We enrolled 92 individuals who met the WHO’s diagnostic criteria for post-COVID syndrome (20), along with 73 control participants matched by age and sex. The controls had contracted SARS-CoV-2 during a similar timeframe but recovered fully without any persistent symptoms or complications. All participants were infected with COVID-19 between February 2020 and June 2022, and blood samples were collected during two periods: April–May 2023 and November–December 2024.

Inclusion criteria required participants to be between 18 and 65 years old and to have had a acute COVID-19 infection that did not necessitate hospitalization. Prior to enrolment, each patient underwent a physical examination, radiological imaging, and laboratory testing to exclude alternative causes of symptoms.

Exclusion criteria included a history of malignancy, autoimmune disorders, chronic or active infections, ongoing treatment with corticosteroids or immunosuppressive agents, or pre-existing chronic cardiovascular disease prior to their COVID infection. Medical records were reviewed to obtain participants’ clinical histories and relevant parameters from the time of their COVID illness.

Post-COVID individuals were recruited through the Uppsala post-COVID outpatient clinic and the National Swedish COVID Association. Matched control participants were selected from the Östhammar primary care center and Praktikertjänst Rosendal Primary Care Center among patients who sought medical attention for non-COVID-related symptoms.

Among the post-COVID group, 35 patients contracted SARS-CoV-2 during Sweden’s first wave (February–August 2020), 22 during the second wave (September 2020–February 2021), 13 during the third wave (March–July 2021), and 22 during the fourth wave (August 2021–April 2022). Those infected during waves 1–3 had not been vaccinated. In contrast, participants infected in the fourth wave, which was predominantly driven by the Omicron variant, had received at least two vaccine doses.

All participants had a PCR-confirmed COVID-19 infection, with the exception of ten post-COVID patients and nine controls who contracted the virus early in the pandemic, when PCR testing was limited and primarily reserved for severely ill individuals.

The mean time interval between COVID-19 infection and sampling was 34.0 ± 12.4 months for patients (range 14–58 months) and 32.0 ± 12.9 months for controls (range 11–61 months).

For participants infected during the fourth wave (Omicron), the mean interval was 21 ± 7.4 months for patients (range 14–34 months) and 24 ± 10.6 months for controls (range 11–41 months). Among participants infected during waves 1–3 (pre-Omicron), the corresponding intervals were 39 ± 10.2 months for patients (range 25–58 months) and 37 ± 11.0 months for controls (range 24–61 months).

The study was approved by the Swedish Ethical Review Authority (2021-06852-0160) and conducted in accordance with the Helsinki declaration. All participants gave written informed consent.

The study was conducted in two phases: the first took place in April–May 2023, and the second in November–December 2024 in which both patients and controls were recruited at the same time. During these periods, clinical data were gathered, plasma was collected and frozen at -80C, and participants completed several standardized assessment tools. These included the Fatigue Severity Scale (FSS) (21) for physical fatigue, the Mental Fatigue Scale (MFS) (22) for cognitive fatigue, the Montgomery-Åsberg Depression Rating Scale (MADRS) (23) for depressive symptoms, and the Hospital Anxiety and Depression Scale (HAD) (24) for evaluating both depression and anxiety.

Cut-off scores, adapted to the Swedish population, were as follows. For the FSS (range 0–63), a score of ≥36 indicated significant physical fatigue. The MFS (range 0–42) used a threshold of ≥10 for mental fatigue. MADRS scores (range 0–54) were interpreted as follows: 0–12 indicating no depression, 13–19 mild, 20–34 moderate, and >35 severe depression. For the HAD scale, both the depression and anxiety subscales (range 0–21) used a cut-off of ≥7.

Post-COVID symptom burden was further assessed using a Symptom Severity Score (SSS), which rated 17 symptoms on a 10-point scale (0 = no symptom, 10 = maximum severity), as presented in previous studies (25, 26).

Plasma samples from the participants were analyzed using the Explore 384 Inflammation assay (Olink Proteomics AB, Uppsala, Sweden) by the analysis service at Olink. The following assays did not meet quality control criteria and were therefore not included in the project: BCL2L11, BID, LTA, PTPRM, RAB6A, CD40LG, IDS, CLEC7A, HGF and MGLL. Data are expressed as normalized protein expression (NPX) values, Olink Protemics’ arbitrary unit, on a log2 scale. NPX values were acquired by normalizing cq-values against extension control as well as an interplate control and a correction factor.

Plasma of 92 patients and 67 controls was analyzed using MSD S-PLEX SARS-CoV-2 Spike kit (Meso Scale Discovery, Rockville, MD). The samples were analyzed according to the manufacturer’s instructions at SciLifeLab Affinity Proteomics (Uppsala University, Sweden). The dynamic range (Limit of Detection (LoD) to highest value) was 104.5 fg/mL to 376–000 fg/mL. For graphing and analysis, any concentrations below the LoD were assigned the LoD value.

Differences in assessment scores between PASC and control group were evaluated using a two-sample t-test assuming equal variance (p-value cutoff 0.05) or Fisher’s exact test (p-value cutoff 0.05). Linear correlations between severity scores and BMI, as well as severity scores and age, were assessed using the Pearson correlation coefficient and reported as R².

Data from the Olink Inflammatory Panel were analyzed using R (The R Foundation for Statistical Computing, Vienna, Austria) with the R package OlinkAnalyze (GitHub: github.com/Olink-Proteomic/OlinkRPackage/tree/master/OlinkAnalyze). Groups were compared using t-tests, and p-values were adjusted for multiple testing using the Benjamini-Hochberg method (adjusted p-value, padj; significance cutoff 0.05). Heatmaps displaying Z-scores for each protein were generated using the R package pheatmap.

A random forest analysis, a supervised machine learning approach, was performed to explore the relationships between protein assays and study groups. This method is well suited for capturing complex, non-linear interactions between biomarkers. The analysis was conducted with the primary aim of ranking biomarkers according to their relative importance rather than developing a predictive model. Variable importance was assessed using permutation importance. Internal model validation was performed using the out-of-bag (OOB) procedure inherent to the random forest algorithm, whereby each tree is trained on a bootstrap sample and evaluated on observations not included in that sample. This approach provides an internal estimate of model error and reduces the risk of overfitting without the need for an explicit train/test split or cross-validation. The random forest analysis was implemented using the ranger package in R.

Gene Set Enrichment Analysis (GSEA) was performed on the differentially expressed proteins using the Hallmark gene sets from the Molecular Signature Database (MSigDB). Additionally, protein-protein interaction networks were analyzed using STRING to explore interactions among the differentially expressed proteins and identify key network hubs.

Further, within the patient subgroup, associations between differentially expressed proteins and age, BMI, symptom severity scores, and log2-transformed spike protein levels were assessed using linear regression. Differences in protein levels between patients infected with the Omicron variant and those infected with earlier variants were assessed with t-tests. P-values were adjusted for multiple comparisons using the Benjamini-Hochberg method.

Detailed clinical information and responses to rating scales of each individual study participant are presented in Supplementary Table S1. The age range of study subjects was between 21 and 64 years, with a mean age of 44.4 ± 10.3 years in post-COVID patients (range 21–63 years), and 44.5 ± 11.5 years in controls (range 21–64 years) (Table 1).

The majority of participants in both groups were females, with 80.4% in the patient group and 82.2% among controls. Patients and controls had a comparable socioeconomic backgrounds and similar levels of education. A high educational level was observed in 65% of patients and in 78% of controls (p = 0.47). Body Mass Index (BMI) was similar across both groups (p = 0.96). Comorbidities at the time of infection were largely comparable between patients and controls, with the exception of ADHD, which was more frequently diagnosed in the patient group (Table 1). Most patients reported a good general health and a physically active life prior to their COVID-19 infection. At the time of blood sampling and assessment for post-COVID symptoms, 65.2% of patients were on sick leave.

All administered rating scales revealed significant differences between the patient and control groups. Patients exhibited markedly higher levels of both physical fatigue (FSS: 55.9 ± 9.5) and mental fatigue (MFS: 21.3 ± 11.5) compared to controls (p < 0.001). Depression ratings, assessed using both MADRS (15.0 ± 7.2) and HAD Depression (7.0 ± 4.1), indicated mild depression among patients. HAD scores for anxiety (6.6 ± 4.1) were just below the cutoff. Both depressive and anxiety symptoms were significantly more pronounced in patients than in controls (p < 0.001) (Table 1). The post-COVID Symptom Severity Score (SSS; range 0–170) was significantly higher in patients (61.0 ± 26.3) than in controls (8.7 ± 11.5) (p < 0.001). The most debilitating symptoms reported by patients, based on a 0–10 scale, were cognitive fatigue (6.8 ± 2.5), physical fatigue (6.6 ± 2.3), headache (4.7 ± 3.2), and muscle pain (myalgia) (4.5 ± 3.2) (Table 1).

There was no correlation between the post-COVID symptom severity score and patients age (R² = 0.0148) or BMI (R² = 0.019) (Supplementary Figure S1). A comparison between patients who were infected with the earlier COVID variants (wave 1-3) and patients who contracted the Omicron variant (fourth wave) revealed no significant differences in self-reported post-COVID symptoms (p-value >0.05).

To determine plasma protein profiles associated with PASC, we performed plasma protein analysis using proximity extension assay technology. The total number of proteins detected in all samples was 358. A principal component analysis (PCA) showed overlapping distributions of PASC and control samples (Supplementary Figure S2). The results of the t-test comparisons between patients and controls for all 358 proteins are presented in Supplementary Table S2.

In total, 26 proteins were differentially expressed between patients and controls: 23 were upregulated and 3 were downregulated (Table 2; Figure 1).

The most highly upregulated proteins were Oncostatin M (OSM), which belongs to the IL-6 family of cytokines and activates JAK signaling (padj = 0.0019), and interleukin-1 receptor antagonist (IL1RN), which blocks the activity of the pro-inflammatory cytokines IL-1α and IL-1β (padj = 0.0019).

The PASC group showed upregulation of IL-6 (padj= 0.0329), interleukin-12 subunit beta (IL12B) (padj=0.0470), and interleukin-2 (IL-2) (padj= 0.0488), as well as proteins related to tissue repair and angiogenesis, such as Transforming Growth Factor Alpha (TGFA) (padj=0.0246) and Angiopoietin-like protein 2 (ANGPTL2) (padj = 0.0122).

The levels of Colony Stimulating Factor 3 (CSF3), also known as Granulocyte Colony-Stimulating Factor and structurally related to the IL-6 superfamily, and Colony Stimulating Factor 1 (CSF1), also known as Macrophage Colony-Stimulating Factor, were both significantly increased in patients (padj=0.0238 and padj=0.0320, respectively).

Similarly, we detected increased levels of lymphotoxin Beta Receptor (LTBR), a member of the TNF receptor superfamily that strongly activates both canonical and non-canonical NF-κB signaling pathways (padj=0.0246) as well as the Major Histocompatibility Complex Class II, DR Alpha (HLA-DRA) (padj=0.0058).

Furthermore, two molecules that may play a role in reducing inflammation were also upregulated: CD83 (padj=0.0413) and Chemokine ligand 22 (CCL22) (padj=0.0116). CD83, expressed by macrophages, has been described as an important immune checkpoint molecule involved in the resolution of inflammation (27). CCL22 has both pro-inflammatory and anti-inflammatory functions, particularly through its role in recruiting regulatory T cells (Tregs). By promoting interactions between dendritic cells (DCs) and Tregs, it regulates the recruitment of Treg distribution in both health and inflammatory conditions (28).

To evaluate the combined discriminatory power of all proteins and to identify key predictors of group classification, we complemented the t-test analysis with a random forest analysis. The proteins with the highest importance in random forest largely overlapped with those from the t-test, supporting the robustness of the findings. The list of ranked importance values, with the top 50 markers highlighted, is provided in Figure 2. The proteins with the highest importance for distinguishing PASC patients from controls were IL1RN, followed by OSM, ANGPTL2, CLEC4A, HLA-DRA, and PKLR.

Notably, levels of Proteasome Assembly Chaperone 3 (PSMG3), a chaperone involved in proteasome biogenesis, and the proinflammatory cytokine Interleukin-16 (IL-16), a chemoattractant for CD4⁺ T cells, monocytes, and eosinophils, were not significantly altered in the t-test after adjustment, but showed discriminatory relevance in the random forest model.

Gene Set Enrichment Analysis (GSEA) was performed using the 26 differentially expressed proteins (DEPs) and the Hallmark gene set collection from MSigDB, resulting in five partially overlapping enriched gene sets: Allograft Rejection, Inflammatory Response, TNF-α Signaling via NF-κB, IL-6/JAK/STAT3 Signaling, and IL-2/STAT5 Signaling. All enriched gene sets contained at least three genes and met the significance threshold (padj < 0.05) (Table 3).

Both the Allograft Rejection and IL-2/STAT5 signaling pathways were enriched, reflecting activation of T cells and the adaptive immune response. Within the Inflammatory Response pathway, upregulation of CSF3, IL-6, CCL22, CSF1, OSM, and IL12B was detected, indicating sustained inflammatory signaling. Additionally, enrichment of the TNF-α signaling via NF-κB and IL-6/JAK/STAT3 pathways was observed, consistent with activation of the innate immune system and inflammatory responses.

A network analysis using STRING was performed with the differentially expressed proteins to further investigate protein-protein interactions. In total 19 of the 26 proteins created a network with hub proteins belonging to the gene sets Inflammatory response (CSF3, IL6, CCL22, CSF1, OSM and IL12B) from the GSEA (Figure 3).

No significant associations were observed between the 26 DEPs and any of the clinical assessment scores, including MADRS, HAD Anxiety, HAD Depression, MSF, and FSS scores. The Post-COVID Symptom Severity (SSS) total score, as well as each of the 17 individual symptoms included in this scale, also showed no significant associations with the 26 DEPs according to linear regression analysis.

Notably, patients with higher BMI exhibited elevated levels of inflammatory proteins (Supplementary Table S3). Linear regression analysis identified positive associations between BMI and 14 of the 26 differentially expressed proteins, including IL1RN, LGALS9, IL6, AMN, LILRB4, ANGPTL2, NCF2, EPHA1, CLEC4D, OSM, CCL22, CSF3, IL12B, and CST7, while SCGB3A2 was inversely associated with BMI.

The comparison between PASC patients infected with the Omicron variant and those infected with earlier SARS-CoV-2 variants revealed no differences among the 26 DEPs (Supplementary Table S4). Likewise, no correlations were detected between DEPs and age (Supplementary Table S5).

We did not detect any significant difference in plasma spike protein levels between patients and controls (Mann–Whitney U test, p = 0.29) (Figure 4). Five outliers were identified in the control group and four in the patient group. None of these individuals recalled having had a recent SARS-CoV-2 infection or COVID-19 vaccination shortly before blood sampling, and the reason for their elevated spike protein levels remains unclear. No correlation was found between spike protein plasma concentrations and the post-COVID Symptom Severity Score, nor with any of the 17 individual symptoms included in this scale. Likewise, linear regression analysis showed no association between spike protein levels and DEPs.