Reduced oxygen delivery to muscles during moderate exercise in adults with long COVID

This single‐center, cross‐sectional, observational study received both NHS (IRAS ID: 313936) and institutional (ETH2324‐1808) research ethics approval. All participants provided written informed consent before their involvement in the study, and the study conformed to the standards set by Good Clinical Practice and by the Declaration of Helsinki (Version 2024).

Participants were asked at each study visit to report any adverse events following any of the study visits; this included contacting the study team directly within the 7 days post completion of study activities. Adverse events were reported and investigated by appropriate clinical personnel (TB).

NIRS‐derived signals for oxyhaemoglobin [O₂Hb], deoxyhaemoglobin [HHb], total hemoglobin (THB = [O₂Hb] + [HHb]) and tissue saturation index (TSI% = [O₂Hb]/[THB] × 100) were continuously monitored pre‐ and post‐exercise during seated rest and unloaded and loaded cycle exercise with an Artinis Portalite MKII device (Artinis Medical Systems BV, Zetten, Netherlands). The NIRS optode was placed on the area of greatest muscle mass on the left gastrocnemius muscle for each participant and was covered/secured using bandages to avoid displacement of the device and to shield against contamination of ambient light. Upon completion of the test on day one, the placement of the optode was marked using an anatomical marker pen so that the optode could be placed in the same position on day two.

CPET data were reviewed for completeness by the test site coordinator (CT). Raw gas analysis data were transformed from breath‐by‐breath to the middle 5‐of‐7 breath and 30‐s moving averages to reduce variability and identify data outliers or missing data. Definitions for VT1 and the respiratory compensation point (VT2) are defined in previous work (Thomas, Kudiersky, et al., 2025; Thomas, Nunes, et al., 2025). VT1 and VT2 were determined using the V‐slope method of the middle 5‐of‐7 breaths data. Standardized American College of Sports Medicine guidelines (Liguori & American College of Sports Medicine, 2020) were used to make exercise threshold decisions by the test site coordinator (CT) with 10% checked by an additional author (RA). Any uncertainty on exercise threshold decisions was resolved through discussion between MF, CT, and RA where consensus originally had not been reached. A detailed account of CPET data processing and analysis can be found in our previous work (Thomas, Kudiersky, et al., 2025). NIRS data were exported at a frequency of 100 Hz and processed using Microsoft Excel. TSI% during seated rest periods before and after cycle‐exercise (5‐min), as well as loaded (1‐min stages) and unloaded pedaling periods (30 s), were calculated.

Compared with day 1, on day 2 heart rate (mean increase 4 ± 5%), minute ventilation (mean increase 6 ± 4%), breathing frequency (mean increase 8 ± 10%), and P_ETO₂ (mean increase 2 ± 8%) for LC were all increased at rest (Table 3). At VT1, V̇O₂ (mean reduction 9 ± 20%), oxygen pulse (mean reduction 9 ± 22%), and V̇CO₂ (mean reduction 9 ± 17%) for LC were all reduced on day 2 (Table 3). No between‐day differences were observed for CPET variables at iso‐time peak for either LC or CON. No between‐day differences were observed for any of the CPET variables at rest for CON. VT1 was not identifiable on both days for any CON participant. There were no between‐day differences in peak blood lactate at rest (p = 0.451) or post‐exercise (p = 0.087) in LC.

Table 5 shows the results recorded between days for LC and CON. No differences were observed for resting values of TSI% for LC (n = 41; 64 ± 5 vs. 63 ± 5%; p > 0.05) or CON (n = 10; 60 ± 4 vs. 62 ± 5%; p > 0.05) cohorts. No differences were observed in TSI% for those with an identifiable VT1 in LC across both days (n = 22; 65 ± 5 vs. 65 ± 4; p > 0.05). This trend continued for each of the exercise minutes and recovery period for both groups (Table 4).

The present study revealed notable between‐group differences in ventilatory pattern. Specifically, LC participants demonstrated elevated V̇E/V̇O2₂, V̇E/V̇CO₂, RER, and P_ETO₂, alongside reduced P_ETCO₂ at iso‐time peak, compared with controls. These findings align with Aparisi et al. who report increased ventilatory inefficiency, characterized by higher V̇E/V̇CO₂ and lower P_ETCO₂ at the anaerobic threshold (AT) in previously hospitalized COVID‐19 patients with persistent dyspnoea compared with asymptomatic controls (Aparisi et al., 2021). Known differences have been reported between those people with LC who had been hospitalized with a SARS‐CoV‐2 infection and those who were not (Pérez‐González et al., 2022), but it appears there may be some interesting overlap between these groups (e.g., inspiratory muscle strength) (Gach et al., 2024; Hennigs et al., 2022; Nagel et al., 2022). The observed increase in P_ETO₂ further suggests dysfunctional breathing and/or a mismatch between ventilation and perfusion in the lungs, supporting the need for future research to investigate these areas in more detail. Notably, unlike previous findings at the AT (Aparisi et al., 2021), the higher iso‐time peak RER in the LC cohort indicates impaired energy substrate utilization for the given absolute exercise intensity. Nevertheless, additional mechanisms within the oxygen delivery and utilization pathway (e.g., impaired cardiac output, microvascular dysfunction, altered muscle fiber composition, and mitochondrial dysfunction) are likely contributing to the early shift toward anaerobic metabolism observed during submaximal exercise in individuals with Long COVID.

SARS‐CoV‐2 infection is well‐documented to cause increased platelet activation and damage to the microcirculation, including degradation of the endothelial glycocalyx layer (Wu et al., 2024). The glycocalyx plays a critical role in regulating microvascular resistance and capillary hemodynamics; thus, COVID‐19 related damage could contribute to tissue hypoxia (Østergaard, 2021). During acute infection, elevated levels of angiopoietin‐2 (ANG‐2) and vascular endothelial growth factors (VEGFs) have been observed, indicating endothelial activation and vascular instability (Vassiliou et al., 2023). These biomarkers increase vascular permeability by promoting leakage across the alveolar‐capillary barrier, contributing to pulmonary oedema and impaired gas exchange (Bellani et al., 2016; Vassiliou et al., 2023; Wu et al., 2024). Furthermore, these elevated biomarkers cause microvascular dysfunction through a reduction in capillary integrity and oxygen delivery and promote a prothrombotic state as endothelial activation can trigger clot formation, leading to microthrombi and ventilation‐perfusion mismatch (Kell & Pretorius, 2022). These changes could amplify inflammatory signaling, perpetuating tissue damage and fibrosis risk, causing the chronic sequelae that is LC. Patel et al. reported that increased ANG‐1 and P‐selectin (P‐SEL) levels could classify LC cases with 96% accuracy, implicating endothelial dysfunction as a potential pathology (Patel et al., 2022). Although the present study did not include blood biomarker analyses, reduced TSI% data suggest impaired oxygen delivery that may be consistent with the hypothesis of capillary damage and increased microvascular resistance in those living with LC. Future research is warranted to examine this proposed relationship and progress this mechanistic knowledge base.

The presence of fibrinaloid microclots could also impair oxygen transport. These heterogeneous, amyloid‐state biomarkers are composed of inflammatory proteins, fibrinogen chains, and antifibrinolytic molecules, which render them highly resistant to proteolytic breakdown (Kell & Pretorius, 2017). As a result, they have been proposed to persist in circulation, obstructing microvascular function, impairing gas exchange, and potentially contributing to the early onset of anaerobic metabolism observed in individuals with LC in this study and elsewhere (Kell et al., 2022; Pretorius et al., 2021; Thomas, Nunes, et al., 2025). Previous work from our lab indicates that microclots may fragment into smaller aggregates following submaximal CPET (Thomas, Nunes, et al., 2025), which may enter and/or impede the microvasculature. This could reduce perfusion, potentially exacerbate post‐exercise dysfunction, and impair physiological function. However, Appelman et al. found that after maximal CPET designed to induce PEM in a LC cohort, substantial amyloid‐containing deposits were not found within capillaries, but rather within the extracellular matrix when compared to controls. This suggests that while microclots may play a role in the development of PEM, their distribution, movement characteristics, underlying mechanisms, and outcomes remain unclear and warrant further investigation.

Invasive CPET (iCPET) offers insight into identifying localized physiological dysfunction and research using this method has shown distinct endotypes of LC (Risbano, 2025; Risbano et al., 2023); for example, preload failure and impaired systemic oxygen extraction. Combining real‐time blood assessment of arterial and venous blood with measures of whole‐body oxygen consumption allows researchers to determine whether oxygen delivery or extraction is impaired within exercising participants (Leitner et al., 2024). Kahn et al. (2024) observed that for 75% of LC participants with exertional intolerance who took part in iCPET, peak systemic oxygen extraction was impaired. This study also observed that this impairment was no different between those who have undertaken supervised rehabilitation, reaffirming the need for targeted management, rehabilitative, and therapeutic strategies for those living with LC. Although the approach in the present study lacks the specificity to localize physiological dysfunction, the reduction in oxygen pulse at VT1 in the LC cohort on day 2 compared with day 1, as well as the elevated P_ETO₂ compared with the control cohort, supports the notion that oxygen is not extracted appropriately in the LC cohort. Combining NIRS with techniques such as venous occlusion induced by a pneumatic cuff would allow for the measurement of peripheral fractional oxygen extraction (Wolfsberger et al., 2022). This would provide greater insight into this dysfunction and allow for better specificity during noninvasive CPET.

SARS‐CoV‐2 exploits host cell mitochondria for viral replication, leading to structural damage characterized by altered mitochondrial morphology, size, and number (Molnar et al., 2024). These disruptions impair normal mitochondrial functioning resulting in reduced energy production, and mitochondrial dysfunction is linked to symptoms of LC (Molnar et al., 2024). For example, differently regulated mitochondrial proteins and structural abnormalities, including mitochondrial swelling and disrupted cristae, have been reported (Molnar et al., 2024; Peppercorn et al., 2023). Additionally, Appelman et al. proposed that reduced maximal mitochondrial respiration and decreased mitochondrial content may be an important consideration in the pathophysiology of PEM. This may be one of many factors that caused elevated blood lactate levels >3.0 mMol/L post‐CPET on both days from the LC cohort in the present study, as seen with myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) patients across 2‐day CPET (Lien et al., 2019). Davenport et al. observed similar bioenergetic patterns between LC and ME/CFS through 2‐day CPET (e.g., significant reductions in V̇O₂ and workload at the ventilatory anaerobic threshold in both LC and ME/CFS compared to non‐disabled controls) (Davenport et al., 2026). However, it is important to note that, although the underlying pathophysiology of both conditions remains unclear, the observed similarities beyond the work by Davenport et al. are largely based on commonly reported symptoms rather than confirmed shared mechanisms. Detailed investigations, including skeletal muscle biopsies, cardiopulmonary imaging, invasive hemodynamic assessments, mitochondrial functional assays, and metabolomic profiling could provide further insight into whether mitochondrial limitations and other peripheral or central factors contribute to the progressive impairment observed during 2‐day CPET in individuals with LC and ME/CFS.

Biomarkers with proinflammatory properties have been commonly identified as part of diagnostic LC reviews (Espin et al., 2023; Lai et al., 2023; Thomas et al., 2024); however, the evidence often varies between participants as no established profile of biomarkers have been identified. It has been suggested that inflammatory subgroups may exist; for example, Talla et al. identified two distinct inflammatory subtypes of LC in their cohort (Talla et al., 2023). This may indicate that subgrouping may be possible for several other pathologies (e.g., endothelial dysfunction, microclotting, and mitochondrial dysfunction); however, given the mechanistic overlap that has been postulated between hypotheses (e.g., inflammation has been associated with viral persistence and microclot breakdown; Pretorius et al., 2021; Proal et al., 2025) these cannot be mutually exclusive. Heightened inflammation has also been correlated with metabolic dysfunction (Aid et al., 2025), and such profiling may optimize patient profiling and improve referral toward therapeutic targets that are more likely to be helpful for those individuals (e.g., antiviral therapy may be beneficial for individuals with persistently heightened inflammatory profiles). CPET and NIRS could be helpful within the context of this pathology; however, targeted blood biomarker profiling of the inflammatory subtypes identified by Talla et al. is required concurrently to support this approach.

Dysautonomia and chronotropic intolerance (e.g., abnormally low heart rate reaction) are common in people living with LC (Mustonen et al., 2024). Using the criteria applied by Mustonen et al., identifying dysautonomia using the protocol in the present study could be calculated with the addition of heart rate recovery data. Meanwhile, alternative criteria of chronotropic intolerance that do not require maximal testing could also be applied (e.g., degree of change in heart rate during the 6MWT) (Provencher et al., 2006). Impaired endothelial and microvascular function indicated by the findings of the present study has been linked to cardiovascular symptoms of LC, parallel to postural orthostatic tachycardia syndrome or as a part of broader cardiovascular dysautonomia (Fedorowski et al., 2023; Oikonomou et al., 2022). Other proposed mechanisms include abnormal sympathetic activity and excess circulating catecholamines, peripheral sympathetic noradrenergic denervation, which is responsible for venous pooling (Vernino et al., 2021). Venous blood pooling has previously been assessed by NIRS of the gastrocnemius muscle (Stone et al., 2016; Truijen et al., 2012) and may offer an additional and potentially more suitable, simple, and inexpensive assessment of dysautonomia for people living with LC.

The variability in probe placement and unaccounted differences in adipose tissue thickness prevented valid between‐group comparisons of TSI% in this study. Increased adipose tissue can distort NIRS signals, leading to an overestimation of TSI%. This occurs because the signal becomes more reflective of the adipose tissue, which typically has a higher TSI% compared to skeletal muscle due to its lower metabolic rate (Niemeijer et al., 2017). However, probe placement location between days was kept identical using an anatomical marker pen on day 1, which allowed for reliable comparisons between day for the same participant. Additionally, the use of continuous‐wave NIRS and the selection of the gastrocnemius muscle rather than the vastus lateralis employed in previous studies introduce further considerations regarding signal quality and interpretation. This study did not include biomarker analyses or additional assessment techniques to confirm endothelial damage or mitochondrial dysfunction, both of which have been hypothesized to impair oxygen delivery and contribute to abnormal energy metabolism and PESE/PEM. Age matching was not applied to recruitment in this study, with the LC group significantly older than the CON group. It is possible that the worsened ventilatory responses observed for LC during incremental CPET may have been impacted by this factor (e.g., V̇E/V̇CO₂ slope increases with age; Phillips et al., 2020), and this uncertainty could be resolved by recruiting similarly aged CON groups in future work.

The present study makes use of submaximal exercise approaches, which identified VT1 in over half of the LC cohort on both days. This response may serve as a practical diagnostic tool for clinicians supporting individuals with LC. Nonetheless, given the overlap of these physiological responses with other conditions such as ME/CFS and the contraindication of such testing for those with moderate‐to‐severe PEM, further diagnostic measures are needed to accurately identify, manage, and treat affected individuals. Future research integrating muscle biopsies and endothelial and mitochondrial biomarkers alongside the current methodology could yield valuable mechanistic insights. However, such investigations must carefully balance the burden placed on participants with the cost‐effectiveness and potential impact of findings. Traditionally, the vastus lateralis has been more commonly used as opposed to the gastrocnemius for NIRS measurements during cycling assessments as this muscle is more heavily involved in cycling tasks (Iskra & Paravlić, 2025). During pilot testing with LC volunteers, we encountered several issues when using the vastus lateralis (e.g., sensor displacement, time to affix the device, and participant comfort) which added to the participant burden in addition to the additional time required for questionnaire completion and cognitive assessment. Accordingly, we chose the gastrocnemius as a convenient, well‐tolerated and previously used muscle of interest to complement the additional measurements. Future use of vastus lateralis NIRS measurements with LC participants during CPET would further LC mechanistic understanding and enable greater comparison to the broader literature. Research designs may seek to reduce participant burden by isolating this measure and remove additional time constraints or complete them remotely on separate days (e.g., questionnaire completion). Furthermore, the use of NIRS for assessment in those living with LC has potential applications for the metabolic assessment of other regions of the body (e.g., brain) (Murkin & Arango, 2009; Simonson & Piantadosi, 1996). Future work may apply this technology without exercising the individual, allowing for inclusion of those who present with a greater risk of more severe PEM, which may provide pathophysiological insight driving their symptom presentation.

Anonymized data can be provided upon request.