Autoantigen profiling reveals a shared post-COVID signature in fully recovered and Long COVID patients

Some individuals do not return to baseline health following SARS-CoV-2 infection, leading to a condition known as Long COVID. The underlying pathophysiology of Long COVID remains unknown. Given that autoantibodies have been found to play a role in severity of COVID infection and certain other post-COVID sequelae, their potential role in Long COVID is important to investigate. Here we apply a well-established, unbiased, proteome-wide autoantibody detection technology (PhIP-Seq) to a robustly phenotyped cohort of 121 individuals with Long COVID, 64 individuals with prior COVID-19 who reported full recovery, and 57 pre-COVID controls. While a distinct autoreactive signature was detected which separates individuals with prior COVID infection from those never exposed to COVID, we did not detect patterns of autoreactivity that separate individuals with Long COVID relative to individuals fully recovered from SARS-CoV-2 infection. These data suggest that there are robust alterations in autoreactive antibody profiles due to infection; however, no association of autoreactive antibodies and Long COVID was apparent by this assay.

consistent with other post-acute sequelae of COVID such as multisystem inflammatory syndrome 74 in children (MIS-C), which also has no association with anti-interferon antibodies (19). 75

76
The identification of previously reported autoantibodies can be performed using targeted assays. 77 However, identifying the full range of novel autoreactive antibodies, whether they are pathological 78 or not, requires technologies capable of high-throughput, unbiased, proteome-wide screens. In 79 this study, we screened a cohort of individuals with prior SARS-CoV-2 infection, many of whom 80 met clinical criteria for Long COVID, to determine whether a consistent pattern of autoreactivity 81 could be identified. This same technology has been previously utilized to discover novel 82 autoantibodies in a wide range of disease contexts (20-23). 83 86 We employed a previously published proteome-wide approach using a T7 phage-display assay 87 with immunoprecipitation and next-generation sequencing (PhiP-Seq) (20-24). We tested sera 88 from 185 otherwise healthy individuals with prior SARS-CoV-2 infection in parallel to sera from 57 89 otherwise healthy individuals collected prior to the known existence of COVID-19 (pre-COVID). 90 Using an unbiased analysis, we identified a distinct pattern of autoreactivity which effectively 91 classifies individuals with prior SARS-CoV-2 infection from individuals not yet exposed to the virus 92 with a logistic regression AUC of 0.90 ( Figure 1A). The protein targets from which these enriched 93 immunoprecipitated peptides were derived are widely varied and lack any apparent shared 94 biological functions or cell type ( Figure 1B). Among the identified targets, a single peptide derived 95 from ARHGAP31 (Figure 2A) displayed the greatest amount of enrichment with 22% of individuals 96 with prior SARS-CoV-2 infection yielding enrichment greater than 6 standard deviations of the 97 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted February 9, 2023. ; https://doi.org/10.1101/2023.02.06.23285532 doi: medRxiv preprint mean of the pre-COVID controls. No difference in enrichment was observed for those diagnosed 98 with Long COVID with respect to Post-COVID ( Figure 2B). with at least 5-fold greater than background (defined as fold-change over mock-IP with protein 119 A/G beads) were compared (Figure 3). Seventeen of the 20 enriched proteins were present in 120 both Long COVID and Convalescent COVID, and none of these enrichments was observed in any 121 of the pre-COVID controls. Overall, there was no significant differences in enrichment between 122 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 9, 2023. ; https://doi.org/10.1101/2023.02.06.23285532 doi: medRxiv preprint Long COVID and Convalescent COVID. Peptides derived from three proteins, TMED10, FUCA1, 123 and POL2RK, were only observed in Long COVID, however these were infrequently observed.  Figure 4A). However, using a strict cutoff of 6 standard deviations above the pre-141 COVID controls to determine positivity, none of these autoantibodies was phenotype specific. By 142 looking at the distribution of these antibodies in Long COVID patients with given sub-phenotype 143 relative to the remaining Long COVID patients and all convalescent COVID patients, it was 144 apparent that the statistical significance in a particular phenotype was driven by either a single 145 individual with extremely high autoantibody signal in the case of KDM3B and TTF2, or higher 146 group signal but not meeting the positive threshold cutoff in the case of FUCA1 ( Figure 4B). 147 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Autoimmunity has been proposed as one potential mechanism driving Long COVID. We applied 162 an unbiased, proteome-wide, validated approach to assess associations between antibody 163 autoreactivity and clinical phenotype. A clear and robust difference in autoreactivity was detected 164 between those infected with SARS-CoV-2 and pre-COVID controls. This difference was 165 constituted by peptides from diverse and varied proteins, most of which are intracellular, 166 suggesting that the origin of the differential enrichment is due to cross-reactivity with SARS-CoV-2 167 directed antibodies in those that were exposed. A sequence comparison between a peptide from 168 the most enriched protein, ARHGAP31, and Orf1a of SARS-CoV-2 supports this notion, but 169 orthogonal validation through fine-scale epitope mapping and antibody cloning would be required 170 to demonstrate this conclusively. While the clinical significance of incidental autoreactivity due to 171 the humoral immune response to SARS-CoV-2 remains largely unknown, prior studies have 172 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted February 9, 2023.  Table 1). Long COVID was defined using study instruments, which have been described in detail We defined 6 groups of symptoms (symptom phenotypes) based on either organ system cluster. depression" on the PHQ-8 (score higher than 10) and "moderate anxiety" on the GAD-7 (score 244 higher than 9) were compared to all participants with scores indicating less severe classifications 245 than "moderate depression" and "moderate anxiety" in the following groups: (1) all participants 246 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted February 9, 2023. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted February 9, 2023. ; https://doi.org/10.1101/2023.02.06.23285532 doi: medRxiv preprint were considered enriched for an antibody, and samples with a FC of 6 standard deviations above 272 the mean of pre-COVID controls were considered positive for an autoantibody. FC values were 273 also used to calculate z-scores for each disease category sample by using each respective control 274 (as specified in figures and results), and for each control sample by using all remaining controls. 275 These z-scores were used for the logistic-regression feature weighting. In the case peptide-level 276 analysis, raw reads were normalized by calculating the number of reads per 100,000 reads. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted February 9, 2023. ; https://doi.org/10.1101/2023.02.06.23285532 doi: medRxiv preprint   is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted February 9, 2023.  Fold Change>Mock-IP <5 >20 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 9, 2023. ; https://doi.org/10.1101/2023.02.06.23285532 doi: medRxiv preprint