Salivary cortisol in long COVID: a marker of broader stress system and circadian rhythm dysregulation

This prospective, single-center study was conducted at the National Institute for Infectious Diseases L. Spallanzani in Rome, Italy. The post-COVID outpatient clinic, established in 2020, initially followed all hospitalized COVID-19 patients. Over time, asymptomatic cases declined, while referrals for LC increased. Between February 2023 and March 2024, 104 consecutive patients were enrolled, referred from home, hospital wards, or the early-treatment SARS-CoV-2 clinic. Inclusion criteria were age >18, confirmed SARS-CoV-2 infection (positive swab ≥28 days before baseline), and signed informed consent. Exclusion criteria were ongoing use of systemic, injectable, or inhaled steroids; oral contraceptives; and a history of uncontrolled rheumatologic or uncontrolled endocrine disease. Uncontrolled rheumatologic disease was defined as active or unstable autoimmune conditions requiring high-dose immunosuppressive therapy or systemic steroids and associated with ongoing systemic inflammation. Uncontrolled endocrinological disorders were defined as a history of an endocrinological condition with abnormal laboratory findings. Eight participants were excluded because did not match with long-COVID definition (N = 3) or were under steroids or oral conceptive pills regimen (N = 5). Data from 96 participants were included in the descriptive analysis. Blood samples were unavailable for five participants; therefore, samples from 91 participants were analyzed (12 APC and 79 LC). For the SC analysis, patients who collected saliva at the wrong time or provided an insufficient sample for analysis were excluded. Details are shown in the study flow-chart (Figure 1).

Acute SARS-CoV-2 infection onset was defined by the first positive molecular or antigenic nasopharyngeal swab. LC was defined as ≥1 new or persistent symptom > 4 weeks post-infection. LC severity was based on the number of symptoms of interest (SOI): fatigue, cognitive deficits, poor exercise tolerance, dyspnea, arthralgia, and dysautonomia. Moderate LC was defined as 1–4 SOI, and severe LC as ≥5. Fatigue was assessed using the 10-item Fatigue Assessment Scale (FAS), which scores both physical and mental fatigue. A total score ≥22 indicates fatigue, and ≥35 indicates severe fatigue.

The primary outcome was to assess daily SC variation in post–SARS-CoV-2 individuals, both symptomatic and asymptomatic, compared to healthy controls (HC). Secondary outcomes included evaluating HPA and HPT axis function, sex hormone levels, and inflammatory and coagulative markers across LC, asymptomatic post-COVID (APC), and HC groups. The study also aimed to identify risk factors associated with Long COVID development.

Participants were jointly assessed by an infectious disease specialist and an endocrinologist at baseline (BL) and at a three-month follow-up (V2). At BL, detailed medical histories were recorded, including acute COVID-19 phase and diagnostic data. Clinical evaluations focused on LC symptom burden, daily functioning, and fatigue, using the Fatigue Assessment Scale (FAS).

Participants were instructed verbally and provided with a written sheet detailing the procedure for saliva collection at fixed clock times (8:00 AM, 3:00 PM, and 11:00 PM) on a typical weekday. They were advised to avoid alcohol, caffeine, physical exercise, stressful activities, and toothbrushing before sampling. Each participant recorded the exact time of collection on the vial label, and samples collected more than ±30 minutes from the scheduled time were excluded from analysis. Samples were refrigerated immediately after collection and returned during the follow-up visit. Adherence was verified through participant logs and verbal confirmation. Wake time and sleep duration were not standardized or recorded, while smoking habits were documented.

Additional diagnostic tests and specialist referrals were provided as clinically indicated. To avoid interference, after saliva and blood collection, all participants with LC were advised to follow a standardized supportive regimen comprising liposomal glutathione, vitamin C, L-arginine, a lactose- and gluten-free low-glycemic diet, yoga nidra, and individualized physical activity tailored to symptom severity. Two weeks post-BL, early morning fasting blood samples (7:30–8:30 AM) were collected for hormonal, inflammatory, and coagulative markers, including cortisol, adrenocorticotropic hormone (ACTH), dehydroepiandrosterone sulfate (DHEAS), sodium, potassium, testosterone (males), thyroid-stimulating hormone (TSH), free triiodothyronine (fT3), free thyroxine (fT4), interleukin-1β (IL-1β), IL-6, tumor necrosis factor-alpha (TNF-α), NOD-like receptor family pyrin domain-containing 3 (NLRP3), IL-8, vitamin D, C-reactive protein (CRP), D-dimer, intercellular adhesion molecule-1 (ICAM-1), and vascular cell adhesion molecule-1 (VCAM-1). A 1 µg ACTH stimulation test was performed in patients with basal cortisol 3–15 µg/dL; hypocortisolism was defined as basal <3 µg/dL or peak <18 µg/dL. SC samples were collected the day before blood sampling, refrigerated, and delivered on the test day. Saliva from seven healthy, asymptomatic controls (either COVID-negative or >6 months post-infection) was also analyzed for comparison. Individuals >6 months after SARS-CoV-2 infection were included only if they had fully recovered and reported no residual or persistent symptoms compatible with Long COVID. The >6-month cutoff was selected to minimize potential short-term post-viral HPA axis alterations. Three HC participants had never been infected with SARS-CoV-2, while the median time since previous infection among the four HC with prior COVID-19 was 627 days (IQR 127). Inflammatory and coagulative markers were analyzed in a subset of participants due to limited reagent availability. Samples were processed in the order of arrival at the laboratory, and laboratory staff were blinded to participant group. At V2, participants underwent repeat clinical and fatigue assessments to track LC symptom progression. Laboratory personnel performing hormonal and cytokine analyses were blinded to participants' clinical classification (LC, APC, or HC) to minimize bias.

Hormonal assays were performed on serum samples using the Atellica IM chemiluminescent immunoassay system (SIEMENS) for cortisol [reference values (rv) 4.3-22.4 µg/dL], testosterone (rv 2.5-8.5ng/ml), fT3 (rv 2.3-4.2 pg/ml), fT4 (rv 0.8-1.6 ng/dl), and TSH (rv 0.4-3.5 µU/ml). DHEAS was measured via chemiluminescence (MAGLUMI 2000 Plus, SNIBE) (rv male 18-30yrs 24-690 µg/dL; 31-50yrs 106-995 µg/dL; 51-70yrs 24-313 µg/dL; >70yrs 5-253 µg/dL)(rv female 18-30yrs 18-391 µg/dL; 31-50yrs 19-266 µg/dL; 51-70yrs 8-188 µg/dL; >70yrs 7-177 µg/dL), and ACTH (rv 7.2-63.3 pg/ml) via electrochemiluminescence (Elecsys ACTH, COBAS E411, ROCHE). Vitamin D concentrations were measured using an electrochemiluminescent immunoassay (ECLIA, Roche) (rv deficient < 25 nmol/L; insufficient 25–75 nmol/L; sufficient 75–250 nmol/L; potential toxicity > 250 nmol/L). All samples were processed within six hours. SC was collected at 8:00 AM, 3:00 PM, and 11:00 PM using Salivette devices (SARSTEDT) and analyzed with the Elecsys Cortisol II assay (ROCHE) on the COBAS E411. Inflammatory and coagulation markers were assessed in 89 participants (77 LC, 12 APC, and 18 controls). Plasma samples were obtained after centrifugation of whole blood for 10 minutes at 1,800 rpm and immediately stored at −80 °C until analysis. Plasma levels of D-dimer, NLRP3, E-selectin (E-Sel), ICAM-1, VCAM-1, IL-1β, IL-6, IL-8, and TNF-α were quantified using an automated ELISA system (ELLA microfluidic analyzer, ProteinSimple, Bio-Techne). The detection limit of these assays was 0.16 pg/mL for IL-1β, 0.28 pg/mL for IL-6, 0.19 pg/mL for IL-8, 0.30 pg/mL for TNF-α and <500 ng/ml for D-Dimer. No established clinical cut-offs exist for NLRP3 (pg/ml), ICAM-1 (pg/ml), VCAM-1 (pg/ml), or E-selectin (pg/ml). Values were therefore interpreted relative to the distribution observed in the healthy control group.

Data from 104 patients was anonymized using patient id at collection. 96 participants were included in the analysis. We summarized the data in descriptive tables (Table 1) stratifying the samples in subgroups and evaluating the statistical difference across all measurements and exams. Statistical significance was assessed using non-parametric tests to avoid requiring normally distributed data. The Mann–Whitney U test was applied for all unpaired data comparisons, and the Wilcoxon signed-rank test was used to compared paired samples (for example for the visits comparison). Categorical variables were compared using Fisher's exact test. All statistical tests were two-tailed. Variables with missing data were excluded from the analysis. Benjamini-Hochberg false discovery rate (FDR) correction was applied to each measurement across all possible pairs (HC, APC, LC, Moderate LC, and Severe LC as appropriate) to account for multiple tests but not to be overly conservative in significance estimation. For example, for the salivary cortisol time course this means that at each time point we corrected the statistical significance across the possible pairs (HC-APC, HC-LC, and APC-LC for Figure 2a). FDR-adjusted p-values < 0.05 were considered statistically significant. We tested several clinical variables (i.e., physical activity, age, gender, smoking status, body mass index, presence of more than three comorbidities, number of comorbidities, number of vaccine doses, hospitalization during the acute SARS-CoV-2 infection, home management, treatment with monoclonal antibodies or antivirals during the acute phase, and type of acute infection symptoms), to evaluate their association with the development of long COVID; the significance of these variable was not corrected for multiple tests. Statistics represented in the plots are as follows: n.s. p≥0.05, *: p<0.05, **: p<0.01, ***: p<0.001, ****: p<0.0001. All analyses were performed in Python, and statistical analysis was carried out using the Statannotations package (Vinas et al., 2023). To ensure the statistical validity of our circadian claims we performed a linear mixed model approach with the statsmodels python package (Skipper Seabold, 2010). In this linear mixed model, we considered the cortisol levels to be:

Then we are interested if the + Time*Group component is statistically significative when compared to the HC behavior, which represents our null hypothesis.

Table 1 presents the demographic and clinical characteristics of the 96 participants. The cohort included 60% female, all of White ethnicity, with a mean age of 58.1 ± 14.8 years. Among the 83 participants classified as LC (86%), 80% had moderate disease (1–4 SOI) and 20% had severe disease (≥5 SOI). Thirteen participants (14%) were classified as APC. A mild or asymptomatic acute SARS-CoV-2 infection managed at home was reported by 64% of LC patients, compared with 54% of APC participants. APC individuals were more often hospitalized and treated with antivirals or required oxygen supplementation (23% vs 9.6%; p=0.039). LC patients exhibited a higher proportion of ≥3 comorbidities (65% vs 38%, p=0.02), prior SARS-CoV-2 infection (44.6% vs 23%, p=0.03), and previous engagement in competitive sports (51.8% vs 30.1%, p=0.04). No significant differences were found in drug exposure between groups (Table 2). Mean FAS scores were significantly higher in LC than in APC (29.3 vs 14.4, p<0.001), including both mental (13.5 vs 7.5, p<0.001) and physical domains (15.7 vs 8, p<0.001). At BL, 66 patients (24 males, 42 females) reported impairment in daily activities due to LC. FAS scores were strongly associated with LC diagnosis and severity (p<0.0001), and with reduced quality of life (p<0.001) (Figures 3A-D). Symptom prevalence at BL and V2 is shown in Figure 4.

The following analyses should be interpreted as exploratory due to the small HC sample. SC levels differed significantly among study groups (Figure 2A). HC exhibited significantly higher median 8:00 AM SC levels (18.54 nmol/L [IQR 3.3]) compared with both APC (13.16 nmol/L [IQR 6.97]; p=0.014) and LC participants (11.72 nmol/L [IQR 5.41]; p = 0.004). At 3:00 PM, median SC values were 2.92 nmol/L [IQR 2.44] in HC, 4.47 nmol/L [IQR 0.93] in APC, and 4.31 nmol/L [IQR 2.47] in LC (both p=1 vs HC). At 11:00 PM, corresponding values were 1.56 nmol/L [IQR 1.24] vs 2.25 nmol/L [IQR 0.8] in APC (p=0.18), and 2.27 nmol/L [IQR 1.65] in LC (p=0.05). When stratified by LC severity (Figure 2B), participants with severe LC exhibited higher median 11:00 PM SC levels (3.15 nmol/L [IQR 2.6]) compared with moderate LC (2.06 nmol/L [IQR 1.3]; p=0.67), APC (2.25 nmol/L; p=1) and HC (1.56 nmol/L; p=0.22). Morning SC levels were significantly lower in LC compared with HC, but no difference was found between moderate and severe LC subgroups (Figure 2B). Our linear mixed model approach revealed that the daily decrease in cortisol is more pronounced in HC than any other population (APC, LC, Severe LC, Moderate LC, Supplementary Figure S1). This result underlines the possibility of a reduction in amplitude of the salivary cortisol oscillations after exposure to SARS-CoV-2.

Morning BC levels did not differ significantly between LC and APC participants (15.2 µg/dL [IQR 6.2] vs 12.3 µg/dL [IQR 7.4] p= 0.1). Median ACTH hormone concentrations were higher in LC compared with APC group (25 pg/mL [IQR 16.1] vs 13 pg/mL [IQR 13.2]; p = 0.003), while remaining within normal laboratory reference ranges (Figures 5A, B). Thirty-two patients (26 with LC [21.6% of all LC patients] and 6 APC [46% of all APC patients]) met the inclusion criteria for the ACTH-test; five declined participation. Among the 27 individuals tested, one female LC patient showed a subnormal cortisol response (BC 0' 9.6 µg/dL, BC 30' 9.2 µg/dL, BC 60' 14.5 µg/dL) and was diagnosed with adrenal insufficiency requiring replacement therapy. No statistically significant differences were found between APC and LC groups for testosterone (males only), DHEAS, vitamin D, TSH, fT3, or fT4 (Figures 6A–F). Median (IQR) values for APC and LC, respectively, were as follows: testosterone 5.35 ng/ml (IQR 0.62) vs 5.4 ng/ml (IQR 2.07) p=0.7; DHEAS 67.05 µg/dL (IQR 106.3) vs 127 µg/dL (IQR 79.5)p=0.14; vitamin D 56 nmol/L (IQR 32.2) vs 67 nmol/L (IQR 30) p= 0.2; TSH 1.65 µU/ml (IQR 0.63) vs 1.95 (IQR 1.32) p=0.29; fT3 3.28 pg/ml (IQR 0.6) vs 3.36 pg/ml (IQR 0.47) p= 0.8; and fT4 1.2 ng/dl (IQR 0.18) vs 1.15 ng/dl (IQR 0.21) p= 0.51.

(Figures 7A–I) presents the distribution of inflammatory and coagulation biomarkers among APC, LC, and HC participants. Both APC and LC subjects showed higher levels of inflammatory and endothelial activation markers compared with HC. Specifically, APC exhibited elevated median IL-8 levels (10.22 pg/ml [IQR 13.8] vs. 3.37 pg/ml [IQR 2.9] in HC; p=0.006) and NLRP3 levels (0.16 pg/ml [IQR 0] vs. 0.16 pg/ml [IQR 0.08] in HC; p=0.04), while LC participants showed similarly increased IL-8 (9.20 pg/ml [IQR 7.12] vs. 3.37 pg/ml [IQR 2.9] in HC; p=0.0001) and NLRP3 concentrations (0.16 pg/ml [IQR 0] vs. 0.16 pg/ml [IQR 0.22] in HC; p=0.003). No significant differences were observed for IL-1β, IL-6, or TNF-α across groups. Regarding coagulation and endothelial markers, LC participants exhibited higher median ICAM-1 levels (375896 pg/ml [IQR 114273] vs. 297954 pg/ml [IQR 89223] in HC; p=0.03) and VCAM-1levels (751944.5 pg/ml [IQR 337504] vs. 637439 [IQR 279772] in HC; p=0.01). APC subjects also showed elevated median ICAM-1 concentrations compared with HC (363694 pg/ml [IQR 1054133] vs. 297954 pg/ml [IQR 89223] in HC; p=0.09). No significant differences were observed for D-dimer or E-selectin across the three groups. No statistically significant differences in inflammatory or coagulation markers were detected between APC and LC participants.

During the study, 4 patients were lost to follow-up, and 92 completed the V2 visit (3 months after BL). Mean total FAS scores significantly improved from BL to V2 (28 vs 24.5; p < 0.001), driven by improvement in the mental component (12.5 vs 10; p = 0.003), while physical FAS remained stable (15 vs 14; p = 0.7). At V2, 76 patients (82%) continued to meet the diagnostic criteria for LC (compared with 86% at BL). Among this, 18% were classified as severe and 81.5% as moderate.

Large-scale studies have underscored the adrenal glands and GR activity as major determinants of systemic circadian alignment (Talamanca et al., 2023), and women show stronger circadian gene expression, which may contribute to their greater susceptibility to LC (Talamanca et al., 2023). SARS-CoV-2 can induce epigenetic changes affecting both GR pathways and circadian genes, potentially prolonging HPA axis dysfunction in vulnerable individuals (Li et al., 2021). Similar GR desensitization has been observed in athletes (Yeager et al., 2011; Lu et al., 2017), among whom persistent symptoms after COVID-19 have been described (Diebold et al., 2025). In addition, SARS-CoV-2 has been shown to impair mitochondrial oxidative phosphorylation and increase ROS production, changes that can reduce cortisol synthesis and affect its intracellular signaling (Midzak and Papadopoulos, 2016; Zhang et al., 2020; Miller et al., 2021; Nunn et al., 2022; Mantle et al., 2024). Furthermore, the GR itself is sensitive to the cellular redox environment (Okamoto et al., 1999; Vandevyver et al., 2014). These redox-dependent mechanisms have been documented in steroid resistance syndromes and could similarly contribute to blunted cortisol action in LC. Beyond direct viral effects, molecular mimicry between ACTH and SARS-CoV-2 antigens has been proposed as an additional mechanism (Wheatland, 2004), whereby cross-reactive antibodies could attenuate ACTH activity, similar to patterns described in other coronavirus infections (Leow et al., 2005). Together, these alterations may help explain blunted cortisol action in LC and warrant targeted investigation.

This study has several limitations. The very small number of healthy controls inherently reduces statistical robustness of our findings as such, the study should be considered exploratory. Home-based salivary cortisol collection may introduce uncontrolled variance due to behavioral factors. Because wake-up time and sleep patterns were not recorded, the observed reduction in morning SC may partly reflect inter-individual variability rather than intrinsic HPA axis dysfunction. Moreover, we did not control for psychological distress or physical activity, both of which influence cortisol secretion. The observed cortisol differences might also reflect secondary consequences of chronic illness rather than primary HPA axis dysregulation. Additionally, the single-center design may limit the generalizability of the data. Longitudinal studies with comprehensive covariate adjustment are needed to disentangle these relationships. The study also relied on an earlier definition of Long COVID (>4 weeks post-infection), which may limit comparability with research using the current WHO definition (within 3 months). Some measures, including smoking status and prior competitive sports participation, were qualitative and not assessed through validated questionnaires, which warrants cautious interpretation. Additionally, antidepressant use may theoretically influence HPA axis activity, but such effects are usually transient and normalize with chronic treatment. As treatment duration was not recorded, residual confounding cannot be fully excluded. Furthermore, several mechanistic aspects discussed, such as glucocorticoid receptor sensitivity, mitochondrial function, and circadian gene regulation, were not directly assessed. Finally, the observational design precludes causal inference, and the associations observed should be interpreted with caution. Despite these limitations, this is, to our knowledge, the first study to comprehensively characterize circadian SC profiles in LC. Future research should aim to confirm these findings in longitudinal cohorts integrating hormonal, immune, and neurocognitive assessments to better delineate post-COVID pathophysiology. Studies evaluating whether normalization of cortisol rhythmicity parallels clinical improvement will be crucial to understanding the reversibility and clinical significance of HPA axis dysfunction in LC. Moreover, to fully understand the abnormalities occurring in the HPA axis during LC and quantify their impact on symptomatology, it is necessary to further investigate genetic and epigenetic markers of GR sensitivity, as well as the mitochondrial function of endocrine cells within the HPA axis in these patients.