Long COVID is linked to ongoing problems with cellular energy production and immune system imbalance

In our previous work, we demonstrated that COVID-19 autopsy tissues—including the heart, kidney, liver, and lymph nodes—exhibited prolonged inhibition of OXPHOS transcripts (26, 28) (Figure 1A). Notably, these transcriptional changes persisted despite the absence of detectable viral transcripts, suggesting that SARS-CoV-2 infection is associated with sustained mitochondrial and immune dysregulation even after viral clearance. This repression was accompanied by increased expression of immune and HIF-1α transcripts across all tissues except the lymph nodes. In the lymph nodes, there was a significant decrease in immune gene transcription and a marked upregulation of the renin-angiotensin-aldosterone system (RAAS) pathway, which coincided with histological scarring. Conversely, lung tissue showed recovery of OXPHOS transcripts but exhibited the highest level of immune activation, characterized by significantly elevated mitochondrial immune and stress responses (26, 28).

These findings suggest that persistent immunometabolic alterations can remain in viral patients long after viral clearance, suggesting that sustained activation of these pathways may contribute to PCS. However, because these samples were derived from individuals who succumbed to acute COVID-19, the observed mitochondrial and immune alterations likely represent the upper bounds of SARS-CoV-2–induced tissue stress and inflammatory activation rather than PCS-specific pathology. Unlike severe human cases, PCS patients do not succumb to infection, limiting access to post-acute human tissues. To investigate these processes in a model more reflective of post-acute disease, we analyzed a hamster model of PCS.

The Syrian hamster is a well-established model that recapitulates key features of mild human PCS (45). These animals, being naturally susceptible to SARS-CoV-2, develop a non-lethal, self-limiting disease, reaching peak lung viral loads at the acute phase (3 dpi) and achieving complete viral clearance in all the tissues by 5 dpi (20, 26, 28). The post-acute/chronic phase (31 dpi) displays respiratory dysfunction and neurological impairment at 31 dpi (20). To build on these findings we performed pathway analysis on RNA-seq data from the heart, kidney, and lungs of SARS-CoV-2-infected hamsters (Figures 1B, C;Supplementary Figure S1).

During the acute phase (3 dpi), we observed upregulation of mitochondrial ribosomal genes in the lungs and heart, along with signatures of enhanced mitochondrial biogenesis in the heart and kidneys (Figure 1C). Non-oxidative metabolic genes were also upregulated in the heart and lung, suggesting a shift away from oxidative metabolism that was previously shown to be crucial for facilitating viral replication in the lungs (3, 27, 46). Robust innate immune activation was also observed across all tissues, with enrichment of pathways related to mtdsRNA and mtDNA signaling. This response was most pronounced in the lungs, followed by the heart and the kidneys. RAAS and senescence-associated pathways were activated in the lungs, heart, and kidneys, while PANoptosis-related signatures were strongly induced in the lungs and heart (Figure 1C).

By the post-acute phase (31 dpi), lung metabolism normalized, aligning with prior human data showing mitochondrial recovery after severe COVID-19 (26) (Figure 1C). However, immune gene expression and mitochondrial stress responses in the lung remained elevated, indicating ongoing inflammatory signaling. In contrast, heart and kidney tissues continued to exhibit suppressed OXPHOS expression, reflecting sustained bioenergetic impairment. While immune signatures had resolved in the kidney, they remained modestly elevated in the heart. The sustained mitochondrial suppression signaling in the heart suggests a greater vulnerability to long-term dysfunction in cardiovascular tissues, and the persistent activation of mitochondrial stress responses in the lung indicates a distinct vulnerability to chronic immune dysregulation.

To investigate chronic fatigue, we analyzed transcriptomic data from hamster quadriceps muscles collected at 3, 31, and 61 dpi (15). To assess translational relevance in muscle tissue to humans, we also examined vastus lateralis muscle biopsies from individuals with PCS that developed chronic fatigue syndrome (PCS-CFS) up to 1-year (369 dpi) post-infection, comparing them with Type 2B muscle fiber atrophy (T2bFA) patients that also suffer from CFS (T2bFA-CFS) without prior SARS-CoV-2 infection (14) (Figure 2A).

At 3 dpi, hamster skeletal muscle, displayed a downregulation of OXPHOS, and upregulation of innate immune and mitochondrial stress response transcripts (Figure 2B;Supplementary Figure S2). OXPHOS gene expression remained suppressed at 31 and 61 dpi, with the later time points also showed persistent extracellular matrix remodeling and, by 61 dpi, increased expression of HIF and mTOR pathways.

In human skeletal muscle, both PCS-CFS and T2bFA-CFS patients exhibited similar transcriptional signatures, including downregulated OXPHOS and elevated HIF signaling. However, PCS-CFS samples also showed heightened activation of the integrated stress response (ISR), unfolded protein response (UPR), and both innate and adaptive immune signaling. T2bFA-CFS patients also exhibited predominant upregulation of mitochondrial stress, ISR, and induced senescence transcripts (Figure 2B). Notably, several immune and stress-response pathways were more pronounced in humans than in hamsters—differences likely reflecting variation in sampling intervals, disease severity, and species-specific immune adaptation.

To provide context for these findings, we developed a custom list of pathways from publicly available gene lists related to muscle function. We then organized these pathways into enriched terms within functional clusters and conducted GSEA analysis (Figure 2C;Supplementary Table S1). In hamster skeletal muscle we observed a progressive disruption of muscle function-related pathways from 3 to 61 dpi. At 3 dpi, pathways related to apoptosis and proliferation were enriched. By 31 dpi, pathways associated with development and differentiation were upregulated, and muscle mass pathways were downregulated. These alterations were maintained at 61 dpi, which also displayed a suppression of contraction-related pathways along with an increase in pathways linked to abnormal pathology, consistent with a decline in muscle function.

In humans, consistent with their pathological phenotype, PCS-CFS samples showed significant downregulation of mitochondrial and contraction pathways, resembling the changes seen in hamster muscles at 61 dpi. T2bFA-CFS samples showed upregulation of developmental and proliferation pathways, alongside a downregulation of mitochondrial function pathways, consistent with the trends seen across all PCS muscle samples (Figure 2C). Notably, the overlap between PCS-CFS and T2bFA-CFS was more pronounced at the level of core transcriptional programs (Figure 2B) than at the pathway level (Figure 2C), indicating that shared mitochondrial repression may be accompanied by context-specific secondary remodeling programs. Overall, this research suggests that the persistent inhibition of OXPHOS transcripts is linked to progressive changes in pathways associated with muscle function decline and fatigue in PCS.

Neurological symptoms are among the most debilitating and persistent in PCS, encompassing anosmia, dysgeusia, brain fog, fatigue, and mood disturbances. These symptoms are thought to arise from a combination of direct viral effects, chronic neuroinflammation, and altered neuro-metabolism (9, 21, 22, 47–49). To investigate, we analyzed RNA-seq data from SARS-CoV-2–infected Syrian hamsters at acute (3 dpi) and chronic (31 dpi) time points across key sensory and cognitive brain regions (20).

To model sensory deficits, we profiled the olfactory epithelium (OE), olfactory bulb (OB), and trigeminal ganglia (TG), tissues responsible for smell and facial sensation (Figure 3A). The OE and OB are first-line sensory interfaces directly exposed to inhaled pathogens, with the OE acting as a mucosal immune sensor and the OB integrating signals via olfactory neurons. By contrast, the TG is more insulated but acts as a secondary sensory relay responsive to both peripheral and central signals (50).

During the acute phase (3 dpi), all sensory tissues displayed robust mitochondrial stress and innate immune activation (Figure 3B), consistent with systemic inflammatory signaling observed in peripheral tissues (Figures 1, 2). In addition, the OB and TG showed strong induction of tissue damage pathways and PANoptosis-associated genes, alongside suppression of OXPHOS and peroxisomal metabolism. By 31 dpi, olfactory tissues remained inflamed, with the OE exhibiting persistent HIF activation and suppression of fatty acid oxidation (FAO). In contrast, immune signaling in the TG was largely resolved, while OXPHOS and mitophagy pathways were upregulated, indicating a regenerative mitochondrial response (Figure 3B;Supplementary Figure S3).

To place these changes in a functional context, we again generated a custom list of pathways by combining publicly available gene lists related to brain function and performed GSEA analysis (Figure 3C;Supplementary Table S1). This revealed sustained suppression of neuronal signaling and synaptic function in the OE and OB from 3 to 31 dpi, providing a mechanistic basis for persistent anosmia and sensory dysfunction in PCS. TG, in contrast, showed only mild, transient changes, aligning with its trajectory toward recovery. Collectively, these findings indicate that the olfactory system, particularly the OE and OB, undergoes early mitochondrial dysfunction and prolonged immune activation, processes that, if unresolved, may underlie long-term sensory deficits observed in post-acute phases.

To investigate cognitive and affective symptoms, we analyzed the striatum, thalamus, medial prefrontal cortex (mPFC), and cerebellum, regions involved in motor, executive, and emotional functions (20) (Figure 4A). Unlike the olfactory tissues, these brain regions are normally protected by the blood–brain barrier (BBB), making them less susceptible to immune activation unless the BBB integrity is compromised (9). Despite this protection, at 3 dpi, all brain regions showed a robust upregulation of innate immunity and mitochondrial stress pathways, raising the possibility of altered BBB integrity (Figure 4B). The thalamus also showed upregulation of PANoptosis and the cerebellum activated senescence pathways. The mPFC and cerebellum exhibited reduced OXPHOS, while the striatum and thalamus showed elevated OXPHOS transcripts. By 31 dpi, the striatum, thalamus, and cerebellum showed reversal of early responses, with decreased immune activity, consistent with greater BBB integrity. This reversal in immune response was associated with a reversal in metabolic response, with the striatum and thalamus now displaying reduced OXPHOS transcripts, and cerebellum upregulation. In contrast, the mPFC worsened, with stronger repression of OXPHOS transcripts (Figure 4B; Supplementary Figure S4).

Immune and metabolic changes were accompanied by altered neuronal function (Figure 4C). At 3 dpi, immune activation and mitochondrial stress correlated with reduced neuronal signaling in the thalamus, striatum, and mPFC. By 31 dpi, persistent OXPHOS inhibition in the mPFC aligned with suppressed neuronal pathways. The striatum and thalamus recovered both metabolically and functionally. Interestingly, the cerebellum showed enhanced neuronal pathways at 3 dpi but suppression at 31 dpi, suggesting incomplete functional recovery, despite greater metabolic recovery later (Figure 4C).

Collectively, our hamster model suggests that while the initial response to SARS-CoV-2 is a systemic, multi-organ inflammatory event, the long-term sequelae are defined by profound, tissue-specific bioenergetic deficits. The persistent mitochondrial suppression in high-energy tissues like skeletal muscle, heart, and specific brain regions may contribute to fatigue, exercise intolerance, and neurological symptoms of PCS. These findings in a controlled model raise the question of whether a similar, persistent pathogenic state can be identified in human patients experiencing PCS.

Building on our previous observations of persistent OXPHOS suppression across multiple COVID-19 autopsy tissues (26, 28) (Figure 1A), we next examined whether similar mitochondrial and immune transcriptional programs are evident in the CNS. To do so, we analyzed publicly available RNA-seq datasets from human brain autopsy samples, specifically the striatum and frontal cortex, collected from patients who died from acute COVID-19 (21, 22, 42) (Figure 4A). While these severe cases may reflect intensive care units (ICU)-associated exposures (e.g., hypoxia, ventilation, sedation), the inclusion of non-infected autopsy controls provides a important baseline for interpreting COVID-19-associated transcriptional changes. In the frontal cortex, we observed pronounced inhibition of OXPHOS transcripts, increased HIF and mTOR signaling, and widespread immune activation, without clear induction of compensatory mitochondrial stress responses—findings consistent with inflammatory remodeling and possible BBB disruption (Figure 4B;Supplementary Figure S4). In contrast, the substantia nigra displayed elevated OXPHOS-associated transcripts with minimal immune activation, suggesting regional metabolic resilience. Functionally, persistent OXPHOS inhibition in the frontal cortex coincided with suppression of neuronal and synaptic signaling pathways, whereas the substantia nigra exhibited few functional pathway changes (Figure 4C). Together, these results are consistent with region-specific metabolic and immune remodeling in the human brain during severe COVID-19, providing context for how acute neuroimmune stress may increase neurological vulnerability and potentially predispose susceptible individuals to persistent post-acute symptoms.

We next expanded our analysis to longitudinal PBMC datasets collected 2–12 months post-infection using single-cell RNA sequencing (scRNA-seq) (16). Importantly, this cohort was stratified by acute disease severity (mild vs severe requiring ICU care) and time since recovery (early: 2–4 months; late: 4–12 months), rather than by the presence of persistent symptoms (Figure 5A). To determine whether mitochondrial and immune dysregulation persists after acute infection, we analyzed PBMCs from mild and severe patients across early and late recovery phases.

Pathway- and transcript-level analyses revealed distinct temporal patterns of mitochondrial regulation across PBMC subtypes (Figure 5B). During the early phase, severe cases demonstrated coordinated downregulation of mtDNA-encoded OXPHOS transcripts across complexes I, III, IV, and V in most immune cell types. This suppression was weaker in mild cases, where reductions were more restricted, for example in B cells and dendritic cells. Mild patients also showed partial metabolic recovery, including increased nDNA- and mtDNA-encoded OXPHOS transcripts in plasmacytoid dendritic cells (pDCs) and hematopoietic stem and progenitor cells (HSPCs).

In contrast, late-phase severe patients showed limited evidence of coordinated recovery, instead exhibiting persistent broad suppression of mtDNA-encoded OXPHOS genes across nearly all PBMC subsets, with only isolated exceptions, such as MT-CO2 and MT-CO3. nDNA-encoded OXPHOS pathways were similarly downregulated, indicating sustained nuclear–mitochondrial transcriptional inhibition. Because late-phase samples reflect individuals longitudinally followed after severe disease, these findings suggest that prolonged mitochondrial suppression can persist for months after recovery, independent of symptom stratification, consistent with durable immunometabolic reprogramming described in this cohort (16). Collectively, these results are consistent with the idea that severe SARS-CoV-2 infection is associated with long-lasting repression of mitochondrial bioenergetic programs in circulating immune cells, providing a potential systemic link through which acute disease severity may contribute to prolonged post-acute vulnerability (Figures 5A, B).

We next expanded our analysis to human plasma and serum samples collected longitudinally at 1 and 20–24 months-post-infection, via proteomics and metabolomics (16, 43, 44). To investigate systemic metabolic alterations in PCS patients, we next analyzed plasma metabolomic data collected from acutely hospitalized COVID-19 patients early after infection, as well as from COVID-recovered individuals and PCS patients sampled 20–24 months post-infection (44) (Figure 5C). Notably, this study (44) observed elevated lactate (via metabolomics) and IL-17 (via biochemical analysis) in PCS patients, consistent with increased glycolytic activity and immune dysfunction.

Our analysis was consistent with these findings. Early-phase samples showed elevated glycolytic intermediates—including glucose and lactic acid—reflecting high metabolic demand, accompanied by decreased TCA cycle intermediates such as glutamic acid, citric acid, and related carbon-flux metabolites. This pattern of increased glycolysis and reduced TCA throughput aligns with the transcriptional evidence of diminished oxidative phosphorylation and is consistent with SARS-CoV-2–associated inhibition of OXPHOS (3, 25–28, 41) (Figure 5D;Supplementary Figure S5).

By 20–24 months post-infection, both COVID-recovered and PCS individuals exhibited partial metabolic normalization (Figure 5D;Supplementary Figure S5D). Notably, PCS patients retained core metabolic changes from the acute phase, including persistent elevations in glucose and reductions in glutamic acid. Glucose also remained elevated in COVID-recovered individuals, but to a lesser extent, suggesting that SARS-CoV-2 infection exerts lasting metabolic effects even after clinical recovery—though these are more pronounced in PCS patients (Figure 5D;Supplementary Figure S5C). Furthermore, both COVID-recovered and PCS patients showed elevated kynurenine, a tryptophan-derived metabolite associated with chronic inflammation, IDO activation, and NAD⁺ pathway remodeling, further linking immune dysregulation to ongoing metabolic strain. These findings suggest that both groups experience persistent mitochondrial metabolic remodeling, characterized by reduced OXPHOS recovery and sustained immune-metabolic activation. The persistence of dysfunction in COVID-recovered patients underscores that metabolic stress and incomplete recovery may be associated with increased vulnerability and potentially predispose susceptible individuals to ongoing post-acute symptoms.

Building upon the persistent mitochondrial abnormalities identified, we next examined circulating proteomic profiles to determine whether long-term systemic protein expression patterns reflect chronic immune and metabolic dysregulation in PCS patients (Figure 6A). In this study (43), patient serum samples were collected at 1 and/or 6 months post-infection for patients tracked for PCS symptom progression at 6 months and 1-year post-infection, enabling the early detection of PCS biomarkers. A key finding was increased complement activation in PCS patients, which we further analyzed to assess mitochondrial-linked immune and metabolic pathways.

Proteomic volcano plots were generated for individuals with PCS and those who had fully recovered at 1- and 6-months post-infection (Figures 6B, C;Supplementary Figure S6), followed by pathway-level categorization of differentially expressed proteins (Figures 6D–G;Supplementary Figure S6). At 1-month post-infection, individuals who later developed PCS displayed a markedly broader and more pronounced pattern of proteomic dysregulation than those who recovered, encompassing exaggerated induction of innate immune proteins, interferon-stimulated factors, components of the complement cascade, and extracellular matrix-associated proteins (Supplementary Figure S6). In contrast, recovered individuals showed a more restricted and resolving antiviral signature, consistent with controlled immune contraction. These early proteomic perturbations in PCS mirror and are consistent with the chronic immune-metabolic signatures observed at 6 months (Figure 6), suggesting that PCS may involve the persistence and amplification of an initially maladaptive systemic response, rather than solely the emergence of a late-onset process.

Compared with recovered individuals, 6-month PCS patients exhibited broad and sustained upregulation of cellular stress-response proteins, particularly chaperones and protein–quality control regulators (Figure 6D). These included HSP40/70 co-chaperones (DNAJB and HSPA families), ER-stress and unfolded protein response mediators, and mitochondrial protein-folding and import regulators (CLPP, LONP1). PCS samples also showed elevated levels of GDF-15, a stress-induced mitokine upregulated during mitochondrial dysfunction. Prior studies have consistently linked elevated GDF-15 with COVID-19 severity (26–28, 51–53). This coordinated enrichment is consistent with persistent proteotoxic stress and disrupted protein homeostasis, consistent with chronic mitochondrial dysfunction and unresolved inflammatory signaling.

In parallel, PCS samples also demonstrated continued elevation of antioxidant and redox-buffering enzymes, including superoxide dismutase (SOD1/SOD2) and glutathione-related enzymes (GLRX1/2/3, GSTA2, GSTK1, GSTP1, GPX3), along with TXNDC family proteins, MSRB1, and ALDH-family detoxification enzymes (Figure 6E). This antioxidant signature is consistent with ongoing mitochondrial and oxidative stress, aligning with earlier observations of impaired mitochondrial respiration and heightened inflammatory signaling. By contrast, recovered individuals showed minimal or no upregulation of these stress-response or redox-regulating proteins, consistent with resolution of oxidative burden.

The persistence of this oxidative and proteotoxic stress, along with the prolonged elevation, likely contributes to sustained immune dysregulation. In support of these findings, PCS patients exhibit broad, coordinated activation of both innate and adaptive immune protein networks (). On the innate immune side, PCS samples showed marked upregulation of neutrophil-associated chemotactic proteins (CXCL8, CCL7), acute-phase reactants (CRP, CHI3L1), macrophage activation markers (CD163, S100A12, ARG1), interferon-stimulated innate immune mediators (IFI16, IRF3), and complement-related factors (C3, CFD). The sustained presence of these proteins is consistent with ongoing low-grade inflammation, persistent myeloid activation, and unresolved innate immune signaling months after infection. In contrast, recovered individuals displayed attenuation of these pathways, consistent with immunologic resolution. 1

Similarly, proteins involved in adaptive immune-activation—including cytokine receptors (IL21R, IL2RA), cytotoxic effector molecules (GZMA, GZMB-related pathways), B-cell and T-cell costimulatory/activation markers (CD22, CD79B, LCK, TNFRSF8), and antigen-presentation components (HLA-DRB3, HLA-LA2)—remained chronically upregulated in PCS but not in recovered controls. This persistent enrichment of lymphocyte activation and MHC class II-associated signaling is consistent with continued antigenic stimulation and incomplete immunologic contraction, potentially driven by residual inflammatory cues or ongoing dysregulated antigen presentation.

OXPHOS-associated proteins revealed a clear divergence between PCS and recovered individuals (Figure 6F). PCS samples showed mixed or decreased abundance of key mitochondrial metabolic enzymes (PDHB, DLST, ACADM), suggesting persistent inefficiency in TCA flux and fatty acid oxidation. Recovered patients showed a more normalized or upregulated mitochondrial protein profile, indicating improved bioenergetic function. Notably, some PCS samples exhibited upregulation of compensatory mitochondrial proteins (ATP50) consistent with a partial but insufficient attempt to counteract respiratory chain inefficiency.

Long-COVID proteomes demonstrated upregulation of several glycolytic enzymes, such as HK2, TPI1, GAPDH, and PGK1 (Figure 6G), suggesting a shift toward glycolytic compensation, potentially driven by impaired mitochondrial ATP production. Recovered patients displayed a more balanced metabolic profile with fewer glycolytic alterations.

Together, these proteomic data are consistent with a model in which 6-month-old PCS is associated with persistent mitochondrial dysfunction (impaired OXPHOS enzyme recovery), chronic oxidative stress (sustained antioxidant protein upregulation), proteotoxic and ER stress activation (upregulated chaperones, UPR mediators), and glycolytic compensation (upregulated glycolytic enzymes). In contrast, individuals who recovered exhibit substantial normalization of mitochondrial, redox, and metabolic protein expression, consistent with restoration of cellular homeostasis.

Our findings are consistent with a hypothesized model in which, in healthy patients, mitochondrial repair programs—including antioxidant defenses, mitophagy, and integrated stress responses—may support restoration of bioenergetic and immune homeostasis following infection (Figure 7). In contrast, persistent post-acute phenotypes may be associated with sustained mitochondrial dysfunction in one or more tissues, potentially associated with unresolved SARS-CoV-2–induced mitochondrial damage or deficient repair mechanisms. This persistent mitochondrial stress could contribute to prolonged immune activation, consistent with a potential feed-forward relationship between bioenergetic decline and chronic inflammation.