For this exploratory study, female participants aged 18 to 59 years were recruited from the ME/CFS clinic of the Charité Fatigue Centre, Berlin. Serum samples were collected from 45 PCS patients suffering from moderate to severe fatigue, with 26 of them fulfilling the Canadian Consensus Criteria (CCC) for the diagnosis of ME/CFS (pcME/CFS). In the case of 2 of 26 pcME/CFS patients and 3 of 19 PCS patients with a disease duration of less than 6 months, the diagnosis was confirmed at month 6. In addition, we included serum samples from 36 other post-infectious ME/CFS (piME/CFS) patients. These patients developed ME/CFS after an infection most commonly following non-SARS-CoV-2 viral respiratory tract infections or EBV infection. Moreover, we included 34 female HCs, of whom 26 reported a SARS-CoV-2 infection at least 6 months prior. Information about a previous SARS-CoV-2 infection were not available for the piME/CFS study group.

Detailed cohort information is displayed in Table 1 ( Table 1 , Supplementary Data 1 ). Disease and symptom severity were assessed in patients with appropriate questionnaires. The functional disability was evaluated using the Bell score, ranging from 0 to 100 (with 100 for no restrictions) (26). Physical function and daily activities were assessed using the Short Form Health Survey 36 (SF-36), ranging from 0 to 100 (greatest to no restrictions) (27). PEM severity was evaluated according to the brief DSQ-PEM questionnaire and scores ranging from 0 to 46 calculated (no to frequent/severe PEM) (28). The severity of the key symptoms fatigue, pain, and cognitive impairment was quantified using a Likert scale (1 = no symptoms to 10 = severe symptoms). Autonomic dysfunction was assessed using the Composite Autonomic Symptom Score 31 (COMPASS-31), ranging from 0 to 100 (no to strongest impairment) (29).

Routine laboratory parameters were determined at the Charité diagnostics laboratory Labor Berlin GmbH (Berlin, Germany). Study data, including clinical and routine laboratory parameters, were collected and managed using REDCap electronic data capture tools hosted at Charité Universitätsmedizin Berlin (30, 31).

This study was approved by the Ethics Committee of Charité Universitätsmedizin Berlin (EA2/067/20, EA2/066/20) and is per the 1964 Declaration of Helsinki and its later amendments. The participants provided written informed consent.

To detect serum IgG reactivity to EBV proteins and potential human autoantigens, sera were tested in a multiplex dot blot assay as described (32). Briefly, 5 µl of concentration-adjusted recombinant His₆-tagged proteins were spotted on a nitrocellulose membrane. Following a blocking step with 5% milk powder in PBS the membrane was co-incubated overnight at 4°C with an anti-His₆ antibody (clone 3D5) and 1:1000 diluted serum in a 3% milk buffer. Following washing, membranes were incubated for 1.5 hours with fluorescence-labeled anti-mouse IgG antibody (LI-COR^® IRDye 680) and anti-human IgG antibody (LI-COR^® IRDye 800) to quantify the amount of spotted proteins and binding of human IgG. The membranes were scanned in a LI-COR^® Odyssey FC scanner that reports results as arbitrary fluorescence units (AFU), returning a CW700 and a CW800 reading for each dot on the membrane corresponding to the protein concentration (anti-His₆) and the human serum reactivity, respectively. A standard curve of recombinant His₆-tagged human IgG, as well as solvent (8 M urea) and Ni-NTA agarose-affinity enriched mock-transfected HEK293T cell lysate, were used for specific standardization and background correction, enabling blot-to-blot comparability.

Autofluorescence signals caused by the nitrocellulose membrane or the solvent as well as any possible fluorescence due to serum reactivity against HEK293T proteins were subtracted from readings for antigenic proteins. Background-subtracted AFU values for proteins in the CW800 and CW700 channels were converted to normalized arbitrary values using a simple linear regression model drawn from values obtained for serial dilutions of recombinant IgG for each individual membrane. Next, the quotient of normalized CW800 and CW700 values was formed to compensate for potential differences in the amount of sample protein spotted on the membrane. This normalized AFU value describes sera IgG binding against EBV and human proteins. Normalized AFUs ≤ 0 means no detectable binding of IgG to the target protein and were set to 0. All normalized AFU values >0 were considered as positive IgG reactivity.

Statistical data analyses were performed using R Version 4.2.1 and GraphPad Prism Version 9.5.1. All scripts for the analyses conducted in R are openly accessible via GitHub at https://github.com/KerstinRubarth/IMMME_SP4, and raw data is accessible as Supplementary Data 1 .

Patient characteristics are presented as median and range for continuous variables stratified by study group. Univariate comparisons of independent groups were done using the Kruskal–Wallis test. Two-tailed tests were used with a significance level of 5%.

For each study group specific IgG binding was descriptively analyzed, presenting the medians and Interquartile Ranges (IQRs) of the untransformed, background-corrected values (peptide IgG binding), or untransformed, normalized values (protein IgG binding), respectively. Due to the skewed distributions of the IgG levels, data were visualized using a log₁₀ transformation, which provided the most effective scaling for the plots. Univariate pairwise comparisons of each patient group with the HC group were performed with the untransformed values using the nonparametric Mann-Whitney test. Two-tailed tests were used with a significance level of 5%.

In addition, IgG binding was analyzed using linear models, with the group as the independent variable and IgG binding as the dependent variable. To account for skewed distributions and outliers in IgG levels, a log-transformation with a base of 2 was applied. Negative values and zeros were set to 0.01 for this analysis. Subsequently, linear models were fitted using the R-function ‘lmFit’ from the R-package ‘limma’, and model parameters were estimated utilizing the empirical Bayes method (R-function ‘eBayes’). The results were presented through volcano plots with a log2 fold change (FC) cutoff at one and a p-value cut-off of 0.1.

The frequency of positive IgG reactivity was expressed as a percentage. Pairwise comparisons of categorical variables between patient groups with HC group, respectively, was done using two-tailed Chi-Squared test with a significance level of 5%.

Correlations of IgG binding and clinical scores were analyzed for each group separately by presenting heatmaps displaying Spearman-type correlation coefficients ≥ ± 0.3, indicating at least small to moderate correlations (33). Spearman correlation was used due to the non-normal distribution of the data. P-values less than 0.05 were considered as statistically significant and were indicated by * in the heatmaps. Due to the exploratory nature of the study, no correction for multiple testing was applied. Instead, the focus was on investigating effect sizes, such as correlations and log-fold changes. p-values are provided for orientation and should be interpreted in a hypothesis-generating rather than a confirmatory manner.

In a previous study employing a regression model for binary outcomes, a significantly elevated IgG reactivity was observed against two arginine-rich EBV-derived EBNA peptides in piME/CFS patients compared to HC (23). Using NCBI´s Protein Blast and the RefSeq_select databases, we identified human proteins with high sequence homology to arginine-rich EBNA6_70/66 peptides. These proteins selected for their potential pathophysiological relevance to ME/CFS include α-adrenergic receptors (ADRA1B, ADRA1D, ADRA2C), the sodium/potassium/calcium exchanger 3 (SLC24A3), the nuclear serine/arginine repetitive matrix protein 3 (SRRM3), the mitochondrial phospholipase D family member 6 (PLD6) and the intracellular testis-specific Y-encoded-like protein 2 (TSPYL2) ( Table 2 ). All aligned sequences share a characteristic poly-R motif. In the α-adrenergic receptors, this motif was localized in the cytoplasmic C-terminus; in SLC24A3, it was found in the cytoplasmic N-terminus. In PLD6, a mitochondrial outer membrane protein, the poly-R sequence was likewise located within the cytoplasmic region. For EBNA4_529, sequence homology was identified with the nuclear testis-specific Y-encoded-like protein 5 (TSPYL5), which contains three arginines ( Table 2 ).

In the present study, we analyzed antibody reactivity to these EBV EBNA peptides and their homologous human counterparts, as well as to the corresponding full-length proteins, in female PCS, pcME/CFS, piME/CFS patients, and HC. The cohorts were age-matched ( Table 1 ). Time since last SARS-CoV-2 infection was comparable between the post-COVID patient groups and the HC group; 26 of HCs reported a previous SARS-CoV-2 infection, with a median of 10.4 months before study participation. However, patients with piME/CFS had a significantly longer disease duration (median 3.1 years) than pcME/CFS (median: 0.75 years) and PCS (median: 0.58 years) patients. ME/CFS patients exhibited greater physical impairment, more severe fatigue, and more pronounced PEM than PCS patients. The EBNA6_740 peptide was included as a positive control, as our previous study had shown that most patients and HCs displayed high antibody levels against this peptide (22). An overview of the study design is provided in Figure 1 .

We detected IgG reactivity against EBNA6 and EBNA4 peptides, as well as against all the homologous human sequence peptides, in the sera of both healthy controls and patients ( Figure 2 ). In PCS patients, IgG binding was significantly increased for the EBNA6_66/70 homologous sequences TSPYL2_185 and SRRM3_383 ( Figure 2A ). Additionally, all patient groups showed significantly elevated levels of IgG against the EBNA4_529 homolog TSPYL5_133 compared to HCs ( Figure 2B ). Linear model analysis using the Limma framework confirmed significantly higher IgG reactivity to SRRM3_383, TSPYL2_185, and TSPYL5_133 in piME/CFS and PCS patients, and to TSPYL5_133 in the pcME/CFS cohort relative to HCs ( Figures 2D-F ).

Analysis of the frequency of IgG-positive individuals within each cohort revealed antibodies against the EBNA6 poly-R sequences, as well as the ADRA and PLD6_23 peptides, in nearly all PCS patients, and in most ME/CFS and HC individuals ( Figure 2G ). IgG reactivity to the SLC24A3_8, TSPYL2_185, and SRRM3_383 peptides was observed in 40 – 70% of the HCs but in 90 – 100% of PCS; for TSPYL2_185 and SLC24A3_8 peptides, reactivity was also significantly more frequent in piME/CFS patients than in HCs ( Figure 2G ). IgG reactivity to EBNA4_529 was found in approximately 75% of all samples, but more patients showed reactivity to the homologous peptide TSPYL5_133 ( Figure 2H ). As observed in our previous study, strong IgG responses to EBNA6_740 were present in 80 – 95% of all subjects ( Figures 2C, I ).

We further explored the potential cross-reactivity of IgG to EBV-derived and homologous human poly-R peptide sequences using Spearman`s correlation analysis ( Figures 3A-D ). Positive correlations between IgG levels targeting EBNA6_66/70 IgG and most homologous human peptides were observed across all cohorts. A correlation of IgG reactivity against EBNA4_529 and TSPYL5_133 was exclusively found in pcME/CFS. No correlations were observed for reactivity with the non-homologous EBN6_740.

Next, we studied the IgG reactivity to full-length EBNA proteins and the human proteins using a multiplex assay. Since no validated cutoff for positive antibody reactivity against these proteins exists to date, we present background-corrected, normalized arbitrary fluorescence units (AFU) for each protein, as described in the method section.

We observed IgG binding to EBNA6 and EBNA4 proteins in most samples, with higher reactivity to EBNA6 in patient cohorts compared to HCs, reaching statistical significance in the pcME/CFS group ( Figure 4A ). IgG responses to the human proteins were detectable in all study cohorts, although signals were generally low in many individuals. Antibodies targeting ADRA proteins were present in 20 – 40% of the HCs, but were significantly more frequent ( Figure 4C ) and showed higher titers in PCS and ME/CFS patients ( Figure 4A ). While the frequency of IgG reactive against TSPYL2 was similar across groups ( Figure 4C ), both ME/CFS cohorts exhibited higher autoantibodies levels compared to HCs ( Figure 4A ). Comparable levels and frequencies of IgG binding to EBNA4 and its homolog TSPYL5 were found across all study groups ( Figures 4B, D ).

We next examined whether autoantibody levels were associated with the severity of key symptoms using Spearman correlation analyses ( Figure 5 ). In PCS patients, significant positive correlations were observed between most antibody levels, including those to EBNA6_66/70, and autonomic dysfunction assessed by COMPASS score. IgG responses to two ADRA peptides also showed significant associations with fatigue and PEM, while several other correlations showed positive trends ( Figure 5C ). Although no significant correlations between antibodies to the full-length proteins and symptom severity were found in PCS, there was a noteworthy trend toward higher IgG levels to all three ADRA proteins being associated with PEM severity ( Figure 5F ).

In the ME/CFS cohorts, no significant positive correlations were observed between autoantibodies to peptides and symptom severity ( Figures 5A, B ). However, in the piME/CFS group, significant positive correlations were found between autoantibodies to SRRM3 protein and cognitive impairment, and between antibodies to TSPYL5 protein and both cognition and muscle pain ( Figure 5D ). In contrast to PCS, the pcME/CFS group exhibited several inverse correlations between IgG reactivity to ADRA2C_422 peptides and symptoms, such as fatigue, cognitive impairment, and PEM ( Figure 5B ). Similar associations were also found when symptom severity was compared between patients with high versus low autoantibody levels ( Supplementary Data 2 ).