Erythropoietin: oxygen sensing, erythroid biology and the pharmacology of erythropoiesis-stimulating agents

How a single glycoprotein couples tissue oxygenation to red cell output — and how three decades of pharmacology have learned to imitate it.

A. R. Whitfield¹, S. Nakamura², D. Okafor¹

¹Institute of Molecular Haematology, Melbourne, Australia · ²Department of Renal Medicine, Kyoto Clinical Research Centre, Kyoto, Japan

Keywords   erythropoietin · hypoxia-inducible factor · erythropoiesis · chronic kidney disease · HIF prolyl hydroxylase inhibitor · blood doping

1. Introduction

The idea that blood formation is controlled by a humoral factor emerged in 1906, when Carnot and Deflandre reported that plasma from anaemic rabbits stimulated red cell production in recipients. Half a century passed before the factor was localised: bilateral nephrectomy abolished the erythropoietic response to haemorrhage, establishing the kidney as the dominant source. Purification from the urine of severely anaemic patients in the 1970s produced vanishingly small quantities of protein, and the cloning of the EPO gene in 1985 turned an endocrine curiosity into a manufacturing programme. Recombinant human EPO entered clinical use in 1989, transforming the management of anaemia in chronic kidney disease and, soon after, in chemotherapy and in the operating theatre.

The physiology is deceptively simple. A sensor measures oxygen delivery, a hormone reports the deficit, and an effector tissue expands the oxygen-carrying compartment. What makes EPO instructive is the precision of the loop. Erythrocytes circulate for roughly 120 days, so the system must integrate oxygen status over weeks rather than minutes, and it must do so without overshooting into hyperviscosity. The resulting dose–response relationship is steep: small changes in haemoglobin produce large reciprocal changes in plasma EPO, and the hormone’s concentration can vary by more than two orders of magnitude between health and severe anaemia.

Three decades of therapeutic use have also made EPO a case study in the gap between a plausible mechanism and a proven outcome. Raising haemoglobin reliably improves symptoms in some patients and improves survival in none of the large trials; in several, aggressive correction increased stroke, thrombosis and possibly tumour progression. This review treats EPO as three interlocking problems — an oxygen-sensing circuit, a cytokine–receptor system, and a drug class with a narrow therapeutic window — and closes with the analytical chemistry that polices its misuse.

2. Oxygen sensing and the regulation of EPO gene expression

EPO is produced in adults chiefly by peritubular interstitial cells of the renal cortex and outer medulla, a fibroblast-like population marked by PDGFRβ, CD73 and, in the adult kidney, a residual neural-crest-like transcriptional programme. These cells contribute approximately 90% of circulating EPO, with hepatocytes supplying the remainder — a contribution that dominates in fetal life and re-emerges in chronic kidney disease as renal architecture is lost. In fibrosis, EPO-producing cells transdifferentiate toward myofibroblasts and lose secretory capacity, which is why the anaemia of advanced kidney disease reflects both loss of parenchyma and loss of identity in the surviving cells.

Transcription is controlled by hypoxia-inducible factor (HIF), a heterodimer of an oxygen-labile α subunit and a constitutive β subunit (ARNT). In the presence of oxygen, prolyl hydroxylase domain enzymes (PHD1–3, encoded by EGLN1–3) hydroxylate two proline residues in the HIF-α oxygen-dependent degradation domain. Hydroxylated HIF-α is recognised by the von Hippel–Lindau E3 ubiquitin ligase and destroyed by the proteasome. The hydroxylases require molecular oxygen, ferrous iron and 2-oxoglutarate; their Michaelis constant for oxygen sits near the physiological tissue range, so their activity falls steeply as oxygen tension drops. This is the sensor: not a receptor, but a set of enzymes operating close to substrate saturation.

The erythroid axis is biased toward HIF-2α. Whereas HIF-1α drives glycolytic and angiogenic programmes, HIF-2α is the isoform required for EPO transcription in the adult kidney and liver, and conditional deletion of Epas1 in renal interstitial cells abolishes the hypoxic EPO response. A second layer of control is provided by factor-inhibiting HIF, an asparaginyl hydroxylase that dampens HIF transcriptional activity under mild hypoxia, sharpening the switch. The EPO enhancer, located downstream of the gene, integrates HIF binding with hepatic nuclear factor 4 and GATA motifs, which is why hepatic and renal regulation diverge in magnitude despite sharing the same core machinery.

Superimposed on oxygen sensing are modulators that matter clinically. Androgens, insulin-like growth factor 1, angiotensin II and thyroid hormone raise EPO output; inflammatory cytokines, particularly interleukin-1 and tumour necrosis factor, suppress it. The latter effect, combined with hepcidin-mediated iron restriction, produces the anaemia of inflammation, in which EPO concentrations are inappropriately low for the degree of anaemia. Plasma EPO in healthy adults is approximately 4–24 mU/mL, with a circulating half-life of 5–9 hours; the short half-life, together with the steepness of the sensor, means a single measurement is a snapshot of a system that responds within hours.

3. Structure of the hormone and its receptor

Human EPO is encoded on chromosome 7q22 and translated as a 193-residue precursor; cleavage of a 27-residue signal peptide yields the 165-residue mature protein of approximately 30.4 kDa. The polypeptide folds into a four-helix bundle (helices A–D) in the type I cytokine topology, with two long overhand loops. Roughly 40% of the molecule’s mass is carbohydrate: three N-linked chains at asparagine residues 24, 38 and 83, and one O-linked chain at serine 126. The glycans are heavily sialylated, and sialic acid content governs circulation time, because terminal galactose residues are recognised by hepatic asialoglycoprotein receptors. This single fact explains most of the pharmacokinetic differences between marketed erythropoietins.

Two disulphide bonds (Cys7–Cys161 and Cys29–Cys33) stabilise the fold, and the receptor-binding surface comprises residues on helices A and D plus the A–B loop, with a second, lower-affinity site mediating receptor dimerisation. The erythropoietin receptor (EPOR) is a 508-residue type I membrane protein with a single transmembrane domain. Unlike the receptors for interleukin-3 or granulocyte–macrophage colony-stimulating factor, EPOR has no α/β subunit split: it is a homodimer, and ligand binding reorients the two subunits into an active configuration that brings their associated Janus kinase 2 (JAK2) molecules into proximity. Structural work has resolved both the inactive and ligand-bound dimer, showing that EPO acts as a conformational clamp rather than an allosteric activator in the classical sense.

Downstream, JAK2 phosphorylates tyrosine residues in the EPOR cytoplasmic tail, recruiting STAT5, which dimerises and translocates to the nucleus to drive survival and differentiation genes. Parallel signalling through phosphatidylinositol 3-kinase/Akt and mitogen-activated protein kinase/extracellular signal-regulated kinase shapes proliferation and maturation, while SHP-1 provides negative feedback. The functional consequence is a survival decision: in the absence of EPO, erythroid progenitors undergo apoptosis within hours; in its presence, they are licensed to divide and differentiate. Receptor density is highest on colony-forming unit–erythroid cells, at roughly one thousand receptors per cell, and falls as maturation proceeds — which is why the hormone’s principal action is early, not late.

EPOR is not confined to erythroid tissue. It is expressed in endothelium, vascular smooth muscle, cardiomyocytes, renal tubular cells, retina and neurons, usually at low abundance and often as heteromers with the common β chain. This extra-erythroid distribution underpins claims for tissue-protective effects that are discussed in section 7, and it also explains some of the off-target effects of high-dose therapy.

4. Erythroid expansion and the iron constraint

Erythropoiesis proceeds from haematopoietic stem cells through a megakaryocyte–erythroid progenitor to the burst-forming unit–erythroid, an early committed cell that proliferates in response to stem cell factor and interleukin-3 over roughly a week. The colony-forming unit–erythroid is the EPO-responsive stage, undergoing several divisions and a thousand-fold amplification before morphological maturation through basophilic, polychromatic and orthochromatic erythroblasts. Terminal maturation involves nuclear condensation and enucleation, loss of organelles, and remodelling of the membrane skeleton; the resulting reticulocyte completes maturation in the circulation over one to two days.

EPO shapes this sequence at two points. First, it suppresses apoptosis in CFU-E and early erythroblasts, largely by inducing the anti-apoptotic protein Bcl-xL and by phosphorylating forkhead transcription factors that would otherwise drive cell death. Second, it accelerates differentiation through the erythroid transcription factors GATA1, KLF1 and TAL1, and increases expression of the transferrin receptor, coupling the cell’s iron appetite to its commitment. The result is not a simple increase in division rate but a change in the probability that a progenitor survives to divide at all — a mechanism that produces very large changes in output from modest changes in hormone concentration.

Expansion is nonetheless capped by iron. A healthy adult produces about 200 billion erythrocytes daily, consuming roughly 20–25 mg of iron, almost all of it recycled from senescent cells by macrophages; only 1–2 mg is absorbed from the diet. The distribution of that iron is controlled by hepcidin, a hepatic peptide that binds the exporter ferroportin and causes its degradation, thereby locking iron inside enterocytes, macrophages and hepatocytes. Erythroblasts counteract hepcidin by secreting erythroferrone, which suppresses hepatic hepcidin transcription during stress erythropoiesis. Erythroferrone therefore acts as the demand signal that matches iron supply to EPO-driven expansion.

The clinical corollary is that EPO without adequate iron produces iron-restricted erythropoiesis: haemoglobin rises less than predicted, reticulocytes are fewer than expected, and the mean corpuscular haemoglobin falls. Functional iron deficiency — sufficient stores but insufficient mobilisation, typical of inflammation — is the commonest cause of a blunted response and is often mistaken for EPO resistance. Inflammatory cytokines simultaneously suppress EPO production, induce hepcidin and shorten red cell survival, which is why inflammatory burden predicts ESA responsiveness better than dialysis adequacy in many cohorts.

5. From native hormone to therapeutic agent

Recombinant human EPO is produced in Chinese hamster ovary cells and marketed as epoetin alfa, beta, zeta and theta, and as biosimilar variants. All share the 165-residue sequence and the same mechanism; they differ in glycosylation microheterogeneity, formulation and, in some cases, immunogenicity. Given subcutaneously, epoetin has a bioavailability of roughly 50% and peak concentrations at 12–18 hours; given intravenously, it peaks immediately and clears with a half-life of 6–8 hours, longer in dialysis-dependent patients. Standard regimens are therefore two or three times weekly, with dose titration by 25% steps.

Half-life extension has been the main pharmacological project of the class. Darbepoetin alfa contains five N-linked glycans instead of three, achieved by site-directed substitution of five amino acids that create two new glycosylation sequons; the extra sialic acid raises the half-life to approximately 25 hours after subcutaneous dosing and permits weekly, fortnightly or monthly administration. Methoxy polyethylene glycol–epoetin beta attaches a single branched PEG chain, extending the half-life to around 130 hours and enabling monthly dosing in dialysis. Peginesatide, a pegylated synthetic peptide dimer unrelated in sequence to EPO, was approved in 2012 and withdrawn in 2013 after hypersensitivity reactions. The trajectory is consistent: the receptor is the target, and the chemistry is chosen almost entirely for duration.

The newest entrants invert the logic. Rather than supplying hormone, HIF prolyl hydroxylase inhibitors stabilise HIF-2α by competing with 2-oxoglutarate, prompting the kidney to secrete its own EPO and, simultaneously, suppressing hepcidin and improving iron mobilisation. Roxadustat, daprodustat, vadadustat, molidustat and enarodustat are oral, effective in both dialysis-dependent and non-dialysis CKD, and comparatively insensitive to inflammation because they act downstream of cytokine suppression. Their liabilities are the liabilities of the pathway: transient supranormal EPO peaks, thrombotic events, and in some trials signals for heart failure and vascular calcification. Regulatory labels in the United States remain narrower than global approvals, generally restricting use to patients already on dialysis.

6. Clinical use and the target haemoglobin problem

In chronic kidney disease, EPO deficiency is the defining lesion of the anaemia, and replacement corrects it predictably. Landmark trials in the 1990s established that treatment raises haemoglobin, reduces transfusion requirement and improves fatigue and exercise capacity, and the agent became one of the most widely prescribed biologics in nephrology. The assumed corollary — that normalising haemoglobin would further improve outcomes — was tested in three large trials and failed.

CREATE randomised patients with CKD to a normal haemoglobin target of 13–15 g/dL or to 10.5–11.5 g/dL and found no reduction in cardiovascular events, with more dialysis initiation in the high-target arm. CHOIR compared 13.5 g/dL with 11.3 g/dL and was stopped early for harm, with increased death and heart failure in the high-target group. TREAT studied darbepoetin in type 2 diabetes with CKD and mild anaemia: stroke doubled, and the composite cardiovascular endpoint was unchanged, while quality-of-life gains were marginal. Together these trials moved guidelines to a target of approximately 10–11.5 g/dL, avoidance of levels above 13 g/dL, and individualised decisions with explicit discussion of thrombotic risk.

In chemotherapy-induced anaemia, ESAs reduce transfusion need in patients with non-myeloid malignancies receiving myelosuppressive chemotherapy, with a target rise of 1–2 g/dL and a ceiling near 12 g/dL. Meta-analyses reported increased thromboembolism and a mortality signal in some tumour types, leading to restricted labelling: use limited to palliative intent, minimum six-week treatment courses, and caution in patients with prior thrombosis. In low-risk myelodysplastic syndrome, ESAs are effective when endogenous EPO is below roughly 500 mU/mL, and in anaemia of prematurity they reduce transfusion in selected infants. Perioperative use, once popular for autologous donation and surgical blood conservation, has contracted sharply with the adoption of tranexamic acid and patient blood management.

Hyporesponsiveness — failure to reach target at conventional doses — should prompt a structured search: iron deficiency (absolute or functional), infection and inflammation, hyperparathyroidism, aluminium overload, occult blood loss, haemolysis, malignancy progression and, rarely, antibody-mediated pure red cell aplasia. The last of these is the class’s most feared complication. Between 1998 and 2004, a cluster of PRCA followed the subcutaneous use of a polysorbate-formulated epoetin, linked to leachates from rubber stoppers and to the subcutaneous route’s greater immunogenicity. Neutralising antibodies abolish both recombinant and endogenous EPO, producing severe transfusion-dependent anaemia; management requires immediate cessation, immunosuppression, and often renal transplantation in dialysis patients. Incidence has since fallen below one case per 10,000 patient-years with formulation and route changes.

Other adverse effects follow from the biology. Hypertension affects a fifth to a third of patients, attributed to increased viscosity and endothelial effects; seizures, vascular access thrombosis and hyperkalaemia occur. The mechanistic thread linking these is that haemoglobin is not the only thing raised: platelet reactivity, blood viscosity and endothelial activation all rise, and the benefit of more oxygen-carrying capacity saturates before the cost of a more thrombotic circulation does.

7. Actions beyond the erythron

EPOR expression outside erythroid tissue prompted two decades of work on tissue protection. In animal models, EPO attenuates ischaemia–reperfusion injury in kidney, heart, brain and retina; it reduces apoptosis, limits inflammatory infiltration and promotes angiogenesis. The observation that high-dose, briefly administered EPO protects tissues without raising haematocrit led to the search for non-erythroid analogues, including carbamylated EPO and asialoerythropoietin, which bind EPOR-containing heteromers but do not drive erythroid progenitors.

Clinical translation has been largely disappointing. Trials in acute ischaemic stroke, cardiac arrest, traumatic brain injury and cardiac surgery showed no consistent benefit, and several suggested harm when high doses and late administration were used. Explanations include poor blood–brain barrier penetration, a narrow therapeutic time window, supraphysiological dosing with thrombotic consequences, and the possibility that rodent protection depends on pathways that differ in humans. Enthusiasm persists in selected niches — retinal protection, kidney transplantation, neonatal neuroprotection — but erythropoietin is not currently a tissue-protective drug in routine practice.

Metabolic and vascular effects are also recognised: EPO influences adipose tissue, insulin sensitivity and endothelial nitric oxide availability in experimental systems. Whether these contribute meaningfully to clinical outcomes is unresolved, and they complicate interpretation of trials in which haematocrit and blood pressure both change.

8. Misuse and detection

Because EPO raises oxygen-carrying capacity, it confers an obvious advantage in endurance sport, and recombinant EPO entered cycling and distance running in the early 1990s. Sustained supranormal haemoglobin increases viscosity and thrombotic risk, and unregulated use produced documented cases of stroke, myocardial infarction and antibody-mediated PRCA in athletes.

Detection rests on distinguishing recombinant from endogenous hormone. Endogenous EPO is a heterogeneous mixture of isoforms differing in sialylation and charge; recombinant products have a narrower, more basic isoform distribution. Isoelectric focusing of urine, introduced in 2000, exploited that difference and produced the first convictions. Sarcosyl–PAGE, which separates by molecular mass, detects recombinant and PEGylated analogues and distinguishes some biosimilars. Because detection windows are short — a few days at most for conventional doses, longer for analogues — testing has been paired with longitudinal surveillance: the haematological module of the athlete biological passport tracks haemoglobin, reticulocytes and the OFF-score against an individual’s own baseline, flagging abnormal stability or oscillation. Microdosing regimens, designed to stay below thresholds, are countered by out-of-competition testing and by mathematical modelling rather than by any single assay.

9. Conclusions and open questions

Erythropoietin remains the cleanest example in physiology of an endocrine loop built around an enzyme rather than a receptor. The PHD–HIF-2α switch converts a continuous variable — oxygen tension — into a steep, reversible transcriptional output, and the short half-life of the hormone keeps the loop responsive. The clinical class built on this mechanism works exactly as predicted, which is why its failures are informative: haemoglobin correction improves symptoms but not survival, and the marginal patient at high target is harmed.

Four questions organise current work. First, which cells produce EPO in the diseased kidney, and can their identity be preserved or restored rather than merely stimulated? Second, can HIF-PHIs be dosed to capture the iron-mobilising benefit without the thrombotic cost of supranormal EPO peaks? Third, is there a genuinely non-erythroid indication for EPOR agonists, and does it require a different molecule rather than a different dose? Fourth, how should biosimilar switching and long-term immunogenicity be monitored across a market with dozens of products? None of these is answered, and all of them are practical.

The therapeutic lesson is narrower than the biology. Raising haemoglobin is easy; choosing the right patient, the right target and the right iron strategy is the actual clinical problem, and the pharmacology of the last thirty years has mostly been a slow, expensive way of learning that distinction.

10. References