key: cord-102502-hroxg2u1
authors: Megremis, Spyridon; Walker, Thomas D. J.; He, Xiaotong; O’Sullivan, James; Ollier, William E.R.; Chinoy, Hector; Pendleton, Neil; Payton, Antony; Hampson, Lynne; Hampson, Ian; Lamb, Janine A.
title: Microbial and autoantibody immunogenic repertoires in TIF1γ autoantibody positive dermatomyositis
date: 2020-03-26
journal: bioRxiv
DOI: 10.1101/2020.03.25.007534
sha: 
doc_id: 102502
cord_uid: hroxg2u1

We investigate the accumulated microbial and autoantigen antibody repertoire in adult-onset dermatomyositis patients sero-positive for TIF1γ (TRIM33) autoantibodies. We use an untargeted high-throughput approach which combines immunoglobulin disease-specific epitope-enrichment and identification of microbial and human antigens. Increased microbial diversity was observed in dermatomyositis. Viruses were over-represented and species of the Poxviridae family were significantly enriched. The autoantibodies identified recognised a large portion of the human proteome, including interferon regulated proteins; these proteins were clustered in specific biological processes. Apart from TRIM33, autoantibodies against eleven further TRIM proteins, including TRIM21, were identified. Some of these TRIM proteins shared epitope homology with specific viral species including poxviruses. Our data suggest antibody accumulation in dermatomyositis against an expanded diversity of microbial and human proteins and evidence of non-random targeting of specific signalling pathways. Our findings indicate that molecular mimicry and epitope spreading events may play a significant role in the pathogenesis of dermatomyositis.

The idiopathic inflammatory myopathies (IIM) are a heterogeneous spectrum of rare autoimmune musculoskeletal diseases characterised clinically by muscle weakness and systemic organ involvement. IIM are thought to result from immune activation following environmental exposures in genetically susceptible individuals. Viral and bacterial infections have been reported in individuals with IIM (Miller et al., 2018) , but their role in disease pathology is unclear.

Autoantibodies are a key feature of IIM, in common with other autoimmune rheumatic diseases such as rheumatoid arthritis, systemic lupus erythematosus and systemic sclerosis. Myositis-specific autoantibodies, present in approximately 60%-70% of individuals with IIM, are directed against a range of cytoplasmic or nuclear components involved in key intracellular processes, including protein synthesis and chromatin re-modelling . These myositis-specific autoantibodies are often associated with particular clinical features. Individuals with autoantibodies targeting the cytoplasmic nucleic acid sensor MDA5 (also called interferon induced with helicase C domain 1, IFIH1) can present with rapidly progressive interstitial lung disease which is associated with high mortality (Abe et al., 2017; Betteridge et al., 2019) . Several members of the tripartite motif (TRIM) protein family are known autoantibody targets in IIM, and there is a strong temporal association between adult-onset dermatomyositis and malignancy onset in individuals with antibodies to transcription intermediary factor 1γ (TIF1γ, TRIM33) (Oldroyd et al., 2019) . Many TRIM proteins are important immune regulators (van Tol et al., 2017; Versteeg et al., 2014) , and dysregulation of TRIM proteins leading to reduced ability to restrict viral infection has been reported in several autoimmune diseases, including systemic lupus erythematosus and inflammatory bowel disease (Oke et al., 2009; Zhou et al., 2018) .

The inducible type I interferon (T1-IFN) cytokine system, part of the innate immune response, also plays a role in autoimmune rheumatic diseases including IIM (Muskardin and Niewold, 2018) . The T1-IFN antiviral response is initiated when pathogen-associated molecular patterns are recognized by host pattern recognition receptors and cytosolic receptors for viral nucleic acid. This broad viral recognition process triggers downstream signalling pathways which lead to interferon transcription, protein production and expression of interferon-induced genes; this further enhances the anti-viral machinery (Muskardin and Niewold, 2018) . Such a response is critical for host protection against pathogen expansion, and limits infection during the window of time needed to mount an effective specific (adaptive) immune response.

Improved understanding of IIM pathogenesis is required to improve both patient stratification and disease management. In adult-onset dermatomyositis, we propose the following potential molecular mechanism of disease pathogenesis: anti-TIF1 autoantibodies reduce ability to restrict viral infection, which leads to either an increased susceptibility to a wider diversity of viral pathogens or increased exposure to specific anti-TIF1-related viruses. To test this hypothesis, we applied a novel highresolution and high-throughput comparative screening pipeline (Serum Antibody Repertoire Analysis, SARA) [manuscript in preparation] to both anti-TIF1 autoantibody-positive dermatomyositis patients and matched healthy control plasma. Using this approach, high-throughput antigen epitope-sequencing was integrated with bioinformatic modules to de-convolute accumulated immunogenic responses against the total microbial 'exposome' (including viruses, bacteria, archaea and fungi) and human proteins. We report the identification of disease-specific microbial and human protein epitopes which have clinical and aetiological relevance to anti-TIF1 autoantibody-positive dermatomyositis.

To describe the accumulated antibodies present in dermatomyositis we use the SARA pipeline which integrates an Escherichia coli FliTrxTM random 12 amino acid (AA) peptide display system with epitope signature enrichment through competitive bio-panning and high-throughput DNA sequencing ( Figure 1 ). Competitive bio-panning was applied to pooled total immunoglobulin fractions (IgA, IgG, IgM) purified from the plasma of twenty anti-TIF1 positive adult-onset dermatomyositis patients (DM) and twenty healthy controls (HC) ( Table S1 ). Four sample pools were generated and paired: the first pair contained 10 pooled samples (P10) of DM (DM P10) used for competitive biopanning against 10 pooled samples of HC (HC P10) ( Table S1 ). The second pair included 20 pooled samples (P20) of DM (DM P20) used for competitive biopanning against 20 pooled samples of HC (HC P20) (Table S1 ). Figure 1 and figure S1 detail achieved metrics while defining the microbial and autoantibody immunogenic repertoires in DM and HC. We retrieved ≈36 million (DM) and ≈24 million (HC) next generation sequencing (NGS) reads which represent 8.7 million (DM) and 4.7 million (HC) expressed epitopes. Cohorts presented highly enriched epitope sequences that are unique to DM (15, 522) and HC (4,817) respectively. Our epitope cohorts retrieved 6.75 million (DM) and 2.25 million (HC) microbial or human protein annotations respectively which mapped to 9,111 (DM) and 4,994 (HC) Distinct species. 6,202 Unique species were identified in DM while 2,085 Unique species were identified in HC.

In the DM group, linear epitopes were identified for a total of 1,560 microbial species ( Figure S2A) compared to 1,498 in the HC group ( Figure S2B ). In both groups the highest level of richness (number of different species) was observed for bacteria, followed by viruses, fungi and archaea (Figure S2A & S2B) . We defined distinct epitopes as being any discrete epitope sequence retrieved from the DM and HC groups. Unique distinct epitopes represent any discrete epitope sequences which were observed exclusively in either DM or HC. Common distinct epitopes were discrete epitope sequences that were observed in both DM and HC groups by sequence and at a minimum of five-fold the NGS enrichment frequency in one group vs the other group. Unique distinct epitopes represented the vast majority of sequences observed (Fig 1) . Based on the distribution of the corresponding microbial species in the pools of 10 and 20 the majority of distinct species are shared between the P20 and P10 samples, whereas the majority of unique species are observed either in P20 or in P10, in both DM and HC ( Figure S2C & S2D, respectively). These data demonstrate the presence of a stably-enriched microbial component in DM ( Figure S2A ) and also in HC ( Figure S2B ).

The increase in the number of plasma samples within the pool used for cross-panning, from a pool of 10 (P10) to a pool of 20 (P20), had a differential effect between the two groups: in DM, an increase in the number of microbial species was observed from 728 (DM P10) to 832 (DM P20), whereas in the HC group a decrease was observed from 780 (HC P10) to 718 (HC P20) species. The number of epitopes per microbial species in both the DM and the HC significantly increased with increasing pooling size ( Figure 2A ). The number of NGS reads per microbial species significantly increased in the DM P20 (95% CI: 3.97-4.13) compared to DM P10 (95% CI: 3.90-3.98), whereas it did not change in the HC P20 (95% CI: 4.09-4.26) compared to the P10 (95% CI: 4.34-4.43) ( Figure 2B ).

The number of distinct epitopes per microbe (epitopes present in both groups at minimum five-fold enrichment) significantly increased in the P20s compared to P10s in both groups ( Figure 2C ); (n=432, 95%CI: 0.86-0.94) to (n=455, 95%CI: 2.02-2.10) in DM and 432 (n=432, 95%CI: 1.29-1.37) to 448 (n=448, 95%CI: 1.60-1.68) in HC. The unique microbial epitopes which were identified only in one of the two groups significantly increased in DM P20 (n=377, 95%CI: 0.69-0.77) compared to DM P10 (n=296, 95%CI: 0.01-0.03) and decreased in HC P20 (n=270, 95%CI: 0.18-0.23) compared to HC P10 (n=348, 95%CI: 0.31-0.36) ( Figure 2E ). Collectively, these data indicate an expansion of the DMspecific microbial diversity with increasing DM plasma sample size, whereas the opposite effect was observed in HC.

The number of enriched NGS reads per distinct microbial species significantly increased in DM P20 (95%CI: 4.83-4.92) compared to DM P10 (95%CI: 4.27-4.36), while no significant difference was observed in HC P20 (95%CI: 4.89-4.98) compared to HC P10 (95%CI: 4.80-4.89) ( Figure 2D ). The number of sequencing reads per unique microbial species did not differ significantly between DM P20 and DM P10 (adjusted p value >0.999), however it significantly decreased in the HC P20 (95%CI: 2.81-3.04) compared to the HC P10 (95%CI: 3.78-3.85) ( Figure 2F ). Overall, the differential effect of increased sample size on the observed microbial exposure between DM and HC suggests there is a higher biological variability in DM than in HC. DM P20 exhibited a significantly higher number of microbial AA epitopes per microbial species compared to the HC P20 ( Figure 2A ) and a higher number of total identified microbial species (832 versus 718). The same difference was observed in both distinct ( Figure 2C ) and unique epitope sequences ( Figure 2E ). Overall, we demonstrate that plasma from anti-TIF1 DM patients contain a higher number of microbial epitopes per species and against a wider microbial repertoire.

We focused our further analysis on the DM P20 and HC P20 subgroups. To stratify the microbial species based on their potential significance in DM, we first normalised the number of NGS reads against the total number of epitopes per microbial species (NGSRe-norm) ( Figure S3A -S3D), and studied the relative ranking of viruses and cellular microbes ( Figure 3A & 3B) . Secondly, we evaluated the mean NGSRe-norm at the viral family taxonomy level ( Figure 4D & 4E) , and, thirdly, we recorded the total number of species contributing to each family ( Figure 4D & 4E) (Table S2 ). Thus, we tested for potential high taxonomy-level organisation of microbes in DM. Ranking the microbial species based on decreasing NSGRe-norm we observed that viruses were over-represented in the top 10% of dominant microbial species relative to the total number of viral species present in both DM and HC ( Figure 3A & 3B). Specifically, 18.47% and 17.93% of viral species were present in the top 10% of microbes in the DM P20 and HC P20, respectively ( Figure The HC viral IgOme profile contained a higher proportion of total double stranded DNA (dsDNA) (NGSRe-norm 46.35% vs. 39.68%) and single stranded RNA (ssRNA) (NGSRe-norm 43.70% vs. 35.85%) than DM ( Figure 4A & 4B). The DM viral IgOme contained a higher proportion of ssDNA viruses (NGSRe-norm 15.95% vs. 3.10%) and RNA reverse-transcribing retroviruses (NGSRe-norm 7.16% vs.

3.19%) ( Figure 4A & 4B) . To test whether these observations were due to a generalised differentiating signal amongst viral species within each viral group we compared the species-specific NGSRe-norm per viral category between DM and HC ( Figure 4C ). The dsDNA and ssRNA NGSRe-norm were elevated in the HC plasma compared to DM, whereas no significant change was observed in ssDNA viruses or RNA reverse-transcribing viruses ( Figure 4C ).

Overall, 47.9% (23 out of 48) of viral families were represented in both groups. We compared the viral IgOme at the family level, and we recorded the mean NGSRe-norm with increasing enrichment for each viral family (Figure 4D & 4E) . In DM, the five richest viral families were Coronaviridae (20 species), Geminiviridae (16 species), Herpesviridae (13 species), Orthomyxoviridae (9 species) and Poxviridae (8 species) ( Figure 4D ) (Table S2 ). Of these, Geminiviridae infect plants, whereas all other viruses can infect and cause disease in humans. In the HC, Picornaviridae (15 species) was the most enriched virus group followed by Caliciviridae (11 species), Orthomyxoviridae (9 species), Coronaviridae (9 species), and Retroviridae (8 species) ( Figure 4E ) (Table S2) .

We evaluated the mean NGSRe-norm for each family. In DM, Nairoviridae (n=1), Poxviridae (n=8), Secoviridae (natural host: plants) (n=1), Caliciviridae (n=4) and Adenoviridae (n=1) were the top 5 ranked families (Table S3) . Poxviridae was the one family that ranked highly regarding both the NGSRe-norm (Table S3 ) and the family richness (n=8, 95%CI: 2.98-3.61) ( Table S2 ). All 8 pox viruses had high NSGRe-norm and were amongst the dominant DM viral species ( Figure 4F ). Specifically, Variola virus had the highest NSGRe-norm amongst all identified viral species ( Figure 4F ). In the HC P20, Polyomaviridae (n=1), Iflaviridae (n=1), Podoviridae (n=1), Tymoviridae (n=2), and Myoviridae (n=3) were the viral families with the highest mean NGSRe-norm. Podoviridae and Myoviridae are prokaryotic viruses infecting bacteria, Tymoviridae infect plants and Iflaviridae infect insects (ViralZone root-ExPASy) (Hulo et al., 2011) . We also used Z transformation of NGSRe-norm to define the precise location and rank of each viral family within the DM and HC distribution ( Figure 4D & 4E). In DM, multiple viral families (Nairoviridae, Poxviridae, Secoviridae, Alloherpesviridae, Caliciviridae and Adenoviridae) with high NGSRe-norm Z score occurred with at least one standard deviation above the group mean. In HC, only Polyomaviridae (n=1) occurred with high NGSRe-norm Z score. In Figure S4 we provide a cladogram of DM-specific viral species.

We queried the NGS reads against the human proteome to identify accumulated autoantibodies in the pooled plasma. In DM P20 we identified TRIM33 (Mean log10: 1.301), in accordance with the autoimmune profile of our selected plasma ( Figure 5A ) which was not present in HC P20. We also identified 11 additional TRIM proteins in DM P20 which were TRIMs 21, 69, 47, 46, 27, 60, 10, 7, 77 , 3 and TRIML2 ( Figure 5A ). Of these, TRIM21 was also detected in the healthy samples but with significantly lower NGS reads, and TRIM25 (Mean log10: 1.778) was observed only in the HC P20 ( Figure 5A ). These results confirm the presence of TRIM33 autoantibodies in the selected twenty anti-TRIM33-positive DM patients and demonstrate the presence of autoantibodies against other members of the TRIM protein family exclusively in DM.

An expanded dermatomyositis-specific and IFN-regulated human proteome From our antibody analysis, in DM P20 we identified a total of 2885 human protein targets, of which 2537 were highly specific sequence annotation hits (Mean Sig. < 0.578). In HC P20 we detected 1450 human protein targets, 811 with high specificity (Mean Sig. < 0.55). Autoantibodies identified in both DM and HC constituted only 13% (n=498) of total proteins and 8.4% (n=260) of highly specific proteins suggesting a strong disease-specific proteome signature. The 2:1 (total proteins) and 3:1 (high specificity) ratios of identified autoantibodies in DM over HC, suggest that an expanded subset of the human proteome is targeted by autoantibodies in DM patients compared to HC.

Due to the role of TRIM proteins in IFN signalling, we asked whether the autoantibody protein-targets are regulated by interferons. In the DM dataset 1,560 proteins were predicted to be regulated by IFNs compared to 518 in HC (Interferome v2.01) (Rusinova et al., 2013) . The vast majority of these proteins are regulated by interferons type I, type II or both ( Figure S5A ). Regardless of the IFN type we observed a higher number of IFN-regulated proteins in DM compared to HC suggesting a DM-specific enrichment of immunoglobulins against IFN-regulated human proteins. We focused our search directly for IFN autoantibodies expressed in our samples. In DM, IFNGR1 autoantibodies were present, whereas they were absent in HC ( Figure S5B ). We expanded our search for autoantibodies against known proteins that are highly ranked within the IFNG signalling pathway (gene rank within the SuperPath; GeneCardsSuite: PathCards) (Belinky et al., 2015) . Autoantibodies against 26 IFNGrelated proteins were observed in DM, whereas only 5 were found in HC ( Figure S5B ). The proteinprotein interaction (PPI) enrichment score in DM was <1.0e-16 versus 0.123 in the healthy sample (STRING 11.0)(Szklarczyk et al., 2019) ( Figure S5C & 5D). Overall, these data suggest the accumulation of antibodies in DM against proteins that strongly contribute to IFNG signalling and the broader antiviral mechanism by IFN-regulated proteins.

To describe the biological functions of all the identified proteins the Gene Ontology (GO) framework was used ( Figure 5B & 5C). In DM P20 eight GO biological processes were highly enriched ( Figure   5B ). These processes were represented by an average of 25.6% of GO-specific proteins (n=996, 95%CI: 21.2%-29.7%) ( Figure S6A ). In HC P20, ten GO processes were enriched with an overall lower average coverage of 14.5% (n=592, range: 12.5%-15.7%), as less proteins mapped to the same GO unit. In the top-ranked GO processes enriched in DM, the GO coverage was higher in DM than HC regardless of the GO process ( Figure S6A ). More than one third (39.2%: 996 out of 2537) of the DM autoantibody targets were part of the 8 identified biological processes; 18.6% of them (186 out of 996) were also present in HC P20. DM processes involved structural elements including microtubulebased processes, actin-filament functions, cell junction organisation, extracellular structure organisation, cell morphogenesis in differentiation and small GTPase mediated signal transduction ( Figure 5B ). PTK2 and ROCK2 were shared between 7 out of 8 functions, and, KIF14, NDEL1, SRGAP2, APP, PRKCZ, and CLASP2 were shared amongst 6 out of 8 functions ( Figure 5D ). None of these proteins were observed in HC. The fact that the identified DM-specific and HC-specific proteins were robustly clustered in biological functions suggests non-random targeting of specific signalling pathways by autoantibodies. In DM, these processes share multiple autoantibody protein targets suggesting the presence of a DM-specific autoantibody-targeted proteome module ( Figure   S6B ).

We identified that epitope sequences against the Poxviridae family of viruses were significantly enriched in DM, including variola virus which had the highest NGSRe-norm (Fig. 4F ). Given that our DM patients were TRIM33 autoantibody positive and due to the fact that TRIM proteins, IFNG and the IFN antiviral mechanism were significantly enriched in our proteome data, we searched for potential links between Variola virus, other members of the Poxviridae family and TRIM proteins.

Since molecular mimicry is a potential mechanism of autoantibody generation, we aligned the identified variola virus and TRIM epitopes; the sequence annotation thresholds that we used in our bioinformatics pipeline guaranteed robust annotation of the epitope sequences. This would affect cross-kingdom epitope alignment, since we have maximized the phylogenetic distance between each microbial epitope and human proteins. To account for this, we started our epitope sequence analysis with the widest possible diversity of identified variola and TRIM epitopes, sacrificing specificity i.e. epitopes with more than 80% match to variola virus, and TRIM epitopes with an identity of more than 50%. The phylogenetic distances are shown in the circular cladogram in Figure 6 . We identified two different clades containing leaf nodes of TRIM and variola epitopes of high similarity (branch lengths <0.1). The first clade involved the TRIM3 epitope "RIPDDVRRRPGC" and three additional epitopes "RI(Q)DDVRRRPGC", "RI(Q)DDV(H)RRPGC" and "RI(Q)DD(V)(S)RRPGC" each of which mapped to VARV GER58 hdlg 202 and VARV GUI69 005 202. The second clade contained the TRIM3 epitope "SSHARYKSVRFS", and "SSHARYKSVRFS", "SSHARYKS(M)RFS", "SSHARYKSLRFS", "SSHARYKS(L)RF", and "SSHARYKSLRF(T)" of a variola virus (unnamed protein product and viral DNA polymerase processivity factor). We reinstated the annotation specificity thresholds for TRIM and variola epitopes to the default levels (strict, high specificity, maximisation of phylogenetic distances) ( Figure S7A ). The epitope sequence "RI(P)DDVRRRPGC" was retained and shared between TRIM3 (branch length: 0.0541) and VARV GER58 hdlg 202 (branch length: 0.0344). The epitope sequences "SSHARYKSLRFS" and "SSHARYKSLRF" were specific for variola virus but were no longer annotated as TRIM epitopes ( Figure S7A ). We asked whether the above epitope sequences are shared amongst different Poxviridae species that we identified in DM P20

( Figure 4F ). The 3 epitopes ("SSHARYKSLRFS", "SSHARYKSLRF", and "RIQDDVRRRPGC") were shared with high homology between variola, vaccinia, ectromelia, cowpox and camelpox viruses ( Figure S7B ). We aligned the DM TRIM proteins that we identified to test the conservation level of the above epitopes in TRIMs ( Figure 7A ). TRIM3 and TRIM33 shared high similarity of AA sequence and seem to diverge from the rest of TRIMs across the region of interest ( Figure 7A ).

Finally, we aligned all DM-enriched viral and TRIM epitopes ( Figure S7C ) and identified a third epitope "KHKGALGGGGNE" of TRIM47 which is shared by Human immunodeficiency virus 1 "KHKGALGGGG(N)E", and "KHKG(D)LGGGG(Y)E" shared by Synechococcus phage syn9 "KHKG(D)LGGGG(Y)E" ( Figure S7D ). The epitope was poorly conserved amongst the DM-specific TRIM proteins ( Figure 7B ). Overall, we have identified two epitope sequences that are shared between Variola virus and TRIM3 but also between members of the Poxviridae family. A third epitope that is shared between TRIM47, HIV-1 and Syn9 was also observed. These findings suggest that molecular mimicry events, at least amongst these specific viral and TRIM epitopes, is a potential mechanism of pathogenesis in DM.

This is the first study to investigate the accumulated microbial and autoantigen antibody repertoire in adult-onset dermatomyositis (DM) patients positive for antibodies against TIF1 (TRIM33). We used an un-targeted high-throughput approach which combines immunoglobulin disease-specific enrichment of immunogenic epitopes, and subsequent identification of anti-microbial antibody and autoantibody richness and abundance. Key findings include that (1) the DM-specific IgOme was characterised by high microbial diversity, (2) antibodies against viruses were over-represented and species of the Poxviridae family were significantly enriched, (3) DM-specific accumulated autoantibodies target a significant portion of the human proteome, including proteins from specific biological processes and interferon regulated proteins, (4) autoantibodies against TRIM33 and eleven further TRIM proteins were identified, and (5) specific TRIM proteins share epitope homology with viral species identified from the plasma of DM patients.

The identified microbial epitopes in both DM and healthy controls (HC) mapped to bacteria, viruses, fungi and archaea demonstrating the human capacity for building protection against a diverse group of species. The presence of a stably-enriched microbial component in DM was identified, characterised by a higher number of epitopes per species and against a wider microbial repertoire. The effect of sample size in our measurements suggests increased microbial exposure and interpersonal variability in the DM compared to the healthy group, leading to expansion of the identified DM-specific microbial signature upon screening additional samples. Even though our observations provide a static snapshot of the IgOme, the immune system has a memory and the accumulation of antibodies takes place from birth over the entire lifetime of the individual. Thus, our data suggest that DM as a clinical entity is characterised by diverse microbial exposure.

We observed that viruses are over-represented amongst the microbial species with the highest measured abundance in DM and HC. Due to the competitive design of the epitope enrichment process this observation suggests that viral exposure has a significant role in DM. We observed enrichment of the Coronaviridae, Herpesviridae, Orthomyxoviridae, Geminiviridae and Poxviridae families in persists for up to 88 years (Taub et al., 2008) , therefore given the age of our donors, it is very likely that the DM associated antibodies to Poxviridae were raised through the vaccination process. This is supported by the fact that VACV antibodies were observed in the healthy sample without being (Gan and Miller, 2011; Marie et al., 2005) .

Interestingly, there are case studies which report development of DM after vaccination against some of these viruses, notably HBV, HPV, influenza and VACV (Ignasi Rodriguez-Pinto, 2015) . Overall, the DM-specific viral signature includes viruses which can either directly infect muscle tissue and/or indirectly sabotage immune homeostasis. The observation that the respiratory and gastrointestinal systems are the primary physiological targets of these viruses agrees with previous observations regarding the high frequency of this type of infection in patients with juvenile-and adult-onset DM (Pachman et al., 2005; Svensson et al., 2017) . Moreover, the age-range of juvenile and adult onset DM coincides with the epidemiological peaks of respiratory infections during life (Kuan et al., 2019) .

In DM, we identified autoantibodies against a very large number of human proteins (>2500), representing about 14% of the current human proteome (UniProtKB/Swiss-Prot database, last modified 29 July 2019). This was double the proportion compared to the HC-enriched proteome, with only 13% of autoantibodies being found in both DM and healthy controls, suggesting accumulation of autoantibodies in DM against a wider group of proteins. Interestingly three times the number of autoantibody targeted proteins were regulated by type I and II interferons in DM than in HC. This underscores the relevance of interferons in this disease (Neely, 2019; Somani et al., 2008) , and indicates the potential of interferon blockade as a therapy. The identification with high confidence of specific biological processes enriched in both DM and HC, suggests an organisational structure within the targets of these accumulated autoantibodies. Although deconvolution to individual samples is not possible with our data, this is consistent with an infection (exposure)-driven pathological autoantibodyproducing mechanism which progressively expands its repertoire (antigens) under the chronic burden of accumulated stimuli (Chan and Gack, 2016; Cusick et al., 2012; Munz et al., 2009; Panoutsakopoulou et al., 2001) . Biological processes enriched in DM are orientated around cytoskeletal organisation, microtubule movement, actin filaments and cell junctions. Eight of the identified protein targets participated in most of these processes indicating a key role in the regulation of multiple signalling pathways including the formation and disassembly of focal adhesions. The enrichment of antibodies against proteins with a role in the regulation of focal cell-cell adhesion in DM is particularly interesting, since disruption of these structures in epithelial cells is known to facilitate the spread of viruses, by allowing their release from the basal to the apical surface or the external environment (Spear, 2002; Torres-Flores and Arias, 2015) . Interestingly some of the DM associated viruses described here are known to manipulate junctional proteins namely Adenoviridae, Flaviridae, Herpesviridae, Retroviridae, Paramyxoviridae, and Picornaviridae which provides a functional link between DM and increased viral exposure (Mateo et al., 2015; Mothes et al., 2010; Zhong et al., 2013) .

We identified autoantibodies against TRIM33 in DM plasma, confirming the anti-TIF1 positive profile of our patients. Furthermore, we identified autoantibodies against 11 other TRIM proteins in DM, of which only TRIM21 was identified in healthy controls but with much lower abundance. TRIM21 (Ro52) is a known autoantigen in IIM and related autoimmune disorders, and acts as the highestaffinity Fc receptor in humans. TRIM21 binds cytosolic antibodies bound to non-enveloped viral pathogens, to trigger antibody-dependent intracellular neutralization and proteasomal degradation of the immune complex (Lee, 2017; Rajsbaum et al., 2014) . The autoantibodies we identified against TRIM proteins, other than TRIM33 and TRIM21, have not been observed previously in IIM. TRIM69 inhibits vesicular stomatitis virus transcription and interacts with dengue virus non-structural protein 3 to mediate its poly-ubiquitination and degradation (Kueck et al., 2019; Wang et al., 2018) . TRIMs also regulate antiviral pathways indirectly by mediating innate immunity, influencing the transcription of TI-IFNs, pro-inflammatory cytokines and interferon-stimulated genes (van Tol et al., 2017) .

Notably, the TRIM protein family expanded very rapidly in evolution, coinciding with development of the adaptive immune system, suggesting that TRIMs may have evolved to fine-tune interactions between the increasingly complex innate and adaptive immune systems Versteeg et al., 2014) . Deregulated activity of approximately one third of the TRIM proteins has been associated with the development of various type of human cancer (Vunjak and Versteeg, 2019) . For example, TRIM47 and TRIM27 have an oncogenic role in colorectal, prostate, oesophageal, ovarian and non-small cell lung cancer, promoting proliferation and metastasis (Han et al., 2017; Liang et al., 2019; Zhang et al., 2018) . Conversely TRIM3 has tumour suppressor activity, inhibiting the growth of liver, colorectal and gastric cancer (Boulay et al., 2009) . The presence of autoantibodies against cancerassociated TRIMs accords with the strong temporal association between myositis and the development of malignancies in adult-onset anti-TIF1γ positive DM (Oldroyd et al., 2019) , consistent with the high proportion of cancer-associated myositis cases in the current study (45%). Overall, the expanded targeting of TRIM proteins we observed in DM plasma supports the role of the TRIM RING-type E3 ubiquitin ligase subfamily as powerful regulators of the immune system through post-translational modification (Hage and Rajsbaum, 2019; van Tol et al., 2017) .

Since Poxviridae epitopes were enriched in DM, we investigated a potential link between pox viruses and TRIM proteins. We identified epitope sequences with high similarity between TRIM3 and Variola virus in addition to other members of the Poxviridae family. Moreover, TRIM47 had epitopes in common with HIV1 and Synechococcus phage. These high similarity epitopes were identified in the C-terminus of the TRIM proteins; between the filamin and NHL 1 domains for TRIM3, and proximal to the B30.2 domain in TRIM47. The C-terminus of TRIM proteins is under evolutionary positive selection and determines ligand binding specificity, function and subcellular localization; notably the B30.2 domain is associated with ability to restrict viral infection, particularly retroviruses (van Tol et al., 2017) . This finding of high similarity epitope sequences suggests that molecular mimicry, at least amongst specific viral and TRIM epitopes, is a potential mechanism of pathogenesis in DM.

The finding of antibodies against a human protein does not mean that this was the antigen against which the antibody was raised, as pathogen-derived proteins can mimic endogenous epitopes (Panoutsakopoulou et al., 2001; Walker and Jeffrey, 1986) . This might explain inconsistencies between the universal cellular distribution of specific antigens and the specific skeletal muscle target of the disease (Walker and Jeffrey, 1986) . In support of this, we observed that certain DM-specific TRIM epitopes are shared amongst different viruses. TRIM epitopes did not form a separate phylogenetic cluster but were dispersed throughout the viral epitope phylogenetic tree, suggesting that accumulated antibodies against TRIM proteins might be the product of a primal molecular mimicry event and subsequent epitope spreading rather than increased TRIM protein production. This is supported by the observation that the gene expression levels of autoantigens in myositis muscle biopsies do not correlate with the levels of circulating autoantibodies recognizing each cognate endogenous autoantigen; instead they are directly associated with the expression of muscle regeneration markers (Pinal-Fernandez et al., 2019) .

We propose a model whereby pathogenesis of DM is dependent on dynamic and personalised viral exposure patterns, which start very early in life. The main two gateways of physiological transmission of viruses to humans are the respiratory and gastrointestinal systems, while non-physiological presentation of viruses is through immunisation. The link between DM and viruses is complex but strikingly present in our data since there is systematic enrichment for human proteome modules that are directly exploited by, or respond to, most of the viruses that we identified. We propose that an initial viral infection or virus delivery system presents the main antigen that induces antigen-specific antibody production in a predisposed immune environment. Since TRIM proteins share epitope homology with multiple viral species, these antibodies can also bind to TRIMs and potentially other homologous proteins. The outcome of this process is influenced by factors including the antibody populations at the site of infection (local antibodies), the status of the innate immune system, the virusspecific antibody concentration in serum (which usually peaks several weeks post infection), and the history of previous exposure primarily to the same or related viral strains (and epitopes) (Rojas et al., 2018) . Since HLA Class II molecules are major regulators of the adaptive immune response to antigenic challenge through T cell repertoire selection, the level of antibody production to different viral antigens also may be mediated by HLA type. We previously identified that adult-onset anti-TIF1 positive DM is associated with HLA-DQB1*02:02 (Rothwell et al., 2019) . For the samples with available data included in this study, we observed that 6/15 (40%) DM and 3/17 (18%) HC are heterozygous for HLA-DQB1*02:02. The finding that in myositis, autoantigen expression correlates with expression of muscle regeneration markers (Pinal-Fernandez et al., 2019) suggests that after initial viral presentation, the increased concentrations of muscle proteins may then be sufficient even in the absence of viral proteins to invoke periodic rises of autoantibodies. Moreover, the continuous activation of innate immunity observed in DM (Neely, 2019; Pinal-Fernandez et al., 2019; Walsh et al., 2007; Wong et al., 2012) can also potentiate autoimmunity through chronic immune-mediated tissue damage resulting in autoantigen release without the need for specific activation of auto-reactive T cells by a microbial mimic (Panoutsakopoulou et al., 2001) . Therefore, following the initial viral exposure there is a high chance that this effect will spread amongst related proteins within specific signalling pathways (since protein homology is related to function) of the initial hit. In this model, in specific autoantibody-positive myositis subgroups, accumulated autoantibodies against proteins that participate in specific biological processes and signalling pathways would be observed. The finding that in healthy controls the protein coverage of the GO processes significantly decreases suggests more random targeting of human proteins by autoantibodies, opposite to that observed in DM and in support of our hypothesis.

The use of pooled plasma is a limitation in our study since we cannot deconvolute the data into individual patients. However, the enrichment process against healthy adults provides a unique opportunity to study DM as a disease system capturing the breadth of microbial antibody and autoantibody accumulation against a very large number of linear epitopes. Apart from antibodies against TIF1 which is a shared feature of the patients in this study, we do not anticipate that our findings will be equally distributed among patients. Indeed, if the shared feature in DM is a molecular mimicry event combined with epitope spreading, then the stochastic nature of microbial exposure, genetic predisposition and the dynamic nature of immunity would not support a single pathotype. Thus, our study provides the first DM-specific "map" of potential routes to disease. For the future, we recommend: clinical records are kept of short and long-term microbial exposure, infections history including type of pathogen, severity scores, number of hospitalisations and vaccination history; which were not available in this study. Records could be expanded in cases of juvenile DM, to include maternal infection and vaccination history to account for acquired transplacental immunity.

Lifelong exposure to viruses and probably to other microbes contributes not only to accumulation of virus-specific antibodies and protection, but generation of autoantibodies against TRIMs and other cellular proteins, which at some point may reach a critical mass and induce disease most likely as a result of seemingly mild or non-symptomatic infection. Molecular mimicry and epitope spreading may play a significant role, instantly raising questions not only concerning the extent but also the sequence of events. This also means that autoantibodies identified in DM to date might only be the tip of the iceberg. 

The authors declare no competing interests. and TRIM47 (sp|Q96LD4); cross-marked. Conservation level is presented as a colour gradient (high; red, low; blue). (A) Partial alignment of the region where the TRIM3-variola high similarity epitopes were identified. The TRIM3 protein sequence is presented next to the alignment plot. Yellow: the two TRIM3 epitopes with high amino acid (AA) similarity to variola and poxviruses ( Figure 6 ). Black font:

identical AAs between TRIM3 and variola virus. Red font; different AAs between TRIM3 and variola virus. Both epitopes are located only 6 AAs apart (relative to TRIM3 protein sequence). TRIM3 presents high similarity with TRIM33 in respect to the specific AA epitopes. (B) Partial alignment of the region where the TRIM47-HIV-syn9 high similarity epitope was identified. The TRIM47 protein sequence is presented next to the alignment plot. Yellow: the TRIM47 epitopes with high AA similarity to HIV and syn9 phage ( Figure S7D ). Black font: identical AAs amongst TRIM47 and the two viruses. Red font; different AAs amongst TRIM47 and the two viruses. The epitope is poorly conserved amongst the DM-specific TRIM proteins.

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Janine Lamb (Janine.Lamb@manchester.ac.uk).

This study did not generate new unique reagents.

Plasma samples were collected from anti-TIF1 positive adult-onset dermatomyositis (DM) patients through the UK Myositis Network, as described previously (Rothwell et al., 2019) (Table S1 ). All individuals fulfilled definite or probable Bohan and Peter classification criteria for dermatomyositis.

Anti-TIF1 autoantibody positivity was identified by immunoprecipitation, and confirmed by ELISA, as described previously (Rothwell et al., 2019) . Gender and age matched (at time of sample collection) healthy controls (HC) were identified through the University of Manchester Longitudinal Study of Cognition in Normal Healthy Old Age cohort (Rabbitt et al., 2004) 

We implemented the "Serum Antibody Repertoire Analysis (SARA)" pipeline (manuscript in preparation). SARA comprises a comprehensive workflow that integrates molecular biology peptide display and epitope signature enrichment through competitive bio-panning and high-throughput DNA next generation sequencing (NGS), alongside in-house computational scripts to reverse engineer high resolution epitope signatures that reflect original in vivo epitopes present in patient sera. SARA was applied to predict the identity and abundance of antibody epitope repertoires enriched in plasma from anti-TIF1 autoantibody-positive DM patients versus plasma from matched HC (see Supplementary   Methods) . This pipeline provides a digital triage of infectious organism epitopes and autoantibodies predicted to be uniquely present or highly enriched in each sera pool. All informatics analysis was carried out using R(R Development Core Team, 2018) 3.2-3.6 and Python (Sanner, 1999) 3.5. We briefly summarise the seven SARA pipeline modules (M1-M7) below in turn.

Plasma total immunoglobulin purification: M1 Total immunoglobulins IgA, IgG, IgM were purified from twenty anti-TIF1 positive adult-onset DM and twenty HC plasma by AdamTech Total Ig Extraction kits (Pesac, FR). For the sample-pools of 10 (P10), 12 µg of isolated Ig per donor were used for a total of 120 µg. For the sample-pools of 20 (P20), 6 µg of isolated Ig per donor were used for a total of 120 µg. Ig yields and purity were assessed by NanoDrop spectrophotometry and polyacrylamide gel electrophoresis.

Competitive biopanning: M2 Separate purified Ig pools from DM and HC were used for competitive bio-panning with the FliTrxTM random 12 amino acid (AA) peptide surface display system (Lu et al., 2003) . Briefly, we first conducted a pre-panning stage to incubate separate DM and HC immobilised Ig pools with induced FliTrxTM E.

Coli cells (Lu et al., 2003) in order to sequester expressed epitopes relevant to the specific plasma cohort. Unbound bacteria per cohort were further incubated with the alternative immobilised Ig pools for the main panning stages (cross-panning). Tethered bacteria were eluted and expanded to repeat the biopanning process 5 times. OD600 was measured after each round of biopanning to ensure comparable efficiency between DM and HC ( Figure S3E ). After competitive biopanning immobilised bacteria were expanded and polyvalent plasmid cohorts were purified (Maxiprep, Qiagen, UK).

Epitope NGS: M3, NGS data processing: M4 Polyvalent plasmid variance regions were PCR amplified and gel-purified. PCR products were validated by Sanger sequencing (Figure S3F) , and NanoDrop. The DNA primer sequences are:

Forward: 5'-ATTCACCTGACTGACGAC-3', FliTrxTM Reverse: 5'-CCCTGATATTCGTCAGCG-3'. Multiplexed DNA fragments were sequenced on the NextSeq 500 platform (Illumina, UK). We retrieved ≈24 million and ≈36 million paired end FASTQ reads for HC and DM respectively (Figure 1 ). The 36-bp variance region sequences were translated with respect to reading frame. < 14% of DM and <13% of HC comprised premature stop codons, indicative of immunologically-relevant bio-panning enrichment ( Figure S1A -S1D). Non-specific HTS sequence noise from residual bio-panning solution was controlled for by our 10-σ-99 noise floor. This retains all expressed epitopes within each respective HC or DM pool that surpass a threshold of 10 standard deviations above the lowest 99% distinct peptide sequence enrichment when ranked by read count (further described in Supplementary figure S1A-S1D).

Epitope signature set analysis: M5

Distinct epitope signature set analysis presented unique and common AA epitope sequence sets with associated NGS enrichment scores. Minimum thresholds of 10AA sequence length and the 10-σ-99 NGS noise filter provided sensitive annotation power while minimising annotation noise ( Figure S1D -S1F). Cleaned distinct epitope sequences were collated. Unique epitope pools were determined by symmetric difference of the HC and DM pools (i.e. specific epitope sequences uniquely found in HC or in DM patients). The intersect between HC and DM pools was probed and a minimum five-fold change threshold applied to partition relevant epitope collections to supplement the HC or DM pools.

These epitope collections were utilised in downstream microbiome and autoantibody annotations.

>8.6x106, and > 4.7x106 distinct epitopes with a minimum length of 6AA length were retrieved for DM and HC pools respectively. The highest NGS read counts achieved were ≈111,000 (DM) and

≈88,000 (HC) (Figure 1 ).

Epitope annotation and rank scoring: M6 15,522 DM-associated and 4,817 HC-associated unique epitopes were retrieved for DM patients. Only 0.65% of the DM pool comprised peptide sequences common to HC. Of these epitopes common by sequence, the DM enrichment score ranged between 5-fold and 546-fold the healthy control score ( Figure 1 ) ( Figure S1A -S1F). Epitopes were annotated by our modified BLASTp (Henikoff and Henikoff, 1992) approach at 100% minimum identity ( Figure S1G -S1I). This facilitates short AAsequence inputs and resolves protein annotation confidence scores per epitope with statistical control of epitope-protein annotation confidence, intra-protein dis-contiguous sequence matches, redundancy of protein isoforms and common sequence matches, single-epitope multi-protein parsing, and set analysis of resultant patient cohort organisms lists. These were tuned per each epitope's NGS read count. 6.745x106 (DM patients) and 2.250x106 (HC) total protein annotations were retrieved ( Figure   1 ).

Phylogenetic and taxonomic analysis: M7

Our scoring system generates aggregate NGS read count enrichment and annotation stringency scores.

We retrieved every known upstream taxonomic ranking level for all microbe IDs and anchored these via NCBI REST and the Taxonomy repository (Benson et al., 2009; E., 2008; Federhen, 2012) .

Phylogenetic trees were generated by PhyloT biobtye (Letunic, 2018) and microbial annotations were assessed by NGS enrichment, epitope mapping confidence, annotation scores, taxonomic ranks, and phylogenetic clustering ( Figure S1G -S1I). We utilised the interactive Tree Of Life iTOL (Letunic and Bork, 2007; Letunic and Bork, 2016) to taxonomically cluster microbial agents for the DM or HC pools with incorporated annotation layers representing the NGS enrichment and annotation scores ( Figure S1I ). 9,111 distinct and 6,202 unique species were annotated for DM patients (4,994 distinct and 2,085 species were annotated for HC) (Figure 1 ).

We identified any autoantibodies against human proteins that may be present. Annotation scores occupy unit interval space [0,1] as surrogate significance values. Protein cohorts were inspected by Metscape (Tripathi et al., 2015) , KEGG (Kanehisa and Goto, 2000) , UniProt (Apweiler et al., 2013) , and the Human Protein Atlas (Uhlen et al., 2010) , to explore functional predictions based on aberrant autoimmune targeting. 435 (DM) and 407 (HC) enriched human biomarkers from autoantibodies were retrieved ( Figure 1 ). 

The code supporting the current study have not been deposited in a public repository but will be deposited following submission of our methods manuscript (in preparation). In the intervening period these will be available from the corresponding author on request. (Fig. S1A ). These were highly enriched by associated NGS read counts and represent polyvalent FliTrxTM peptide 12 AA coding sequences from DM and HC biopanned total Ig pools.

DM NGS data processing (M4); DM epitope signature set analysis (M5)

We retrieved 13,385,371 distinct expressed AA epitopes that represent the epitope signatures present within the DM and HC Ig pools (Fig. S1A ). Associated NGS read frequencies reflected the epitope enrichment process achieved during biopanning, and Fig. S1B & C confirms that the 10-σ-99 NGS filter optimally controlled biopanning-induced sequencing noise to successfully provide 15,522 DMassociated and 4,817 HC-associated unique epitope sequences. Preferential enrichment for epitopes of length 11 and 12 AAs (Fig. S1D ) confirmed genuine biopanning capture and is concordant with published peptide epitope ranges of 4 to 12 AAs (Buus et al., 2012; Hopp and Woods, 1981) . Minimum epitope lengths of 9AA applied with the 10-σ-99 NGS filter (above) proved optimal for annotation ( Fig. S1E ) using the unique sequence epitope sets of 15,522 for DM and 4,817 for HC individuals (Fig.   S1F ). Ig repertoire sequence overlap determined that only 0.65% of DM epitope set were present in healthy controls (fold-change (FC) values of >=5x to 546x abundance vs HC individuals) while 6.44%

of healthy epitope set was retained in the DM group (FC values of >=5.3x to 1240x vs DM) (Fig. S1F ).

Our modified BLASTp approach retrieved 6.75 million (DM) and 2.25 million (HC) microbial or human protein annotations respectively which mapped to ≈14,000 microbial organisms. Enrichment plots revealed 4,994 Distinct infectious agents in HC individuals and 9,111 infectious agents in DM patients (Fig. S1G ) of which 2,085 and 6,202 Unique infectious agents were identified in healthy individual and DM patients respectively (Fig. S1H) . These signatures presented 377 highly enriched infectious agents in DM and 270 highly enriched agents in healthy individuals ranked by phylogeny ( Fig. S1I ).

Supplementary figure 1: Immunoglobulin epitope enrichment in DM and HC.

Next generation sequencing run data processing: The enrichment fold-change differences of these common sequences ranged from 5x:1,240x compared with the opposite set (0.65% and 6.44% the proportion of the respective unique epitope sets).

Enriched Distinct and Unique microbial agents in DM: Distinct and unique organisms at a 100% BLAST annotation threshold. Epitope sets comprised 10-12 AA sequence lengths following the 10-σ- 

Clinical characteristics and change in the antibody titres of patients with anti-MDA5 antibody-positive inflammatory myositis

Update on activities at the Universal Protein Resource (UniProt) in 2013

PathCards: multi-source consolidation of human biological pathways

Frequency, mutual exclusivity and clinical associations of myositis autoantibodies in a combined European cohort of idiopathic inflammatory myopathy patients

Loss of heterozygosity of TRIM3 in malignant gliomas

NAViGaTOR: Network Analysis, Visualization and Graphing Toronto

Highresolution mapping of linear antibody epitopes using ultrahigh-density peptide microarrays

Viral evasion of intracellular DNA and RNA sensing

Bacterial, fungal, parasitic, and viral myositis

Nonbacterial myositis

Molecular mimicry as a mechanism of autoimmune disease

E-utilities Quick Start Entrez Programming Utilities Help NCBI (NCBI)

The NCBI Taxonomy database

State of the art: what we know about infectious agents and myositis

To TRIM or not to TRIM: the balance of host-virus interactions mediated by the ubiquitin system

TRIM47 overexpression is a poor prognostic factor and contributes to carcinogenesis in non-small cell lung carcinoma

AMINO-ACID SUBSTITUTION MATRICES FROM PROTEIN BLOCKS

Prediction of protein antigenic determinants from amino acid sequences

ViralZone: a knowledge resource to understand virus diversity

Myositis and Vaccines

KEGG: Kyoto Encyclopedia of Genes and Genomes

A chronological map of 308 physical and mental health conditions from 4 million individuals in the English National Health Service

Vesicular Stomatitis Virus Transcription Is Inhibited by TRIM69 in the Interferon-Induced Antiviral State

A review of the role and clinical utility of anti-Ro52/TRIM21 in systemic autoimmunity

PhyloT -A phylogenetic tree generator, based on NCBI taxonomy

Interactive Tree Of Life (iTOL): an online tool for phylogenetic tree display and annotation

Interactive tree of life (iTOL) v3: an online tool for the display and annotation of phylogenetic and other trees

TRIM47 is up-regulated in colorectal cancer, promoting ubiquitination and degradation of SMAD4

Using bio-panning of FLITRX peptide libraries displayed on E. coli cell surface to study protein-protein interactions

Opportunistic infections in polymyositis and dermatomyositis

Connections matter--how viruses use cell-cell adhesion components

Risk factors and disease mechanisms in myositis

Virus cell-to-cell transmission

Antiviral immune responses: triggers of or triggered by autoimmunity?

Type I interferon in rheumatic diseases

Gene Expression Meta-Analysis Reveals Concordance in Gene Activation, Pathway, and Cell-Type Enrichment in Dermatomyositis Target Tissues

High Ro52 expression in spontaneous and UV-induced cutaneous inflammation

The temporal relationship between cancer and adult onset anti-transcriptional intermediary factor 1 antibody-positive dermatomyositis

History of infection before the onset of juvenile dermatomyositis: results from the National Institute of Arthritis and Musculoskeletal and Skin Diseases Research Registry

Analysis of the relationship between viral infection and autoimmune disease

Myositis Autoantigen Expression Correlates With Muscle Regeneration but Not Autoantibody Specificity

R: A Language and Environment for Statistical Computing

Practice and drop-out effects during a 17-year longitudinal study of cognitive aging

TRIMmunity: the roles of the TRIM E3-ubiquitin ligase family in innate antiviral immunity

Molecular mimicry and autoimmunity

Focused HLA analysis in Caucasians with myositis identifies significant associations with autoantibody subgroups

Interferome v2.0: an updated database of annotated interferon-regulated genes

The evolution of poxvirus vaccines. Viruses

Python: A programming language for software integration and development

Severe dermatomyositis triggered by interferon beta-1a therapy and associated with enhanced type I interferon signaling

Viral interactions with receptors in cell junctions and effects on junctional stability

Infections and respiratory tract disease as risk factors for idiopathic inflammatory myopathies: a population-based case-control study

STRING v11: protein-protein association networks with increased coverage, supporting functional discovery in genome-wide experimental datasets

Immunity from smallpox vaccine persists for decades: a longitudinal study

Tight Junctions Go Viral! Viruses

Meta-and Orthogonal Integration of Influenza "OMICs'' Data Defines a Role for UBR4 in Virus Budding

Towards a knowledge-based Human Protein Atlas

The TRIMendous Role of TRIMs in Virus-Host Interactions. Vaccines (Basel)

InTRIMsic immunity: Positive and negative regulation of immune signaling by tripartite motif proteins

TRIM proteins

Polymyositis and molecular mimicry, a mechanism of autoimmunity

Type I interferon-inducible gene expression in blood is present and reflects disease activity in dermatomyositis and polymyositis

Critical role for cholesterol in Lassa fever virus entry identified by a novel small molecule inhibitor targeting the viral receptor LAMP1

Interferon and biologic signatures in dermatomyositis skin: specificity and heterogeneity across diseases

Necrotizing myositis causes restrictive hypoventilation in a mouse model for human enterovirus 71 infection

TRIM27 functions as an oncogene by activating epithelial-mesenchymal transition and p-AKT in colorectal cancer

Cell-to-cell transmission of viruses

Tripartite motif-containing (TRIM) 21 negatively regulates intestinal mucosal inflammation through inhibiting TH1/TH17 cell differentiation in patients with inflammatory bowel diseases