Mechanisms underlying end-organ injury in sleep apnoea

Abstract

Obstructive sleep apnoea is a prevalent chronic condition characterised by repetitive upper airway collapse that promotes the occurrence of gas exchange abnormalities reflected as intermittent hypoxia along with heightened risk for the occurrence of end-organ morbidity. Here, we examine the molecular and cellular mechanisms driving obstructive sleep apnoea-induced morbidity. We describe the maladaptive responses to chronic intermittent hypoxia, including stress programmes, primarily driven by bursts of reactive oxygen species that overwhelm antioxidant defences and trigger robust, NF-κB-mediated inflammatory cascades (e.g. tumour necrosis factor-α, interleukin-6). These responses, strikingly different from the adaptive responses to sustained hypoxia, lead to systemic consequences, including endothelial dysfunction, hypertension and profound metabolic dysfunction with insulin resistance. Understanding this pathophysiology is complicated by marked cellular and tissue heterogeneity, with different cell populations (e.g. endothelium, adipose tissue or different brain regions) exhibiting divergent, context-dependent responses to intermittent hypoxia (i.e. inflammation versus repair). Traditional bulk-tissue analyses and clinical metrics, such as the apnoea–hypopnoea index and hypoxic burden, fail to capture in their entirety this cellular and tissue heterogeneity or the critical kinetics of intermittent hypoxia, particularly during reoxygenation. Critical knowledge gaps remain, including the need to standardise intermittent hypoxia exposure metrics (capturing cycle frequency, hypoxic depth and reoxygenation kinetics), integrate circadian context and other obstructive sleep apnoea-related stressors (e.g. episodic hypercapnia, fragmented sleep), account for key biological modifiers (sex, age, genetic background, comorbidities) and determine the potential reversibility of intermittent hypoxia-induced injury. Addressing these gaps will be essential to advance obstructive sleep apnoea diagnostic and therapeutic approaches. Integrating multi-omics profiling and physiological modelling within standardised intermittent hypoxia paradigms offers a pathway towards patient-tailored interventions.

Shareable abstract

This review explores the mechanisms leading to the vast phenotypic heterogeneity of patients with obstructive sleep apnoea and formulates future research avenues for this highly prevalent disease https://bit.ly/3Z1s9uG

Introduction

Obstructive sleep apnoea (OSA) is characterised by repetitive upper airway collapse during sleep, resulting in cycles of oxygen desaturation and reoxygenation, known as intermittent hypoxia (IH). These cycles often occur with concurrent hypercapnia and are frequently terminated by arousals that disrupt sleep continuity and lead to sleep fragmentation (SF). OSA is a highly prevalent condition affecting nearly 1 billion people worldwide, with representation across the lifespan from infants to advanced ages [1–3]. The important clinical relevance of OSA resides in its causal associations with other chronic diseases, such as neurobehavioural and mood deficits and cardiometabolic diseases, culminating in increased healthcare utilisation and overall mortality (figure 1) [4, 5]. In this review, we evaluate the molecular and cellular mechanisms leading to end-organ injury in OSA by highlighting the divergent characteristics of IH and sustained hypoxia (SH), and the impact of cellular and tissue heterogeneity on OSA responses. We discuss how such complexities affect OSA research while creating opportunities for future discovery.

FIGURE 1.

Obstructive sleep apnoea (OSA) is associated with multisystem comorbidities. CKD: chronic kidney disease; ESRD: end-stage renal disease; GORD: gastro-oesophageal reflux disease; IBD: inflammatory bowel disease; MAFLD: metabolic dysfunction-associated fatty liver disease; MASH: metabolic dysfunction-associated steatohepatitis.

Molecular responses to IH versus SH

In contrast to IH in OSA, SH is a continuous reduction in oxygen availability, as seen in untreated chronic lung disease, long-term high-altitude sojourns and cyanotic congenital heart disease. Some patients may have both IH and SH, e.g. in obesity hypoventilation syndrome. The IH patterns in OSA usually consist of relatively brief desaturations and reoxygenation events that can occur hundreds of times per night, whereas SH persists across day and night. IH is most prominent in moderate to severe OSA in adults and children, but adipose tissue in obese individuals or malignant tumours exhibits periodic fluctuations in tissue oxygen tension, albeit at much lower frequencies [6, 7]. The varying patterns of IH, specifically variations in timing, amplitude, stability and chronicity, result in marked variability in the cellular and molecular responses. In this review, we will first examine some of the salient differences in the mechanisms activated by IH and SH (figure 2).

FIGURE 2.

Schematic comparison of sustained hypoxia (SH) versus intermittent hypoxia (IH). Left panel (SH): continuous low O₂, robust hypoxia-inducible factor (HIF)-1α, activation of vascular endothelial growth factor (VEGF), erythropoietin (EPO), glycolysis, enhanced angiogenesis and erythropoiesis, moderate inflammation or anti-inflammatory bias, shift toward glycolysis. Right panel (IH): transient O₂ dips, blunted HIF-1α, mitochondrial dysfunction leading to increased reactive oxygen species (ROS) bursts, NF-κB activation, increased cytokines (tumour necrosis factor-α (TNF-α), interleukin-6 (IL-6)) and increased transforming growth factor-β (TGF-β) signalling leading to fibrosis. ARNT: aryl hydrocarbon receptor nuclear translocator; ECM: extracellular matrix; HRE: hypoxia-response element; IKK: IκB kinase; MMP-9: matrix metalloproteinase-9.

Hypoxia-inducible factor signalling

A core response to low oxygen is stabilisation of hypoxia-inducible factor-1α (HIF-1α). Under SH, reduced oxygen prevents prolyl hydroxylation, leading to HIF-1α accumulation and subsequent translocation to the nucleus, activating canonical targets that enhance oxygen delivery or metabolic adaptation (vascular endothelial growth factor (VEGF), erythropoietin, glycolytic enzymes, inducible nitric oxide synthase) [8–10]. The hypoxic response is more complex than a mere HIF-1α stabilisation, and includes concepts such as “unused oxygen” [11], von Hippel–Lindau-mediated oxygen sensing [12] and several other molecular cytosolic- and organelle-specific events that are beyond the scope of this review. In IH, rapid reoxygenation between the short hypoxic bouts restores prolyl hydroxylation of the HIF-α subunits, leading to their ubiquitination and proteasomal degradation, and subsequently blunting HIF-1α transcriptional activity [13, 14]. Some examples from in vitro and in vivo animal studies support this central discrepant response to hypoxia between the two hypoxic exposure patterns. For example, a recent in vitro study examined endothelial transcriptomics following SH and IH. SH, but not IH, elicited an HIF-1α transcriptomic programme. Consequently, human aortic endothelium exposed to SH showed nuclear HIF-1α and target induction, whereas matched IH cycles did not [15]. In another study involving mice exposed to either short-term or long-term IH or SH, adipose tissue responses were assessed using HIF chromatin immunoprecipitation sequencing. After 6 h of either cycling or continuous hypoxia of similar magnitude, a similar stabilisation and DNA binding of HIF-1α emerged between the two hypoxia modalities, with corresponding activation of downstream transcriptional programmes. However, when examined after 6 weeks of exposures, SH continued activating HIF-1α, albeit to a lesser degree, whereas in IH-exposed mice, the HIF-1α response was virtually absent [16]. Taken together, these findings suggest that when IH induces HIF-1α, the response tends to be short-lived and partial. Thus, SH engages classical hypoxia pathways, whereas IH shifts cells toward alternative stress programmes.

Angiogenesis and vascular adaptation

VEGF is a target of HIF-1 and a key driver of angiogenesis. SH exposures stabilise HIF-1α, increase VEGF expression and support capillary growth, improving oxygen delivery over time [17]. Under hypoxic conditions, erythropoiesis is also stimulated: HIF-α subunits (primarily HIF-2α for erythropoietin (EPO) regulation) escape prolyl hydroxylase-mediated degradation, translocate to the nucleus, and activate EPO gene transcription in renal interstitial fibroblasts and, developmentally, in hepatic and neural crest cells. This leads to increased EPO secretion, stimulation of erythroid progenitors in the bone marrow, and a rise in red cell mass and haematocrit. SH results in persistent HIF stabilisation, continuous EPO production, and a gradual increase in haematocrit [10, 18]. Real-life examples of such responses include high-altitude residential periods, during which increased capillary density and red blood cell mass occur in humans [19, 20]. In IH exposures in rodents, despite the cumulative time spent with reduced oxyhaemoglobin saturation and tissue oxygenation [21], angiogenic responses are blunted. Similarly, in healthy adults, short IH protocols can acutely increase EPO levels, but the magnitude and duration of erythropoietic response are less pronounced than with SH [22]. OSA-like IH exposure consistently fails to increase VEGF in peripheral tissues and, in fact, may be accompanied by vascular rarefaction in metabolically active tissue depots. In mice, IH but not SH reduced adipose capillarity despite similar average oxygen deficits. Reduced adipose VEGF and HIF-1 α activity in IH were associated with vessel loss and visceral fat “whitening ”, whereas SH maintained HIF-1 α, increased VEGF, and promoted healthy angiogenesis and “beiging ” [16, 23]. These changes are also apparent in patients with OSA, who have been reported to have microvascular rarefaction and impaired angiogenic signalling [24]. At the macrovascular level, IH raises sympathetic activity and vasoconstrictors, elevates endothelin-1, and reduces nitric oxide bioavailability, fostering hypertension and endothelial dysfunction [25, 26]. Conversely, SH more often raises pulmonary arterial pressure but can induce systemic vasodilation through nitric oxide and angiogenic remodelling [10, 27].

Oxidative stress

IH triggers initiation and propagation of oxidative stress through recurrent hypoxia–reoxygenation cycles. Each reoxygenation phase generates bursts of reactive oxygen species (ROS) that can overwhelm antioxidant defences [28–36]. Animal studies show increased ROS across organs with chronic IH [37]. ROS is also an upstream driver of downstream inflammatory responses in IH, including NF-κB activation, linking oxidative stress with increased inflammation under IH (see below) [38, 39]. The increased oxidative stress induced by IH has also been linked to downstream endothelial dysfunction [40]. In patients with OSA, elevated urinary 8-oxo-2′-deoxyguanosine (8-OHdG) and malondialdehyde (MDA) (markers of oxidative stress and DNA damage) have been observed, and these levels normalise with continuous positive airway pressure [41]. In contrast, SH elicits a more gradual oxidative stress challenge, allowing for progressive upregulation of antioxidant enzymes and metabolic shifts that over time reduce radical generation and mitigate its consequences. Although SH can cause oxidative damage, the kinetics of SH-associated adaptation often permit compensation and recruitment of antioxidant protective mechanisms such as nuclear factor erythroid 2-related factor 2 (NRF2) [42]. In its milder forms, SH has also shown promise in targeting mitochondrial disease associated with increased oxidative stress [11]. In summary, IH is a potent generator of oxidative stress, which overwhelms the cellular antioxidant defence mechanisms, whereas SH promotes a more controlled, sometimes adaptive, redox state. We should emphasise that SH is in no way an innocuous state and, if severe from its onset, it can elicit substantial cellular and organ injury, which can be further exacerbated by reperfusion in the context of ischaemia or by reoxygenation [43–47]. Conversely, IH can elicit, particularly when short-lived, a phenomenon defined as preconditioning that provides protection against more severe hypoxic-ischaemic insults or results in improved exercise performance [48–53]. In addition, the transcriptional response to hypoxia is contextually sensitive and manifests as a cell-, tissue- and organ-specific response that also exhibits circadian dependencies [54, 55]. These issues have prompted recommendations aimed at long-term multifunctional physiological monitoring in preclinical animal models [56].

Inflammatory pathways

One of the hallmark features of OSA is the widespread inflammation that is detectable both in the circulation and within various organs (e.g. liver, adipose tissue, brain). In this setting, IH is a potent driver of this inflammatory cascade. Experimental data demonstrate that IH robustly increases NF-κB DNA-binding activity and nuclear translocation of p65, accompanied by phosphorylation of inhibitor of κB (IκB), in human adipocytes and endothelial cells. This activation is associated with increased secretion and mRNA expression of pro-inflammatory cytokines, including tumour necrosis factor-α (TNF-α) and interleukin (IL)-6 [57–61]. Mechanistically, both mitochondrial and NADPH oxidase-derived ROS are implicated in IH-induced NF-κB activation. Inhibition of mitochondrial complex I or ROS scavenging abolishes hypoxia-induced NF-κB activation and TNF-α gene transcription, indicating a critical role for ROS in this pathway [62–64]. Human adipocytes and monocytes show strong IH-induced NF-κB activation (IL-6, IL-8), but limited activation under SH [65]. Additional evidence in the context of non-alcoholic liver steatohepatitis as triggered by chronic IH implicates mitochondrial dysfunction and reprogramming of nuclear respiratory factor (NRF)-dependent gene expression pathways, a process that is further subjected to additional metabolic reprogramming upon reoxygenation [66, 67]. Similar, albeit organ-specific, NRF-dependent pathways conferring either protection or potentially facilitating age-dependent mitochondrial metabolic dysregulation have also been identified in brain and myocardial tissues [68, 69]. In vivo, these processes emerge as systemic inflammation in OSA, including elevated C-reactive protein and pro-inflammatory cytokines whose levels correlate with IH severity [70–73]. Chronic SH does not typically produce similar systemic inflammation and can shift toward HIF-2α effects and M2-like anti-inflammatory programmes [74]. The mitochondrial dysfunction observed in IH seems to be less potent under SH. As mentioned above, diseases characterised by mitochondrial dysfunction seem to benefit from exposure to mild SH in rodent models and even in human experimental settings [11, 75–78]. Overall, IH acts as a strong pro-inflammatory stimulus; SH emphasises hypoxia-specific programmes with lower inflammatory tone.

Distinct transcriptional programmes and metabolic consequences

The canonical hypoxic signalling is consistently activated under SH [79]. By contrast, global analyses of the transcriptome under IH conditions seem to align with a reduction or even abolition of hypoxic signalling activation. In the adipose tissues, pathways such as fatty acid oxidation, adipogenesis, oxidative phosphorylation and DNA damage repair are prominent [23]. In endothelial cells, IH activates cytokine-mediated TNF-α signalling via NF-κB, Wnt/LDL-related protein (LRP)/Dikkopf (DKK) signalling and cell cycle regulation [15]. In the liver, IH promotes similarly maladaptive programming, resembling that of the inflammatory signature of steatohepatitis [80] and fibrosis [81]. We acknowledge that some organs (e.g. kidney) may also be more susceptible to hypoxia signalling activation under IH, but the literature is lacking a systematic characterisation of this effect.

These differences in transcriptional programmes map to divergent physiological effects. Chronic IH leads to systemic insulin resistance and dyslipidaemia, whereas chronic SH can improve insulin sensitivity [16, 81–85]. IH-exposed mice develop hyperlipidaemia, hepatic steatosis and impaired glucose tolerance despite concomitant weight loss; SH often preserves or improves systemic glucose tolerance. Mechanisms include adipose inflammation and adipocyte dysfunction under IH (reduced glucose uptake, increased lipolysis), versus HIF-1-mediated maintenance of adipose vascularisation and function under SH. Only IH induced marked whitening of visceral fat with reduced adipose insulin signalling. In a head-to-head comparative exposure with identical nadirs in oxyhaemoglobin saturation and hypoxia duration, only IH induced visceral adipose insulin resistance, with macrophage infiltration and capillary loss [16]. Overall, IH drives a maladaptive programme of inflammation, oxidative stress and lipid-centric metabolism, whereas SH evokes a more adaptive hypoxia-response profile.

Current knowledge gaps and future directions

1. Timing and profile of IH: We need dose–response data for cycle length, nadir depth and reoxygenation rate on ROS, NF-κB and HIF stabilisation. Clinical IH is heterogeneous in patients, whereas experimental paradigms often focus on a single profile mimicking moderate to severe OSA. The newly developed hypoxic burden metric integrates depth and duration, but equal burden delivered intermittently versus continuously is likely to impose divergent biological pathways.

2. Circadian context: IH during sleep co-occurs with arousal, autonomic nervous system surges and altered hormonal rhythms. Daytime versus nocturnal IH may lead to different transcriptional programmes, and in vitro models often fail to capture the circadian aspects of OSA.

3. Contributing factors: OSA physiology includes hypercapnia, intrathoracic pressure swings and arousals with SF. Very few studies use the combination of IH with either one or all these factors, thus failing to faithfully recapitulate OSA.

4. Heterogeneity and reversibility: Inter-individual variability suggests genetic and epigenetic moderators of IH sensitivity. DNA methylation changes in antioxidants and inflammatory genes in rodents and humans exposed to IH have been described, but the time course of reversal after eliminating IH is incompletely defined. Some changes reverse with weeks of normoxia, while others persist, indicating a long-lasting epigenetic memory. Studies examining early treatment may identify specific windows to prevent irreversible impairments (see below).

Summary

IH initiates a pathophysiological cascade distinct from SH, with prominent roles for ROS, NF-κB, impaired angiogenesis and lipid-centric metabolism that drive insulin resistance and vascular dysfunction. SH engages HIF-1 pathways that support angiogenesis and, in some contexts, improve metabolism. Clarifying dose, timing and combined stressors will refine the mechanism and guide therapy, including antioxidants, pathway modulators and approaches that preserve beneficial SH-like adaptations while avoiding IH injury. Table 1 illustrates the complex and expansive mechanism explored to date in the context of IH as a model of OSA.

TABLE 1.

Summary of major end-organ morbidities associated with obstructive sleep apnoea/intermittent hypoxia, highlighting key injury pathways and corresponding endogenous protective mechanisms

Organ system	End-organ morbidities	Key injury mechanisms	Endogenous protective mechanisms
Cardiovascular and cerebrovascular	Hypertension, CAD, heart failure, arrhythmias, stroke	Mitochondrial and NOX-derived ROS → eNOS uncoupling [28]; NF-κB/AP-1 → TNF-α, IL-6, MMP-9 [86]; sympathetic overdrive via carotid chemoreceptor sensitisation [87]	Nrf2 → HO-1, SOD2 [88]; SIRT1 → mitochondrial biogenesis [69]; preserved eNOS-derived NO [89]
Neurocognitive	Cognitive decline, dementia, white matter lesions	AQP4 dysregulation → impaired glymphatic clearance [90]; HIF-1α → microglial IL-1β, TNF-α [91]; BBB disruption (ROS, MMP-9) [30]	BDNF–TrkB pathway [92]; low-level HIF-1α → VEGF, EPO [93]; IL-10, TGF-β expression [68]
Metabolic	Insulin resistance, type 2 diabetes, dyslipidaemia	HIF-1α in adipose → TNF-α/IL-6 → IRS-1 phosphorylation [94]; sympathetic lipolysis → FFA release [95]; ceramide accumulation → impaired Akt signalling [96]	Adiponectin → AMPK and PPAR-α [97]; AMPK → FA oxidation [98]
Pulmonary	Pulmonary hypertension, RV dysfunction	PASMC K+-channel inhibition → ↑Ca2+ [99]; HIF-1α → PDGF/TGF-β remodelling [100]; ET-1 upregulation [101]	eNOS-derived NO [89]; prostacyclin (PGI2) [102]; angiogenic remodelling [103]
Renal	CKD, proteinuria	Tubular hypoxia → HIF-1α → CTGF, collagen [104]; RAAS activation → ATII, TGF-β [105]; NADPH oxidase → podocyte ROS injury [106]	EPO production [22]; HO-1 antifibrotic effects [107]; natriuretic peptide signalling [108]
Hepatic	MAFLD/MASH	HIF-2α → SREBP-1c, FAS-driven lipogenesis [109]; Kupffer TLR4 → IL-1β, TNF-α → stellate cell activation [110]; mitochondrial ROS → lipid peroxidation [111]	PPAR-α → FA oxidation [112]; parkin/PINK1 → mitophagy [113]; FXR signalling [114]
Gastrointestinal	Barrier dysfunction, dysbiosis, ulcers	HIF-1α → ZO-1, occludin disruption [115]; dysbiosis → LPS → TLR4 activation [116]; ROS-induced epithelial damage [115]	HIF-mediated barrier regulation [117]; mucin/AMP expression [116]; SCFA anti-inflammatory effect [118]
Skin	Impaired healing, inflammatory dermatoses, ageing	ROS → MMP activation → ECM degradation [119]; cytokines (IL-1β, TNF-α) → dermal inflammation [120]; ↓ perfusion via microvascular dysfunction [121]	Nrf2 activation in keratinocytes [122]; VEGF-mediated angiogenesis [123]; melatonin/GF release [124]
Cancer	Tumour growth, metastasis, resistance	HIF-1α → VEGF-mediated angiogenesis [125]; ROS → DNA damage [126]; immune evasion via PD-L1 [127]	p53 pathway [128]; NK cell surveillance [129]; PARP and DNA repair [6]
Reproductive	ED, infertility, preeclampsia, hypogonadism	↓ NO bioavailability in corpus cavernosum [130]; placental hypoxia → trophoblast apoptosis [131]; Leydig cell dysfunction via ROS [132]	VEGF/PlGF in placenta [133]; eNOS-derived NO [130]; GPx, SOD activity [134]
Musculoskeletal	Sarcopenia, osteoporosis, myopathy	ROS and TNF-α → ubiquitin–proteasome [135]; HIF-1α → osteoblast/osteoclast imbalance [136]; mitochondrial dysfunction in muscle [137]	Satellite cell activation [138]; autophagy and mitophagy [139]; osteoprotegerin [140]
Ocular	Glaucoma, retinopathy	Perfusion fluctuation → ischaemia [141]; ROS → pericyte apoptosis [141]; VEGF overexpression [142]	Nrf2 in retina [141]; PEDF versus VEGF balance [142]; vasodilatory autoregulation [143]
Immune and inflammatory	Systemic inflammation, autoimmune modulation	NLRP3 inflammasome → IL-1β, IL-18 [144]; ↑Th17 and ↓Treg balance [145]; NF-κB activation [146]	IL-10 [147], IL-37 secretion [148]

AMPK: AMP-activated protein kinase; AP-1: activator protein-1; AQP4: aquaporin-4; ATII: angiotensin II; BBB: blood–brain barrier; BDNF: brain-derived neurotrophic factor; CAD: coronary artery disease; CKD: chronic kidney disease; CTGF: connective tissue growth factor; ECM: extracellular matrix; ED: erectile dysfunction; eNOS: endothelial nitric oxide synthase; EPO: erythropoietin; ET-1: endothelin-1; FA: fatty acid; FAS: fatty acid synthase; FFA: free fatty acid; FXR: farnesoid X receptor; GF: growth factor; GPx: glutathione peroxidase; HIF: hypoxia-inducible factor; HO-1: heme oxygenase-1; IL: interleukin; IRS-1: insulin receptor substrate-1; K: potassium; LPS: lipopolysaccharide; MMP: matrix metalloproteinase; MAFLD: metabolic dysfunction-associated fatty liver disease; MASH: metabolic dysfunction-associated steatohepatitis; NADPH: nicotinamide adenine dinucleotide phosphate (reduced form); NF-κB: nuclear factor κ-light-chain-enhancer of activated B-cells; NLRP3: NOD-like receptor family pyrin domain containing 3; NK: natural killer; NO: nitric oxide; NOX: NADPH oxidase; Nrf2: nuclear factor erythroid 2-related factor 2; PASMC: pulmonary artery smooth muscle cell; PD-L1: programmed death-ligand 1; PDGF: platelet-derived growth factor; PEDF: pigment epithelium-derived factor; PGI: prostacyclin; PINK1: PTEN-induced kinase 1; PlGF: placental growth factor; PPAR-α: peroxisome proliferator-activated receptor-α; RAAS: renin–angiotensin–aldosterone system; ROS: reactive oxygen species; RV: right ventricle; SCFA: short-chain fatty acid; SIRT1: sirtuin 1; SOD: superoxide dismutase; SREBP-1c: sterol regulatory element-binding protein-1c; TGF-β: transforming growth factor-β; Th17: T-helper 17 cell; TLR4: toll-like receptor 4; TNF-α: tumour necrosis factor-α; TrkB: tropomyosin receptor kinase B; Treg: regulatory T-cell; VEGF: vascular endothelial growth factor; ZO-1: zonula occludens 1.

Tissue and cellular heterogeneity in OSA and end-organ morbidities

In adult patients, OSA directly involves upper airway tissues (i.e. soft palate, tongue and surrounding pharyngeal tissues), resulting in tissue relaxation and collapse, which obstructs airflow [149]. In paediatric patients, the most common risk factor underlying anatomical narrowing of the upper airway is adenotonsillar hypertrophy [150, 151]. Regardless of the cause, OSA imposes systemic effects on both adults and children, affecting virtually every organ and system in the human body, leading to numerous end-organ morbidities [152–154]. Importantly, the widespread distribution of OSA-induced physiological dysfunction is also reflected at the tissular and cellular levels [155].

However, heterogeneity of the biochemical and molecular reactions to the stimuli triggered by OSA (e.g. IH, activation and propagation of oxidative stress, activation of inflammatory pathways) is universally present among similar and, of course, different cellular populations. Hence, different cell populations in the same tissue may react differently upon exposure to the same IH stimulus. Reciprocally, the same cell type may react differently when exposed to the same stimulus according to the organ and physiological contextual setting (figure 3).

FIGURE 3.

Schematic representation of types of cellular heterogeneity in obstructive sleep apnoea (OSA) effects. a) Different cell types in a tissue may react differently to OSA stimuli. b) The same cell type may react differently to the OSA stimuli while being located in different tissue types.

To expand on this, different cell populations within the same tissue often exhibit remarkably varied responses to IH or SH. A variety of interplaying factors, including the presence of subclonal genetic variants, epigenetic status, baseline gene expression, distinctive metabolic profiles and specific cellular functions, will determine the sensitivity of a cellular population to the hypoxic insult [156]. IH can specifically activate some endothelial cell populations in the vascular endothelium, leading to a pro-inflammatory state that involves the expression of adhesion molecules (e.g. vascular cell adhesion molecule 1 (VCAM-1) and intercellular adhesion molecule 1 (ICAM-1)). These mediate the recruitment and attachment of inflammatory cells [89] and increase the production of ROS, contributing to oxidative stress and endothelial dysfunction [157]. However, IH may also activate protective or adaptive mechanisms in other vascular endothelial cell populations [158]. In these populations, IH can stimulate the release of endothelial progenitor cells and endothelial cell-colony forming units, which contribute to vascular repair and regeneration, potentially mitigating damage caused by IH itself [159]. Indeed, evidence of discrepant recruitment of progenitor cells into circulation is associated with the presence or absence of vascular deficits in children with OSA [160]. Such cell-specific and seemingly contradictory reactions to IH stimulus are also observed in the brain. IH exposures will result in neuronal damage and neurocognitive impairment by triggering multiple signalling pathways in neurons, leading to increased excitotoxicity, apoptosis, neuroinflammation, oxidative stress and permeability of the blood–brain barrier (BBB) [31, 92, 161–168]. Nevertheless, IH may also induce phenotypic changes in glial cells (e.g. astrocytes or microglia), which may become either neuroprotective or neuroinflammatory, protecting against ischaemic injury [169–174]. This phenotypic shift resulting in cellular heterogeneity in the response to IH appears to be related to the duration and severity of the IH insult. While chronic hypoxia leads to the activation of both astrocytes and microglia, leading to neuroinflammation and degeneration, acute hypoxia induces microglial activation and an increase in pro-inflammatory markers and cytokines [158, 167, 175–177].

Notably, even the same cell type can exhibit remarkably different reactions to IH according to the physiological context of the organ it is located in, the baseline physiological conditions and the cellular milieu. For example, the response of an endothelial cell in the cerebral vasculature might differ from that of an endothelial cell in the pulmonary circulation when exposed to the same pattern of oxygen fluctuations [178]. Many cellular and molecular mechanisms are associated with this divergent response. Endothelial cells have different functional requirements in various vascular beds; while cerebral endothelial cells form the BBB, pulmonary endothelial cells are involved in gas exchange and pulmonary homeostasis, and glomerular endothelial cells in the kidney influence filtration [179]. Such functional specialisation is reflected in the proteomic and gene expression profiles of the same cell types located in different organs. For example, brain endothelial cells have tighter junctions and membrane transporters, enabling BBB function, whereas pulmonary endothelial cells adapt to control vascular permeability and angiogenesis in response to injury, and vascular endothelial cells in other organs require resistance to haemodynamic shear stress [180–183]. Furthermore, hypoxia-mediated changes in gene and protein expression are governed by epigenetic regulation and differential activation of HIFs. Thus, the timing of the HIF activation along with their downstream targets will vary according to the oxygen levels, the physiological context and the specific functional requirement in each organ [184, 185]. In addition, the organisation and intercellular localisation within human tissues will determine the functionality of the same cell type and its reaction to pathological insults. A recent study identified 12 distinct cross-tissue coordinated cellular modules using a comprehensive single-cell transcriptomics atlas compiled from 35 human tissues and encompassing over 2 million cells from 706 healthy samples, demonstrating intercellular communication and spatial organisation [186]. Given the systemic effects of OSA, it is plausible that a similar multicellular coordination occurs in OSA-associated morbidities.

From a translational medicine standpoint, unravelling the heterogeneous landscapes and multicellular communications that underly the response to OSA-mediated systemic physiological dysfunction is of paramount importance for: 1) understanding disease pathophysiology, 2) developing precise diagnostic methodologies, and 3) designing targeted and effective therapeutic approaches.

Cellular heterogeneity within an affected tissue or set of tissues adds immense complexity to the understanding of diseases by increasing the variability of phenotypic responses that mask the underlying pathological mechanisms [187]. Different cell types within a tissue have distinctive genomic and epigenomic landscapes exhibiting diverse gene expression profiles, metabolic states and functional behaviours. Such diversity is further augmented in disease, with distinct subpopulations of cells responding differently to factors contributing to the development of the disease as well as therapeutic interventions [188, 189]. In this regard, traditional methods based on bulk large-scale tissue studies (i.e. “omics” studies based on tissue samples without distinguishing among cell types) average cell-specific differences within the tissue, often resulting in an incomplete and in fact rather superficial picture of the disease aetiology and pathophysiological mechanisms involved in the pathological phenotype.

Likewise, cellular heterogeneity presents significant challenges for the development of molecular diagnostics tests, negatively impacting biomarker accuracy, sensitivity and specificity [190]. Specimens collected from patients (e.g. blood samples or tissue biopsies) consist of a mixture of cells of which only a subset may carry the molecular signal associated with the disease. Hence, the averaged readout per sample will dilute the signal, compromising the accuracy of the test when the ratio of affected cells is low. Moreover, cellular heterogeneity affects the development of effective therapeutic approaches, in particular in complex diseases such as OSA and end-organ morbidities [155, 191, 192]. Drugs are usually designed to target a specific molecular pathway, which may lead to the elimination of a particular subpopulation of diseased cells, whereas other subpopulations, also functionally affected but bearing genomic and epigenomic differences, remain resistant to the therapy. As a result, patients may develop partial responses to therapy, relapse or manifest residual disease due to incomplete ablation of diseased cells, and even develop more severe phenotypes or comorbidities due to the selection and proliferation of drug-resistant clones [193, 194]. To overcome these challenges, it is necessary to deeply understand the different cellular types affected by the disease and comorbidities, identify druggable pathways in each of them, and develop combination therapies or adaptative treatment strategies accounting for the heterogeneous cellular populations in the affected tissues.

Understanding and interpreting this intrinsic cellular variability requires the evaluation of molecular profiles at a single-cell level. Such level of precision requires the physical separation and analysis of individual cells (e.g. by fluorescent-mediated cell sorting) or the application of single-cell and spatially resolved technologies, as well as algorithms for the deconvolution of cell-specific signals retrieved from bulk “omics” studies [190, 195, 196]. Initial studies of cellular heterogeneity in OSA involved the use of single-cell separation techniques and the interrogation of markers for known OSA-associated pathways, such as inflammation and oxidative stress. Anderson et al. [197] investigated the inflammatory response in tonsillar tissues of paediatric OSA patients by conducting fluorescence-activated cell sorting and ELISA in single-cell suspensions. Inflammatory response in these tissues was characterised by increased effector CD4⁺ T-cells and decreased FoxP3 CD8⁺ T-cells. By using flow cytometry analysis, Gileles-Hillel et al. [198] showed that IH induced a pro-inflammatory shift in adipose tissue macrophages isolated from visceral white adipose tissue (vWAT) characterised by increased Ly6c^hi+ and CD36⁺ cells and accompanied by increased vWAT insulin resistance. In more recent years, considerable efforts have been made towards the application of single-cell and spatial transcriptomics across different tissues in animal models and clinical samples of OSA. Using single-cell RNA-sequencing (scRNA-seq) in lung samples of mice exposed to IH and controls (i.e. over 12 000 cells per group), Wu et al. [199] identified 19 distinct cell types relating to the IH exposure that were associated with 40–60 marker genes for each cluster. Differentially expressed genes in each cluster were mainly associated with response to hypoxia, inflammation, cell adhesion and angiogenesis. Among all cell clusters, the highest levels of transcriptional changes were found in endothelial cells, whereas markers in myofibroblast cells were associated with pulmonary hypertension and were targets for chronic pulmonary hypertension drugs. Aiming to understand the cellular heterogeneity in the context of OSA comorbidities, Khalyfa et al. [84] showed that IH induces metabolic dysfunction in vWAT and identified significant transcriptional changes in vWAT samples of mice exposed to IH compared to controls at the single-cell level. Using scRNA-seq, the authors detected significant transcriptional changes across 14 different cell types in vWAT, reporting 298 commonly regulated genes across those clusters showing significant differences between IH and control samples and which were associated with metabolic pathways. More recently, Cortese et al. [200] explored the application of scRNA-seq in clinical samples, evaluating its application to refine molecular diagnostics of OSA in children. They studied scRNA-seq profiles of peripheral blood mononuclear cells in paediatric OSA patients and controls, and reported that OSA is associated with specific changes in immune cell populations, such as a decline in T-helper cells and an increase in classical monocytes and atypical B-cells. The percentage of these populations varied according to the severity of the disease. Applying machine-learning approaches to the scRNA-seq results to combine cell-specific markers with those differentially expressed in OSA allowed the development of a molecular signature consisting of 32 genes that can precisely (i.e. >95% accuracy) distinguish between children with and without OSA [200]. Even though the studies of cellular heterogeneity in OSA and its associated morbidities are in the infancy stage, the available evidence clearly shows the significance of multicellular communication in the pathophysiology of the disease and demonstrates the unique translational medicine value of such approaches.

Summary

Given the profound and widespread impact of OSA at the systemic, tissular and cellular levels, a comprehensive understanding of its pathophysiology demands an in-depth study of the intricate heterogeneity inherent in biological systems. The emerging insights from single-cell and spatially resolved omics technologies, coupled with powerful analytics, are pivotal in unmasking these complex cellular responses and intercellular communications, moving beyond the limitations of bulk molecular analyses. As demonstrated by recent studies, these advanced approaches are not only refining our molecular understanding of OSA and its numerous comorbidities but are also opening new avenues for the development of highly precise diagnostic tools and truly targeted therapeutic strategies that account for the diverse cellular populations and their context-dependent reactions to OSA's multifaceted insults.

Why has OSA research struggled to move forwards towards improved outcomes?

IH epitomises a dose-dependent paradox: when mild and transient, IH triggers adaptive, hormetic pathways that precondition tissues against subsequent insults; when severe, frequent or prolonged, IH unleashes maladaptive cascades culminating in end-organ injury [201]. In low-dose, short-lasting IH paradigms (nadir inspiratory oxygen fraction (F_iO2) ≈10–12%, 20–30 cycles·h⁻¹, 15–30 s·cycle⁻¹), IH stabilises HIF-1/2, augments antioxidant defences (e.g. superoxide dismutase, catalase), promotes mitochondrial biogenesis and angiogenesis, and enhances ischaemic tolerance in cardiac, neural and vascular beds, mirroring classic ischaemic preconditioning [152, 202]. By contrast, high dose IH (i.e. severe desaturations (F_iO2 ≤5–8%), rapid cycles (>60 events·h⁻¹) and exposures spanning weeks to months can overwhelm redox homeostasis, activate NF-κB-mediated inflammation, disrupt endothelial barrier integrity, exaggerate tonic and reactive sympathetic tone, and accelerate hypertension, atherosclerosis, metabolic dysfunction and neurocognitive decline [152, 158, 201].

Clinically, the apnoea–hypopnea index (AHI) historically guided OSA severity assessment, but its failure to capture desaturation depth and duration and the relatively weak associations between AHI and end-point outcomes spurred the emergence of alternative metrics, such as the “hypoxic burden” (area under the nocturnal desaturation curve) [203]. Indeed, the hypoxic burden correlates more robustly with cardiovascular mortality in large cohorts (Sleep Heart Health Study) than AHI [204]. However, with it being a purely mathematical construct, the hypoxic burden reduces complex, nonlinear oxygen–tissue interactions to a unidimensional number, glossing over key physiological nuances: it assumes uniform tissue vulnerability and a linear dose–response, and neglects reoxygenation kinetics that drive ROS bursts [205]. Microvascular beds in heart, brain and kidney exhibit distinct oxygen-dissociation curves and temporal sensitivities to hypoxia–reoxygenation cycles [9, 26], yet hypoxic burden applies identical weighting across organs. Moreover, the hypoxic burden metric omits critical physiological features such as peak reoxygenation slopes, key drivers of oxidative injury and the modulatory effects of concurrent hypercapnia, all of which are common in clinical OSA. Without accounting for real-time tissue oxygenation dynamics, fluctuations in ROS and organ-specific metabolic responses, the hypoxic burden oversimplifies the complex pathophysiology of OSA and may misrepresent both individual patient risk and the true impact of IH. These issues regarding dosage of hypoxia and its presentation have been summarised in several contextual settings as related to specific organs affected by IH [56, 152, 206–208]. Furthermore, considerations related to reproducibility of the findings across various laboratories despite employing a priori identical conditions have prompted the implementation of control variations in the protocol, targeted division of the total cohort into several cohorts, or alternatively an assessment of genotype–phenotype interactions [209–213].

A second major blind spot that has hampered progress in OSA research is duration. When exactly does IH/OSA cross the threshold from “beneficial or protective” preconditioning to pathological stress? Preclinical studies variably define chronic IH from mere days (2–7 days) up to 90+ days, often invoking murine–human lifespan conversions (e.g. 7 rodent days≈1 human year) but without stringent empirical validation [201]. We should emphasise that the onset time and cumulative duration of OSA in patients is unknowable (no time stamp marks the first airway collapse), so aligning rodent chronicity to human disease remains inherently speculative. As discussed above, the cellular kinetics of HIF stabilisation, ROS generation and inflammatory priming are nonlinear; transient oxidative bursts may initiate adaptive gene programmes, whereas persistent ROS exposures precipitate epigenetic modifications that lock in maladaptive phenotypes [37, 214–217]. Clinical attempts to infer duration effects rely on patient recall, surrogate vascular stiffness measures or cross-sectional polysomnography, but these cannot resolve whether injury accrues over weeks, months or decades [218]. Until validated biomarkers or real-time tissue partial oxygen pressure (P_O2) sensors demarcate the adaptive–maladaptive inflection, preclinical duration windows remain arbitrary and risk either underestimating or overestimating the true impact of IH.

Beyond “dose” and “duration”, biological modifiers (sex, age, genetic background and comorbidities) further dictate IH/OSA outcomes. Approximately 80% of rodent IH studies employ young, male C57BL/6 mice, neglecting female-specific resilience mediated by oestrogenic antioxidant upregulation [219], age-related mitochondrial dysfunction and impaired endothelial repair in older animals [220, 221], and sub-strain metabolic/inflammatory divergences [201]. Discrepancies across studies are likely due to unstandardised IH dose and duration and animal demographics. Moreover, OSA rarely occurs in isolation: obesity, hypertension, diabetes and SF interact with IH to exacerbate metabolic, cardiovascular and cognitive pathology. Multiple hit models combining IH with a high-fat diet or chronic SF reveal complex amplification of dysfunctional consequences yet remain underutilised. We should not omit the importance of control conditions. In the process of establishing a consensus IH protocol, conditions such as airflow, humidity, noise, light exposure, temperature and feeding patterns need to be closely matched to those operative in the context of the IH regimen, because all of these confounders can influence physiological outcomes.

OSA is not simply chronic IH. Although IH clearly plays a major role in the pathophysiology of the end-organ morbidity induced by OSA, other elements that characterise OSA need to be brought into consideration, namely, the episodic hypercapnia (EHC) that accompanies some of the events, and of course the SF that characterises the termination of a vast proportion of the respiratory events. Although these perturbations have not been as extensively studied in rodent models of OSA when compared to IH, evidence has accumulated pointing to the independent detrimental effects of chronic SF on mood and neurocognitive, cardiovascular and metabolic functions. Indeed, SF, even in the absence of sleep curtailment, induces increased sleep propensity that appears to be mediated, at least in part, by enhanced oxidative stress and cytokine release in the central nervous system, likely due to activation of microglia, while increased permeability of the BBB further accentuates the deterioration in memory and learning functions [222–231]. Since the initial observations by Carreras et al. [232] of the presence of large vessel alterations following prolonged SF exposures, more recent evidence has not only corroborated such findings but further extended them to smaller vascular trees such as the coronary arteries [233]. Similarly, SF imposes robust changes in metabolism that encompass the emergence of increased food consumption, with weight gain ultimately leading to obesity, and the emergence of insulin resistance as well as adipose tissue inflammation, seemingly via oxidative pathways, adipocyte-progenitor dysregulation, central leptin-related pathways and alterations in the gut microbiome [234–240]. Regarding EHC, there is limited evidence to enable any putative inferences as to the independent role of EHC. Some studies by the Haddad laboratory (UC San Diego) consisting of EHC combined with IH exposures have shown equivalent findings to those reported using IH alone and have implicated both neuronal and vascular beds along with possible involvement of the gut microbiome [241–245]. There is no doubt that incremental effort is needed to further demarcate the relative contributions of each one of these constitutive characteristics of OSA while assessing their interactions in vivo. For the sake of completeness, the mechanistic role of the gut microbiome in the modulation and mediation of some of the phenotypic expression elements traditionally explored in preclinical models of OSA deserves mention. Indeed, since the initial description of the changes in gut microbiota induced by IH, the contributions of such changes per se to increased sleepiness, and to the vascular and metabolic morbidities in mice exposed to either IH or SF, have recently gained credence and have clearly designated the gut microbiome as a targetable therapeutic site for mitigation of end-organ morbidity [85, 116, 242, 246–255].

Another important aspect that has only recently begun exploration involves the reversibility of IH-induced injury, i.e. as a proxy for CPAP efficacy, and is a topic that remains somewhat contentious. Some studies show limited endothelial or cognitive recovery after IH cessation, implying irreversible remodelling [256]; other studies with shorter IH exposures report full restoration of vascular and metabolic homeostasis under comparable exposures [257]. These divergent outcomes likely reflect the heterogeneity in IH “dose”, cycle timing, chronicity definitions, recovery durations and end-points measured [201]. Harmonising these variables and scaling rodent “chronic” definitions to human decades will clarify whether maladaptive changes represent fixed injuries or reversible plasticity amenable to therapeutic interventions.

To reconcile decades of discordant findings across both clinical and preclinical OSA research, the field must adopt a hybrid strategy that combines standardised frameworks with individualised, physiologically relevant modelling. In preclinical science, this means establishing consensus IH protocol tiers that clearly define “mild”, “moderate” and “severe” exposures based on nadir F_iO2, cycle frequency, desaturation duration, total exposure time and tissue-specific P_O2 dynamics. Such standardisation would enable reproducibility and facilitate meta-analyses across laboratories. In the clinical realm, a parallel effort is needed to move beyond relatively simplistic metrics such as the AHI, Oxygen Desaturation Index or even the hypoxic burden.

Advances in high-resolution oximetry, wearable sensors and real-time gas exchange monitoring now make it possible to reconstruct patient-specific hypoxic signatures. These profiles can be used both to stratify patients in clinical studies and to inform the design of programmable gas-mixing systems in animal models that more faithfully replicate human pathophysiology, accounting for sleep stage (rapid eye movement versus non-rapid eye movement), comorbidities (e.g. obesity, hypertension) and biological variables like age and sex. Centralised repositories linking animal and human IH datasets, including multi-omics, vascular imaging and dynamic physiological readouts, should allow for cross-species integration using machine learning, and could help define the molecular and phenotypic signatures that distinguish adaptive from maladaptive responses. Crucially, both experimental and clinical efforts must move beyond static metrics like area under the curve and towards developing a physiologically grounded “dynamic hypoxic index” that incorporates not just how much hypoxia occurs, but how fast, how often, and in what tissue context it operates.

Summary

Here, we have identified the dose-dependent paradox of IH, the limitations of current severity metrics such as the AHI and hypoxic burden, the problem of OSA duration uncertainty, and the varying influences of biological modifiers such as sex, age, genetics and comorbidities. In addition, the critically important issue of the reversibility of the IH-induced injury and its determinants remains virtually unexplored (figure 4).

FIGURE 4.

Conceptual model of the barriers limiting progress in obstructive sleep apnoea (OSA)/intermittent hypoxia (IH) research. IH exerts dose- and duration-dependent effects, ranging from adaptive responses to maladaptive end-organ injury, which are not fully captured by simplified indices such as hypoxic burden (area under the curve (AUC)). Outcomes are further modified by biological factors (e.g sex, age, diet, comorbidities) and uncertainty in exposure chronicity and reversibility. This complexity underscores the need for standardised, physiologically grounded metrics to better relate IH patterns to clinical outcomes. S_pO2: peripheral capillary oxygen saturation.

Conclusions

We have here provided a comprehensive overview of the pathophysiological mechanisms driving OSA-induced morbidities, with a critical assessment of the extant and extensive evidence published to date. In addition, we formulate novel conceptual frameworks that could guide the field into prospective research opportunities aimed at improved understanding of the phenotypic heterogeneity of OSA and the formulation of more precise therapeutic strategies targeting improved mitigation or even reversal of the adverse consequences of OSA.