Comprehensive antigen profiling predicts post-surgical neuropathic pain in women treated for breast cancer.

Introduction
Neuropathic pain (NP) is defined as pain caused by a lesion or disease of the somatosensory nervous system1. Surgical nerve injuries can lead to persistent peripheral NP (chronic post-surgical NP, CPSNP). CPSNP is especially prevalent (14–31%) in patients operated on for breast cancer (BC)2,3, injury to the intercostobrachial nerve (ICBN) during axillary surgery being an important cause. However, not all ICBN injuries lead to CPSNP. The biological mechanisms explaining why similar nerve injuries become painful in some, but not in all, still need to be elucidated.
Development and maintenance of pain following the onset of peripheral nerve lesion is a complex process involving multiple peripheral and central mechanisms4,5. Extensive preclinical and growing clinical evidence shows that neuroimmune interactions play a key role in this process and involve both innate and adaptive immune systems6–8. However, antibodies are less studied in the context of NP, although it has been shown that antibodies and their complexes have the potential to induce nociceptor hyperexcitability through multiple mechanisms9,10. In the context of nerve injury, pre-existing antibodies derived from previous exposures to environmental pathogens11, and autoantibodies, could enhance nociception due to trauma-related events9,10. Also, tissue injury could lead to exposure of hidden antigens and the generation of autoantibodies with pro-nociceptive properties12,13. However, the role of humoral immune responses in painful peripheral nerve injury has largely remained unexplored and clinical studies are scarce.
In this explorative study, we used mimotope variation analysis (MVA)14, a next generation random 12-mer peptide phage display method, for large-scale assessment of the antibody epitope repertoire in CPSNP after BC surgery, which had been complicated by nerve injury. The data generated by MVA immunoprofiling have successfully been used to identify novel antibody epitope biomarkers for vaccination-induced narcolepsy, multiple sclerosis, and COVID-1914–17. We compared the targeted peptide epitope profiles (immunoprofiles) in the clinical samples of patients with persistent painful versus painless surgical injuries to ICBN. In this, our aim was to assess whether specific immunodominant antibody epitopes associate with CPSNP after BC treatments. We analyzed samples from two timepoints: before surgery and at follow-up (4–9 years from surgery). In these settings, we addressed two questions: (1) Do patients with painful and painless surgical nerve injury present with a pre-existing distinct antibody epitope spectrum before nerve injury? (2) Are there CPSNP-associated changes in the immunodominant antibody response that emerges after surgical nerve injury and BC treatment? With this approach, our aim was to increase understanding of the immunological features in the pathophysiology of CPSNP and identify novel biomarkers for CPSNP.

Results

Patient characteristics
We used the same study group that was previously described by Lötsch et al.18 and Mustonen et al.19. We included 57 patients, of whom 27 had CPSNP, with a matched control group of 30 non-CPSNP patients (Fig. 1 and Table S1). The patient characteristics, including the type of cancer and cancer treatments, have been previously reported elsewhere19. At the follow-up visit, patients in the CPSNP group were on average approximately four years younger (mean age 60,3 vs. 64.0 years) than patients in the non-CPSNP group (Fig. 2b).

The groups had no significant differences in surgical and cancer treatments or in cancer type. Mastectomy was performed on 16 (59%) patients in the CPSNP group and 20 (67%) patients in the non-CPSNP group. Most patients (96% in CPSNP group and 83% in non-CPSNP group) had undergone axillary lymph node clearance. Chemotherapy was administered to 23 (85%) patients in the CPSNP group and 25 (83%) patients in non-CPSNP group. Docetaxel was included in the chemotherapy regimen in all but one patient in CPSPN and two patients in non-CPSNP group. Presurgical samples were taken from both groups (Timepoint 1 [T1]); the mean follow-up time was 6.5 years, ranging from 4 to 9 years (Timepoint 2 [T2], Fig. 2a). In the CPSNP group, the median worst pain past week score was 5/10 NRS, with overall pain intensity ranging from 4 to 8/10 NRS. For CPSNP patients, the median score in the Douleur Neuropathique 4 (DN4) screening questionnaire was 6/10 ranging from 3 to 8/10. At the follow-up visit, four patients in the CPSNP group reported on-demand use of paracetamol or NSAID, one patient reported on-demand use of paracetamol-codeine combination drug, and one patient used regular neuropathic pain medication (gabapentinoid and amitriptyline). None of the patients in the non-CPSNP group reported any use of pain medications. None of the patients had received any neuromodulatory pain treatments.

The most abundant antibody epitopes are uniquely individual and consistently stable
To identify the antibody epitope response repertoire, we performed MVA immunoprofiling of serum samples from CPSNP and non-CPSNP groups collected at two timepoints (n = 126; Fig. 2c, Fig. S1). Data analysis when comparing the individual pre-operative and follow-up samples resulted in defined sets of peptides that were highly individual-specific but similar at both timepoints. The cosine similarity index (CSI) was remarkably high on average (0.79) for paired samples (n = 59 comparisons), while CSI was significantly lower on average (0.04) in unpaired samples (Fig. 2d, Fig. S2). These results demonstrated that the seroreactivity against major antibody epitopes remained similar in time in the same subject but differed between subjects.

BC patients with CPSNP have distinct pre-surgery IgG profiles
To define differences in antibody epitope profiles among study groups, we performed an unsupervised clustering analysis of the most abundant (immunodominant) peptides detected by MVA and determined group-specific epitopes (Fig. S1). To further investigate the origin of these immune features, we aligned the epitope sequences (n = 9344) to the peer-reviewed epitopes from the IEDB with exact matching. This annotation analysis identified 1882 epitopes showing high sequence similarity with the reported epitopes of 79 common human pathogens (Table S2). Table S2 shows annotated linear epitopes associated with pathogen antigens. Antibody immune response to these epitopes was detected in both pre-surgery and follow-up (Timepoints T1 and T2) samples from the patients with CPSNP (Fig. 3a). Pairwise data analysis confirmed the persistence of immune response to the selected epitopes over time (Spearman rho and p-values between timepoint samples showed high correlation: ρ = 0.88, p = 1.7*10− 6 (CPSNP) and ρ = 0.80, p = 1.6*10− 6 (non-CPSNP), Fig. 3b). Further clustering analysis resulted in 17 distinct pathogen-associated epitope clusters (Fig. 3c). Annotation analysis indicated that, while most clusters were heterogenous and contained epitopes of different viral antigens (Cluster 4 and 15, for example, Table S2), some clusters were very clearly linked to a specific epitope of a specific pathogen (Fig. 3c). For example, Clusters 5 and 8 were strongly associated with the human cytomegalovirus (CMV) antigen of phosphoprotein 150 (pp150). Clusters 10 and 12 contained epitopes associated with the glycoprotein D (gpD) of human herpes simplex viruses (HSV1 and HSV2). Cluster 2 was more distinct for non-CPSNP than for CPSNP and was enriched in epitopes associated with enteroviruses (Fig. 3c and Table S2). We then quantified the antibody response against these major epitope clusters (Fig. 3c) that were specifically associated with CPSNP. Amongst these, antibody responses against the epitopes of seven viruses emerged as the most discriminating (Tables S2- S3), specifically against antigenic determinants of the polyprotein of human rhinovirus (HRV C3); regulatory protein E2 of human papilloma virus 16 (HPV16); polyprotein of coxsackievirus B3 (CVB3); pp150 of CMV; glycoprotein D of HSV1 and HSV2; capsid protein VP26 of EBV (Fig. 3d). We further hypothesized that some of the identified epitopes related to these common pathogens could be linked to autoimmune mimicry (Table S4). We observed accumulations of immune response targeting extracellular domains of various human protein or secretory protein antigens, as well as those that could directly influence pain signaling. Among others, for example, the epitope of HRV-C3/HPV16 (Cluster 11/Cluster 6) showed mimicry of the epitope of the human collagen alpha-3(VI) chain, a beaded filament collagen shown to have a role in cancer and acute wound healing (UNIPROT ID P12111) (Table S4). These results indicated that a high pathogen load was a dominating feature differentiating patients with or without CPSNP.

Viral seropositivity by ELISA confirm the MVA data
We analyzed overall seropositivity to EBV, CMV, HSV1 and HSV2 and observed that those for EBV and CMV were prevalent across all study groups, whereas seropositivity for HSV1, and particularly HSV2, were less common (Fig. 4a, Table S1). The percentage of HSV2 seropositive individuals was higher in the CPSNP group than in the non-CPSNP or CTRL groups (HSV2: 48.2%/20.0%/33.3%, Fisher’s Exact test p value 0.027) (Fig. 4a). Next, because the annotation analysis of the antibody profiles to IEDB-reviewed epitopes revealed clear links with common human pathogens, we used the overall seropositivity results to validate the findings of MVA. We assessed the potential of seroreactivity to herpesviral antigens (pp150 of CMV, gpD of HSV1, gpD of HSV2) for accurate classification (Fig. 4b). The antibody response to the examined epitopes was significantly higher in the seropositive samples (Wilcoxon Rank Sum test, FDR-adjusted p < 0.01, Fig. 4b). Additionally, the response to gpD antigen of HSV2 was greater in CPSNP samples than in non-CPSNP samples (Wilcoxon Rank Sum p < 0.05, Fig. S3). Finally, an epitope-specific ELISA using peptides corresponding to epitopes of pp150 CMV (892GRGSRVGVPSLKPTLGGKAV911) and gpD HSV2 (368MAPKRLRLPHIRDDD383) independently confirmed that the MVA findings were genuine targets of the IgG response associated with these pathogens and potentially linked to neuroinflammation (Fig. 4c, Fig. S4).

Predictive value of immune load to pathogens for CPSNP in breast cancer patients
The marked difference in seroreactivity to herpesviruses between the CPSNP and non-CPSNP groups led us to explore the potential of using seroreactivity to antigenic epitopes of the seven common viruses, both against individual viruses and collectively for classification. Individually, antibody responses to single epitopes of CMV, EBV, HRV, HPV16, HSV1 and HSV2 were notably higher in pre-surgery samples of CPSNP than non-CPSNP samples (Fig. 5a). Anti-CVB3 was an exception, as the responses against its epitope were lower in CPSNP than in non-CPSNP samples. In post-operative samples collected 4–9 years after surgery, the seroresponse to most epitopes remained similarly associated with CPSNP (Fig. S5).

We then analyzed all seven features to develop a logistic regression model aimed at predicting CPSNP in BC patients. These data showed that the seroresponse to the defined set of viral epitopes of CVB3, EBV, CMV, HPV-16, and HSV-2, presented a novel predictive biomarker that distinguished CPSNP patients from non-CPSNP patients pre-surgery, achieving a balanced accuracy of 84.6% (Fig. 5b). The observed cross-validated AUC was 0.85 and was significantly greater than expected by chance based on a permutation test (p = 0.00089). Baseline clinical characteristics were comparable between the CPSNP and non-CPSNP groups19. The variables demonstrating minor difference between the two groups were age and BMI before (T1) and after surgery (T2). The reported biomarker associations remained stable after age and BMI adjustments (Fig. S6).
This model was validated on an independent control cohort16,17 comprising age-, sex-, and country-matched individuals (n = 28, achieving AUC = 0.806, with a sensitivity of 0.73 and a specificity of 0.75, Fig. S7). Altogether, our findings suggested that immunity load to pathogens may play a significant role in modulating pain signaling and could provide new avenues for therapeutic interventions.

Discussion
In this study, we present a hypothesis-free analysis of antibody epitopes in patients with painful (CPSNP) and painless (non-CPSNP) nerve injury after BC surgery, using MVA immunoprofiling. MVA uncovered highly individualized immunoprofiles in the paired samples across study groups (pre-operative and follow-up), while also identifying shared immunoreactive features. We identified 1882 epitopes that were differently targeted by antibody response in the two study groups, and these epitopes showed high sequence similarity to antigens of common human pathogens. These included epitope mimics of the polyprotein of HRV-C3, regulatory protein E2 of HPV16, phosphoprotein 150 of CMV, capsid protein VP26 of EBV and envelope glycoprotein D of HSV1 and HSV2, whereas epitopes of polyprotein of CVB3 were associated with painless nerve injury. Similar immune responses to the epitopes of these common human viruses were detected in the samples collected before surgery and after the follow-up showing that the findings persisted over time. Five of these specific epitopes (epitopes of CMV, EBV, HSV2, CVB3, and HPV-16) could be potentially clinically useful as preoperative predictors of CPSNP in patients undergoing BC surgery (AUC of 0.9, 95% CI, 0.794–0.963). This suggests that immune load linked to common viruses is associated with an increased risk of developing CPSNP.
The antibody repertoire reflects both the past and current activity of the adaptive immune system. Immune response against previously encountered environmental pathogens has been identified as a potential risk factor for various autoimmune disorders20, as well as in other contexts, including psychiatric disorders21. Recently, serological evidence of multiple infections was associated with multisite chronic pain22. To the best of our knowledge, this is the first study to report an association between antibody responses against common and latent viruses and the risk of developing persistent pain after surgical nerve injury.
Neuropathic pain (NP) is caused by disease or a lesion in the somatosensory system. However, not all nerve lesions become chronically painful. Immune system has a major contribution in the resolution on pain after nerve injury but also in the maladaptive mechanisms that lead to persistent pain23,24. Moreover, cellular immune response to nerve injury is regarded as a possible therapeutic target25. The crosstalk between immune system and NP includes interactions with both innate and adaptive immunity. This process involves almost every immune system component and cell type, including macrophages, neutrophils, T cells, glial cells, cytokines, chemokines and inflammatory mediators7,13,23–25. However, the role of humoral immune mechanisms has been less studied in the context of painful nerve injury, although growing evidence implicates its potential contribution in certain persistent pain conditions9,10,26,27.
Antibodies may exert pronociceptive effects through immune-complex formation, complement activation, and Fc-receptor–mediated sensitization of peripheral nociceptors, providing a potential mechanistic link between adaptive immunity and pain processing9,10. For example, IgG from complex regional pain syndrome (CRPS) and fibromyalgia patients can transfer pain-like hypersensitivity to rodents26,27. Furthermore, in a subset of fibromyalgia patients, autoantibodies bind to satellite glia cells in the dorsal root ganglia (DGR) of the sensory nerves28. In the context of nerve injury, a mouse model showed accumulation of IgG in DRG and dorsal spinal cord with potential pro-nociceptive effect13. A clinical study of spinal cord injury patients showed an antibody-mediated autoimmune response that was associated with a higher need of NP medication12. However, there is a lack of clinical studies concerning the role of antibodies in painful peripheral nerve injury.
Our data suggest an elevated immune response to multiple human herpesviruses as a risk factor for developing persistent NP after breast cancer surgery. Human herpesvirus infections or their post-infection syndromes have been associated with persistent NP in different contexts, with post-herpetic neuralgia the most well-known. Apart from these, the association of latent herpesvirus infections and other NP conditions is less studied. A rare unilateral pain syndrome with NP features has been reported in association with recurrent HSV infections29. Indeed, HSV1 and − 2 remain latent in the ganglia of sensory nerves with the potential of clinical or subclinical reactivation later in life. Although these viruses are known to reside primarily in trigeminal (HSV1) and sacral (HSV2) ganglia, HSV1 can sometimes reside in thoracic ganglia30 suggesting a plausible anatomical link to the CPSNP-related findings in the present study.
EBV infects mainly B cells and epithelial cells, CMV infects monocytes, lymphocytes and epithelial cells. After primary infection the reactivations of CMV and EBV are usually asymptomatic. Previously, seropositivity to EBV has been associated with chronic multisite musculoskeletal pain and inflammatory joint pain22,31. Other chronic pains are common comorbidities in NP, and they associate with risk of developing CPSNP after surgery32,33. High antibody titers to both EBV and CMV have been associated with elevated levels of other inflammatory mediators such as CRP and IL-6 indicating that the activity of multiple herpesviruses can drive general inflammation34. Furthermore, general inflammation and higher levels of pro-inflammatory cytokines have been associated with NP in different contexts in both pre-clinical and clinical studies6,7,23. This is in line with our data showing elevated antibody response to multiple herpesviruses in CPSNP. High antibody levels to latent herpesviruses are indirect indicators of viral reactivations and they also serve as an indicator of immune dysregulation, due to their associations with elevated levels of inflammation34,35. The compromised immune function which predisposes individuals to recurrent and prolonged herpesviral reactivation specifically could be the underlying factor for the elevated risk of CPSNP following nerve injury36. However, our study does not provide direct information about immune function, inflammation, or herpesvirus reactivations and further studies are needed to confirm this hypothesis.
Notably, compared with the CPSNP group, the non-CPSNP group patients exhibited an elevated IgG response targeting epitopes that mimic the polyprotein of CVB3, a common human enterovirus causing flu-like symptoms and sometimes myocarditis. It remains unclear how antibody response to CBV3 could associate with favorable outcome in terms of pain development after surgical nerve injury. The clinical relevance of this finding needs further evidence. Interestingly, in a recent MVA study, robust antibody response to a conserved enteroviral epitope was associated with better cardiovascular health17.
Our modelling results suggest that antibody response to specific antigen epitopes of CMV, EBV, HSV2, CVB3, and HPV-16 prior to surgery could be useful predictors of CPSNP in patients undergoing BC surgery. There are several mechanisms that could be hypothesized. Firstly, antibodies could be nociceptive per se, like in CRPS and fibromyalgia. Secondly, antibodies may jointly induce general inflammation34, which interferes with normal nerve regeneration and leads to pain sensitization24,36. Thirdly, antibodies may be markers of compromised immune functions, which would be the biological cause of lack of regeneration and pain. Lastly, reverse causality is also worthy of consideration, because chronic NP is associated with increased stress and psychological burden32,37 and stress is known to induce reactivations of herpesviruses. All in all, our findings support the hypothesis that prior infections and viral re-activations potentially prime neuroimmune pathways that facilitate the development and persistence of NP following surgical injury. Since our study does not provide direct information about the underlying mechanisms, further research is necessary to clarify the role of antibody responses in NP development.
Strengths of this study include a prospective setting of thoroughly characterized study groups of painful and painless nerve injury patients after BC surgery. Our data is unique, because a similar setting is very difficult to obtain in most other NP etiologies where the onset of nerve lesion and NP are usually unpredictable. Another strength of this study is its comprehensive immune response analysis. The interaction between the immune system, inflammation, and pain is intricate. MVA allowed us to screen a vast set of peptides, accelerating the discovery process of antigens associated with chronic pain at the epitope level. However, our study does not provide direct information on active infection, pathogen load, or viral reactivation. The patient cohort is small, but the patient groups were carefully characterized with a detailed assessment of NP along with preoperative and follow-up samples with timespan up to nine years. These together add considerable depth to our study. Our results may not be directly generalizable to men and to other NP etiologies, because our study groups comprised of BC treated women with surgical nerve injury.
Our analysis demonstrated that the antibody response profiles in BC patients experiencing CPSNP showed reactivity to multiple viral antigens. These immune response patterns may be candidates as predictive markers assessing the risk of pain development during the preoperative stage. This finding could pave ways to new therapeutic opportunities, such as use of antiviral drugs, for the prevention and management of neuropathic pain. Since our results from clinical patients are preliminary, we encourage further studies to elucidate the multidisciplinary mechanisms and potential treatment avenues in persistent NP.

Patients and methods

Patient selection
We have previously described the diagnosis of NP in 251 women whose treatment for BC involved axillary surgery and ICBN resection (as reported by the surgeons). We examined the patients for the presence or absence of CPSNP of the ICBN 4–9 years after surgery32. All patients were recruited from a previous prospective cohort of 1000 women operated on for BC at the Helsinki University Hospital during 2006–201038. Thus, we were able to use data and samples collected preoperatively from the patients in the current study. During surgery, the operating surgeon reported the handling of ICBN (totally resected, partially resected, spared or not identified). Patients with total or partial ICBN resection were invited for the follow-up research visits for the current study. Follow-up research visits took place during 2014–2016, 4–9 years after surgery32. The visits included thorough clinical sensory examination by a neurologist (HH) for diagnosis of NP according to stepwise grading criteria (unlikely, possible, probable, definite NP)39, and collection of blood samples. The contents of sensory examination and NP grading have previously been reported in detail elsewhere32. To fulfill the criteria for definite NP, the pain and sensory alterations had to be located within the area of ICBN innervation and the ICBN resection had to be confirmed by the operating surgeon. Because of the surgeon’s report of nerve lesion, no additional diagnostic tests are needed to fulfill the criteria of definite NP39. We used the “worst pain past week” in the Brief Pain Inventory (BPI) long form to measure pain intensity on a Numerical Rating Scale (NRS, 0–10). NRS ≥ 4/10 was considered to reflect at least moderate pain intensity and clinically meaningful pain40. All patients were asked for the presence of pain in other body locations (back, neck, joints, head, other) and use of current pain medications. In addition, Douleur Neuropathique 4 (DN4) questionnaire41 was completed for each patient with pain.
For the current study assessing the role of antibody response profiles in CPSNP, we selected subgroups using the following criteria: (1) the CPSNP group included patients fulfilling the criteria of definite NP39 and pain intensity ≥ 4/10 on the NRS and (2) the non-CPSNP group included patients with ICBN resection with no CPSNP and no other pains. Patients with ongoing cancer treatment, neurological or autoimmune diseases were excluded from the present analysis. Patient selection is presented in Fig. 1. For technical control purposes, preoperative and follow-up samples from six BC-operated patients without ICBN injury and without pain were included in the MVA studies. These patients were from the same original cohort as the analyzed groups32.
The Coordinating Ethics Board of the Helsinki and Uusimaa Hospital District approved the study (149/13/03/00/14) which was also registered in ClinicalTrials.gov (NCT02487524). This study was performed in accordance with the Declaration of Helsinki. All patients gave written informed consent.

Breast cancer treatments
The cancer treatment data were accessible from the previous study38. All patients had been operated on for unilateral BC, and none had received neoadjuvant treatment. Breast surgery was either mastectomy or breast conserving surgery (BCS) accompanied with axillary surgery, which was either sentinel lymph node biopsy (SLNB) or axillary lymph node dissection (ALND). The resection of the ICBN (either total or partial) was verified by the operating surgeon. Oncological treatments, including radiotherapy, chemotherapy, and endocrine treatment, were administered according to national treatment protocols. The chemotherapy regimen comprised a combination of docetaxel and CEF (cyclophosphamide, epirubicine, and 5-fluorouracil). Trastuzumab was administered to the patients with HER2 (human epidermal growth factor 2) -positive BC. Tamoxifen (premenopausal women) or aromatase inhibitor (postmenopausal women) comprised the endocrine treatment.

Sample collection
Plasma samples for the MVA were collected at two separate timepoints: at induction of anesthesia for the BC surgery (Timepoint 1 [T1]) and at the research visit 4–9 years after surgery (Timepoint 2 [T2]). Plasma from blood samples collected into ethylenediaminetetra-acetic acid (EDTA) tubes was extracted by centrifugation at 3000 revolutions per minute (RPM) for 10 min, and transferred to cryogenic vials, which were immediately frozen and stored at −80 °C. Both the preinjury and follow-up samples were drawn and prepared by the same research nurse. The follow-up research visit protocol included the analysis of basic laboratory parameters and inflammatory markers, including high-sensitivity C-reactive protein (hs-CRP). The collection of study samples has been reported previously32.

Mimotope variation analysis
Sample (n = 126) processing by MVA was performed according to standard MVA protocols14. In brief, plasma samples (n = 126, 2 µl of plasma per analysis) were incubated with Escherichia coli M13 phage library displaying random 12-mer peptides (complexity 1 × 109 peptides; NEB, E8111L). IgG–phage complexes were isolated with protein G-coated magnetic beads (NEB, S1430) and amplified by Phusion High Fidelity PCR (Thermo Scientific) for next-generation sequencing. An average of 3 million peptide-encoding DNA sequences per sample was achieved: 5000 most abundantly detected peptides by abundance values (read counts) from each sample were taken into immunoprofile-similarity analysis. For comparisons, the normalized scalar products of peptide count vectors were calculated for the cosine similarity index (R package lsa) that ultimately resulted in a 126 × 126 sample immunoprofile-similarity matrix. Due to not meeting technical criteria, four samples (two controls, one CPSNP and one non-CPSNP), were excluded from downstream group comparison and regression analyses (Table S1).

Delineating cohort specific epitope responses

Cohort-specific epitope selection
To delineate cohort-specific antibody response, the data comparison of the most immunodominant epitopes between CPSNP and non-CPSNP groups was carried out (see schematic overview of the data analysis workflow, Fig. S1). The most immunodominant epitopes were determined by clustering analysis (defining 5–11 positions-long epitopes that all contained five fixed amino acids) and hypergeometric test analysis where SPEXS2 Software was used14. Samples of these two groups at either timepoint (T1, before surgery; T2, at the 4–9 years follow-up) were compared with each other using optimal cut-point analysis (R package cutpointr, maximizing the Youden metric) and Student’s t-test (R Stats Package, log-transformed). Based on the group comparisons, 9344 epitopes were chosen for further analysis (t-test p < 0.05 and Youden > 0.25).

Alignment analysis on infectious epitopes of IEDB and marker epitope selection
To look for potential mimicry with human and pathogen antigens, group-specific (n = 9344) epitopes were exactly aligned with sequences of T and B cell peer-reviewed epitopes in a public database (Immune Epitope Database [IEDB], version: epitope_full_v3.tsv, 2,225,965 epitopes of which 2,214,634 were linear, date accessed: 15.04.2024) (see schematic overview of data analysis workflow, Fig. S1). Based on the relevance (determined by assessed probability of pathogen exposure) and alignment scores, the most relevant microorganisms were selected (n = 79, Table S2). To evaluate the statistical significance of specific epitope–motif alignments, we performed permutation-based tests for each aligned epitope–motif pair (n = 6254). For each pair, the epitope sequence was held constant while the amino acid positions of the matched motif were randomly shuffled 10,000 times to generate permuted motif variants preserving amino acid composition. Each permuted motif was aligned to the original epitope sequence, and the number of exact matches was recorded. The empirical p-value was calculated as: (number of permuted matches + 1)/(number of permutations + 1) and p-values were adjusted with FDR. 1882 epitopes that showed sequence similarity with the revised epitopes from IEDB, were visualized and clustered using R packages Rtsne (where t-SNE is stochastic neighbor embedding)42–44 and dbscan45. For clustering analysis, the relative values of immune response to the 1882 epitopes in T1 samples (n = 63) were log-transformed. Epitope clusters were identified using the Rtsne function from R’s Rtsne package (dims = 2, perplexity = 15, max_iter = 500) and the dbscan function from the dbscan package (epsilon = 2, minPts = 10). Altogether, 17 different clusters were identified (Table S2). Further, the most relevant epitopes and antigens identified by the study cohort were described by alignment scores (Table S3) where the seven most interesting leads (with the highest alignment scores) were selected (see schematic overview of data analysis workflow, Fig. S1). The total immune response value for each epitope was calculated by summarizing the counts of all unique peptides containing the aligned motifs and log-transformed per sample (Wilcoxon Rank Sum test, p-values unadjusted). Following regression and receiver operating characteristic (ROC) analyses, corresponding visualizations were created using MedCalc Statistical Software (version 17.0.4, MedCalc Software Bvba, Belgium) (see schematic overview of data analysis workflow, Fig. S1). For independent validation (regression and ROC analyses), previously published immunoprofiles were used for country- (Finland), age- (52.9 ± 6.21 years) and sex- (female) matched clinical samples (n = 26) with no diagnoses of BC and NP pain16.

Alignment analysis on autoantigenic epitopes of IEDB
Autoimmune-associated epitopes were filtered from the original IEDB database (version: epitope_full_v3.tsv, export on 15.04.2024) alignment with 1882 queried epitopes (see schematic overview of data analysis workflow, Fig. S1). All autoimmune-associated peer-reviewed B cell epitopes containing 82 infection-related epitope mimics were included in the study (Table S4). For the antigens identified, cellular locations were gained from the UNIPROT database (column intracellular/extracellular, column features – cytoplasm, endoplasmic reticulum, cell membrane, secreted, cell junction, cell projection, chromosome, Golgi apparatus, lysosome, mitochondrion, nucleus, recycling endosome, NA – data not available) (Table S4).

Epitope validation
For epitope validation, in-house epitope-specific enzyme-linked immunosorbent assay (ELISA) was used to analyze randomly selected samples with proportional seronegative and -positive sample representation (human cytomegalovirus [CMV] pp150, n = 23; herpes simples virus 2 [HSV2] gpD, n = 23). Peptides containing target epitopes identified by MVA (for CMV pp150 901SLKPTLGGK910, for HSV2 gpD 370PKRLRLPHIRDD383) and control epitope (1AASAASAA8) were synthesized with C-terminal addition (GGGDYKDDDKK(biotin), CASLO ApS). Nunc Immobilizer Streptavidin-coated 96-well plates (Thermo Scientific, 436015) were coated with 0.03 µg/ml (3 ng/well) peptides in 1xPBS for 1 h at room temperature and washed with 0.1% PBS-Tween. After blocking unspecific binding with 5% BSA for 1 h and washing with 0.1% PBS-Tween, diluted plasma samples (plasma dilution 1:1000) in 1% BSA 0.1% PBS-Tween were allowed to bind to peptides overnight at + 4 °C. Next day, the plate was washed with 0.1% PBS-Tween and incubated with secondary antibodies (polyclonal anti-human-HRP (1:1000; goat anti-human IgG (H + L) (Invitrogen, 31410) for 1 h at room temperature. Finally, the plate was washed with 1x phosphate-buffered saline (PBS), and 20x dilution of chemiluminescent substrate (SuperSignal™ ELISA Femto Substrate (Thermo Scientific, 37075)) was added. Results were measured within 5 min using a multimodal microplate reader (Hidex Oy). The signal intensities were calculated as Signal/Background ratio (relative light unit [RLU]) (Wilcoxon Rank Sum test p-values, Pearson correlation coefficients).

EBV, CMV, HSV1 and HSV2 seropositivity
Commercial ELISA tests were used to determine the serology of 59 samples (CTRL [control, n = 4], CPSNP [n = 26], non-CPSNP [n = 29]) to herpesviruses Ebstein-Barr virus (EBV; Anti-EBV-CA IgG ELISA, 2791–9601 G), CMV (Anti-CMV, EI 2570–9601 G), HSV1 (Anti-HSV1-gG1, EI2531-9601-2G), HSV2 (Anti-HSV2-gG2, EI2532-9601-2G) by analyzing samples from T1 (Table S1). Analyses were carried out in accordance with the manufacturers’ specifications. Absorbency was measured at 450 nm multimodal microplate reader (Hidex Oy).

Statistics
Statistical analyses were conducted with R statistical programming language (v.4.2.3) and RStudio environment (v. 2024.09.1 Build 394). Data were analyzed, graphs were produced and visualized using R packages tidyverse46, ggpubr47, ggsci48 and scales49. Cosine similarity indices (CSIs) for sample comparisons based on the top 5000 peptide abundance values and composition were calculated with the cosine function in R package lsa50. Boxplots were generated using the style of Tukey with R packages ggpubr and ggplot251. In figures, the upper, middle and lower boxplot lines represent the 75th, 50th, and 25th percentiles, while whiskers represent the largest or smallest value within 1.5x interquartile range above the 75th percentile or below the 25th percentile, respectively. The p-values of Wilcoxon Rank Sum tests were visualized with R ggpubr or ggplot2 packages, and statistical significance is shown where applicable. Student’s t-tests (base R Stats Package), along with each epitope’s Youden index calculation analysis (R package cutpointr52, maximizing the Youden metric), were used to identify 9344 unique group-differentiating features. MedCalc® Statistical Software (v.20.121.2, www.medcalc.org; 2022) was used to conduct logistic regression and ROC analyses of five pathogen epitopes were used as combinational tests. The proportion of seropositive and -negative samples was evaluated with Fisher’s Exact Test from the R Stats Package53. Spearman’s ρ statistic for epitope-specific ELISA and MVA data correlation were calculated with cor.stat from the R Stats Package. Age adjusted logistic regression, ROC/AUC and cross-validation calculations were performed in R with glm() function from “stats”, “pROC”54 and “caret”55 packages. Model performance was assessed using 5-fold cross-validation. Statistical significance was evaluated using a permutation test in which outcome labels were randomly shuffled 10,000 times, generating a null distribution of cross-validated AUC values. The permutation p-value was calculated as the proportion of permuted AUCs greater than or equal to the observed AUC obtained from the unpermuted data.

Supplementary Information
Below is the link to the electronic supplementary material.