in response to both physiological hypoxia and pathological hypoxia^13,14,15. Recent studies have made significant progress in revealing that although they lack a nucleus, mitochondria, and other complex organelles, RBCs are well equipped to promote O₂ release in normal individuals under high-altitude hypoxia or in patients with pathological hypoxia owing to tightly regulated HIF-independent but glucose-dependent metabolic reprogramming^16,17,18,19. However, owing to a lack of transcriptional activity, specific age-induced changes in RBCs cannot be effectively detected by current high-throughput omics profiling methods, such as transcriptomics and single-cell sequencing^20,21,22. Although research regarding RBC metabolomics in the context of aging and related diseases is limited, emerging studies highlight the metabolic changes that occur in RBCs during aging. For example, using high-throughput metabolomics, D’Alessandro’s team revealed that glycolysis is influenced by sex, age, and race in stored RBCs and confirmed that ATP and hypoxanthine can serve as biomarkers for the recovery of RBC function post-transfusion¹⁸. Independent metabolomics research revealed that arginine metabolism in RBCs is a biomarker of organismal aging²³, suggesting new potential targets for addressing the sequelae of aging. Notably, early parabiosis studies have indicated that multiple plasma circulating molecules, including chemokines, proteins, and enzymes, present in young blood have the potential for rejuvenation²⁴. However, functional and metabolic changes that occur in erythrocytes, including their ability to deliver O₂ and act as a hypoxia sensor, during natural aging, the specific mechanisms underlying these functional and metabolic alterations in RBCs in response to aging and their contribution to aging-related tissue hypoxia and dysfunction remain unclear. Thus, a specific means to target RBC function and counteract aging-related tissue functional decline is lacking.

Therefore, we sought to fill these important knowledge gaps with the goal of promoting healthy aging by employing multidisciplinary approaches, including accurately defining RBC O₂ delivery capability in a large human cohort across the lifespan, multiple sophisticated genetically modified mice, nonbiased high-throughput metabolomic profiling, cutting-edge isotopic labeling and tracing experiments with inosine, preclinical studies, molecular modeling, mutagenesis and competitive enzymology assays. Here, we define decreased erythrocyte bioenergetics as a new hallmark of aging; reveal new significant systemic, cellular, metabolic, and molecular insights into age-induced RBC metabolic and functional decline and their contribution to aging-induced peripheral tissue hypoxia and dysfunction; and develop an innovative approach to increase RBC metabolism and O₂ delivery for rejuvenation. Overall, our study highlights a new but compelling concept that “aging is not merely peripheral tissue hypoxia and dysfunction but also a systemic erythroid hypoxia–metabolic syndrome” and that targeting RBCs is a new approach to promote healthy aging.

Results

Decreased RBC oxygen delivery capacity is an early hallmark of aging linked to age-dependent tissue dysfunction

To determine whether decreased O₂ delivery capacity is a gradual, early and common pathogenic feature that occurs with aging, we recruited an independent cohort of 301 human subjects (Supplementary Table S1) ranging in age from 20 to 85 years (168 males, 133 females) and measured functional changes in their RBCs (Fig. 1a). With respect to RBC function, we measured the P50, which is important for assessing the ability of RBCs to release O₂, by accurately measuring Hb-O₂ binding affinity. We observed a significant yet gradual decrease in P50 with age (Fig. 1b), which represents increased O₂ binding affinity and reduced O₂ release, with no significant differences between males and females (Supplementary Fig. S1a). These results show that reduced RBC O₂ delivery ability is a common and early feature of the aging process in both sexes and raises the possibility that RBC functional decline (i.e., a reduced ability to deliver O₂ to peripheral tissues) promotes aging-related tissue functional decline.

**Fig. 1: Age-dependent changes in RBC function and metabolism are linked to changes in tissue function.**

To assess this possibility, we conducted linear regression analysis of age and clinical parameters (adjusted for BMI and sex) to identify clinical parameters associated with age-dependent functional decline. We found that 14 clinical characteristics were positively correlated with age, while 11 clinical features were negatively associated with age (Supplementary Fig. S1b). Notably, cardiovascular parameters (increased blood pressure), liver function indicators (elevated AST and TBA levels), renal function indicators (increased blood uric acid and urea levels and a decreased eGFR) and metabolic impairment (increased glucose and triglyceride levels) are clinical parameters commonly associated with age-dependent functional decline (Fig. 1c). Next, to precisely define and compare the interactive links between RBC functional decline and multiple tissue dysfunctions during aging, we used the GSVA R package to calculate the age-related RBC, renal, liver and blood pressure (BP) functional scores (Supplementary Fig. S1c). We observed that RBC and renal functional scores decreased with age, whereas the BP score significantly increased with age (Fig. 1d), which further explains the decrease in tissue function during aging. Further pairwise comparisons of the calculated specific functional scores revealed that the RBC functional score was significantly positively correlated with the renal function score and significantly and negatively correlated with the BP score (Fig. 1e; Supplementary Table S2). Thus, we determined that the gradual decrease in RBC O₂ release capacity is an early aging-related cellular dysfunction that is strongly linked to the progression of aging-related tissue dysfunction.

Age-dependent decreases in BPGM-mediated glucose metabolism underlie RBC functional decline

Owing to the lack of transcriptional and translational activity, metabolomics is the most appropriate omics method for identifying changes associated with RBC function during aging. Thus, we conducted high-throughput unsupervised metabolomics analysis of RBCs isolated from human subjects ranging in age from 20 to 85 years. Approximately 537 metabolites were detected in the RBCs (Supplementary Fig. S1d), and the QC samples in the PCA score plot clustered well, with 90% of the metabolites having a coefficient of variation of less than 30%, ensuring the high quality of our RBC metabolomic profiling data (Supplementary Fig. S1e, f). The PLS-DA score plot revealed differences in the metabolic fingerprints during aging (Fig. 1f). We identified 111 metabolites that were positively correlated and 81 metabolites that were negatively correlated with age (Pearson’s correlation, BH-adjusted P < 0.05, adjusted for BMI and sex; Supplementary Fig. S1g, Data S1). Pathway enrichment analysis of aging-related metabolites revealed that glycolysis, pentose phosphate pathway (PPP), purine metabolism, and glutathione metabolism were markedly altered during aging (Fig. 1g). These metabolic pathways are closely related to the regulation of O₂ release, energy balance, and ROS homeostasis in erythrocytes. With respect to glycolysis, we observed significant increases in the levels of glycolytic intermediates, including glucose/fructose-6-phosphate (G6P/F6P), fructose-1,6-bisphosphate (F-1,6-BP), NAD⁺, ATP, pyruvate, and lactate, during aging (Fig. 1h, i; Supplementary Fig. S1h). With respect to the levels of PPP intermediates, including 6-phosphogluconate (6PG), glutathione (GSH), sedoheptulose-7-phosphate (Sed7P), and ribose-1/5-phosphate (R1P/R5P), significantly increased with age (Fig. 1h, i). We also observed an increase in RBC reactive oxygen species (ROS) levels with aging (Fig. 1j). These results suggest that the PPP pathway is significantly activated during aging to counteract the increase in ROS content. Additionally, we observed that the level of 2,3-bisphosphoglycerate (2,3-BPG, a glycolytic metabolite generated by the RBC-specific Rapoport–Luebering shunt (RLS) and a known allosteric modulator that regulates Hb-O₂ binding affinity²⁵) gradually decreased with aging, with a trajectory similar to that of P50 (Fig. 1k). These findings prompted us to measure the activity of BPG mutase (BPGM), the first enzyme in the RLS. BPGM activity was significantly lower in the RBCs of elderly individuals than in those of young individuals (Fig. 1l). These results suggest that despite the increase in RBC glycolysis with aging, the suppression of the RLS accounts for the reduction in 2,3-BPG activity.

Similar to the findings in humans, we observed that the P50 of RBCs in 15-month-old WT mice was significantly lower than that in young 4-month-old WT mice, while the ROS level was significantly greater (Fig. 1m, n). Consistent with RBC human metabolomics signature, metabolomic profiling of mouse RBCs revealed significant differences in the metabolic fingerprints between young and aged mice (Fig. 1o), which was characterized by elevated lactate and pyruvate levels but decreased 2,3-BPG levels in aged mice compared with young WT mice (Fig. 1p; Supplementary Fig. S1i). Moreover, the ratio of GSH to GSSG significantly increased in aged mice (Fig. 1p; Supplementary Fig. S1i). Thus, our data demonstrated that humans and mice experienced similar age-dependent decreases in O₂ delivery capability and 2,3-BPG activity along with increased glycolysis and PPP activity to combat increased oxidative stress (Fig. 1q).

RBC-specific genetic ablation of BPG mutase reduces O2 release and accelerates aging

Extending upon our data from human studies, we conducted proof-of-principle genetic studies to determine the essential role of erythrocyte BPGM with advancing age. Briefly, we generated mice with RBC-specific ablation of BPGM (eBpgm^–/– mice) by mating Bpgm^f/f mice with epoR-Cre mice (for details, see the Materials and methods) and compared RBC function and aging-related phenotypes, including primarily cognitive ability, muscle function, and glucose tolerance, between the control (Bpgm^f/f) and eBpgm^–/– mice at 4 and 15 months of age (Fig. 2a). Western blot analysis confirmed that knockout of BPGM was specific to erythrocytes, as BPGM expression was unchanged in other tissues (Supplementary Fig. S2a). No significant differences in body weight were detected between Bpgm^f/f and eBpgm^–/– mice at 4 and 15 months of age (Supplementary Fig. S2b). The P50 of eBpgm^–/– mice was significantly lower than that of Bpgm^f/f mice at both 4 and 15 months (Fig. 2b), whereas the ROS levels in the RBCs of eBpgm^–/– mice, but not in the RBCs of Bpgm^f/f mice, significantly increased with age, and the ROS levels were significantly greater in the RBCs of 15-month-old eBpgm^–/– mice than Bpgm^f/f mice of the same age (Fig. 2c). These results indicate that genetic ablation of BPGM in RBCs leads to a decrease in O₂ release and an increase in oxidative stress. Consistent with the findings of a previous study from the Mayo Clinic Experience²⁶, RBC counts and Hb and HCT levels were higher in eBpgm^–/– mice than in Bpgm ^f/f mice at both 4 and 15 months of age (Supplementary Table S3), indicating that the genetic ablation of erythrocyte-specific BPGM leads to erythrocytosis as a compensatory response to overcome insufficient O₂ delivery from mature RBCs.

**Fig. 2: Genetic ablation of BPGM in RBCs accelerates aging.**

Next, we applied the Barnes maze (BM) test and novel object recognition (NOR) test to evaluate the differences in cognition (i.e., learning and memory abilities) between Bpgm^f/f and eBpgm^–/– mice (Fig. 2d). Compared with 4-month-old eBpgm^–/– mice, 15-month-old eBpgm^–/– mice took more time to find the escape box on Day 5 following the 4-day training period, but the time needed by Bpgm^f/f mice did not significantly change between 4 and 15 months of age (Fig. 2e). Compared with 4-month-old eBpgm^–/– mice, 15-month-old eBpgm^–/– mice exhibited worse short-term and long-term memory, but the memory of Bpgm^f/f mice did not significantly change during aging (Fig. 2f). Moreover, as Bpgm^f/f mice aged, they did not show a significant change in terms of exploring new objects, whereas new object exploration by eBpgm^–/– mice significantly decreased. Similar to the BM test results, eBpgm^–/– mice exhibited less interest in investigating novel objects than Bpgm^f/f mice did at 15 months (Fig. 2g). We also assessed muscle function using the rotarod test and grip strength test. Compared with 4-month-old eBpgm^–/– mice, 15-month-old eBpgm^–/– mice had a reduced ability to remain on the rotarod, while there was no significant difference in duration on the rotarod between 15-month-old and 4-month-old Bpgm^f/f mice (Fig. 2h). The grip strength test revealed that the 15-month-old eBpgm^–/– mice displayed significantly lower grip strength than age-matched Bpgm^f/f mice (Fig. 2h). Additionally, we analyzed glucose homeostasis in the mice using an intraperitoneal glucose tolerance test. We observed that the ability of Bpgm^f/f mice to metabolize glucose did not change significantly between 4 and 15 months of age, whereas the ability of eBpgm^–/– mice to metabolize glucose decreased significantly with age (Fig. 2i). Moreover, glucose metabolism by 4-month-old Bpgm^f/f and eBpgm^–/– mice did not differ, but compared with age-matched Bpgm^f/f mice, 15-month-old eBpgm^–/– mice metabolized glucose less efficiently (Fig. 2i). Additionally, the hypoxyprobe staining results indicated that hypoxia in the muscles and hippocampal CA3 region significantly increased in eBpgm^–/– and Bpgm^f/f mice with age, that hypoxia in the muscles and hippocampal CA3 sections was significantly greater in 4- and 15-month-old eBpgm^–/– mice than that in age-matched Bpgm^f/f mice (Fig. 2j, k), and that there were no significant differences in the levels of hypoxia in the hippocampal CA1 and DG regions between eBpgm^–/– and Bpgm^f/f mice (Fig. S2c). Research has shown that the CA3 region of the hippocampus is closely related to learning and memory ability²⁷. Finally, metabolomics analysis of RBCs between 15-month-old eBpgm^–/– and Bpgm^f/f mice was performed to evaluate differences in key aging-related metabolites. We observed significant differences in the RBC metabolic fingerprints between 15-month-old eBpgm^–/– and Bpgm^f/f mice of the same age (Fig. 2l). In particular, compared with those in 15-month-old Bpgm^f/f mice, the 2,3-BPG, ATP, 6PG levels and the ratio of GSH to GSSG in the RBCs of eBpgm^–/– mice were significantly lower (Fig. 2m), indicating that the antioxidant stress response in and O₂ release capacity of 15-month-old eBpgm^–/– mice were reduced. In addition, the levels of inosine in RBCs were significantly lower in 15-month-old eBpgm^–/– mice than in 15-month-old Bpgm^f/f mice, whereas the Sed7P, Ery4P and R5P/R1P levels were significantly greater in 15-month-old eBpgm^–/– mice than 15-month-old Bpgm^f/f mice (Fig. 2m). These results suggest that inosine metabolism and the nonoxidative pathway of PPP are more active in 15-month-old eBpgm^–/– mice than in 15-month-old Bpgm^f/f mice, similar to the results observed in aging humans. Taken together, our findings provide genetic evidence that decreased BPGM-mediated 2,3-BPG production in erythrocytes underlies aging-induced erythrocyte glucose metabolic impairment and functional decline and drives peripheral tissue hypoxia and dysfunction with aging.

Increased PNP activity induces inosine-derived R1P as a compensatory fuel for RBCs in elderly individuals

In addition to the glucose metabolic reprogramming observed in RBCs during aging, our metabolomics results revealed that purine metabolism in RBCs was also significantly impacted during aging (Fig. 1g). Specifically, the level of inosine in RBCs significantly decreased with age (Fig. 3a), similar to the gradual decreases in the 2,3-BPG level and P50. The decrease in inosine content in RBCs as a function of age was accompanied by an increase in hypoxanthine and ribose-1/5-phosphate (R1P/R5P) (Fig. 3a). Similarly, we detected significant changes in purine metabolism in aged mice. In particular, the levels of hypoxanthine and R1P/R5P were significantly increased in aged mice compared with young mice (Supplementary Fig. S3b), indicating inosine catabolism increases with age in mice. These findings prompted us to measure the activity of PNP, the enzyme that catalyzes the conversion of inosine to hypoxanthine and R1P. The results revealed that PNP activity gradually increased during aging (Fig. 3b). Our results demonstrated that increases in PNP activity and inosine catabolism to increase R1P/R5P levels are common features of aging in both humans and mice. Early studies reported that the ribose produced from inosine degradation is an important carbon source for the body, which can undergo glycolysis to provide ATP or enter the pentose phosphate pathway²⁸. Thus, a gradual increase in RBC PNP activity with age contributes to the aging-dependent catabolism of inosine, indicating that inosine-derived ribose may be an additional fuel involved in the enhanced glycolysis and the PPP pathway activity in the RBCs of elderly individuals.

**Fig. 3: The eENT1-PNP axis provides inosine-derived R1P as a compensatory fuel to improve RBC function and combat aging.**

To address this possibility, we conducted in vitro experiments in which RBCs from 28-year-old human volunteers and 2-month-old mice were incubated with inosine to determine the effects of inosine on RBC function (Fig. 3c). We observed that inosine significantly increased the P50 of cultured human RBCs in a dose-dependent manner within 1 h, peaking at 100 μmol (Fig. 3d). Moreover, inosine significantly reduced the ROS levels in cultured human RBCs (Fig. 3d). Given that ENT1 is the major transporter for inosine uptake in RBCs²⁹, we hypothesized that ENT1 is a potential candidate for the uptake of inosine. To test this hypothesis, we treated cultured primary human and murine RBCs with a specific ENT inhibitor in the presence of inosine. We found that the inosine-mediated increase in P50 and decrease in ROS levels were significantly blocked by dipyridamole (DIP, a specific ENT inhibitor) in both cultured human and murine RBCs (Fig. 3e, f). Thus, we provide both human and murine evidence that inosine enters RBCs through the ENT1 transporter, directly increasing O₂ release and reducing ROS levels. Next, given that PNP is a key enzyme that drives inosine catabolism and on the basis of our finding that PNP activity increases during aging, we determined the importance of PNP in the ENT1-dependent uptake of inosine-mediated increased O₂ delivery and anti-oxidative stress capability, in primary cultured RBCs from young (n = 4, 30 years) and elderly individuals (n = 4, 71 years) in the presence of PNP inhibitors. Intriguingly, forodesine (For, a specific PNP inhibitor) significantly attenuated the inosine-mediated increase in P50 and decrease in ROS levels in cultured RBCs from both young and elderly individuals (Fig. 3g). Notably, the P50 induced by inosine in RBCs from elderly individuals was greater than that in RBCs from young individuals, which further supports our findings that elderly individuals have increased RBC PNP enzyme activity (Fig. 3g).

Moreover, early studies showed that adenosine promotes RBC function via activation of the A2B receptor (ADORA2B)³⁰_, and in view of the fact that adenosine can be converted to inosine via adenosine deaminase (ADA) catalysis, we were prompted to further delineate whether inosine-induced RBC function is dependent on adenosine generated by ADA signaling. In contrast to the gradual increase in PNP activity (Fig. 3b), there were no significant changes in ADA activity during aging (Supplementary Fig. S3c). Moreover, adenosine treatment directly increased P50 and reduced ROS levels in cultured primary RBCs from both young and elderly individuals, whereas neither forodesine nor 2’-deoxycoformycin (DCF, an ADA inhibitor) abolished these adenosine-mediated effects (Fig. 3h, i), ruling out the possibility that the adenosine-mediated increase in RBC function is dependent on the activity of ADA or PNP. In contrast, adenosine-induced functional changes in RBCs, including changes in the P50 and anti-ROS capability, were significantly reduced in cultured primary RBCs from WT mice treated with PSB1115 (an ADORA2B inhibitor) and in RBCs from Adora2b^–/– mice (Supplementary Fig. S3d, e). These results indicate that adenosine primarily improves the P50 and ROS levels in erythrocytes through ADORA2B, providing both pharmacological and genetic evidence that adenosine exerts its effects through ADORA2B in RBCs. Thus, we demonstrated that the regulatory effects of adenosine and inosine on erythrocyte function differ: adenosine functions through the A2B receptor, whereas inosine acts through PNP-mediated inosine metabolism.

Next, to directly elucidate how inosine is metabolized inside RBCs during aging, [¹³C₅-ribose]-inosine metabolic flux tracing analysis was performed. Briefly, we cultured primary human RBCs isolated from young (n = 3, 30 years) and elderly (n = 3, 71 years) individuals in the presence of isotopically labeled inosine and unlabeled glucose for up to 2 h, followed by flux analyses to determine the fate of the inosine-derived ¹³C₅-ribose (Fig. 3j; Supplementary Data S2). Initially, we observed that the [¹³C₅]-inosine concentration in the supernatant was significantly lower in the elderly group than in the young group after 0.5–1 h of incubation (Supplementary Fig. S3f), whereas the level of inosine in the RBCs did not significantly differ (Supplementary Fig. S3g); however, the levels of [¹³C₅]-R1P/R5P were significantly higher in RBCs from the elderly group than in RBCs from the young group after 1–2 h of incubation (Fig. 3k). These findings are consistent with the increased activity of PNP in RBCs from elderly individuals (Fig. 3b), which converts [¹³C₅]-inosine to [¹³C₅]-R1P. Further tracing analyses revealed that inosine-derived R1P was metabolized into glycolysis intermediates, with significantly higher levels of [¹³C₃]-F-1,6-BP, [¹³C₃]-G3P, [¹³C₃]-2-PG/3-PG, [¹³C₃]-PEP and [¹³C₃]-lactate in RBCs from older individuals (Fig. 3k). Inosine-derived R1P also fueled the expansion of the PPP in RBCs from the elderly group, as evidenced by the increased incorporation of [¹³C₆] into 6-PG and [¹³C₄]-Sed7P (Fig. 3k). Finally, we found that the increase in the level of inosine-derived R1P/R5P also resulted in increased incorporation of R1P-derived ¹³C into 2,3-BPG in RBCs from the elderly group (Fig. 3k). Overall, we revealed that the ENT1-mediated uptake of inosine coupled with increased PNP activity-induced inosine-derived R1P production is an additional mechanism that counteracts age-related glucose metabolism impairment as a compensatory mechanism by promoting glycolysis, PPP and even RLS in the RBCs of elderly individuals (Fig. 3l).

eENT1-dependent inosine uptake is essential for promoting RBC function and combating aging

To assess the importance of ENT1 in RBC function and its contribution to peripheral tissue hypoxia and functional decline during aging, we generated erythrocyte-specific ENT1-deficient mice (eEnt1^–/–) and assessed its effects on O₂ release and aging-related phenotypes (Fig. 4a). First, complete blood count tests revealed that compared with Ent1^f/f mice, eEnt1^–/– mice had lower RBC counts and Hb and HCT levels at 15 months, whereas the MCV were larger than that of Ent1^f/f mice (Supplementary Table S4), indicating that 15-month-old eEnt1^–/– mice exhibited an anemic phenotype, which is consistent with previous reports³¹. Next, we observed that eEnt1^–/– mice exhibited significantly reduced O₂ release and significantly increased ROS levels during aging, whereas no significant difference was detected between 4- and 15-month-old Ent1^f/f mice. At 15 months of age, eEnt1^–/– mice had significantly lower P50 values than Ent1^f/f mice did, while ROS levels were significantly higher in eEnt1^–/– mice than in Ent1^f/f mice (Fig. 4b). Additionally, inosine failed to induce O₂ delivery and anti-ROS capacity in RBCs from 4-month-old eEnt1^–/– mice (Supplementary Fig. S4a), providing direct evidence that eENT1 is essential for the uptake of inosine and its subsequent effects on improving RBC O₂ delivery and anti-ROS capability.

**Fig. 4: Genetic ablation of ENT1 in RBCs accelerates aging.**

Finally, we compared the cognitive function of young (4-month-old) and aged (15-month-old) male Ent1^f/f and eEnt1^–/– mice by performing the BM test and novel object test. The results of the BM test revealed that compared with 4-month-old eEnt1^–/– and Ent1^f/f mice, 15-month-old eEnt1^–/– and Ent1^f/f mice took more time to find the escape exit, whereas at 4 months of age, eEnt1^–/– and Ent1^f/f mice did not show significant differences in the time taken to find the escape exit (Fig. 4c). However, at 15 months of age, compared with age-matched controls, eEnt1^–/– mice took significantly more time to find the escape exit on Days 3 and 4 of the training period (Fig. 4c). Short-term memory and long-term memory were significantly reduced in aged eEnt1^–/– and Ent1^f/f mice, whereas compared with age-matched Ent1^f/f mice, 15-month-old eEnt1^–/– mice exhibited worse short-term memory and long-term memory (Fig. 4d). The NOR test results revealed that the interest of eEnt1^–/– and Ent1^f/f mice in exploring new objects did not significantly differ at 4 months of age, but at 15 months, eEnt1^–/– mice demonstrated significantly less interest in exploring new objects than age-matched Ent1^f/f mice did (Fig. 4e). Furthermore, the rotarod test and grip strength test were performed to assess muscle function. Compared with 15-month-old Ent1^f/f mice, 15-month-old eEnt1^–/– mice spent less time on the rotarod, but the difference was not significant at 4 months of age (Fig. 4f). The grip strength tests revealed that the grip strength of 15-month-old eEnt1^–/– mice was significantly worse than that of age-matched Ent1^f/f mice (Fig. 4g). No difference was observed among 4-month-old mice. A glucose tolerance test revealed that the ability of Ent1^f/f mice and eEnt1^–/– mice to metabolize glucose did not significantly differ at 4 months of age (Fig. 4h). Moreover, the glucose metabolism of Ent1^f/f mice did not significantly change with age (Fig. 4h). However, compared with 4-month-old eEnt1^–/– mice, 15-month-old eEnt1^–/– mice metabolized glucose less efficiently (Fig. 4h). Further tissue staining with a hypoxia probe revealed that hypoxia in the muscles and hippocampal DG, CA1, and CA3 regions significantly increased with age in both Ent1^f/f and eEnt1^–/– mice, whereas 15-month-old eEnt1^–/– mice exhibited more severe hypoxia in the muscles and hippocampal DG region than Ent1^f/f mice did (Fig. 4i, j; Supplementary Fig. S4b). Additionally, metabolomic analysis revealed significant differences in RBC metabolic fingerprints between 15-month-old eEnt1^–/– and Ent1^f/f mice (Fig. 4k). The levels of 2,3-BPG, ATP, 6PG and GSH in the RBCs of 15-month-old eEnt1^–/– mice were significantly lower than those of 15-month-old Ent1^f/f mice (Fig. 4l), indicating that the O₂ release capacity and anti-oxidative stress capability of 15-month-old eEnt1^–/– mice were impaired. Furthermore, the levels of inosine, hypoxanthine, Sed7P, Ery4P and R5P/R1P in the RBCs of 15-month-old eEnt1^–/– mice were significantly lower than those in the RBCs of 15-month-old Ent1^f/f mice (Fig. 4l). These results suggest that inosine metabolism and the nonoxidative pathway of the PPP activity are lower in 15-month-old eEnt1^–/– mice than in 15-month-old Ent1^f/f mice, similar to the results observed in aging humans. Thus, we provided genetic evidence that RBC ENT1 is essential for counteracting the age-dependent decline in RBC function (i.e., O₂ release), cognitive ability, muscle function and glucose and inosine metabolism.

Preclinical studies: Inosine supplementation improves aging-related multifunctional decline

Extending upon the results from our genetic studies, we conducted preclinical studies in WT mice to determine whether inosine supplementation can combat the aging-related decreases in RBC, vascular, muscular, and cognitive function and general tissue hypoxia. Specifically, 16-month-old mice were injected intraperitoneally with inosine every day for 4 weeks, and their P50, blood pressure, body weight, muscle strength and cognitive function were monitored (Fig. 5a). Compared with those in the vehicle group, the RBCs in the inosine-injected group showed an increase in the P50 and a reduction in ROS levels (Fig. 5b), but the blood pressure and weight were not significantly different (Supplementary Fig. S5a, b). Thus, this in vivo preclinical study provides evidence that inosine increases O₂ release capacity and reduces ROS levels in the RBCs of 16-month-old mice.

**Fig. 5: Inosine supplementation impedes age-related RBC and tissue functional decline.**

Aging is accompanied by a progressive loss of muscle strength³². In this study, forelimb grip strength was measured in naturally aged animals with or without inosine treatment (Fig. 5c). Compared with vehicle-injected age-matched control mice, 16-month-old mice treated with inosine exhibited improved forelimb grip strength. Additionally, as shown in Fig. 5c, inosine treatment improved the performance of aged mice in the rotarod test, as they remained on the rotarod significantly longer. Finally, to explore the effects of inosine on spatial recognition and memory, control and inosine-injected mice were subjected to the BM test and NOR test. Compared with age-matched controls, inosine-treated mice were more interested in investigating novel objects (Fig. 5d). Spatial learning and memory were assessed using the BM test (Fig. 5e). Compared with the vehicle-injected control group, the inosine-treated group took significantly less time to find the escape box (Fig. 5f, g). Additionally, we found that the level of hypoxia in the hearts, kidneys, and muscles of the inosine-treated mice was significantly reduced (Fig. 5h). Thus, we provide preclinical evidence that inosine supplementation improves the aging-related decreases in cognitive function and muscle strength and reduces the onset of tissue hypoxia.

Finally, to elucidate the potential cellular effects in the brain mediated by inosine treatment, we conducted H&E staining, hypoxia probe analysis, and immunostaining of the hippocampus, a major region of the brain for learning processing and long-term memory storage. H&E staining of tissue from saline-treated mice revealed extensive necrosis in three regions of the hippocampus accompanied by pyknosis (Fig. 5i). In contrast, inosine treatment reduced pyknosis and necrosis in brain tissue. Additionally, Nissl staining revealed that, compared with saline treatment, inosine treatment increased the density of neurons in the dentate gyrus (DG) and cornu ammonis (CA1 and CA3) regions of the hippocampus (Fig. 5j). The hypoxyprobe staining results of the DG and CA1 and CA3 regions revealed that the staining intensity of the inosine-treated group was significantly lower than that of the saline-treated group (Fig. 5k, m). GFAP staining revealed no obvious difference in astrocyte antigen expression between the saline-treated and inosine-treated groups (Supplementary Fig. S5c; Fig. 5m). We then evaluated the changes in the number and morphology of microglia through Iba1 antibody staining in the DG and CA1 and CA3 regions of the hippocampus (Fig. 5l, m). Compared with those in the saline-treated group, the number of microglia in the inosine-treated group was significantly lower, and more branches were observed, indicating a more active morphology (Fig. 5k, l). Taken together, the results of our preclinical studies demonstrate that inosine treatment promotes O₂ release from and reduces ROS levels in RBCs; counteracts age-related heart, kidney, muscle, and hippocampus hypoxia, inflammation, and neuron degeneration; and improves cognitive ability and muscle strength.

2,3-BPG competes with Pi to bind to the phosphate-binding domain of PNP as an endogenous inhibitor

Given our findings that increased PNP activity-mediated R1P production from inosine is an important and compensatory mechanism for maintaining RBC bioenergetics during aging, we aimed to define the molecular basis underlying PNP activation during aging. Therefore, we revisited our metabolomics profiling results and conducted correlation studies of PNP activity with all the significantly altered metabolites during aging. Intriguingly, we found that the 2,3-BPG level was significantly negatively correlated with PNP activity during aging in humans (Fig. 6a), suggesting a new but compelling possibility that 2,3-BPG, as a small metabolite, inhibits PNP activity. Extending upon the results from these correlation studies in humans, we took advantage of our erythrocyte-specific BPGM knockout mice (eBpgm^–/–). Like that in humans, PNP activity was significantly elevated in RBCs of 15-month-old Bpgm^f/f mice compared with those of 4-month-old young mice. Surprisingly, we observed significantly higher PNP activity in eBpgm^–/– mice than in Bpgm^f/f control mice at 4 months of age and a further increase in PNP activity in eBpgm^–/– mice at 15 months of age compared with the controls (Fig. 6b). Thus, we concluded that compared with control mice, mice lacking the ability to produce 2,3-BPG in their RBCs had greater PNP activity at both 4 and 15 months of age, immediately suggesting that 2,3-BPG is likely a metabolite that can inhibit PNP activity in RBCs.

**Fig. 6: 2,3-BPG affects the degradation of inosine in RBCs by inhibiting PNP activity.**

To determine whether 2,3-BPG can directly inhibit PNP activity, we conducted an in vitro assay to quantify the PNP activity in the absence and presence of different physiologically relevant concentrations of 2,3-BPG. The results (Fig. 6c) revealed that 2,3-BPG inhibited PNP activity in a concentration-dependent manner. Next, to determine whether 2,3-BPG inhibits PNP under more physiological conditions, we prepared RBC extracts from 4-month-old eBpgm^–/– mice and measured PNP activity in the presence of different concentrations of 2,3-BPG. The results (Fig. 6d) revealed that 2,3-BPG inhibited PNP activity in RBC extracts in a dose-dependent manner. Thus, we demonstrated that 2,3-BPG directly inhibits PNP activity in both a purified PNP system and RBC extracts within the concentration range normally found in RBCs.

Having demonstrated the inhibitory effect of 2,3-BPG on PNP activity, we sought to determine the underlying molecular mechanism. First, we performed a molecular docking analysis to assess the potential binding between PNP and 2,3-BPG. The results revealed that 2,3-BPG interacted with a cavity on the protein surface with a binding energy of –7.3 kcal/mol, indicating the high binding affinity of 2,3-BPG to PNP (Fig. 6e). Next, we examined the interactions of the phosphate groups of 2,3-BPG, which formed four hydrogen bonds with the Ala116, Ser33, and His86 residues of PNP (Fig. 6e). PNP active site prediction analysis revealed that the predicted active pocket of PNP contains the Ala116, Ser33, and His86 residues (Supplementary Fig. S6a), further supporting the possibility that 2,3-BPG inhibits PNP activity by binding to its active pocket. Thus, we performed molecular dynamics simulations to determine the binding stability between 2,3-BPG and the active site of human PNP. The root mean square deviation (RMSD) values are often used to determine the stability of protein–ligand complexes, with a smaller RMSD value representing a more stable complex³³. Thus, quantification of the RMSD values by constructing RMSD curves revealed that the curves of the 2,3-BPG and PNP complexes nearly overlapped, fluctuating within 0.5 nm and stabilizing at 0.38 ± 0.15 nm, indicating the strong binding stability of 2,3-BPG to the PNP protein (Fig. 6f). Given that the number of hydrogen bonds reflects the strength of protein–ligand binding³⁴, the results of the atomic analyses further supported the high binding affinity between 2,3-BPG and PNP owing to the high hydrogen bond density and strength (Fig. 6g). Moreover, the MM/GBSA method of GROMACS was applied to calculate the free binding energies of the 2,3-BPG and PNP complexes to describe the lowest energy conformations from the molecular dynamics simulations. If protein‒ligand interactions are weak, multiple rough low-energy clusters are present in the free energy distribution. Conversely, strong and stable interactions lead to the formation of an almost single and smooth energy cluster in the potential energy distribution³⁴. Consistent with the RMSD curve results (Fig. 6g), the free energy distribution diagrams revealed that the 2,3-BPG and PNP complex formed a single minimum energy cluster with high complex stability (Fig. 6h). Therefore, we provide multiple lines of evidence that 2,3-BPG strongly binds to the active site of PNP.

Early studies demonstrated that the active site of PNP includes residues for binding purine, ribose, and phosphate (Pi)³⁵. Among all of these residues, Ala116, Ser33, and His86 in PNP were found to interact with its endogenous substrate, phosphate (Pi)³⁶. Given that the concentrations of both 2,3-BPG and Pi in RBCs are in the millimolar range³⁷ and on the basis of our genetic, molecular modeling, and atomic analyses, we hypothesized that 2,3-BPG inhibits PNP activity by competing with Pi to bind to these phosphate combined residues. To test this hypothesis, we utilized an in vitro competition assay to quantify PNP activity in the presence of free Pi and different concentrations of 2,3-BPG. Using this assay, we observed a decrease in activity when 2,3-BPG was added to the RBC lysate of 4-month-old eBpgm^–/– mice, and increasing the Pi concentration reduced the inhibitory effect (Fig. 6i). This Lineweaver–Burk double reciprocal plot revealed a common intercept on the y axis, indicating competitive inhibition, which is consistent with the active site docking prediction.

Finally, to determine whether 2,3-BPG has an inhibitory effect on PNP activation by competing for binding to these three amino acid residues in the active site of PNP identified by molecular docking and molecular dynamics simulations, we mutated Ser33 and His86 of PNP to alanine and Ala116 to phenylalanine, after which we performed protein expression experiments in HEK293T cells (Fig. 6j). Western blot analyses confirmed the successful expression of the WT and mutated PNP proteins in HEK293T cells (Supplementary Fig. S6b). Consistent with previously published studies³⁸, we confirmed that mutation of Ser33, His86 or Ala116 led to a significant decrease in PNP activity in HEK293T lysates (Fig. 6k). To determine which residues are responsible for 2,3-BPG binding to the active sites of PNP and in turn contribute to the inhibitory activity of 2,3-BPG, we added different concentrations of 2,3-BPG to the lysates of HEK293T cells expressing the WT and mutated PNP proteins and subsequently measured PNP activity. The inhibitory effect of 2,3-BPG on PNP activity was maintained in lysates from cells expressing the WT or H86A-mutated PNP protein but completely abolished in lysates from cells expressing the A116F or S33A mutants, indicating that 2,3-BPG exerts an inhibitory effect by interacting with the Ser33 and Ala116 residues of PNP (Fig. 6l, m). Taken together, the results of sophisticated multidisciplinary molecular experiments, including molecular docking, molecular dynamics simulations coupled with in vitro competition functional assays and mutation analyses, revealed that 2,3-BPG acts as an endogenous inhibitor of PNP by competing with free Pi to bind to residues Ser33 and Ala116 in the PNP active site. The gradual decrease in the level of 2,3-BPG in RBCs with age allows for increased PNP activity, thereby promoting inosine catabolism to produce an additional carbon source during aging (Fig. 6n).

Discussion

Although RBCs are the only cell type that delivers O₂, the significance of RBCs in aging remains enigmatic. Here, we report that decreased erythroid BPGM-mediated 2,3-BPG production, as a newly identified hallmark of aging, contributes to RBC glucose metabolic impairment, RBC functional decline and peripheral tissue hypoxia and dysfunction with aging (Fig. 7). Moreover, we revealed glucose and purine metabolic crosstalk via the 2,3-BPG-PNP axis during aging and highlight inosine as an alternative fuel to combat impaired glucose metabolic reprogramming and initiate rejuvenation to increase RBC metabolism and oxygen delivery (Fig. 7). Overall, our work adds RBC metabolic and functional decline to a list of hallmarks of aging and positions inosine as an innovative intervention for age-related hypoxia and tissue dysfunction.

O₂ (known as the “elixir of life”) is essential for maintaining tissue energy metabolism, function and survival³⁹. As the most abundant cell type, RBCs are very sensitive to hypoxia, and their role in aging and the underlying molecular mechanisms remain unknown. We report here that the RBC O₂ release capacity gradually decreases from an early age, with no significant differences between sexes; these findings are consistent with those of our previous study⁴⁰. Similar to earlier findings, the elderly group in our cohort displayed aging-related clinical outcomes, including higher levels of inflammation⁴¹, liver and kidney dysfunction⁴², and impaired glucose‒lipid metabolism⁴³, than young individuals did. Additionally, we revealed that decreased O₂ release capacity is strongly linked to age-related metabolic impairment and tissue dysfunction. Our findings highlight the novel and compelling concept that an age-dependent decline in RBC function is a previously unrecognized hallmark of aging that is strongly linked to metabolic impairment and age-related functional decline in multiple tissues.

Mature RBCs lack nuclei and mitochondria⁴⁴ and act as both a deliverer and sensor of O₂⁴⁵, which is vital for maintaining the normal function and survival of every cell within the body. Owing to the large amount of O₂ carried by RBCs, they contain a well-defined system to defend against oxidative stress⁴⁶. Thus, RBCs rely on sophisticated glucose metabolic reprogramming, including glycolysis, RLS and the PPP, to effectively deliver O₂ and maintain their anti-oxidative stress capacity²⁵. Notably, one of the most well-known allosteric modulators is 2,3-BPG, a byproduct of glycolysis known for regulating hemoglobin O₂ binding affinity, which is generated primarily by BPGM, a key enzyme in the RLS that is highly enriched in RBCs^45,47. Consistent with our comprehensive metabolomics findings, early studies reported that the RBC 2,3-BPG concentration in humans decreases with age⁴⁸. In contrast, we revealed that glycolysis, the PPP and purine catabolism are significantly activated in RBCs during aging. These findings are strongly supported by early studies that revealed that the activity of hexokinase and pyruvate kinase, which are involved in glycolysis, did not significantly change in RBCs during aging but is significantly increased in patients with Alzheimer’s disease⁴⁷. Here, we demonstrated that BPGM activity decreases with increasing age in humans. Functionally, we provide genetic evidence that erythroid-specific ablation of BPGM in mice decreases 2,3-BPG production and O₂ delivery and accelerates age-related functional decline. Thus, both human and mouse genetic studies have added a significant new chapter to our understanding of erythrocyte BPGM-mediated 2,3-BPG production during aging. Furthermore, the PPP is critical for RBCs to produce sufficient amounts of NADPH and thus maintain an abundance of GSH, which acts as a ROS scavenger to defend against cellular oxidative stress⁴⁸. Consistent with the findings of early studies⁴⁹, we observed elevated ROS levels along with increased PPP activity and GSH levels in our aging study. Our metabolomics results revealed that the reduction in 2,3-BPG production caused by decreased BPGM activity is key to the decrease in oxygen release by RBCs, whereas the activation of the PPP is the compensatory metabolic basis for resistance to the increased levels of ROS. Further genetic studies in mice provided proof-of-principle evidence that decreased erythrocyte BPGM-mediated 2,3-BPG production is an essential component underlying aging-induced erythrocyte glucose metabolic impairment, RBC functional decline and peripheral tissue hypoxia and dysfunction with aging. This work highlights the novel and compelling concept that erythroid BPGM-mediated glucose metabolic impairment is a central mechanism and “previously overlooked” hallmark of aging underlying reduced RBC function and thereby driving age-dependent tissue hypoxia and dysfunction.

Glucose is the major fuel that maintains RBC bioenergetics, and our findings, including those from both the human and mouse genetic studies, show that inosine, which is taken up by an ENT1-dependent mechanism, is an additional fuel for RBCs owing to increased PNP-driven inosine catabolism, compensating for the impaired glucose metabolism in these cells during aging. Our results show that in RBCs from older adults, inosine-derived ribose is directed toward glycolysis, PPP and RLS to counteract the age-dependent decline in RBC function. The concentration of inosine in RBCs is in equilibrium with that of extracellular inosine via the action of ENT1. Early studies indicated that ENT1 is highly expressed in RBCs³¹ and that ENT1-mediated adenosine uptake plays an essential role in increasing O₂ delivery by inducing A2B-AMPK-dependent BPGM activation in the contexts of chronic kidney disease²⁹ and fetal growth restriction⁵⁰. Notably, earlier studies revealed that inosine or adenosine helps preserve ATP and 2,3-BPG levels in stored RBCs⁵¹, suggesting that these compounds may serve as preservatives for the storage of both blood and other organs⁵². Recent studies have demonstrated that inosine increases fat burning or provides energy in adipocytes, cancer cells and T cells through either adenosine receptors or the transporter ENT1^53,54,55. In contrast to the increased PNP activity-mediated inosine catabolism, we detected no changes in ADA activity during human aging. These findings led to our discovery that adenosine-induced erythrocyte O₂ delivery and anti-oxidative stress capacity are independent of ADA and PNP and further confirmed an early report that adenosine functions primarily via its surface receptor, ADORA2B, to improve RBC function³⁰. Our findings are strongly supported by those of previous studies showing that unlike inosine, adenosine does not serve as an alternative carbon source to enhance glycolysis and the TCA cycle in T cells⁵⁵. However, inosine-mediated metabolic reprogramming in erythrocytes and its specific function and underlying mechanisms during aging remained unresolved prior to our current studies. Here, we provide initial in vitro and in vivo evidence that ENT1-PNP-facilitated inosine uptake and metabolism are essential for directly inducing O₂ offloading and anti-ROS activity. We also provide in vivo evidence that specific genetic ablation of ENT1 in mouse RBCs led to a decrease in O₂ release from RBCs and accelerated aging. There is currently no approach to promote oxygen delivery to combat aging. Extending upon the results from genetic studies, we provide both in vitro and in vivo preclinical evidence that inosine can directly increase O₂ delivery from RBCs, mitigate tissue hypoxia and slow aging-related functional decline. Therefore, our findings support the novel concept that glucose is not the only fuel for RBCs and that inosine-derived R1P metabolized by PNP owing to its increased activity is an additional fuel that can compensate for improper glucose metabolism to combat aging-related tissue hypoxia, organ damage and dysfunction. These findings are extremely significant and provide a foundation for future translational studies on rejuvenation and even all types of physiological and pathological hypoxia.

Emerging studies have reported that 2,3-BPG is involved in the regulation of various protein functions in addition to directly binding to Hb, particularly in the areas of O₂ transport and metabolic regulation. For example, recent research revealed that 2,3-BPG is involved in phosphorylating a histidine residue in phosphoglycerate mutase 1 (PGAM1), which catalyzes the conversion of 3-PG to 2-PG, increasing its activity and thus regulating energy metabolism⁵⁶. Additionally, hypoxanthine, IMP, and inosine are components of the oxypurine cycle in RBCs. Early studies reported that the oxypurine cycle is significantly correlated with 2,3-BPG levels⁵⁷. In addition, previous research revealed that 5’-nucleotidase might be activated by 2,3-BPG, which promotes IMP phosphorylation⁵⁸. Furthermore, early studies have shown that 2,3-BPG can inhibit AMP deaminase activity in various animal tissues⁵⁹. These studies suggest that 2,3-BPG is involved in regulating purine metabolism. However, the role of 2,3-BPG in the regulation of PNP activity remained unknown prior to the current study. Our human metabolomics results unexpectedly revealed that R5P or R1P and elevated PNP enzyme activity are significantly negatively correlated with the level of 2,3-BPG. Our human metabolomics findings are strongly supported by early studies showing that during storage, BPGM activity and 2,3-BPG levels decrease, while PNP enzyme activity increases⁶⁰. Here, we provide proof-of-principle genetic evidence that PNP enzyme activity is significantly increased in mice with erythroid-specific BPGM mutation. The active site of PNP can be divided into three binding regions depending on the moiety it accommodates: the purine binding and sugar binding regions accommodate substrate moieties, while the phosphate binding region accommodates inorganic phosphate. Crystal structure studies have revealed that the residues Ser33, His86, and Ala116 of PNP, especially Ser33 and His86, interact with Pi through direct hydrogen bonding^38,61. Early studies demonstrated that acyclovir diphosphate can bind directly to the phosphate binding site of PNP and compete with Pi, thereby reducing the reaction rate⁶². The phosphate-containing drug developed by Zlatko Janeba also binds to these sites, forming hydrogen bonds and increasing the efficacy of PNP inhibitor drugs⁶³. Intriguingly, our molecular docking and molecular dynamic simulations revealed that 2,3-BPG has the potential for strong binding to the Pi binding site and that the two phosphate groups of 2,3-BPG provide the possibility for phosphate competition. Our phosphate competition experiments confirmed the competitive inhibitory effect of 2,3-BPG on PNP activation. Moreover, PNP site-directed mutagenesis provided functional evidence that 2,3-BPG inhibits PNP enzyme activity primarily by binding to Ser33 and Ala116. Overall, we determined that 2,3-BPG acts as an endogenous inhibitory modulator of PNP activity by competitively binding to the Pi site in mature RBCs. Thus, the 2,3-BPG-induced decrease in PNP activity is a compensatory response to increase inosine catabolism and generate additional carbon sources to combat glucose metabolic impairment during aging. These findings are extremely innovative and add a significant new knowledge regarding RBC biology and metabolism during aging.

In conclusion, in both human and murine studies, we revealed that the BPGM-dependent decline in RBC bioenergetics is a significant and novel hallmark of aging and plays a critical role in driving age-induced tissue hypoxia and dysfunction. At the cellular and metabolic levels, we further revealed that the ENT1-dependent uptake of inosine and increase in PNP-mediated R1P production from inosine is an alternative fuel to counteract glucose metabolism impairment, induce oxygen delivery by RBCs and combat age-induced tissue hypoxia and tissue dysfunction. At the molecular level, we revealed that the erythroid glycolytic intermediate 2,3-BPG is an endogenous inhibitory modulator of PNP, a key enzyme of purine metabolism. Thus, the age-dependent decrease in 2,3-BPG content increases erythroid PNP activity and inosine catabolism to produce an alternative fuel to overcome impaired glucose metabolism and promote oxygen delivery as novel compensatory glucose-mediated purine metabolism machinery. Overall, our study significantly advances our understanding of aging from a completely new angle, elucidating the dysfunction of RBCs during aging from a metabolic perspective and proposing a novel concept in which tissue damage during aging is a result of the impairment of oxygen release from erythrocytes. By repositioning RBCs as active regulators of healthy aging, we open new avenues for rejuvenation.

Materials and methods

Volunteer recruitment and data collection

The Hunan Aging Cohort Study was established in Hunan Province, China, and was conducted with the approval of the Research Ethics Committee of Xiangya Hospital of Central South University with informed consent from volunteers (202310207). The collection of biological samples and data in this study was performed in accordance with the guidance of the Human Genetic Resource Administration, Ministry of Science and Technology of the People’s Republic of China.

The blood samples and clinical information from 301 recruited individuals are shown in Supplementary Table S1. All volunteers were considered qualified and included in this study on the basis of the following criteria: (1) healthy adults over 18 years of age; (2) no severe diseases (such as cancer, cardiovascular disease, diabetes mellitus, autoimmune/inflammatory, or severe gynecological disease); (3) no persistent drug or alcohol abuse; (4) no other clinical trial participation within the past three months; (5) no use of hormone medication (past six months), anti-platelet drugs, or cholinesterase inhibitors for the treatment of Alzheimer’s disease; and (6) no extreme vigorous exercise, acute injury or other abnormal behaviors for one month before sample collection. For young females with regular menstrual cycles, blood was collected one week before menstruation.

Animals

All animal protocols were performed in accordance with the National Institutes of Health guidelines and approved by the experimental animal ethics committee of Central South University, Changsha, Hunan Province, China (CSU-2023-0336). All experimental animals were housed under controlled conditions at a temperature of 22 °C with 50% humidity and lights programmed to be on from 7 am to 7 pm. Food and water were available ad libitum. Mice with erythrocyte-specific deletion of BPGM (eBpgm^–/– mice) were generated by crossing mice that were homozygous for a floxed Bpgm allele with EpoR-Cre⁺ mice. Mice with erythrocyte-specific deletion of ENT1 (eEnt1^–/– mice) were generated by crossing mice homozygous for a floxed Ent1 allele with EpoR-Cre⁺ mice. Four-month-old and 15- and 16-month-old C57BL/6 WT mice were purchased from Aniphe Biolaboratory, Inc. C57BL/6 WT mice were bred in a pathogen-free, temperature- and humidity-controlled facility on a 12-hour light and dark cycle. Sick animals were excluded from the study. Mice were labeled with ear tags, and all the researchers were blinded to the experimental procedures.

Human biological sample pretreatment and storage

All blood samples were collected after an overnight fast to avoid dietary influence. Each 10 mL venous whole blood sample was collected from individuals by a professional nurse. One 5 mL blood sample was collected in an EDTA anticoagulant tube, while another 5 mL sample was collected in a standard tube. After collection, the tubes were immediately sent to the laboratory and stored at 4 °C, after which the biochemical tests were performed. The blood in the EDTA tube was centrifuged at 4 °C (500× g, 10 min) to separate the plasma and blood cells. Plasma and RBCs were further centrifuged at 4 °C (4000 rpm, 10 min) to obtain the plasma, platelets, and RBCs. All blood samples were stored at –80 °C for long-term preservation.

Measurement of P50 and ROS levels of RBCs

To measure the P50, 20 μL of whole blood was added to 3 mL of Hemox buffer (TCS Scientific Corporation) with 5 μL of anti-foaming reagent (TCS Scientific Corporation) and 5 μL of 22% BSA in PBS. The mixture was then analyzed with a Hemox analyzer (TCS Scientific Corporation) for data measurement at 37 °C and subsequent construction of the oxygen equilibrium curve. The Hemox analyzer was used to determine the P50 from the oxygen equilibrium curve at 37 °C. The ROS level was measured using a ROS Assay Kit (BL714A, Biosharp) according to the manufacturer’s protocol. Briefly, 1 μL of RBC pellet was incubated with a ROS fluorescence probe (H2DCFDA, 2 μM) for 45 min at 37 °C in the dark. The samples were then washed 3 times with PBS and resuspended in 200 μL of PBS. ROS levels were determined by measuring the fluorescence intensity with a microplate reader (Synergy HTX Multi-Mode Reader, BioTek).

Intraperitoneal administration of inosine or vehicle to mice

The effect of inosine was investigated in aged 16-month-old C57BL/6 male mice. Mice were injected intraperitoneally with vehicle (90% saline) or inosine (20 mg/kg body weight) daily for 1 month and culled at the age of 17 months for terminal assays. The dosage of inosine was determined according to published articles related to inosine treatment⁶⁴. For consistency, rigor, and reproducibility, mice were injected at the same time each day (10:00 am), each physiological assay was performed at the same time of the day (indicated separately for each assay), and all animals were fed the same normal chow.

Health analysis

The physiological functions and health of various organs in 17-month-old mice were assessed after inosine supplementation and included the following: body weight, blood pressure, rotarod and grip strength tests to indicate neuromuscular strength, and the NOR and Barnes maze tests to examine behavior (functional tests are described below). Blood collected from 17-month-old mice was subjected to biochemical testing and P50 measurement.

Blood pressure measurement

Mouse blood pressure was measured using a noninvasive BP system (CODA High Throughput System with 8 Activated Channels; Kent Scientific; Torrington, CT) as previously described¹⁹. Briefly, mice were placed on a temperature-controlled platform such that the tail temperature was between 32 °C and 35 °C, and blood pressure was monitored using a volume pressure recording sensor. Training was conducted once a day for 3 days before minipump implantation. For each mouse blood pressure measurement, the first 5 cycles were considered the adaptation period, followed by 20 additional cycles of recorded blood pressure measurements. The average of these 20 cycles was used for blood pressure analysis.

Rotarod test

Motor coordination and balance were tested using a rotarod apparatus as described previously between 10:00 am and 12:00 noon⁶⁵. The apparatus consists of a plastic rod (diameter of 6 cm) and five equal sections separated by six disks (each with a diameter of 25 cm), which allows five mice to be tested simultaneously. Mice were trained on the rotarod apparatus for 3 consecutive days before the test day. On the test day, mice were acclimated to the environment for 30 minutes. The time at which the mice fell off the rod (rotating at a speed of 10 or 30 rpm) was recorded. There was a 5 min break between each speed test. The mean latency time of falling off the apparatus was taken as the performance value.

Grip strength measurement

The grip strength test was used to measure the maximal muscle strength of the forelimbs. The test was based on assessing a mouse grasping a grid with sensors, as previously described⁶⁵.

NOR test

The NOR test was performed as described previously⁶⁶. The mice were placed in an open-top, transparent acrylic resin box with bedding that was divided into 4 quadrants (Q1, Q2, Q3, and Q4). Each mouse was initially placed in the empty chamber and allowed to freely explore the arena without objects for 10 minutes to adapt to the test environment. Next, the mice started the training procedure, in which each mouse was placed in the arena with 2 identical objects (Q2Block+Q3Block) for 5 minutes each day for a total of 3 days. On Day 3, the final trials were performed by placing the mice in the arena and exposing them to one of the previous objects and a novel object (Q2Block+Q3Cylinder) for 5 minutes. The recognition times and motion trajectories associated with the novel object were recorded by TopScan 2.0 software (CleverSys) and used for subsequent analysis.

Barnes-maze test

The Barnes maze test was performed as described previously⁶⁶. A circular platform with 40 holes and surrounded by visual clues was set as the arena. One stationary escape box was set on one hole that was not changed throughout the experiment for mice the to navigate to, with negative stimuli (buzzer and light) provided as clues to exit. On Day 1, each mouse was placed in the area and guided by the researcher to the escape box without negative stimuli for familiarization. Next, 2 rounds of adaptation were performed, which were the same as the familiarization procedure but included the negative stimuli. Two rounds of acquisition trials were completed by placing the mice in the arena and allowing them to explore freely to find the exit for no more than 3 minutes. Four acquisition trials were completed on Days 2 to 5. Days 1 to 4 of the trials were termed the training period. Then, the next day after training, the short-term memory was assessed, and the long-term memory was assessed one month or one week after treatment. Primary latency to reach the escape box was recorded by the TopScan system and used to assess spatial learning and memory.

Glucose tolerance tests

For the glucose tolerance tests, glucose (2 g/kg BW) was injected intraperitoneally (i.p.) after an overnight (16 h) fast, and blood glucose was monitored using blood glucose strips and an Accu-Check glucometer (Roche) at indicated times as described previously⁶⁷. Afterward, the area under the curve was calculated using the R software.

Detection of tissue hypoxia

The level of hypoxia in heart, liver, kidney, muscle, and brain tissue was assessed using a Hypoxyprobe kit (Hypoxyprobe Omni Kit, Hypoxyprobe, Inc.) according to previously described methods²⁹. Briefly, mice were intraperitoneally injected with pimonidazole (50 mg/kg body weight) 30 min before anesthesia. Transcardiac perfusion was performed before organ collection, after which the tissues were fixed in 10% buffered formalin and embedded in paraffin to prepare 4-micron sections. The sections were deparaffinized by a series of immersions in xylene and rehydrated in a graded series of alcohol and distilled water. Antigen retrieval was performed by boiling the sections in sodium citrate buffer (pH 6.0) at 95 °C for 30 min. After blocking with 3% BSA in PBS at room temperature for 1 h, the sections were incubated overnight with rabbit anti-pimonidazole antibody (PAb2627AP, 1:100 in 0.5% BSA) in a humid environment at 4 °C. Following primary antibody incubation, incubation was performed with the Alexa Fluor 594-labeled goat anti-rabbit secondary antibody (A-11037; Thermo Fisher; 1:1000 in 3% BSA in PBS). Negative control samples were incubated with the secondary antibody only. The sections were mounted with ProLong Gold antifade reagent containing DAPI (P36935; Invitrogen). The images were imported into ImageJ, and the percentage of the area with positive staining for a certain antibody relative to the total structural area was calculated under the same threshold settings. Afterward, the fluorescence signal intensity was quantified.

Histological analysis

Mouse tissues (heart, liver, kidney, muscle, and brain) were fixed in 10% buffered formalin and sent to Servicebio Biotechnology Co., Ltd. (Hubei, China) to be prepared into 4-micron sections and subsequent hematoxylin‒eosin staining or Nissl staining. The slides were examined under a light microscope to detect tissue damage. After being embedded in Tissue-Tek optimal temperature compound (Sakura Finetek, Torrance, CA, USA), 10 μm sections of the brains were prepared on a cryostat, mounted on slides, and stored at –20 °C. Coronal brain sections with similar anatomical locations near the bregma (−2 mm) were selected according to the mouse brain atlas. The frozen sections were washed with PBS, blocked, and permeabilized with 0.3% Triton X-100 (Sigma, St. Louis, MO, USA) and 10% BSA (Sigma) in PBS at room temperature for 2 h. Sections were incubated with primary antibodies diluted in blocking buffer (0.3% Triton X-100 and 10% BSA in PBS) overnight at 4 °C. After being washed 3 times with PBS for 5 minutes each, the sections were incubated with secondary antibodies diluted in blocking buffer for 1 h at room temperature. After being washed 3 times (5 min each), the sections were mounted with ProLong Gold Antifade Mountant with DAPI (P36935; Invitrogen). Primary antibodies used were anti-GFAP (Invitrogen, Cat# Pa1-10019; 1:500; Waltham, MA, USA) and anti-Iba1 (Wako, Cat# 019-19741; 1:500; Osaka, Japan). Image acquisition was performed using a Zeiss LSM 880 confocal microscope, and the images were processed using Zen Blue Edition Lite 3.0 software (Carl Zeiss, Oberkochen, Germany) to generate maximum intensity projections of each image using a 20× objective lens through the tiling and positioning module and Z-stacks. Afterward, the Iba1 and GFAP fluorescence signal intensities were quantified using ImageJ software. The images were imported into ImageJ, and the percentage of the area with positive staining for a certain antibody relative to the total structural area was calculated under the same threshold settings.

Western blotting

Western blot analysis was performed as previously described²⁹. Briefly, 25 mg of tissue was ultrasonically lysed in cold water supplemented with 1× protease inhibitor cocktail (Roche, Cat# 04906837001) at a ratio of 1:10, after which 10× PBS was added for equilibration. Protein concentration was measured using a BCA protein assay kit (ThermoFisher Scientific). The samples were loaded onto 10% SDS‒PAGE gels for Western blot analysis. Total protein was incubated with primary antibodies against BPGM (Proteintech, #17173-1-AP, 1:2000) or β-actin (Proteintech, #66009-1-Ig, 1:3000), followed by incubation with secondary antibodies.

Metabolomics analysis

The RBC samples were thawed at 4 °C, and 20 μL aliquots were mixed with 200 μL of a mixture of cold methanol/acetonitrile (methanol:acetonitrile:water 5:3:2 v/v/v) to remove the protein. The mixed samples were then vortexed continuously for 30 min at 4 °C and then centrifuged at 18,000× g for 10 min at 4 °C. Supernatants were injected onto the ultrahigh-pressure liquid chromatography‒mass spectrometry (UHPLC‒MS) system composed of a Vanquish UHPLC coupled to a Q Exactive mass spectrometer (Thermo Fisher Scientific, Bremen, Germany) to obtain RBC metabolic fingerprints using a 5-minute gradient as previously described²⁹. Briefly, the metabolites were separated with a Kinetex C18 column (Waters, 150×2.1 mm, 1.7 μm; Phenomenex, 00F-4475-AN). The injection volume was 5 μL, and the flow rate was 0.4 mL/min. The column temperature was maintained at 45 °C, and the autosampler temperature was set at 4 °C. For detection in positive ion mode, the mobile phase consisted of 0.1% formic acid in water (A) and 0.1% formic acid in acetonitrile (B) with gradient elution as follows: 0–0.5 min at 5% B; 0.5–1.1 min from 5–95% B; 1.1–2.75 min at 95% B; 2.75–3 min from 95–5% B; and 3–5 min at 5% B. In negative ion mode, mobile phases were composed of 1 mM ammonium acetate in 95% water (A, 5% acetonitrile) and 1 mM ammonium acetate in 95% acetonitrile (B, 5% water) with gradient elution as follows: 0–0.5 min at 0% B; 0.5–1.1 min from 0–100% B; 1.1–2.75 min at 100% B; 2.75–3 min from 100–0% B; and 3–5 min at 0% B. The mass spectrometer was operated in full MS mode at a resolution of 70,000, the scan range was 65–900 m/z, the maximum injection time was 200 ms, 2 microscans were performed, automatic gain control was set to 33106 ions, the source voltage was 4.0 kV (for both positive and negative ion modes), the capillary temperature was 320 °C, and the sheath gas (45), auxiliary gas (15), and sweep gas (0) were all nitrogen. Samples were randomly injected in positive and negative ion modes independently. The QC sample was injected six times prior to sample analysis and then injected at regular intervals (every ten samples) thereafter to monitor the stability of the system. The QC samples were prepared as composite samples and contained 5 μL of each sample.

The raw MS data were first converted to mzXML files using RawConverter (Scripps Research Institute) and then subjected to peak picking (mz diff = 10 ppm) and grouping (bw = 5, mz window = 0.025, minfrac = 0.2) with EI-Maven software (version 0.12) on the basis of the endogenous metabolite database HMDB. The isotopic abundance ratios of metabolites were calculated using EI-Maven software. Metabolite identification was conducted by matching the metabolite isotopic relative abundance ratio values and comparing the accurate m/z values with MS/MS spectra with an in-house database.

In vitro RBC intervention experiments

RBCs were isolated from the blood of young (30 years) and elderly (71 years) humans, WT mice, and RBC-specific knockout mice using sodium heparin. For in vitro glucose flux experiments in which the RBCs were treated with inosine, erythrocytes were isolated from the blood of young volunteers using sodium heparin as an anticoagulant. The erythrocytes were subsequently washed 3 times with F-10 Nutrient Mix (Invitrogen, #0905074EF), suspended to 10% hematocrit, and then transferred to 1.5 mL tubes in triplicate. For metabolite treatment, different concentrations of inosine (5 μM, 50 μM, 100 μM, and 500 μM) were added to the erythrocytes, which were then incubated for 1 h, after which the P50 was measured to determine the optimal concentration. Then, the erythrocytes were incubated with 100 μM inosine ± 5 μM dipyridamole (an ENT1 inhibitor), 100 μM inosine ± 5 μM forodesine (a PNP inhibitor), 100 μM adenosine, 100 μM adenosine ± 5 μM forodesine, 100 μM adenosine ± 5 μM 2’-deoxycoformycin (an ADA inhibitor), or adenosine ± 20 μM PSB1115 (an ADORA2B inhibitor) for 1 h at 37 °C. Finally, the erythrocytes were washed three times in cold PBS, after which the P50 and ROS concentrations were measured.

¹³C₅-Inosine tracing experiments with cultured RBCs

Erythrocytes were isolated from the blood of young (30 years) and elderly (71 years) humans using sodium heparin as an anticoagulant for inosine metabolic flux experiments. The erythrocytes were washed 3 times with F-10 Nutrient Mix (Invitrogen), suspended to 10% hematocrit, and then transferred to 1.5 mL tubes in triplicate. For certain experiments, erythrocytes (20 μL) were cultured in the presence of 100 μM [¹³C_5-ribose]-inosine (MedChemExpress) for 0.5 h, 1 h, or 2 h at 37 °C (3 samples at each time point). Next, the erythrocytes were washed three times with ice-cold PBS and subjected to metabolic quenching in cold acetonitrile. Finally, erythrocytes and supernatants were collected and lysed as described above and analyzed by the Vanquish UHPLC coupled to a Q Exactive MS (Thermo Fisher Scientific, Bremen, Germany). Metabolite identification and isotopologue distributions were performed with EI-Maven software (Princeton, NJ) and the R package AccuCor⁶⁸.

PNP activity assay

PNP enzyme activity was determined according to previous methods⁶⁹. Briefly, RBCs or 293 T cells were diluted (1:1) in PBS containing 0.3% Triton X-100. PNP activity was determined spectrophotometrically by adding 1 μL of the cell lysate to 200 μL of reaction buffer containing 50 mM KH₂PO₄ (pH 7.5), 1 mM inosine, and 10 mU xanthine oxidases (Sigma). Reaction progress was monitored at 293 nm for 30 min at 30 °C. After the reaction was complete, RBC hemoglobin concentration was determined by measuring the absorbance at 414 nm, and the PNP activity was normalized to the hemoglobin concentration. The total protein levels in the cell lysates were determined using a BCA protein assay kit, and the PNP activity of the 293 T cells was normalized to the total protein concentration.

ADA activity assay

ADA activity was determined by measuring inosine production by coupling the ADA reaction with purine nucleoside phosphorylase and xanthine oxidase reactions⁷⁰. ADA activity was assayed under saturated substrate conditions at 37 °C in a reaction mixture containing 2 mM adenosine, 10 mU xanthine oxidases, 10 mU purine nucleoside phosphorylase, and 50 mM sodium phosphate (pH 7.4), in a total volume of 200 μL, of which 1 μL was the cell lysate. The increase in absorbance at 293 nm resulting from the deamination of adenosine to yield inosine and urate was continuously monitored with a spectrophotometer, and the rate of inosine production was calculated by linear regression.

2,3-BPG competition PNP enzyme assays

A 0.1 mg/mL solution of purine nucleoside phosphorylase (PNP; 1ku; HARVEY BIO) in 100 mM Tris/HCl (pH = 7.5) was prepared. The enzymatic activity assay was performed, and the enzyme unit was defined as previously described⁷¹. The enzymes were assayed spectrophotometrically at 25 °C. The standard reaction mixture contained 1 mM inosine in sodium phosphate buffer (different concentrations, pH 7.5) and xanthine oxidase (0.05 U/mL) in a final volume of 0.2 mL. The phosphorolysis of inosine was similarly assessed by measuring the absorbance at 293 nM in the presence of xanthine oxidase (0.05 U/mL) and different concentrations of 2,3-BPG. The reaction rate, as determined by the measured increase in A293 due to the formation of uric acid, was proportional to the enzyme concentration. One enzyme unit is defined as the amount of enzyme that catalyzes the phosphorolysis of 1 μmol of inosine/min, corresponding to an increase in A293 under the conditions described. Specific activity is expressed as units/mg of protein or Hb.

BPG mutase activity assay

Ten microliters of erythrocytes were lysed in Tris-HCl buffer (pH 7.4) containing 0.5% Triton X-100 and protease inhibitors (Roche, Cat# 04906837001). Then, 5 μL of erythrocyte lysate was incubated at room temperature for 30 min in a 100 μL reaction mixture containing 100 mM triethanolamine (pH 7.6), 1 mM magnesium sulfate, 4 mM adenosine triphosphate (ATP), 3 mM 3-phosphoglycerate, and 10 units of phosphoglycerate kinase. Afterward, 5 μL of 11.63 M perchloric acid was added to stop the reaction, and the sample was subsequently centrifuged at 10,000× g for 5 min at 4 °C. The supernatant (8 μL) was mixed with 10 μL of 2.5 M K₂CO₃, vortexed, and centrifuged at 10,000× g for 5 min at 4 °C. Finally, 10 μL of the supernatant was used for 2,3-BPG quantification using a commercially available kit as previously described⁷².

Molecular docking prediction of the 2,3-BPG binding site in PNP

Molecular docking was used to predict the binding affinity between 2,3-BPG and PNP. The three-dimensional PNP structure was obtained from the UniProt database, and the two-dimensional structure of 2,3-BPG was obtained from the PubChem database (https://pubchem.ncbi.nlm.nih.gov/) and converted into a three-dimensional structure using Chem3D software. All the small molecule was converted into Pdb format using Open Babel for energy minimization in a pH 7.4 environment, followed by hydrogenation with PyMOL (https://www.pymol.org/), and converted into PDBQT format. Afterward, the affinity between 2,3-BPG and PNP was calculated using AutoDock Vina (https://vina.scripps.edu/)⁷³. An affinity ≤ –5.0 kcal/mol indicated a strong interaction. PyMOL software was used to visualize the results. The CavityPlus online tool (https://www.pkumdl.cn/cavityplus) was used to predict the active pockets of the PNP protein, and the best active pocket was determined on the basis of DrugScore and Druggability rankings⁷⁴.

Molecular dynamics simulations

Molecular dynamics simulations of the 2,3-BPG-PNP complexes were performed using GROMACS 2022.4 software⁷⁵. Amber Tools was used to add GAFF force field to the small molecule, whereas Gaussian 16 W is used to add hydrogen atoms to the small molecule and calculate the RESP potential⁷⁶. Potential data was added to the topology file of the molecular dynamics system. The simulations were carried out at a static temperature of 300 K and atmospheric pressure (1 bar). Amber99sb-ildn was used as a force field, water molecules were used as solvent (TIP3P water model), and the total charge of the system was neutralized by adding an appropriate number of Na⁺. The steepest descent method was applied to minimize the energy, and then equilibration was performed with the isothermal isovolumic (NVT) ensemble and isothermal isobaric ensemble (NPT) ensemble for 100,000 steps, with a coupling constant of 0.1 ps for a total duration of 100 ps. Finally, free molecular dynamics simulations were performed. The process consisted of 5,000,000 steps, the step length was 2 fs, and the total duration was 100 ns. After the simulation was completed, the built-in tool of the software was used to analyze the trajectory, and the RMSD, root-mean-square fluctuation (RMSF), and protein radius of gyration of each amino acid trajectory were calculated and combined with the free energy (MMGBSA), free energy topography, and other data.

Correlation analysis

Clinical phenotype and metabolomics data correlation analyses were performed using the pcor.test function of the ggcor R package (version 1.1) via the Pearson correlation method to determine aging-related characteristics and the relationship between erythrocytes and organ function.

Calculation of age-related change score

The age-related change score measures the change tendency of group features with age and is calculated on the basis of different functional signatures. The GSVA algorithm⁷⁷ from the GSVA R package was applied to evaluate the age-related change score of each sample.

Statistical analysis

All the data were analyzed using the R language (version 4.4.1). Each dataset was normalized to the sum of the spectral integrals and subjected to Pareto scaling. Statistical significance was determined using an unpaired two-tailed Student’s t test or one-way ANOVA with Dunnett’s post hoc multiple comparisons analysis for comparisons of three or more groups. Pathway enrichment analysis of the metabolites was conducted using the clusterProfiler R package⁷⁸. The visualization, correlation network, and pathway enrichment results were performed using the igraph packages and ggplot2 R packages, respectively.

Data availability

Differential metabolite datasets and tracing datasets underlying individual display items in the manuscript are included in Supplementary Data S1, S2. This paper does not report the original code.

References

Chen, C., Ding, S. & Wang, J. Digital health for aging populations. Nat. Med. 29, 1623–1630 (2023).
Article PubMed CAS Google Scholar
Aujla, N. et al. New horizons in systems engineering and thinking to improve health and social care for older people. Age Ageing 53, afae238 (2024).
Article PubMed PubMed Central Google Scholar
Reicher, L. et al. Phenome-wide associations of human aging uncover sex-specific dynamics. Nat. Aging 4, 1643–1655 (2024).
Article PubMed PubMed Central Google Scholar
Zhang, S. et al. A metabolomic profile of biological aging in 250,341 individuals from the UK Biobank. Nat. Commun. 15, 8081 (2024).
Article PubMed PubMed Central CAS Google Scholar
Yeo, E.-J. Hypoxia and aging. Exp. Mol. Med. 51, 1–15 (2019).
PubMed PubMed Central CAS Google Scholar
Kusumbe, A. P. et al. Age-dependent modulation of vascular niches for haematopoietic stem cells. Nature 532, 380–384 (2016).
Article PubMed PubMed Central CAS Google Scholar
Yamamoto, T. & Isaka, Y. Pathological mechanisms of kidney disease in ageing. Nat. Rev. Nephrol. 20, 603–615 (2024).
Article PubMed Google Scholar
Chen, X., Shi, C., He, M., Xiong, S. & Xia, X. Endoplasmic reticulum stress: molecular mechanism and therapeutic targets. Signal Transduct. Target. Ther. 8, 352 (2023).
Article PubMed PubMed Central CAS Google Scholar
Arjamaa, O., Nikinmaa, M., Salminen, A. & Kaarniranta, K. Regulatory role of HIF-1alpha in the pathogenesis of age-related macular degeneration (AMD). Ageing Res. Rev. 8, 349–358 (2009).
Article PubMed CAS Google Scholar
Nisar, A. et al. Hypoxia and aging: molecular mechanisms, diseases, and therapeutic targets. MedComm (2020) 5, e786 (2024).
Article PubMed PubMed Central CAS Google Scholar
Semenza, G. L. Hypoxia-inducible factors in physiology and medicine. Cell 148, 399–408 (2012).
Article PubMed PubMed Central CAS Google Scholar
Dandavate, V. et al. Hepatic BMAL1 and HIF1α regulate a time-dependent hypoxic response and prevent hepatopulmonary-like syndrome. Cell Metab. 36, 2038–2053.e5 (2024).
Article PubMed CAS Google Scholar
Sun, K. et al. Sphingosine-1-phosphate promotes erythrocyte glycolysis and oxygen release for adaptation to high-altitude hypoxia. Nat. Commun. 7, 12086 (2016).
Article PubMed PubMed Central CAS Google Scholar
Guilliams, K. P. et al. Red cell exchange transfusions lower cerebral blood flow and oxygen extraction fraction in pediatric sickle cell anemia. Blood 131, 1012–1021 (2018).
Article PubMed CAS Google Scholar
Nemkov, T. et al. Genetic regulation of carnitine metabolism controls lipid damage repair and aging RBC hemolysis in vivo and in vitro. Blood 143, 2517–2533 (2024).
Article PubMed PubMed Central CAS Google Scholar
Xu, P. et al. Erythrocyte transglutaminase-2 combats hypoxia and chronic kidney disease by promoting oxygen delivery and carnitine homeostasis. Cell Metab. 34, 299–316.e296 (2022).
Article PubMed PubMed Central CAS Google Scholar
Nemkov, T. et al. Biological and genetic determinants of glycolysis: Phosphofructokinase isoforms boost energy status of stored red blood cells and transfusion outcomes. Cell Metab. 36, 1979–1997.e1913 (2024).
Article PubMed PubMed Central CAS Google Scholar
Chen, C., Liu, W. & Xia, Y. Gene-metabolite linkage marks stored red blood cell quality. Cell Metab. 36, 1905–1907 (2024).
Article PubMed Google Scholar
Xie, T. et al. Erythrocyte metabolic reprogramming by sphingosine 1-phosphate in chronic kidney disease and therapies. Circ. Res. 127, 360–375 (2020).
Article PubMed CAS Google Scholar
Schaum, N. et al. Ageing hallmarks exhibit organ-specific temporal signatures. Nature 583, 596–602 (2020).
Article PubMed PubMed Central CAS Google Scholar
Ma, S. et al. Spatial transcriptomic landscape unveils immunoglobin-associated senescence as a hallmark of aging. Cell 187, 7025–7044.e34 (2024).
Article PubMed CAS Google Scholar
Shen, X. et al. Nonlinear dynamics of multi-omics profiles during human aging. Nat. Aging 4, 1619–1634 (2024).
Article PubMed PubMed Central Google Scholar
Reisz, J. A. et al. Arginine metabolism is a biomarker of red blood cell and human aging. Aging Cell 24, e14388 (2025).
Article PubMed CAS Google Scholar
Zhang, B. et al. Multi-omic rejuvenation and life span extension on exposure to youthful circulation. Nat. Aging 3, 948–964 (2023).
Article PubMed PubMed Central CAS Google Scholar
Chatzinikolaou, P. N. et al. Erythrocyte metabolism. Acta Physiol. 240, e14081 (2024).
Article CAS Google Scholar
Gangat, N. et al. Erythrocytosis associated with EPAS1(HIF2A), EGLN1(PHD2), VHL, EPOR or BPGM mutations: The Mayo Clinic experience. Haematologica 107, 1201–1204 (2022).
Article PubMed PubMed Central CAS Google Scholar
Watson, J. F. et al. Human hippocampal CA3 uses specific functional connectivity rules for efficient associative memory. Cell 188, 501–514.e518 (2025).
Article PubMed CAS Google Scholar
Pfeifer, A., Mikhael, M. & Niemann, B. Inosine: novel activator of brown adipose tissue and energy homeostasis. Trends Cell Biol. 34, 72–82 (2024).
Article PubMed CAS Google Scholar
Chen, C. et al. Erythrocyte ENT1-AMPD3 Axis is an Essential Purinergic Hypoxia Sensor and Energy Regulator Combating CKD in a Mouse Model. J. Am. Soc. Nephrol. 34, 1647–1671 (2023).
Article PubMed PubMed Central Google Scholar
Zhang, Y. & Xia, Y. Adenosine signaling in normal and sickle erythrocytes and beyond. Microbes Infect. 14, 863–873 (2012).
Article PubMed PubMed Central CAS Google Scholar
Mikdar, M. et al. The equilibrative nucleoside transporter ENT1 is critical for nucleotide homeostasis and optimal erythropoiesis. Blood 137, 3548–3562 (2021).
Article PubMed PubMed Central CAS Google Scholar
Lai, Y. et al. Multimodal cell atlas of the ageing human skeletal muscle. Nature 629, 154–164 (2024).
Article PubMed PubMed Central CAS Google Scholar
Shirts, M. R. et al. Lessons learned from comparing molecular dynamics engines on the SAMPL5 dataset. J. Comput. Aided Mol. Des. 31, 147–161 (2017).
Article PubMed CAS Google Scholar
Pei, J. et al. Analyzing the potential targets and mechanisms of chronic kidney disease induced by common synthetic Endocrine Disrupting Compounds (EDCs) in Chinese surface water environment using network toxicology and molecular docking techniques. Sci. Total Environ. 958, 177980 (2025).
Article PubMed CAS Google Scholar
dos Santos, D. M. et al. Crystal structure of human purine nucleoside phosphorylase complexed with acyclovir. Biochem Biophys. Res. Commun. 308, 553–559 (2003).
Article PubMed Google Scholar
Erion, M. D. et al. Purine nucleoside phosphorylase. 1. Structure−function studies. Biochemistry 36, 11725–11734 (1997).
Lichtman, M. A. & Miller, D. R. Erythrocyte glycolysis, 2,3-diphosphoglycerate and adenosine triphosphate concentration in uremic subjects: relationship to extracellular phosphate concentration. J. Lab. Clin. Med. 76, 267–279 (1970).
PubMed CAS Google Scholar
Canduri, F. et al. Structure of human PNP complexed with ligands. Acta Crystallogr. D. Biol. Crystallogr. 61, 856–862 (2005).
Article PubMed Google Scholar
Biddle, C. Oxygen: the two-faced elixir of life. AANA J. 76, 61–68 (2008).
PubMed Google Scholar
Yu, F. et al. Longevity Humans Have Youthful Erythrocyte Function and Metabolic Signatures. Aging Cell 24, e14482 (2025).
Article PubMed PubMed Central CAS Google Scholar
Yu, L. et al. IgG is an aging factor that drives adipose tissue fibrosis and metabolic decline. Cell Metab. 36, 793–807.e795 (2024).
Article PubMed PubMed Central CAS Google Scholar
Cai, Y. et al. Elimination of senescent cells by β-galactosidase-targeted prodrug attenuates inflammation and restores physical function in aged mice. Cell Res. 30, 574–589 (2020).
Article PubMed PubMed Central CAS Google Scholar
Chung, K. W. Advances in Understanding of the Role of Lipid Metabolism in Aging. Cells 10, 880 (2021).
Article PubMed PubMed Central CAS Google Scholar
Daugas, E., Candé, C. & Kroemer, G. Erythrocytes: Death of a mummy. Cell Death Differ. 8, 1131–1133 (2001).
Article PubMed CAS Google Scholar
D’Alessandro, A. & Xia, Y. Erythrocyte adaptive metabolic reprogramming under physiological and pathological hypoxia. Curr. Opin. Hematol. 27, 155–162 (2020).
Article PubMed PubMed Central Google Scholar
D’Alessandro, A. et al. Donor sex, age and ethnicity impact stored red blood cell antioxidant metabolism through mechanisms in part explained by glucose 6-phosphate dehydrogenase levels and activity. Haematologica 106, 1290–1302 (2021).
Article PubMed PubMed Central Google Scholar
Kosenko, E. A., Aliev, G. & Kaminsky, Y. G. Relationship between chronic disturbance of 2,3-diphosphoglycerate metabolism in erythrocytes and Alzheimer disease. CNS Neurol. Disord. Drug Targets 15, 113–123 (2016).
Article PubMed CAS Google Scholar
Orrico, F. et al. Oxidative Stress in Healthy and Pathological Red Blood Cells. Biomolecules 13, 1262 (2023).
Article PubMed PubMed Central CAS Google Scholar
Domingo-Ortí, I. et al. Metabolic footprint of aging and obesity in red blood cells. Aging 13, 4850–4880 (2021).
Article PubMed PubMed Central Google Scholar
Sayama, S. et al. Maternal erythrocyte ENT1-mediated AMPK activation counteracts placental hypoxia and supports fetal growth. JCI Insight 5, e130205 (2020).
Article PubMed PubMed Central Google Scholar
Valeri, C. R. & Usnr, M. C. Cryobiology overview of red cell preservation: achievements and prospective. Prog. Clin. Biol. Res 11, 55–87 (1976).
PubMed CAS Google Scholar
Ely, S. W. & Berne, R. M. Protective effects of adenosine in myocardial ischemia. Circulation 85, 893–904 (1992).
Article PubMed CAS Google Scholar
Niemann, B. et al. Apoptotic brown adipocytes enhance energy expenditure via extracellular inosine. Nature 609, 361–368 (2022).
Article PubMed PubMed Central CAS Google Scholar
Crunkhorn, S. Inosine boosts fat burning. Nat. Rev. Drug Discov. 21, 633 (2022).
Article PubMed CAS Google Scholar
Wang, T. et al. Inosine is an alternative carbon source for CD8(+)-T-cell function under glucose restriction. Nat. Metab. 2, 635–647 (2020).
Article PubMed PubMed Central Google Scholar
Oslund, R. C. et al. Bisphosphoglycerate mutase controls serine pathway flux via 3-phosphoglycerate. Nat. Chem. Biol. 13, 1081–1087 (2017).
Article PubMed PubMed Central CAS Google Scholar
Ipata, P. L. & Tozzi, M. G. Recent advances in structure and function of cytosolic IMP-GMP specific 5′nucleotidase II (cN-II). Purinergic Signal. 2, 669–675 (2006).
Article PubMed PubMed Central CAS Google Scholar
Van Den Berghe, G., Bontemps, F. & Vincent, M. F. Cytosolic purine 5′-nucleotidases of rat liver and human red blood cells: Regulatory properties and role in AMP dephosphorylation. Adv. Enzym. Regul. 27, 297–311 (1988).
Article Google Scholar
Daniels, I. S. et al. AMP Deaminase 3 Deficiency Enhanced 5′-AMP Induction of Hypometabolism. PLoS One 8, e75418 (2013).
Nishino, T. et al. Dynamic Simulation and Metabolome Analysis of Long-Term Erythrocyte Storage in Adenine–Guanosine Solution. PLoS One 8, e71060 (2013).
Walter et al. Crystal structure of human purine nucleoside phosphorylase at 2.3 Å resolution. Biochem. Biophys. Res. Commun. 308, 545–552 (2003).
Tuttle, J. V. & Krenitsky, T. A. Effects of acyclovir and its metabolites on purine nucleoside phosphorylase. J. Biol. Chem. 259, 4065–4069 (1984).
Article PubMed CAS Google Scholar
Pugmire, M. J. & Ealick, S. E. Structural analyses reveal two distinct families of nucleoside phosphorylases. Biochem. J. 361, 1–25 (2002).
Coolen, E. J. et al. Oral bioavailability of ATP after prolonged administration. Br. J. Nutr. 105, 357–366 (2011).
Article PubMed CAS Google Scholar
Kim, M. G. et al. Analgesic effect of simultaneously targeting multiple pain processing brain circuits in an aged humanized mouse model of chronic pain by transcranial focused ultrasound. APL Bioeng. 9, 016108 (2025).
Article PubMed PubMed Central Google Scholar
Qiang, Q. et al. Erythrocyte adenosine A2B receptor prevents cognitive and auditory dysfunction by promoting hypoxic and metabolic reprogramming. PLoS Biol. 19, e3001239 (2021).
Article PubMed PubMed Central CAS Google Scholar
Singh, P. et al. Taurine deficiency as a driver of aging. Science 380, eabn9257 (2023).
Article PubMed PubMed Central CAS Google Scholar
Perez-Ramirez, C. A. et al. Atlas of fetal metabolism during mid-to-late gestation and diabetic pregnancy. Cell 187, 204–215.e214 (2024).
Article PubMed CAS Google Scholar
Krais, J. J., De Crescenzo, O. & Harrison, R. G. Purine nucleoside phosphorylase targeted by annexin v to breast cancer vasculature for enzyme prodrug therapy. PLoS One 8, e76403 (2013).
Article PubMed PubMed Central CAS Google Scholar
Detter, J. C., Stamatoyannopoulos, G., Giblett, E. R. & Motulsky, A. G. Adenosine deaminase: racial distribution and report of a new phenotype. J. Med. Genet. 7, 356–357 (1970).
Article PubMed PubMed Central CAS Google Scholar
Savage, B. & Spencer, N. Rabbit erythrocyte purine nucleoside phosphorylase Differential-inactivation studies. Biochem. J. 179, 21–27 (1979).
Article PubMed PubMed Central CAS Google Scholar
Liu, H. et al. Beneficial Role of Erythrocyte Adenosine A2B Receptor-Mediated AMP-Activated Protein Kinase Activation in High-Altitude Hypoxia. Circulation 134, 405–421 (2016).
Article PubMed PubMed Central CAS Google Scholar
Trott, O. & Olson, A. J. AutoDock Vina: improving the speed and accuracy of docking with a new scoring function, efficient optimization, and multithreading. J. Comput. Chem. 31, 455–461 (2010).
Article PubMed PubMed Central CAS Google Scholar
Xu, Y. et al. CavityPlus: a web server for protein cavity detection with pharmacophore modelling, allosteric site identification and covalent ligand binding ability prediction. Nucleic Acids Res. 46, W374–w379 (2018).
Article PubMed PubMed Central CAS Google Scholar
Van Der Spoel, D. et al. GROMACS: fast, flexible, and free. J. Comput. Chem. 26, 1701–1718 (2005).
Ramsey, S. et al. Solvation thermodynamic mapping of molecular surfaces in AmberTools: GIST. J. Comput. Chem. 37, 2029–2037 (2016).
Article PubMed PubMed Central CAS Google Scholar
Hänzelmann, S., Castelo, R. & Guinney, J. GSVA: gene set variation analysis for microarray and RNA-seq data. BMC Bioinforma. 14, 7 (2013).
Article Google Scholar
Wu, T. et al. clusterProfiler 4.0: A universal enrichment tool for interpreting omics data. Innovation 2, 100141 (2021).
PubMed PubMed Central CAS Google Scholar

Download references

Acknowledgements

This work was supported by the National Natural Science Foundation of China (NSFC) (82230023, 82221002 to Y.X.) and Feifan Scholar Fund of Xiangya Hospital of Central South University (to Y.X.); the NSFC (youth program, 82401827 to W.L., 82301514 to F.Y., 82100788 to T.X., 82400873 to C.C. and 82405222 to Z.Y., 82401260 to Y.C.); the Natural Science Foundation of Hunan Province (youth program, 2023JJ40919 to W.L., 2023JJ41014 to Z.Y. and 2023JJ41018 to F.Y.); the China Postdoctoral Science Foundation (2023M743968 to W.L., 2021M693571 to T.X., BX2021382 to T.X., and 2023M733980 to Z.Y.); the Postdoctoral Fellowship Program (Grade B) of the China Postdoctoral Science Foundation (GZB20240866 to C.C.); Changsha outstanding innovative youth training program (kq2209005 to T.X.).

Author information

These authors contributed equally: Wuping Liu, Zhaoyu Yang, Changhan Chen, Fang Yu.

Authors and Affiliations

National Medical Metabolomics International Collaborative Research Center, Xiangya Hospital, Central South University, Changsha, Hunan, China
Wuping Liu, Zhaoyu Yang, Changhan Chen, Fang Yu, Mengzhi Wu, Zhouzhou Yao, Yuhua Fan, Tingting Xie, Linlin Wan, Tiansheng Chou, Xianjing Feng, Hao Qi, Yuyu Chou, Juan Liu, Zhiyu Yang, Yujin Zhang, Rodney E. Kellems & Yang Xia
National Clinical Research Center for Geriatric Disorders, Xiangya Hospital, Central South University, Changsha, Hunan, China
Wuping Liu, Changhan Chen, Mengzhi Wu, Zhouzhou Yao, Tiansheng Chou, Hao Qi, Yuyu Chou, Juan Zhao, Juan Liu, Zhiyu Yang & Yujin Zhang
Institute of Integrative Medicine, Department of Integrated Traditional Chinese and Western Medicine, Xiangya Hospital, Central South University, Changsha, Hunan, China
Zhaoyu Yang
Department of Neurology, Xiangya Hospital, Central South University, Changsha, Hunan, China
Fang Yu, Xianjing Feng & Yang Xia
Department of Otolaryngology Head and Neck Surgery, Xiangya Hospital, Central South University, Changsha, Hunan, China
Yuhua Fan & Yang Xia
Department of General Medicine, Xiangya Hospital, Central South University, Changsha, Hunan, China
Tingting Xie
Department of Radiology; Xiangya Hospital, Central South University, Changsha, Hunan, China
Linlin Wan
Department of Laboratory Medicine, Xiangya Hospital, Central South University, Changsha, Hunan, China
Juan Zhao
Department of Biochemistry and Molecular Biology, The University of Texas McGovern Medical School at Houston, Houston, TX, USA
Rodney E. Kellems
Furong Laboratory, Changsha, Hunan, China
Yang Xia

Authors

Wuping Liu
View author publications
Search author on:PubMed Google Scholar
Zhaoyu Yang
View author publications
Search author on:PubMed Google Scholar
Changhan Chen
View author publications
Search author on:PubMed Google Scholar
Fang Yu
View author publications
Search author on:PubMed Google Scholar
Mengzhi Wu
View author publications
Search author on:PubMed Google Scholar
Zhouzhou Yao
View author publications
Search author on:PubMed Google Scholar
Yuhua Fan
View author publications
Search author on:PubMed Google Scholar
Tingting Xie
View author publications
Search author on:PubMed Google Scholar
Linlin Wan
View author publications
Search author on:PubMed Google Scholar
Tiansheng Chou
View author publications
Search author on:PubMed Google Scholar
Xianjing Feng
View author publications
Search author on:PubMed Google Scholar
Hao Qi
View author publications
Search author on:PubMed Google Scholar
Yuyu Chou
View author publications
Search author on:PubMed Google Scholar
Juan Zhao
View author publications
Search author on:PubMed Google Scholar
Juan Liu
View author publications
Search author on:PubMed Google Scholar
Zhiyu Yang
View author publications
Search author on:PubMed Google Scholar
Yujin Zhang
View author publications
Search author on:PubMed Google Scholar
Rodney E. Kellems
View author publications
Search author on:PubMed Google Scholar
Yang Xia
View author publications
Search author on:PubMed Google Scholar

Contributions

W.L., Zhaoyu Y., C.C. and F.Y. designed the experiments and performed the data analysis; J.L., Zhiyu Y. and C.C. performed the metabolomics screening; C.C, Y.F., J.Z., Y.F. and W.L. collected the clinical blood samples and information; Y.Z. and T.X. participated in study design; C.C. and Y.F performed P50 measurements; M.W. performed cell experiments; W.L., Zhaoyu Y., L.W., T.C., X.F., Y.C. and H.Q. performed animal studies; W.L., Zhouzhou Y., L.W. and Y.X. were responsible for figure preparation; W.L. wrote the manuscript; R.E.K. provided expertise in proofreading the manuscript; W.L., Zhaoyu Y., C.C., F.Y. and Y.X. designed and supervised the human subject; Y.X. was responsible for the study conceptualization, design, writing and final proofreading of the manuscript. All authors approved the final version of the manuscript.

Corresponding author

Correspondence to Yang Xia.

Ethics declarations

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Supplemental Material (download PDF )

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Cite this article

Liu, W., Yang, Z., Chen, C. et al. Inosine promotes erythrocyte metabolic reprogramming and restores oxygen release for rejuvenation via 2,3-BPG-PNP axis. Cell Discov 12, 19 (2026). https://doi.org/10.1038/s41421-026-00877-6

Download citation

Received: 19 July 2025
Accepted: 04 February 2026
Published: 17 March 2026
Version of record: 17 March 2026
DOI: https://doi.org/10.1038/s41421-026-00877-6