The effect of pre-existing sleep disturbance on T cell responses to SARS-CoV-2 variants, pro-inflammatory and pro-resolving mediators, and glucocorticoid sensitivity in Long COVID

Abstract

Sleep disturbance is highly common in Long COVID (LC), and is known to worsen infection outcomes and hinder full recovery from infections. We here investigated whether sleep disturbance prior to SARS-CoV-2 infection compromised inflammatory responses in LC. Blood of 74 participants from the National Institute of Health RECOVER Adult clinical cohort were analyzed at the 6-month time point following the infection. Participants were categorized into groups of Likely, Possible, and No LC based on the Long COVID Research Index. Findings show that Likely LC did not differ from Possible or No LC with respect to interferon-gamma expression in CD4 and CD8 T cells stimulated with omicron spike-peptide variants, the expression of inflammatory mediators by monocytes (IL-6, TNF, COX-2), the ability of glucocorticoids (GC) to suppress inflammatory expression in monocytes, or levels of inflammatory pro-resolving mediators (SPMs). 53% of the Likely LC group reported pre-existing sleep disturbance, compared to 30 and 23% of Possible and No LC groups, respectively. In the Likely LC group, reduced ability of GCs to suppress inflammation was associated pre-existing sleep disturbance, suggesting compromised anti-inflammatory control. Research on the type of sleep disturbance (insomnia, hypersomnia) driving altered GC sensitivity will help to identify LC subtype-specific intervention targets.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-47039-y.

Introduction

Sleep disturbance is among the most frequently reported symptoms in Long COVID (LC), also called post-acute sequelae of SARS-CoV-2 infection (PASC), with an average prevalence of 31% across studies¹. A diverse range of sleep disturbances has been reported in LC. Insomnia and unrefreshing sleep are the most commonly reported symptoms^2,3, in addition to reports of hypersomnia³, breathing-related disorders⁴, circadian rhythm disorders⁴, REM behavior disorders⁴, and restless legs syndrome⁵. Sleep is well known for its impact on the immune system and plays a critical role in infection risk, progression, and outcome⁶. With respect to SARS-CoV-2 infection, poor sleep, as reflected by short sleep duration, insomnia symptoms, poor circadian alignment, among others, is a risk factor for the acquisition of infection, infection severity, and mortality^7,8. Pre-existing insomnia, poor sleep quality, and short sleep (< 6 h) has been reported to increase the risk of developing LC by up to 2.7 fold^9,10.

Mechanistically, sleep disturbance may contribute to an increased risk for developing LC by affecting aspects of the immune system. There is good evidence that sleep in humans supports optimal development of B and T cell responses to pathogens, including the response to SARS-CoV-2^11–13. Longer objective sleep duration following mRNA booster vaccinations against SARS-CoV-2 enhanced the B cell antibody response in healthy humans¹⁴. In health care workers, shift work and short sleep durations (< 7 h/night) were both associated with lower anti-S1-receptor-binding domain (RBD) IgG levels following a booster dose of a mRNA SARS-CoV-2 vaccine¹¹. With respect to T cell responses, acute sleep loss, i.e. staying awake for one night following vaccination against hepatitis A, decreased antigen-specific effector and persisting CD4 T cells, an effect that was still evident one year after vaccination¹⁵. T cell responses play a critical role in cellular immunity towards many viruses, including SARS-CoV-2^16,17. Vaccinations against SARS-CoV-2 lead to robust T cell responses, including the production of interferon (IFN)-γ by both CD4 and CD8 T cells¹⁶. There is increasing evidence that T cells play an important role in the development and persistence of LC. Using a deep phenotypic approach, increased frequencies of activated SARS-CoV-2 specific CD4 and CD8 T cells, and higher levels of SARS-CoV-2 antibodies have been found in individuals with LC compared to fully recovered individuals, which may contribute to immune dysregulation, inflammation, and clinical symptoms associated with LC¹⁸. High spontaneous IFN-γ secretion has been further shown to be associated with the persistence of symptoms, i.e., fatigue, suggesting that IFN-γ secretion may play a causal role in LC recovery processes¹⁹.

Short or disturbed sleep further affect numerous inflammatory signals⁶, overall leading to a more pro-inflammatory state. This manifests in the activation of the nuclear factor-κB (NF-κB) pathway, resulting in elevated levels of interleukin (IL)−6, tumor necrosis factor (TNF), among other cytokines, as well as activation of the cyclooxygenase (COX) pathway, leading to increased COX-2 enzyme expression and subsequent production of various prostaglandins^20–22. Such changes in the inflammatory milieu affect numerous physiological processes, including functioning of B and T cells²³, and are a pathway through which deficient sleep increases the risk for the many diseases involving immunopathology²⁴. Dysregulation of various inflammatory mediators have been repeatedly observed in LC, in particular dysregulations of IL-6, TNF, and IFN-γ^18,25–27. With respect to mediators that control or resolve inflammation, glucocorticoids (GC), such as the corticosteroid cortisol or the synthetic GC dexamethasone, display potent immunosuppressive properties²⁸. The extent of their anti-inflammatory effect is regulated by GC blood concentrations, but also the sensitivity of tissues and immune cells towards GCs²⁹. Altered GC sensitivity of immune cells has been observed following stressful challenges³⁰, including acute sleep restriction or disruption^31,32. In chronic infection-associated illnesses, such as myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS), studies have reported increased GC sensitivity of immune cells^33,34. Changes in GC sensitivity of immune cells could be one mechanism by which pre-existing sleep disturbance contributes to inflammatory dysregulation in LC.

Beside GC concentrations and sensitivity as modulators of inflammation, certain lipid-derived so-called specialized pro-resolving mediators (SPMs) are actively involved in the resolution of inflammation³⁵. These mediators promote the return to inflammatory homeostasis by cessation of neutrophil infiltration, enhanced efferocytosis, and counter-regulation of pro-inflammatory chemokines and cytokines, among many other actions³⁵. Failure to mount an adequate resolution response promotes chronic inflammatory dysregulation³⁶. Exposing healthy individuals to experimental sleep disturbances leads to a profound decrease in D-series resolvins, which are SPMs deriving from the omega-3 fatty acid docosahexaenoic acid (DHA)³⁷. Targeting inflammatory resolution physiology has been suggested as interventional strategy in the treatment of LC³⁸. However, knowledge on the type and degree of SPM dysregulation in LC is limited.

These immune domains, including IFN-γ expression by T cells, cytokine expression by monocytes, and activation of SPMs, cross-modulate each other, such that changes in one domain can propagate alterations in other domains. IFN-γ expression by T cells, for example, can prime monocytes, leading to elevated expression of inflammatory mediators upon stimulation³⁹. Inflammatory upregulation, if persistent, can delay the initiation of the resolution phase mediated by SPMs, which, in turn, can modulate IFN-γ expression by T cells⁴⁰. Given that sleep disturbance has the potential to affect each of these domains, it may therefore disrupt the balance between inflammation and its resolution through various pathways in Long COVID.

We investigated whether the presence of sleep disturbance prior to infection is associated with the development of LC via (a) compromised IFN-γ secretion of T cells stimulated with specific SARS-CoV-2 variants, (b) increased expression of inflammatory mediators by monocytes and the ability of GC to suppress inflammation, and (c) disrupted levels of inflammatory pro-resolving mediators (SPMs).

Results

Study design and participants

The study leveraged participant enrollment and biospecimen/data collection from the RECOVER Adult protocol. For the current sub-study, participants were selected from the acutely infected cohort, i.e., participants who were enrolled within 30 days of acute SARS-CoV-2 infection, and from the non-infected cohort. A symptom survey was collected at times of enrollment and every 3 months following enrollment (Fig. 1). Biospecimens (PBMCs and plasma) for this sub-study were analyzed for a single timepoint following enrollment, either at the 6- or 12-month collection time point. The ‘Long COVID Research Index’ (LCRI) as described by Geng and colleagues⁴¹ was used to group participants by LC status (Likely LC (LCRI ≥ 11), Possible LC (0 < LCRI < 11), and No LC (LCRI = 0)) at the biospecimen collection time. Seventy-four participants, enrolled as acutely infected or uninfected participants, were selected for this study. Table 1 shows participant characteristics for groups of Likely LC, Possible LC, and No LC. Major comorbidities, SARS-CoV-2 vaccination status, and a demographic breakdown by pre-existing sleep disturbance within LC groups (Likely and Possible LC) can be found in Supplementary Tables S1, S2, and S3, respectively.

Fig. 1.

Study Design. Data and samples were collected from the acute cohort of the RECOVER Adult clinical protocol (OTA-21-015B). The presence of pre-existing sleep disturbance was assessed retrospectively for the year before enrollment into the acute cohort. Self-reported symptoms were assessed throughout the study at time of enrollment and every 3 months thereafter. Symptom surveys were used to categorize participants into groups of Likely LC, Possible LC, and No LC based on the ‘Long COVID Research Index’ (LCRI)⁴¹. Samples (PBMC/plasma) collected either at 6 months (T6) or 12 months (T12) after enrollment were analyzed for the present study.

Table 1.

Participant characteristics overall and by LC status.

Characteristic	Likely LC n = 19	Possible LC n = 24	No LC n = 31	Overall N = 74
Age, mean (SD)1	44 (12)	44 (13.1)	47 (12.9)	46 (12.6)
Sex
Female, n (%)	14 (74%)	20 (83%)	23 (74%)	57 (77%)
Male, n (%)	5 (26%)	4 (17%)	8 (26%)	17 (23%)
Race/Ethnicity
White Non-Hispanic, n (%)	17 (89%)	17 (71%)	23 (74%)	57 (77%)
Black Non-Hispanic, n (%)	1 (5%)	0 (0%)	0 (0%)	1 (1%)
Asian Non-Hispanic, n (%)	0 (0%)	2 (8%)	3 (10%)	5 (7%)
Hispanic, n (%)	0 (0%)	2 (8%)	4 (13%)	6 (8%)
Multiracial/Other/Not Provided, n (%)	1 (5%)	3 (12%)	1 (3%)	5 (7%)
Sleep Disturbance (Pre-Infection)
Yes, n (%)	9 (47%)	7 (29%)	7 (23%)	23 (31%)
No, n (%)	8 (42%)	16 (67%)	24 (77%)	48 (65%)
Not Provided, n (%)	2 (11%)	1 (4%)	0 (0%)	3 (4%)
Continued Sleep Disturbance (At Sample)2
Yes, n (%)	5 (26%)	5 (21%)	0 (0%)	10 (14%)
Probable SARS-CoV-2 Infection Status
Infected, n (%)	19 (100%)	20 (83%)	16 (52%)	55 (74%)
Uninfected, n (%)	0 (0%)	4 (17%)	15 (48%)	19 (26%)
Time (Days) At Sample Collection Since Infection, mean (SD)	251 (90)	246 (105)	252 (118)	250 (106)

¹Age information provided for all participants of the study; ²Portion of participants, who have continued sleep disturbance at sample, among those participants reporting sleep disturbance pre-infection.

No difference in T cell responses to SARS-CoV-2 variants by LC status (Likely, Possible, No LC)

To investigate SARS-CoV-2 variant specific intracellular expression of IFN-γ in CD4 and CD8 T cells, samples of cryopreserved PBMCs were thawed and stimulated with specific overlapping Omicron spike peptide variants BA.5 or XBB or left unstimulated. Figure 2 and Table S4 show the results for IFN-γ expression of CD4 and CD8 T cells, stimulated with BA.5 or XBB, or left unstimulated by LC status. IFN-γ expression of CD4 and CD8 T cells stimulated with BA.5 or XBB variants or left unstimulated did not differ between groups of Likely LC, Possible LC, and No LC.

Fig. 2.

No difference in T cell responses to SARS-CoV-2 variants by LC status. Percentage of IFN-γ positive CD4 and CD8 T cells, stimulated with BA.5 or XBB variants, or left unstimulated in groups of Likely LC, Possible LC, and No LC. Assay results on log-transformed scale presented as boxplots with original values. Analyses included N = 74 participants (Likely LC n = 19, Possible LC n = 24, No LC n = 31).

IFN-γ positive BA.5- and XBB-stimulated CD4 T cells nominally lower in Likely LC with pre-existing sleep disturbance

The primary outcome for this analysis was the effect of sleep disturbance prior to SARS-CoV-2 infection on the difference in analyte levels between those participants who developed LC at 90 days or more after infection and those who did not develop LC. The effect of pre-existing sleep disturbance was assessed using linear models with factors for LC group (Likely LC, Possible LC, No LC), presence of sleep disturbance prior to SARS-CoV-2 infection (Yes, No), and an interaction effect between LC group and sleep disturbance. Unadjusted estimates were generated from these models for group comparisons of interest. The effect of pre-existing sleep disturbance by LC status on IFN-γ expression of CD4 and CD8 T cells stimulated with peptide variants BA.5 or XBB or left unstimulated is presented in Fig. 3 and Table S5. In the Likely LC group, IFN-γ positive BA.5- and XBB-stimulated CD4 T cells were nominally lower (p < 0.10) in participants with pre-existing sleep disturbance compared to those without sleep disturbance. This effect was not observed in the Possible and No LC (p > 0.10) groups. When comparing the sleep disturbance effect between groups, the effect in the Likely LC group was not different from the effect in the No LC group or the combined group of Possible and No LC (p > 0.10). There was no effect of pre-existing sleep disturbance on other T cell responses in the Likely LC group.

Fig. 3.

IFN-γ positive BA.5- and XBB-stimulated CD4 T cells nominally lower in Likely LC with pre-existing sleep disturbance. Estimated effect of pre-existing sleep disturbance on the percentage of IFN-γ positive CD4 and CD8 T cells, stimulated with BA.5 or XBB variants, or left unstimulated in groups of Likely LC, Possible LC, and No LC. Estimated contrasts on log-transformed scale presented as estimated mean with 95% CI. Analyses included N = 71 participants (Likely LC n = 17, Possible LC n = 23, No LC n = 31). ⁽*⁾ p < 0.10 for comparison of sleep disturbance effect within LC groups. A positive value indicates that the analyte is higher in those with sleep disturbance; a negative value indicates that the analyte is lower in those with sleep disturbance.

No difference in total IL-6, TNF, COX-2 expression in monocytes and GC sensitivity of monocytes by LC status

To investigate IL-6, TNF, COX-2 expression in monocytes, samples of cryopreserved PBMCs were thawed and stimulated with lipopolysaccharide (LPS) or left unstimulated. To assess GC sensitivity of these monocytes, dexamethasone (DEX) was added to an additional LPS-stimulated well. Figure 4 and Table S6 show the results for IL-6, TNF, COX-2 expression in monocytes and their GC sensitivity by LC status. There were no group differences in monocytic expression of total IL-6, TNF, and COX-2. Exploratory analyses of the different inflammatory marker expressing subsets of monocytes revealed that COX-2/TNF double positive and TNF single positive monocytes were higher in Likely LC compared to Possible and No LC combined (p = 0.035 and p = 0.045, respectively). In contrast, COX-2/IL-6 double positive monocytes were nominally lower in Likely LC compared to Possible/No LC (p = 0.056). There was no difference in GC sensitivity by LC status (all p > 0.10), except for a nominally higher GC sensitivity in COX-2/TNF double positive monocytes in Likely LC compared to No LC (p = 0.093).

Fig. 4.

No difference in total IL-6, TNF, COX-2 expression in monocytes and in the GC sensitivity of monocytes by LC status. (A) Percentage of total IL-6, TNF, and COX-2 positive LPS-stimulated monocytes in groups of Likely LC, Possible LC, and No LC. Assay results on log-transformed scale presented as boxplots with original values. Analyses included N = 64 participants (Likely LC n = 17, Possible LC n = 20, No LC n = 27). (B) GC sensitivity of monocytes expressed as delta value (LPS-stim. minus LPS + DEX-stim.) of the percentage of total IL-6, TNF, and COX-2 positive monocytes in groups of Likely LC, Possible LC, and No LC. Assay results presented as boxplots with original values. Analyses included N = 64 participants (Likely LC n = 17, Possible LC n = 20, No LC n = 27). A higher value indicates higher GC sensitivity.

Total TNF expression in monocytes nominally (p < 0.10) higher in Likely LC with pre-existing sleep disturbance

The effect of pre-existing sleep disturbance by LC status on IL-6, TNF, and COX-2 expression in LPS-stimulated monocytes is presented in Fig. 5A and Table S7. There was no significant pre-existing sleep disturbance effect observed for IL-6 and COX-2 within the LC groups or between the LC groups (all p > 0.10). Total TNF expression in monocytes was nominally (p = 0.091) higher in Likely LC with sleep disturbance compared to Likely LC without sleep disturbance, while this effect was not observed for groups of Possible and No LC. Exploratory analyses revealed that COX-2/TNF double positive monocytes were higher in Likely LC with sleep disturbance compared to Likely LC without sleep disturbance (p = 0.013). This effect of sleep disturbance was larger in Likely LC compared to the combined group Possible/No LC (p = 0.010). In contrast, COX-2/IL-6 double positive and IL-6 single positive monocytes were lower in Likely LC with sleep disturbance compared to Likely LC without sleep disturbance (p = 0.045 and p = 0.018, respectively). The effect of sleep disturbance on IL-6 single positive monocytes in Likely LC was larger compared to the combined group of Possible/No LC (p = 0.046).

Fig. 5.

Total TNF expression nominally (p < 0.10) higher and GC sensitivity of monocytes significantly lower in Likely LC with pre-existing sleep disturbance. (A) Estimated effect of pre-existing sleep disturbance on the percentage of total IL-6, TNF, and COX-2 positive LPS-stimulated monocytes in groups of Likely LC, Possible LC, and No LC. Estimated contrasts on log-transformed scale presented as estimated mean with 95% CI. Analyses included N = 61 participants (Likely LC n = 15, Possible LC n = 19, No LC n = 27). ⁽*⁾ p < 0.10 for comparison of sleep disturbance effect within the LC groups. A positive value indicates that the analyte is higher in those with sleep disturbance. (B) Estimated effect of pre-existing sleep disturbance on GC sensitivity expressed as delta value (LPS-stim. minus LPS + DEX-stim.) of the percentage of total IL-6, TNF, and COX-2 positive monocytes in groups of Likely LC, Possible LC, and No LC. Estimated contrasts presented as estimated mean with 95% CI. Analyses included N = 61 participants (Likely LC n = 15, Possible LC n = 19, No LC n = 27). ⁽*⁾ p < 0.10 for comparison of sleep disturbance effect within the LC groups. A positive value indicates that the analyte is higher in those with sleep disturbance. ^#p < 0.05 and ^(#)p < 0.10 for comparison of sleep disturbance effect between groups of Likely LC and combined Possible/No LC.

Lower GC sensitivity of monocytes in Likely LC with pre-existing sleep disturbance

With respect to the GC sensitivity of monocytes (presented in Fig. 5B and Table S7), the DEX-induced suppression of IL-6 expression, i.e., the GC sensitivity, was nominally (p = 0.056) lower in Likely LC participants with sleep disturbance compared to those without sleep disturbance. When comparing the effect of sleep disturbance between groups, the effect of GC on total IL-6, TNF, and COX-2 suppression differed between Likely LC and the combined group of Possible/No LC (p = 0.010, 0.034, 0.099, respectively). This indicates lower GC sensitivity of monocytes in Likely LC across all three investigated inflammatory markers. Exploratory analysis showed that GC sensitivity was nominally (p = 0.056) lower for IL-6/TNF/COX-2 triple positive monocytes in Likely LC participants with sleep disturbance compared to those without. This effect was larger in Likely LC compared to the combined group of Possible/No LC (p = 0.008).

No difference in SPM precursors by LC status

Targeted lipidomics analyses were performed using HPLC–MS/MS to determine plasma levels of the precursors of main SPMs and related bioactive lipid mediators and metabolites. Analytes for the present study contained arachidonic acid (AA)-derived fatty acids, including the lipoxin precursor 15-hydroxyeicosatetraenoic acid (15-HETE), eicosapentaenoic acid (EPA)-derived fatty acids, including the E-series resolvin precursor 18-hydroxyeicosapentaenoic acid (18-HEPE), docosapentaenoic acid (DPA)-derived fatty acids, and docosahexaenoic acid (DHA)-derived fatty acids, including the D-series resolvin precursor 17-hydroxydocosahexaenoic acid (17-HDHA). Figure 6 and Table S8 show the results for SPM precursors and lipid mediators by LC status. None of the main SPM precursors, i.e., 15-HETE, 18-HEPE, and 17-HDHA, differed significantly between groups by LC status (all p > 0.10). In the other investigated lipid mediators and metabolites, nominally lower levels were observed for the AA-derived 12-HETE in Likely LC compared to Possible/No LC (p = 0.084); all other comparisons had a p value above 0.1.

Fig. 6.

No difference in SPM precursors and bioactive lipid mediators by LC status. SPM precursors of lipoxins (15-HETE), E-series resolvins (18-HEPE), and D-series resolvins (17-HDHA) for groups of Likely LC, Possible LC, and No LC. Assay results on log-transformed scale presented as boxplots with original values. Analyses included N = 74 participants (Likely LC n = 19, Possible LC n = 24, No LC n = 31).

No effect of pre-existing sleep disturbance on SPM precursors in Likely LC

The effect of pre-existing sleep disturbance on the main SPM precursors and other bioactive lipid mediators and metabolites by LC status is presented in Fig. 7 and Table S9. There was no significant sleep disturbance effect in SPM precursor levels in the Likely LC group (p > 0.10). Nominally higher levels of the lipoxin precursor 15-HETE were observed in the Possible LC group with sleep disturbance compared to Possible LC without sleep disturbance (p = 0.095). Exploratory analyses revealed a significant sleep disturbance effect on pro-inflammatory PGF_2α in the Likely LC group, indicating that Likely LC participants with sleep disturbance had higher PGF_2α levels than those without sleep disturbance (p = 0.036). This effect of sleep disturbance was nominally larger than in the No LC group (p = 0.084). Furthermore, there was a sleep disturbance effect observed in the pro-resolving mediator RvD2 in the Likely LC group, suggesting that Likely LC participants with sleep disturbance have nominally higher RvD2 levels than those without sleep disturbance (p = 0.062). This sleep disturbance effect in RvD2 in the Likely LC group was greater than the effect in the No LC group (p = 0.032) and combined group of Possible/No LC (p = 0.088). Sleep disturbance effects were further observed for the lipids 5-HETE and 20-carboxy-LTB₄ between groups of Likely LC and No LC (p = 0.082 and p = 0.098, respectively).

Fig. 7.

No difference in SPM precursors between Likely LC with and without pre-existing sleep disturbance. Estimated effect of pre-existing sleep disturbance on SPM precursors of lipoxins (15-HETE), E-series resolvins (18-HEPE), and D-series resolvins (17-HDHA) in groups of Likely LC, Possible LC, and No LC. Estimated contrasts on log-transformed scale presented as estimated mean with 95% CI. Analyses included N = 71 participants (Likely LC n = 17, Possible LC n = 23, No LC n = 31). ⁽*⁾ p < 0.10 for comparison of sleep disturbance effect within LC groups. A positive value indicates that the analyte is higher in those with sleep disturbance.

Discussion

The current investigation examined the effect of pre-existing sleep disturbance on specific T cell responses to SARS-CoV-2 Omicron spike peptide variants, monocytic expression of inflammatory markers, their suppression by glucocorticoids (GCs), as well as specialized pro-resolving mediators (SPMs). Overall, individuals with Likely LC did not differ in these three investigated immunologic domains from individuals categorized as Possible LC or No LC. However, the presence of sleep disturbance prior to SARS-CoV-2 infection affected some of the examined immune measures. In particular, the presence of sleep disturbance in Likely LC compromised the capacity of GCs to inhibit pro-inflammatory IL-6, TNF, and COX-2 expression in monocytes to a greater extent than in groups of Possible and/or No LC. T cell responses and levels of SPMs, on the other hand, were overall less affected by the presence of sleep disturbance in Likely LC compared to Possible and/or No LC. These exploratory findings suggest that pre-existing sleep disturbance may be associated with compromised glucocorticoid mediated inflammatory suppression in a subset of individuals classified as Likely LC. This might indicate that targeting GC signaling pathways aimed to promote anti-inflammatory regulation could be beneficial for this particular LC subtype, i.e., LC with pre-existing sleep disturbance.

With regards to T cell responses in LC and the modulating role of pre-existing sleep disturbance, we found that IFN-γ positive BA.5- and XBB-stimulated CD4 T cells were nominally (p < 0.10) lower in Likely LC participants who reported pre-existing sleep disturbance compared to those without sleep disturbance. Other T cell indices in the Likely LC group were not affected by sleep disturbance, including BA.5- or XBB-stimulated or spontaneous expression of IFN-γ by CD8 T cells. While T cells and their responses to SARS-CoV-2 variants have not been studied in relation to sleep disturbance in LC, studies in healthy individuals have shown that lack of sleep compromised CD4 T cell specific antigen responses to vaccination¹⁵ and prevented the migration of T cells towards lymph nodes, a process happening during undisturbed good quality sleep⁴². Independent of the sleep disturbance status, the current study did not find that stimulated or spontaneous expression of IFN-γ by T cells differed between participants with Likely LC compared to those with Possible or No LC. In comparison, a previous study reported that the spontaneous expression of IFN-γ by PBMCs, in particular CD8 T cells, was higher in participants with LC compared to never-infected participants¹⁹. The lack of an effect of LC on T cell responses and modulation by pre-existing sleep disturbance likely relates to the large heterogeneity of LC⁴³ that makes it challenging to identify changes within a relatively small sample size.

While monocytic expression of total TNF was nominally (p < 0.10) higher in Likely LC with pre-existing sleep disturbance compared to those without, no difference was found for total IL-6 and total COX-2 expression. Exploratory analyses on monocytes positive for multiple inflammatory markers, which have been suggested to offer better protection against viruses than single positive cells⁴⁴, revealed higher COX-2/TNF double positive monocytes in Likely LC with sleep disturbance, whereas COX-2/IL-6 double positive and IL-6 single positive monocytes were lower. Persistent pro-inflammatory cytokine release has been suggested to contribute to Long COVID pathophysiology^43,45. Current observations point to a modulating role of pre-existing sleep disturbance in LC on the pro-inflammatory cytokine response. Given the exploratory nature, future investigations are needed for improved understanding of the role of pre-existing sleep disturbance on inflammatory responses in LC.

With respect to the suppression of inflammatory signals by GCs, current findings suggest that pre-existing sleep disturbance is associated with a weakened inflammatory control by GC in LC. Specifically, the extent to which pre-existing sleep disturbance compromised the capacity of GC to suppress IL-6 and TNF was greater in Likely LC compared to Possible and/or No LC, and a similar trend was observed for COX-2. Similarly, when looking at monocytes that concurrently expressed IL-6, TNF, and COX-2 (i.e., triple positive monocytes), pre-existing sleep disturbance in Likely LC compromised inflammatory suppression by GC compared to those without sleep disturbance, and this effect was greater compared to groups of No LC and the combined group of Possible and No LC. Beside concentrations of GCs in blood, GC sensitivity of tissues and cells determines the magnitude of the immunosuppressant effect of GC²⁹. The reduced GC sensitivity, or increased GC resistance, may indicate a compromised inflammatory control by GCs in a subgroup of LC, i.e., those who experience sleep disturbance. Sleep disturbance has been observed to alter GC sensitivity in the experimental setting. Two studies reported increased GC sensitivity following experimentally induced acute sleep restriction or disruption of about 2 weeks in healthy individuals^31,32, while GC sensitivity was unchanged in chronic insomnia disorder⁴⁶. Comparing these findings is to some extent difficult, because the current study implementation did not reveal any information on chronicity, type, or severity of reported sleep disturbance, nor on treatment and treatment adherence. While GC sensitivity has not been assessed in LC, reduced GC sensitivity has been observed in several chronic inflammatory and immune diseases⁴⁷, for which sleep disturbances are common⁶. In LC, several studies have reported on circulating levels of GC, though not in relation to sleep. Reduced cortisol levels have been found among individuals with LC compared to convalescent controls⁴⁸. In this study, cortisol was also the best predictor in distinguishing LC from control participants⁴⁸. Another study assessing cortisol at a standard morning time did not report changes in LC participants compared to those recovered from the infection, nor did the cortisol rise following ACTH injection differ between groups⁴⁹. In another study enrolling participants with fatigue-dominant LC, lower cortisol levels were found to be associated with higher fatigue levels⁵⁰. Findings of the current study add to previous knowledge on the role of GCs in LC, and suggest that inflammatory control is compromised in a subset of participants with LC, i.e., those who experienced sleep disturbance prior to infection. The mechanisms underlying this observation are not known. Several factors can modulate GC sensitivity⁵¹. Chronic exposure to pro-inflammatory mediators as well as chronically elevated cortisol levels have been shown to relate to impaired functioning of the GC receptors^29,51,52. Disturbed and/or short sleep are well established to promote a pro-inflammatory state⁶ and to dysregulate the hypothalamus–pituitary–adrenal axis, frequently manifested in elevated cortisol levels⁵³. These dysregulations may contribute to altered GC sensitivity, and improved understanding of these pathways could help inform therapeutic strategies aimed at normalizing GC sensitivity.

Inflammatory control also involves various pro-resolving lipid mediators (SPMs) that are known to actively resolve inflammation³⁵. Pre-existing sleep disturbance did not affect the primary SPM outcomes of this study, i.e., the precursors of the D- and E-series resolvins and of the lipoxins. Exploration of other lipid mediators revealed higher pro-inflammatory PGF_2α and nominally higher resolvin D2 levels in Likely LC with pre-existing sleep disturbance. This was somewhat surprising, given that two previous studies suggested that sleep disturbance was associated with compromised inflammatory resolution physiology, as manifested in the reduction of certain D- and E-series resolvins and their precursors^37,54. The current study findings do not support a substantial role of pre-existing sleep disturbance on SPMs, which may relate to the limited information on type and severity of sleep disturbance and whether or not the disorder was successfully treated. Independent of a potential role of pre-existing sleep disturbance on SPMs in LC, groups of Likely, Possible, and No LC did not differ in SPMs, except for nominally lower levels of the omega-6 fatty acid-derived 12-HETE in Likely LC. This was again unexpected, given that inflammatory resolution physiology has been suggested early on to play a central role in the persistence and non-resolution LC³⁸. Two previous studies reported higher levels of the D-series resolvin precursor 17-HDHA in LC compared to fully recovered or never-infected healthy individuals^54,55, as well as elevated levels of the resolvins RvD4, 17R-RvD5_{n-3 DPA}, and RvE2⁵⁴. In addition, pro-inflammatory lipid mediators, including PGE₂ and LTB₄ have been found elevated in LC compared to individuals who fully recovered from the infection^54–57. Various methodological factors may have contributed to the lack of differences in pro-inflammatory and pro-resolving lipid mediators between groups in the current investigation, including the data-driven categorization of individuals into groups of Likely, Possible, and No LC⁴¹, and the relatively small sample size to investigate a highly heterogeneous syndrome.

The present study has certain limitations. One limitation is the type of sleep data available for this investigation, which was a single question on the presence or absence of sleep disturbance, defined by a diverse range of symptoms (i.e., stopping breathing during sleep or sleep problems, such as snoring, trouble falling asleep, nighttime awakenings, or trouble staying awake during the day, 3 or more times a week). Specific information on the type of sleep disturbance (insomnia, hypersomnia, and breathing-related disorders), the severity of the sleep disturbance, and/or objective wearable or EEG-based sleep data would likely have provided more precise information on the role of sleep disturbance on various immune indices in LC. We here assessed the frequency of cytokine-producing cells, which does not necessarily reflect the actual amount of cytokines produced. The assessment of cytokine concentrations in the blood circulation in future studies will provide additional valuable information on the potential role of sleep in modulating inflammatory status in LC. Related to this is that our study investigated only a limited set of immune components; there is a need for future studies to include additional cellular and humoral measures to more comprehensively capture the multifaceted immune response to SARS-CoV-2, such as antibody titers, neutralizing capacity, and/or antibody affinity maturation. In addition, SARS-CoV-2 specific T cell responses are highly variable between individuals⁵⁸, which will require replication in larger Long COVID samples. Another limitation that applies to LC research in general is the enormous heterogeneity of the syndrome, suggesting multiple subtypes of LC that are potentially driven by different biological mechanisms^43,59. The sample size of the current study did not allow for a subtyping approach, which should be considered in future research to gain deeper mechanistic insights. The small sample size also limited the adjustment of models for multiple potential confounders, which would have compromised the stability of the model estimates. Future investigation will require to control for potential confounders, including the presence of comorbidities and medications. Further, the sample size did not allow for utilizing more advanced analysis approaches, such as causal discovery models or multivariate logistic regression models that provide insight into causal links and how factors influence a set of outcomes. Another potential limitation is that participants were categorized into groups of Likely LC, Possible LC, and No LC based on the ‘Long COVID Research Index’ (LCRI)⁴¹. Symptoms contributing to the LCRI best discriminate between infection with SARS-CoV-2 and no infection, and include post-exertional malaise (PEM), fatigue, brain fog, dizziness, palpitations, change in smell or taste, thirst, chronic cough, chest pain, shortness of breath, and sleep apnea. This data-driven classification may differ from physician-based clinical diagnosis of LC and may affect findings. Future studies may compare different LC classification approaches.

To conclude, findings suggest that the presence of pre-existing sleep disturbance, which has been identified as a risk factor for the development of LC^9,10, may compromise certain immune indices in LC, in particular the ability of glucocorticoids (GC) to suppress inflammation. This suggests reduced inflammatory control by GC in a subtype of LC, i.e., those presenting with pre-existing sleep disturbance. Research on the type of sleep disturbance (e.g., insomnia, hypersomnia, breathing-related problems) reducing GC sensitivity in large LC cohorts may help to identify subtype-specific intervention targets.

Methods

Study design

The study leveraged participant enrollment and biospecimen/data collection from the RECOVER Adult clinical protocol (OTA-21-015B).The RECOVER Adult clinical protocol was approved by the institutional review board (IRB) at New York University (IRB# S21-01,226). The protocol of the current sub-study was determined as exempt by the local IRB at Beth Israel Deaconess Medical Center (IRB# 2023P000828). Informed consent was obtained from all participants prior to enrollment into the RECOVER Adult clinical protocol. All methods were performed in accordance with relevant guidelines and regulations. For the current sub-study, participants were selected from the acutely infected cohort of the RECOVER Adult clinical protocol, i.e., participants who were enrolled within 30 days of acute SARS-CoV-2 infection, and from the non-infected cohort. Enrollment for the current sample took place between January 2022 and the end of February 2023, during the Omicron variant time period. A symptom survey was collected at times of enrollment and every 3 months following enrollment (Fig. 1). Biospecimens (PBMCs and plasma) for this sub-study were analyzed for a single timepoint following enrollment, either at the 6- or 12-month collection time point. LC status was determined at each biospecimen/data collection time point (see below).

Participants

Seventy-four participants enrolled as acutely infected or uninfected participants were selected for this study. Inclusion criteria were enrollment into the acutely-infected cohort (within 30 days of SARS-CoV-2 infection) or non-infected cohort, females and males ≥ 18 years of age, and PBMC processing at site of enrollment (excluding delayed processing following overnight shipment of whole blood to central processing site). Exclusion criteria included having a history of primary immune deficiency diseases prior to SARS-CoV-2 infection, intake of immune modulators or suppressants, or being currently pregnant or lactating.

Participants were categorized into groups of Likely LC, Possible LC, and No LC based on the ‘Long COVID Research Index’ (LCRI)⁴¹, as outlined in the following. Likely LC: LCRI score of 11 or greater. Possible LC: LCRI score of greater than 0 and less than 11. No LC: LCRI score of 0. LC group was defined independent of prior SARS-CoV-2 infection status, and the Possible LC and No LC groups include participants who were enrolled as uninfected (see Table 1). While individuals enrolled as uninfected were confirmed to be seronegative for nucleocapsid antibodies at baseline, they were not re-tested at follow-up time points of biospecimen collection. The approach of categorizing individuals enrolled as uninfected into LC groups was chosen because data suggest that the prevalence of SARS-CoV-2 infection in the population is substantially higher than previously estimated, as indicated by the detection of nucleocapsid-specific immune responses in individuals who were never diagnosed or who tested seronegative for nucleocapsid antibodies using conventional assays⁶⁰.

All participants were chosen to have no reported re-infections within 3 months prior to biospecimen collection. A single sample collection timepoint was analyzed for each participant and the self-reported symptoms survey at that timepoint was used to define the LC group.

Assessment of sleep disturbance

At baseline, participants were asked to report on the presence of sleep disturbance prior to infection (assessed retrospectively for the year before infection). The question on sleep disturbance was part of the LC symptom form that was sent out to participants electronically via the REDCap application, and asked for ‘Stopping breathing during sleep or sleep problems (such as snoring, trouble falling asleep, nighttime awakenings, or trouble staying awake during the day) 3 or more times a week’. Participants could select the following response options: ‘No, I have NOT had this symptom; Yes, I DID have it in the YEAR BEFORE infection; Yes, I DID have it AROUND the time of infection; Yes, I have it NOW; I don’t know or prefer not to answer’. For this study, the presence or absence of sleep disturbance prior to SARS-CoV-2 infection (‘Yes, I DID have it in the YEAR BEFORE infection’) were analyzed.

Quantification of analytes in peripheral blood mononuclear cells (PBMCs) and plasma

All laboratory methods and procedures used for the quantification of analytes in peripheral blood mononuclear cells (PBMCs) and plasma were carried out in accordance with established protocols as described in the following.

Cryopreservation of peripheral blood mononuclear cells (PBMCs)

Whole blood was collected via direct venipuncture into 8 ml cell preparation tubes (CPT) with sodium citrate (BD Vacutainer, Franklin Lakes, NJ, USA) and directly processed on site for peripheral blood mononuclear cell (PBMC) isolation and cryopreservation according to the RECOVER standard protocol using aseptic techniques under a biosafety cabinet. Briefly, after centrifugation of CPT tubes and removal of half the plasma layer, isolated PBMCs were diluted 1:1 with phosphate buffered saline (PBS), washed once in 10 ml PBS, and counted. Cell viability (Cellometer K2 or other automated cell counter) for the present study was 96.9 ± 0.3% (mean ± SEM). PBMCs were resuspended in freezing medium (10% DMSO in complete medium, i.e., RPMI 1640 medium without L-glutamine, 10% fetal bovine serum (FBS), 1% penicillin–streptomycin-glutamine 100X) to a minimum concentration of 5 million cells per ml and aliquoted into cryovials (Nunc internally-threaded barcoded universal tubes, Thermo Scientific, Waltham, MA, USA). Subsequently, samples were placed in a slow rate freezing container (Corning CoolCell, Nalgene Mr. Frosty, or slow rate freezer) and stored at −80 °C up to one week until shipment to the central storage facility on dry ice. Aliquots of cryopreserved PBMCs were stored over liquid nitrogen and shipped on dry ice to the Beth Israel Deaconess Medical Center for immune functional testing.

Intracellular expression of IFN-γ in T cells

Assessment of intracellular expression of interferon (IFN)-γ in T cells was performed using an established protocol as previously described⁶¹. For the stimulation of T cells, 10⁶ PBMCs/well were re-suspended in 80 µL of R10 medium (RPMI 1640 medium with 10% FBS) supplemented with anti-CD28, CD49d monoclonal antibody (mAb, 1 µg/ml). Each sample was assessed with mock (100 µL of R10; background control) or specific overlapping Omicron spike peptide variants (BA.5 or XBB, 2 µg/ml) and incubated at 37 °C for 1 h. The peptide pools used in the current study were designed to measure both CD4 and CD8 T cell responses⁶². T cell responses to spike peptide variants are highly cross reactive⁶², suggesting high generalizability to variants other than those used in this study. After incubation, 0.25 µL of GolgiStop and 0.25 µL of GolgiPlug (BD Biosciences) in 40 µL of R10 was added to each well and incubated at 37 °C for 8 h and then held at 4 °C overnight. The next day, the cells were washed twice with 1X PBS, stained with LIVE/DEAD Fixable Aqua dye (Invitrogen) for 10 min and then stained with predetermined titers of mAbs against CD4 (clone L200, BV711, BD) and CD8 (clone SK1, BUV805, BD), among other markers, for 30 min. Cells were then washed twice with 2% FBS/DPBS buffer and incubated for 15 min with 200 µL of BD CytoFix/CytoPerm Fixation/Permeabilization solution. Cells were washed twice with 1X Perm Wash buffer (BD Perm/Wash Buffer 10X in the CytoFix/CytoPerm Fixation/Permeabilization kit diluted with MilliQ water) and stained intracellularly with mAbs against IFN-γ (clone B27, BUV395, BD) and CD3 (clone SP34.2, Alexa 700, BD), among other markers, for 30 min. Cells were washed twice with 1X Perm Wash buffer and fixed with 200 µL of freshly prepared 1.5% formaldehyde (Polysciences, Warrington, PA). Fixed cells were transferred to 96-well round bottom plates and analyzed using a BD FACSymphony A5 Cell Analyzer (BD Biosciences). Gating analysis was performed using FlowJo 10.8.1 software (BD Biosciences). The analysis was performed in the Center of Virology and Vaccine Research at BIDMC, Boston. The gating strategy for T cell analysis can be found in Supplementary Figure S4. The following T cell-related variables were determined for the present study: IFN-γ positive BA.5-stimulated CD4 T cells [%], IFN-γ positive XBB-stimulated CD4 T cells [%], IFN-γ positive unstimulated CD4 T cells [%], IFN-γ positive BA.5-stimulated CD8 T cells [%], IFN-γ positive XBB-stimulated CD8 T cells [%]. IFN-γ positive unstimulated CD8 T cells [%].

Intracellular expression of IL-6, TNF, and COX-2 in monocytes and GC sensitivity of monocytes

Assessment of intracellular expression of interleukin (IL)−6, tumor necrosis factor (TNF), and cyclooxygenase (COX)−2 in monocytes and glucocorticoid (GC) sensitivity was performed using an established protocol developed in our laboratory³² and adapted for the use of PBMCs. Frozen samples were thawed in R10 medium, i.e., RPMI 1640 medium with GlutaMAX and 25 mM HEPES (Gibco Life Technologies, Grand Island, NY, USA) supplemented with 10% FBS (Gibco), washed, resuspended in R10 and rested for 2 h at 37 °C in a humidified 5% CO₂ atmosphere. Cells were counted and viability, i.e., 84.7 ± 1.3% (mean ± SEM), was determined using Guava ViaCount staining according to manufacturer’s instructions (Luminex Guava easyCyte, Austin, TX). After preincubation, 1 million viable cells per well were plated into a 96-well PCR plate (Thermo Scientific). Then, cells were stimulated with lipopolysaccharide (LPS) from Escherichia coli O127:B8 (final conc. 50 pg/ml in R10, L3137, Sigma-Aldrich, St. Louis, MO) or left unstimulated (R10 only). To assess GC sensitivity of monocytes, dexamethasone (final conc. 50 nM in R10, D8893, Sigma-Aldrich) was added to an additional LPS-stimulated well. Cells were then incubated for 20 min at 37 °C in a humidified 5% CO₂ atmosphere. Subsequently, brefeldin A (final conc. 10 μg/ml, BioLegend) was added to all wells and cell were incubated for 4 h at 37 °C in an electronic plate incubator (EchoTherm). Plates were kept at 4 °C overnight. Cells were washed twice with DPBS (Gibco), treated with Benzonase (Sigma-Aldrich, incubation 15 min at 37 °C) and EDTA (Sigma-Aldrich, incubation 5 min at room temperature) to reduce clumping and detach cells. Subsequently cells were washed twice with DPBS, stained with viability dye (LIVE/DEAD Fixable Near-IR 775, Invitrogen), and incubated for 10 min at room temperature in the dark. Fluorescence-conjugated surface antibodies were added (CD45 PerCP (clone HI30), CD14 APC (clone 63D3) [BioLegend, San Diego, CA]) and cells were incubated for 30 min at room temperature in the dark. Following two washes with Facs buffer (DPBS containing 2% FBS and 0.09% NaN₃ [Sigma-Aldrich]), cells were fixed and permeabilized (BD Cytofix/Cytoperm, BD Biosciences, San Diego, CA), and incubated for 15 min at room temperature in the dark. Cells were washed twice with BD Perm/Wash buffer (BD Biosciences), fluorescence-conjugated intracellular antibodies were added (IL-6 FITC (clone MQ2-13A5), TNF BV421 (clone MAb11) [BioLegend], COX-2 PE (clone AS67) [BD Biosciences]), and cells were incubated for 45 min at room temperature in the dark. Following two washes with BD Perm/Wash buffer, cells were resuspended in DPBS containing 1.5% formaldehyde (Polysciences, Warrington, PA) and stored at 4 °C in the dark until flow cytometric measurement (BD LSR II, BD Biosciences, Flow Cytometry Core, Center for Virology and Vaccine Research, BIDMC). Preparations were measured within 20 h and about 250,000 total leukocytes were acquired per sample. Gating analysis was performed using FlowJo 10.10.0 software (BD Biosciences). The gating strategy for monocyte analysis can be found in Supplementary Figure S5. The mean viability of cells after incubation was 77.4 ± 0.7% (mean ± SEM). Percentage of IL-6, TNF, and COX-2 positive monocytes was quantified among all CD14 positive monocytes. The GC sensitivity was expressed as delta value subtracting percentage cytokine/enzyme expressing monocytes in LPS-stimulated wells with DEX addition from the percentage in LPS-stimulated wells without DEX addition. Higher values indicate higher GC sensitivity of monocytes.

Targeted lipidomics

Plasma was analyzed by the Lipidomics Core Facility at Wayne State University, Detroit, MI, following procedures as previously described^63–65. Samples were diluted in phosphate buffer (pH 7.2), spiked with a mixture of internal standards containing 15S-HETE-d₈, leukotriene B₄-d₄, resolvin D2-d₅, 14,15-EpETrE-d₁₁, and prostaglandin E₁-d₄ (1 ng each) for recovery and quantitation, and mixed thoroughly. Polyunsaturated fatty acid (PUFA) metabolites were then extracted from the samples using C18 extraction columns as described earlier^63–65. Briefly, the internal standard-spiked samples were applied to conditioned C18 cartridges, washed with water followed by hexane and dried under vacuum. The cartridges were eluted with 0.5 ml methanol. The eluate was dried under a gentle stream of nitrogen. The residue was re-dissolved in methanol-25 mM aqueous ammonium acetate (1:1) and subjected to LC–MS/MS analysis. High performance liquid chromatography (HPLC) was conducted using a Luna C18 (3 µm, 2.1 mm × 150 mm) column. The mobile phase consisted of a gradient between A: methanol–water-acetonitrile (10:85:5 v/v) and B: methanol–water-acetonitrile (90:5:5 v/v), both containing 0.1% ammonium acetate. The gradient program with respect to the composition of B was as follows: 0–1 min, 50%; 1–8 min, 50–80%; 8–15 min, 80–95%; and 15–17 min, 95%. The flow rate was 0.2 ml/min. The eluate from HPLC was directly introduced to the ESI source of a QTRAP 7500 mass analyzer (SCIEX, Framingham, MA) in the negative ion mode with following conditions: Curtain Gas and GS1: 40 psi, GS2: 70 psi, Temperature: 500 °C, Ion Spray Voltage: −2500 V, Collision Gas: 12 psi, Declustering Potential: −60 V, and Entrance Potential: −7 V. The eluate was monitored by the Multiple Reaction Monitoring (MRM) method to detect unique molecular ion-daughter ion combinations for each of the 125 transitions (to monitor a total of 156 lipid mediators, but only 64 lipid mediators were investigated for the present study). The MRM was scheduled to monitor each transition for 120 s around the established retention time for each lipid mediator. Optimized Collisional Energies (18–35 eV) and Collision Cell Exit Potentials (7–10 V) were used for each MRM transition. Mass spectra for each detected lipid mediator were recorded using the Enhanced Product Ion (EPI) feature to verify the identity of the detected peak in addition to MRM transition and retention time match with the standard. The data were collected and MRM transition chromatograms quantitated using the SCIEX OS 3.4 software. The internal standard signals in each chromatogram were used for normalization for recovery as well as relative quantitation of each analyte. Details on the MRM-internal standard mapping used for quantitation can be found in Supplementary Table S13. Intra-assay and inter-assay variabilities for the method are < 3%. Lower level of quantification (LLOQ) and level of detection (LOD). Details on methods can be found in Supplementary Table S14.

Analysis for the present study was restricted to 37 lipid mediators and metabolites, which contained at least 25% of values above the LOD (see Supplementary Table S14). Samples with undetectable lipid mediator or metabolite levels were substituted with the respective LOD divided by the square root of 2 for statistical analysis⁶⁶. Analytes for the present study contained lipid mediators and metabolites from the following PUFA classes: Omega-6 PUFA arachidonic acid (AA)-derived fatty acids, including the lipoxin precursor 15-hydroxyeicosatetraenoic acid (15-HETE); omega-3 PUFA eicosapentaenoic acid (EPA)-derived fatty acids, including the E-series resolvin precursor 18-hydroxyeicosapentaenoic acid (18-HEPE); omega-3 PUFA docosapentaenoic acid (DPA)-derived fatty acids; and omega-3 PUFA docosahexaenoic acid (DHA)-derived fatty acids, including the D-series resolvin precursor 17-hydroxydocosahexaenoic acid (17-HDHA).

Statistical methods

The primary analysis was the effect of sleep disturbance prior to SARS-CoV-2 infection on the difference in analyte levels between participants grouped by LC status. LC status was split into 3 levels: Likely LC (‘Long COVID Research Index’ (LCRI) ≥ 11 at sample collection), Possible LC (0 < LCRI < 11 at sample collection), and No LC (LCRI = 0 at sample collection). Estimates were generated to compare the Likely LC group to the No LC group and the Likely LC group to the combination of the Possible LC and the No LC groups. Descriptive statistics were also generated for all three LC groups separately. Additionally, LC status and sleep disturbance effects on analyte levels were explored separately.

Primary pre-existing sleep disturbance and LC status models: The model for the primary analysis had the original or log-transformed analyte as the outcome in a linear model with factors for LC group (Likely LC, Possible LC, No LC), presence of sleep disturbance prior to SARS-CoV-2 infection (Yes, No), and an interaction effect between LC group and sleep disturbance. Unadjusted estimates were generated from this model for group comparisons of interest. Exploratory models: Exploratory models were generated to evaluate the separate effects of LC status and sleep disturbance on each original or log-transformed analytes.

Due to sample size limitations analyses were not adjusted for confounding variables. All analyses were considered exploratory and evaluated at an alpha of 0.05. Sample size was smaller for some analytes, due to insufficient aliquot volume or technical problems related to cryopreservation (no viable cells/no monocytes present). Total sample sizes and sample sizes per group were provided in the figure legends. Analyses were carried out in the R statistical software (version 4.4.0) on the Seven Bridges Platform with RECOVER data from a data lock on December 5, 2024. Boxplots are given for original and log-transformed assay results with the median and interquartile range represented by the box and whiskers representing up to 1.5 times the interquartile range above the respective quartile depending on presence of data. All data points are plotted over the boxplots for clarity. Dotplots are given for original and log-transformed assay results with 95% confidence interval bars added from model estimates of group specific confidence intervals.

Author contributions

M.H.: Conceptualization, Methodology, Writing—Original Draft, Writing—Review & Editing, Supervision, Project Administration, Funding Acquisition. J.C.: Methodology, Formal Analysis, Resources, Data Curation, Writing—Review & Editing, Visualization, Supervision. L.C.E.: Investigation, Data Curation, Writing—Review & Editing. R.D.: Data Curation, Writing—Review & Editing. H.W.: Formal Analysis, Data Curation, Writing—Review & Editing, Visualization. E.N.B.: Investigation, Writing—Review & Editing. K.S.: Investigation, Writing—Review & Editing. J.L.: Investigation, Data Curation, Writing—Review & Editing. H.S.: Formal Analysis, Writing—Review & Editing. W.G.: Formal Analysis, Writing—Review & Editing. E.W.K.: Resources, Writing—Review & Editing. A.A.P.: Conceptualization, Methodology, Writing—Review & Editing, Funding Acquisition. D.H.B.: Conceptualization, Methodology, Resources, Writing—Review & Editing, Supervision, Funding Acquisition. J.M.M.: Conceptualization, Methodology, Writing—Review & Editing, Supervision, Funding Acquisition. All authors approved the final version of the manuscript.

Funding

Funding was provided by the following grants: NHLBI OTA No. PATHO-PH2-SUB_01_23 to MPIs MH, AAP, DHB, and JMM; NIH/NCRR S10RR027926, NIH/OD S10OD032292 to the Wayne State University Lipidomics Core Facility.

Data availability

All data and analyses supporting the findings of the present study are included in the manuscript and/or the supplementary file. Original raw data are not publicly available due to local data policy regulations but will be made available for research upon reasonable request to the corresponding author.

Declarations

Competing interests

The authors declare no competing interests.