Article Text
PDF
Abstract
Objectives Sjögren disease (SjD) diagnosis often requires either positive anti-SSA antibodies or a labial salivary gland biopsy with a positive focus score (FS). One-third of patients with SjD lack anti-SSA antibodies (SSA−), requiring a positive FS for diagnosis. Our objective was to identify novel autoantibodies to diagnose ‘seronegative’ SjD.
Methods IgG binding to a high-density whole human peptidome array was quantified using sera from SSA− SjD cases and matched non-autoimmune controls. We identified the highest bound peptides using empirical Bayesian statistical filters, which we confirmed in an independent cohort comprising SSA− SjD (n=76), sicca-controls without autoimmunity (n=75) and autoimmune-feature controls (SjD features but not meeting SjD criteria; n=41). In this external validation, we used non-parametric methods for binding abundance and controlled false discovery rate in group comparisons. For predictive modelling, we used logistic regression, model selection methods and cross-validation to identify clinical and peptide variables that predict SSA− SjD and FS positivity.
Results IgG against a peptide from D-aminoacyl-tRNA deacylase (DTD2) bound more in SSA− SjD than sicca-controls (p=0.004) and combined controls (sicca-controls and autoimmune-feature controls combined; p=0.003). IgG against peptides from retroelement silencing factor-1 and DTD2 were bound more in FS-positive than FS-negative participants (p=0.010; p=0.012). A predictive model incorporating clinical variables showed good discrimination between SjD versus control (area under the curve (AUC) 74%) and between FS-positive versus FS-negative (AUC 72%).
Conclusion We present novel autoantibodies in SSA− SjD that have good predictive value for SSA− SjD and FS positivity.
- Sjogren's syndrome
- antibodies
- autoantibodies
Data availability statement
Data are available on reasonable request. All data relevant to the study are included in the article or uploaded as supplementary information. All data relevant to the study are included in the article or uploaded as supplementary material but data are available on reasonable request.
Statistics from Altmetric.com
WHAT IS ALREADY KNOWN ON THIS TOPIC
Seronegative (anti-SSA antibody negative (SSA−)) Sjögren disease (SjD) often requires a labial salivary gland biopsy for diagnosis, which is challenging to obtain and interpret.
WHAT THIS STUDY ADDS
We identified novel autoantibodies in SSA− SjD that, when combined with readily available clinical variables, provide good predictive ability to discriminate (1) SSA− SjD from control participants and (2) abnormal salivary gland biopsies from normal salivary gland biopsies.
HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
This study provides novel diagnostic antibodies addressing the critical need for improvement of SSA− SjD diagnostic tools.
Introduction
Sjögren disease (SjD) is an autoimmune exocrinopathy with characteristic focal lymphocytic infiltrate of salivary glands (SGs) that results in symptoms of oral and ocular dryness. Although patients most commonly experience exocrine gland-related symptoms, over 40% of individuals have extraglandular systemic organ involvement.1
The diagnosis of SjD is challenging. Dryness is common, present in up to 65% of the general population2; however, SjD has a prevalence of <1%.3 Unlike dryness attributed to many other causes, SjD is an autoimmune exocrinopathy. Detecting autoimmunity, and thus diagnosing SjD, commonly requires either a positive anti-SSA antibody test or a labial SG biopsy with a focus score (FS) ≥1 (ie, ≥1 foci (50 mononuclear cells) per 4 mm2 of tissue).4 5
Accurate diagnostic testing is critical because patients with SjD need to be followed longitudinally for extraglandular organ involvement and appropriate targeted therapeutic intervention. Anti-SSA antibodies (Ro52 and/or Ro60) are present in 40%–68% of patients with SjD.1 Thus, about one-third of patients with SjD are anti-SSA antibody negative (seronegative or SSA−). This ‘seronegative’ patient population often requires a labial SG biopsy for diagnosis.5 6 A specialist is required to perform the biopsy and pathologists experienced in FS calculation must interpret results. The latter requirement is often overlooked, but re-evaluation of SG biopsies by expert pathologists results in a diagnostic revision in over half of cases.7 Moreover, a labial SG biopsy is invasive with a rare risk of local numbness. Understandably, patients can be reluctant to undergo this procedure. Thus, procuring and interpreting labial SG biopsies are limiting steps toward a timely diagnosis of SjD.8
To address the challenges associated with labial SG biopsies, a major need in the SjD community is new biomarkers to diagnose SSA− SjD. Ideally, biomarkers have high sensitivity and specificity, and use specimens that are readily available (eg, blood, tears or saliva). We identified novel autoantibodies in seronegative SjD sera using a whole human peptidome array, confirmed our results with ELISA, and applied our results towards predicting SjD.
Methods
Population
For the human peptidome array and validation ELISAs, we used sera from eight anti-SSA antibody negative (SSA−; meaning negative for Ro52 and/or Ro60) participants with SjD meeting SjD American College of Rheumatology (ACR)/EULAR 2016 criteria9 and eight age-matched and sex-matched controls without autoimmune disease (table 1) as previously described from the University of Wisconsin (UW) Rheumatology Biorepository.10 11
Demographics of participants
For external validation, we used samples from the Sjögren’s International Collaborative Clinical Alliance (SICCA) registry and biorepository, a multisite international registry housed at the University of California, San Francisco. Participants were enrolled in the SICCA registry if they had (i) known SjD, (ii) SG enlargement, (iii) repeated dental caries without risk factors or (iv) abnormal serology (anti-SSA or anti-SSB antibody, antinuclear antibody (ANA) or rheumatoid factor (RF)). Registry details can be found at https://siccaonline.ucsf.edu.9 12 13
All participants with SjD from the SICCA registry met the 2016 ACR/EULAR criteria.14 We compared participants with SSA− SjD (n=76) with sicca-controls (n=75) and participants with autoimmune-features (AF-controls) (n=41 table 1). Sicca-controls had symptoms or signs of dryness but lacked autoimmunity (ANA <1:320, negative RF, negative anti-SSA antibodies and FS <1 on labial SG biopsy).15 AF-controls had ANA ≥1:320, positive RF or FS ≥1 on labial SG biopsy but did not meet the 2016 ACR/EULAR criteria for SjD.4 These subjects might represent a control, non-specific ‘undifferentiated connective tissue disease’, or preclinical autoimmune disease. Anti-SSB antibody positivity was similar between these groups. The SICCA registry recorded rheumatoid arthritis (RA) and systemic lupus erythematosus (SLE) diagnosed by the treating physician but did not independently confirm these diagnoses. Additional analyses were performed comparing samples from the SICCA registry for SSA+ SjD (n=75) and from the UW Rheumatology Biorepository for rheumatologist-diagnosed SLE (n=20), and rheumatologist-diagnosed subjects with RA (n=20).11
Patient and public involvement
Patients/Public were not involved in the design, conduct or reporting of the manuscript.
Whole peptidome array and statistical analysis
To evaluate autoantibody reactivity, identify common features of antigens and better understand SjD, we detected serum IgG binding to a whole human peptidome array (Roche NimbleGen, Madison, Wisconsin, USA).16 The peptidome array comprised over 5.3 million overlapping 16 amino acid peptides tiled at 2 amino acid intervals across the human proteome. We describe our whole peptide array analysis and peptide selection in the online supplemental methods.
Supplemental material
Enzyme-linked immunosorbent assay
ELISA was optimised for serum concentration, peptide concentration and incubation duration and performed as described in the online supplemental methods. We used the following controls: (1) blank wells, (2) wells with peptide coating but no sera, (3) wells with sera but no peptide coating and (4) two positive controls on each plate used to normalise for plate-to-plate variation. Homologous and heterologous inhibition assays were performed as described in the online supplemental methods. We reported the coefficient of variation (CV) (the ratio of the SD to the mean).
Immunohistochemistry
Description present in the online supplemental methods.
Statistical analysis and model building from ELISA results
Our statistical approach is described in detail in our online supplemental methods. To build predictive models incorporating clinical variables, we used adaptive lasso for clinical variable selection (predicting SjD vs sicca-control and FS positive vs negative) (Jmp Pro V.17, Cary, North Carolina, USA). Of 21 clinical variables (online supplemental table A), the top 6 significant factors identified by adaptive lasso regression included ocular staining score ≥5, platelet count, IgG, ANA ≥1:320, RF and unstimulated whole salivary flow. Because ocular staining scores are not readily available to most clinicians, we included platelet count, IgG, ANA, RF and unstimulated whole salivary flow into predictive model calculations that incorporate the new peptides entering external validation.
Supplemental material
Separate logistic regression models were created to predict odds of SjD or positive FS as a function of adjusted optical density (OD) for the peptides and clinical variables identified from the adaptive lasso. Graphical exploration of continuous features suggested some could benefit from transformation (log or square-root) prior to inclusion in models. Continuous features (or transformations thereof) were initially modelled using restricted cubic splines (3 knots) to allow for potential non-linear associations. Performance was quantified with receiver operating characteristic (ROC) curve (C-statistic) and further adjusted for model optimism.17 Final model construction and validation was performed using R (V.4.2.2)18 and the associated rms package.19 Nagelkerke’s R-squared (R2 N) measure20 was used to determine optimism adjusted values. Absolute reduction in the area under the ROC curve (AUC) or in R2 N (vs the full model with all relevant (transformed) predictors) is given for each single-term deletion. Additionally, we used random subsampling (ie, Monte Carlo cross-validation) to check the capacity of novel peptide binding data to improve outcome prediction beyond the use of clinical variables alone21 22; we used 10 000 random splits and an 80/20 training test ratio, although results were relatively insensitive to that ratio. Within each training set, we used marginal prescreening and stepwise model selection to obtain separate logistic regressions using clinical variables only or clinical variables and peptide variables, and we compared prediction accuracy on the test sets via differences in AUC.
The prevalence of positive FS among patients referred for minor SG biopsy (prevalence of 0.173 (95% CI 0.113 to 0.254)),23 together with positive and negative likelihood ratios (at various cut-points) was used to compute positive and negative predictive values. CIs for positive and negative predictive values were developed from the separate CIs for the prevalence and likelihood ratios.24
Results
Whole human peptidome array analysis
Of >5.3 million peptides, our analysis yielded 469 peptides bound more by IgG in SSA− SjD sera than controls and 431 peptides bound less by SSA− SjD serum IgG than controls (figure 1). We identified five motifs from the peptides bound more in SSA− SjD than control participants (figure 2A). Of these five motifs, three had hits on PROSITE to proteins relevant in SjD. Motif (HYA)-G-(YW)-G-(QG)-(ADT)-(NG)-(DTA)-(AT)-(SND)-(SYK) is found in heterogeneous nuclear ribonucleoprotein (hnRNP), A-kinase anchor protein 8-like and serine protease 55. Motif (MP)-(GE)-F-(RP)-(GD)-(NLK)-(PD)-G-(NQK)-(FD)-(VG) is found in complement c1q tumour necrosis factor-related protein (CTRP)2 and collagenα6(IV) chain. Motif (Y-H-P-I-P-Q-E-N-T-G-V) matches to serine/threonine protein kinase Nek5 (part of Never in Mitosis A Kinases), which controls cell cycle progression, protein quality control and mitochondrial DNA remodelling.25 Of the four motifs from peptides bound less in SSA− SjD than controls (figure 2B), none could be matched on PROSITE to <100 known proteins.
Consort flow diagram demonstrating selection of top peptides for ELISA confirmation. Starting with over 5.3 million peptides on the human peptidome array, we reduced our peptides of interest to 469 peptides bound more in SSA− SjD than controls and 431 peptides bound less in SSA− SjD than controls with our whole peptidome array analysis. This was narrowed to 22 peptides bound more in SSA− SjD than controls and 2 peptides bound less in SSA− SjD than controls by narrowing to those peptides with a fold increase in SSA− SjD versus control of at least 10, requiring at least two significant peptides bound in the same protein, and at least half of participant bound more than a threshold (threshold defined as mean+1 SD of all peptide signals on the array). 15 candidate peptides were ultimately selected for external validation after removing peptides where less than half of the SjD values were beyond the SEM of control participants. Controls had no known autoimmune disease; SSA− SjD met 2016 ACR/EULAR criteria for SjD but were anti-SSA antibody negative. ACR, American College of Rheumatology; OD, optical density; SjD, Sjögren disease.
Peptide motifs, Gene Ontology and functional analysis of proteins for which peptides were bound by IgG more or less in SSA− SjD than non-autoimmune controls. (A) Peptide motifs bound more by SSA− SjD IgG than control IgG (n=8 participants, n=469 peptides); (B) peptide motifs bound less by SSA− SjD IgG than control IgG (n=8 participants, n=431 peptides); (C) Gene Ontology of peptides bound more by SSA− SjD IgG than control IgG; (D) Gene Ontology of peptides bound less by SSA− SjD IgG than control IgG; (E) Functional protein domain binding analysis of peptides bound more by SSA− SjD IgG than healthy control sera; (F) Functional protein domain binding analysis of peptides bound less by SSA− SjD IgG than control IgG. Control had no known autoimmune disease; SSA− SjD met 2016 ACR/EULAR criteria for SjD but were anti-SSA antibody negative. ACR, American College of Rheumatology; ES, enrichment score; SjD, Sjögren disease.
Among peptides bound more by SSA− SjD IgG, Gene Ontology (GO) analysis of biological processes, cellular components and molecular functions showed a top enriched cluster of postsynaptic/cell junction (figure 2C; online supplemental table B). Among peptides bound less with SSA− SjD IgG, the top enriched GO cluster was sarcoplasmic reticulum (figure 2D; online supplemental table B). We also evaluated protein domains. Top protein domains identified from peptides bound more by SSA− SjD IgG include RNA binding, zinc finger and alpha actinin (figure 2E; online supplemental table C). The top protein domains from peptides bound less by SSA− SjD IgG include Ca2+ channel signalling, WD-40 repeats and ankyrin repeats (figure 2F; online supplemental table C).
Antibodies to D-aminoacyl-tRNA deacylase 2 and retroelement silencing factor 1 are higher in participants with SSA− SjD than control participants
We validated our top candidate array peptides (n=24) with ELISA using the same participant sera that was used for the array (‘internal validation’). Based on the results of our internal validation (figure 3 and online supplemental figure A), we selected 15 peptides for external ELISA validation using different participant sera. Dot plots of external validation findings are shown in online supplemental figure B. Using two-sided Wilcoxon rank-sum tests, we found IgG binding to peptides from D-aminoacyl-tRNA deacylase 2 (DTD2) had an estimated 64% chance of an adjusted OD higher for a SSA− SjD than a sicca-control participant (95% CI 54% to 72%; p=0.004; average CV 4.5%; figure 4A). We found IgG binding to peptides from retroelement silencing factor 1 (RESF1) had an estimated 59% chance of an adjusted OD higher for a SSA− SjD than sicca-control participants (95% CI 50% to 68%; p=0.047; average CV 8.8%; figure 4A). CV (%) for the OD of all the peptides are mentioned in online supplemental table D. Results remained significant after excluding anti-SSB antibody (SSB+) positive subjects.
Supplemental material
Supplemental material
Internal validation of peptides identified from the array as bound more by SSA− SjD sera than control sera. ELISA results of IgG binding from SSA− SjD sera versus control sera to peptides from different proteins. The participants used for these ELISAs were the same as those used on the array (n=8 age of participants with SSA− SjD, sex, race matched to n=8 control participants). P values reported in each panel were determined by Mann-Whitney U test. Control participants had no known autoimmune disease; SSA− SjD met 2016 ACR/EULAR criteria for SjD but were anti-SSA antibody negative. ACR, American College of Rheumatology; DTD2, D-aminoacyl-tRNA deacylase 2; OD, optical density; RESF1, retroelement silencing factor 1; SjD, Sjögren disease.
IgG from SjD and FS-positive participants binds DTD2 peptides more than controls and DTD2 is high in SSA− SjD salivary glands. IgG from participants with SSA− SjD bind peptides from DTD2 and RESF1 more than control IgG. IgG from FS-positive participants bind peptides from RESF1, DTD2 and SCRB2 more than FS-negative IgG. (A) AUC of the adjusted optical density of peptide groups between SSA− SjD (n=76) and sicca-controls (n=75); (B) AUC of the adjusted optical density of peptide groups between SSA− SjD (n=76) and combined sicca-controls and autoimmune-feature-controls (n=116); (C) one-sided Wilcoxon rank-sum test with Benjamini-Hochberg correction (q-value) for SjD versus combined control participants; (D) Kruskal-Wallis of SSA− SjD (n=76), SSA+ SjD (n=75), combined control (n=116), SLE (n=20) and RA (n=20) IgG binding to DTD2 peptide; (E) immunohistochemistry analysis using anti-DTD2 protein antibody shows significantly higher DTD2 in SSA− SjD (n=4) compared with sicca-control (n=3) salivary glands. No primary antibody with secondary only (−) and anti-DTD2 plus secondary antibody (+). Representative images at 10× magnification; (F) AUC comparing distributions of adjusted optical density of peptide groups comparing between FS-positive versus FS-negative biopsies (n=85 FS positive and n=107 FS negative). The forest plot shows the degree of IgG binding to the peptide of interest differed between FS-positive and FS-negative comparisons; (G) one-sided Wilcoxon rank-sum test with q-value of binding from FS-positive versus FS-negative participants. SSA− SjD met 2016 ACR/EULAR criteria for SjD but were anti-SSA antibody negative. Sicca-controls=subjects with symptoms or signs of dryness but negative ANA, RF, SSA and FS <1; autoimmune-feature-controls=subjects with symptoms/signs of dryness and ANA ≥1:320, positive RF or FS ≥1 on labial salivary gland biopsy but not meeting 2016 ACR/EULAR SjD criteria; combined controls=sicca and autoimmune-feature-controls combined; subjects with SLE and RA were diagnosed by a rheumatologist. ACR, American College of Rheumatology; ANA, antinuclear antibody; AUC, area under the receiver operating characteristic curve; DTD2, D-aminoacyl-tRNA deacylase 2; FS, focus score; RA, rheumatoid arthritis; RESF1, retroelement silencing factor 1; RF, rheumatoid factor; SjD, Sjögren disease; SLE, systemic lupus erythematosus.
Next, we compared antibody binding to our top candidate peptides in SSA− SjD with a combination of sicca-controls and AF-controls (combined controls). As above, peptides from DTD2 and RESF1 were bound more by SSA− SjD than combined controls IgG (p=0.003 and p=0.033, respectively; figure 4B). Results remained significant after excluding SSB+ participants. IgG binding to DTD2 had a 63% chance of observing a higher adjusted OD in participants with SSA− SjD than combined controls (95% CI 54% to 70%). Binding to RESF1 had a 59% chance of being higher in SSA− SjD than combined control participants (95% CI 51% to 67%). Recognising the directional changes from the array data, we performed a one-sided Wilcoxon rank-sum test with Benjamini-Hochberg correction (q-value) to control the false discovery rate (figure 4C). IgG binding to peptides from DTD2 survived at 5% (q=0.021). Binding to DTD2 was similar between SSA− SjD and SSA+ SjD but higher in participants with SjD than SLE and RA (figure 4D; online supplemental figure C). When sensitivity analysis was performed including only participants with RF-negative SjD, eliminating the risk of RF-related interference, IgG binding to DTD2 remained significantly higher in SjD than control subjects (online supplemental figure D).
Supplemental material
Supplemental material
Anti-DTD2 antibody is specific to DTD2 peptide and DTD2 is increased in salivary gland tissue
The binding specificity of IgG to the DTD2 peptide was assessed by incubating two SjD serum samples with DTD2 peptide or a control peptide prior to ELISA to detect IgG binding to plate-bound DTD2 peptide. We observed a dose-dependent inhibition from 13% to 91% for pre-incubation with 2–100 µg/mL of DTD2 peptide with no substantial inhibition for the control peptide (online supplemental figure E). Furthermore, more DTD2 is present in SSA− SjD than sicca-control SG tissue (figure 4E; online supplemental figure F).
Supplemental material
Supplemental material
Antibodies to RESF1, DTD2 and SCRB2 are higher in participants with labial salivary gland biopsies with an FS ≥1 than biopsies with an FS <1
Because a surrogate marker for a positive or negative labial SG biopsy is a significant clinical need, we evaluated whether autoantibody binding to the 15 peptides differed between participants who had a positive biopsy (FS ≥1) compared with a negative FS on biopsy (FS <1). We found that IgG from FS ≥1 participants bound peptides from RESF1, DTD2 and SCRB2 more than control FS <1 participants (p=0.010, p=0.012, p=0.027, respectively; figure 4F). Results remained significant after excluding SSB+ participants. IgG to RESF1 and DTD2 both had an estimated 61% chance that adjusted OD would be higher for a positive than a negative FS (95% CI 53% to 68% and 52% to 68%, respectively). IgG to SCRB2 had an estimated 59% chance that adjusted OD would be higher for a positive than negative FS (95% CI 51% to 67%). We performed one-sided Benjamini-Hochberg correction to control false discovery rate. Peptides from RESF1 and DTD2 survived at 5% (q=0.044; q=0.044; figure 4G).
A predictive model incorporating clinical variables shows good discrimination between participants with SSA− SjD and combined control participants
We generated a regression model to predict SSA− SjD by incorporating IgG binding to our peptides into a model with clinical variables. After model selection, the predictive model included IgG binding to DTD2 (square-root transformed), unstimulated salivary flow (square-root) and ANA (other peptide binding and clinical factors did not add to the model; figure 5A). This SjD prediction score discriminated between participants with SSA− SjD and control participants. AUC (C-index) was 73.5% (95% CI 66.0% to 79.9%), which decreased to 72.2% after correcting for optimism. Unstimulated salivary flow contributed the most to the model (single term deletion of unstimulated salivary flow yielded a >5.5 percentage point reduction in AUC) and second most important was binding to DTD2 (single term deletion of DTD2 binding yielded a >3.6 percentage point reduction in AUC). The model calculates a prediction score that is higher in participants with SjD than combined control participants (figure 5B). On cross-validation, we showed that models using clinical predictors plus IgG binding to DTD2 had better overall prediction accuracy than models that used only the clinical variables (figure 5C). Thus, inclusion of peptide binding improves the performance of models predicting SjD versus control.
Models that incorporate binding to a peptide from DTD2 have good predictive ability for SjD. (A) The selected predictive model incorporated three predictors (IgG binding to a peptide from DTD2, unstimulated salivary flow and high ANA) with an AUC of 73.5% (95% CI 66.0% to 79.9%), which decreased to 72.2% after correcting for optimism. The table shows estimated model coefficients and their SEs as subscripts. The effects of single term deletion are shown; (B) dot plot showing the separation between SSA− SjD and combined controls by SjD prediction model score; (C) in separate Monte Carlo cross-validation, we repeatedly and randomly split the external validation data into 80% training and 20% testing; we used model selection on each training set, separately for clinical only variables or clinical plus peptide variables, and we built prediction rules on the test set. The improvement in AUC by including peptide variables is expressed in the shift above the diagonal line. Levels refer to histogram bin frequency in 10 000 training/test splits; (D–E) specificity and sensitivity graphed separately for cut-points of the score ranging from −1.6 to 1.6. Optimism-corrected values as dotted lines closely track the original values; (F–G) positive and negative predictive value graphed separately. SjD=subjects with SSA SjD who met 2016 ACR/EULAR criteria for SjD but were anti-SSA antibody negative; controls=combined controls (sicca combined with autoimmune-feature-controls); sicca-controls=subjects with symptoms or signs of dryness but negative ANA, RF, SSA and FS <1; autoimmune-feature-controls=subjects with symptoms/signs of dryness and ANA ≥1:320, positive RF or FS ≥1 on labial salivary gland biopsy but not meeting 2016 ACR/EULAR SjD criteria. ACR, American College of Rheumatology; ANA, antinuclear antibody; AUC, area under the receiver operating characteristic curve; DTD2, D-aminoacyl-tRNA deacylase 2; FS, focus score; NPV, negative predictive value; PPV, positive predictive value; RF, rheumatoid factor; SjD, Sjögren disease; UWS, unstimulated whole salivary flow.
Sensitivity, specificity, positive predictive value and negative predictive value are shown in figure 5D–G. Using the selected predictive model, we can select thresholds that are either highly specific or highly sensitive, potentially confirming an SSA− SjD diagnosis without the need for biopsy in 5% of participants (n=4/76) or avoiding the need for a biopsy in 13% of controls that will not achieve an SjD diagnosis (n=15/116). Clinical features of the highest quartile of subjects by DTD2 binding compared with the lowest three quartiles are similar (online supplemental table E).
A predictive model incorporating clinical variables shows good discrimination between FS-positive and FS-negative participants
We generated a regression model incorporating IgG binding to our peptides with clinical variables. The selected predictive model included IgG binding to DTD2 (square-root), unstimulated salivary flow (square-root), platelet count (log transformed) and ANA (figure 6A). The C-index of the model was 71.6% (95% CI 63.9% to 78.2%) and decreased to 69.3% after correcting for optimism. Binding to DTD2 contributed the most to the model (single term deletion of DTD2 yielded a >3.9 percentage point reduction in AUC) and the second most important was unstimulated salivary flow (single term deletion of unstimulated salivary flow yielded a 3.3 percentage point reduction in AUC). This final ‘FS prediction score’ discriminated between FS positive and FS negative (figure 6B). On cross-validation, we showed that models using clinical predictors plus IgG binding to DTD2 had overall better prediction accuracy than models that used only the clinical variables (figure 6C).
Models that incorporate binding to a peptide from DTD2 have good predictive ability for FS positivity. (A) The selected predictive model incorporated four predictors (IgG binding to a peptide from DTD2, unstimulated salivary flow, platelet count and high ANA) with an AUC of 71.6% (95% CI 63.9% to 78.2%). The table shows estimated model coefficients and their SEs in subscript. The effects of single term deletion are shown; (B) dot plot showing the separation of a model score between positive and negative FS groups; (C) in separate Monte Carlo cross-validation, we repeatedly and randomly split the external validation data into 80% training and 20% testing; we used model selection on each training set, separately for clinical only variables or clinical plus peptide variables, and we built prediction rules on the test set. The improvement in AUC by including peptide variables is expressed in the shift above the diagonal line. Levels refer to histogram bin frequency in 10 000 training/test splits; (D–E) specificity and sensitivity graphed separately for cut-points of the score ranging from −1.6 to 1.6. Optimism-corrected values as dotted lines and differ from original values by at most 2.6 or 1.8 percentage points for sensitivity and specificity, respectively; (F–G) PPV and NPV graphed separately. ANA, antinuclear antibody; AUC, area under the receiver operating characteristic curve; CV, coefficient of variation; DTD2, D-aminoacyl-tRNA deacylase 2; FS, focus score; NPV, negative predictive value; PPV, positive predictive value; UWS, unstimulated whole salivary flow.
We calculated sensitivity and specificity for FS prediction score cut-points (range −1.6 to 1.6; figure 6D–E). Positive and negative predictive values are shown (figure 6F–G). Positive likelihood ratios could only be computed for cut-points ranging between −1.6 and 1.0, since none of the FS-negative group had calculated scores over 1.02; consequently, positive predictive values were defined over this limited range as well. If we select a stringent positive score indicating a biopsy will result in a positive FS (with no FS false positives), we could avoid the need for an SG biopsy in 14% of patients (n=12/85). If we select a stringent negative score indicating an SG biopsy will be negative (no false negatives), we could avoid the need for a biopsy in 5% of patients (n=5/107).
Discussion
We describe new autoantibodies targeting peptides from DTD2 and RESF1 that are higher in SSA− SjD than relevant sicca and AF-controls. We also describe new autoantibodies targeting peptides from DTD2, RESF1 and SCRB2 that are higher in FS-positive than FS-negative participants. When DTD2 binding was combined with clinical features, we achieved good predictive discrimination between SSA− SjD and control participants and also between FS-positive compared with FS-negative participants. Higher abundance of DTD2 protein in SSA− SjD SGs adds biological validity to our serological findings.
Novel autoantibodies that help diagnose SSA− SjD fill a major gap in care. The current standard for diagnosis of SSA− SjD frequently requires a labial SG biopsy. There are well-recognised barriers to pursuing this biopsy including (1) concern from the patient about potential adverse effects, namely permanent numbness at the biopsy site; (2) finding a practitioner for biopsy performance and (3) identifying a pathologist with experience calculating an FS. Given these barriers, developing novel diagnostic tests using readily available sources is a major unmet clinical need. We showed that antibodies targeting peptides from DTD2 can be used along with standard clinical metrics to detect SSA− SjD or a positive FS with good discrimination. Indeed, binding to a peptide from DTD2 was the most important single term in our final model for FS prediction. Furthermore, we identified score cut points that yield a high specificity or positive predictive value. Patients with a high SjD or FS prediction score might not need a labial SG biopsy to confirm their diagnosis of SjD. On the other hand, we can select cut points with high sensitivity or negative predictive value. Patients with a very low score might not need to proceed to labial SG biopsy because ultimately, they will not achieve criteria for SjD or have a positive FS.
Given the need for novel diagnostic testing for SSA− SjD, others have also sought to detect autoantibodies. Longobardi et al used a human proteome array with 19 500 proteins to identify 11 antibodies targeting novel proteins that were confirmed on a discovery and validation dataset.26 Using a panel of 12 antigens, they developed a predictive model with an AUC of 0.88. Unfortunately, none of the proteins targeted in the constructed model predicted SSA− SjD on external validation. Another common panel is the ‘early antibody’ or ‘Sjo’ panel comprising SG protein-1 (SP1), carbonic anhydrase 6 (CA6) and parotid secretory protein (PSP) antibodies. Initially identified in mice, an early publication found higher positivity of these antibodies in SjD compared with matched controls; however, only six subjects with anti-SSA/SSB antibody negative SjD were included in this study. The largest study of these antibodies came from the University of Pennsylvania SICCA cohort that compared SjD (n=81) with non-SjD (n=129) and found no difference in overall anti-SP1 or anti-CA6 antibody positivity between groups.27 A significant difference was reported in anti-PSP antibody with 35% positivity in SjD compared with 21% positivity in non-SjD (p=0.04).27 Again, only six subjects with SSA− SjD were included in the SjD group, limiting the generalisability of these findings to this population. Additional studies have failed to show an increase of these antibodies in children.28 Aquaporin 5 antibodies are higher in subjects with SjD than subjects without SjD29 30 and correlate with FS,30 but the performance of the assay in subjects with SSA− SjD remains unclear. Other autoantibodies have been described in SjD but to the authors’ knowledge no other autoantibodies have been validated as diagnostic in SSA− SjD.31
We found that autoantibodies in SSA− SjD bound a peptide from DTD2 significantly. DTD2 recycles D-aminoacyl-tRNA to D-amino acids and free tRNA molecules, preventing D-type amino acids from forming proteins.32 DTD2 might act as a proofreading mechanism since defective tRNA synthetase in mice results in neurodegeneration from misfolded proteins.33 34 We also found SSA− SjD bound a peptide from RESF1 more than controls. RESF1 regulates gene expression and repressive epigenetic modifications. Specifically, it recruits SETDB1 for endogenous retrovirus silencing.35 RESF1 also promotes embryonic stem cell self-renewal.36 Finally, SCRB2 was bound more in FS-positive than FS-negative labial biopsies. SCRB2 is a lysosomal receptor for glucosylceramidase37 and a receptor for enterovirus.38 The absence of SCRB2 decreases macrophage and T-cell response in mouse models of crescentic glomerulonephritis39 and Listeria infection.40 It is unclear how IgG that recognises linear epitopes in these proteins might contribute to SjD; indeed, it is not known if full-length proteins are bound by these autoantibodies. However, like other autoantigens in systemic autoimmunity (eg, Ro and histones), DTD2 and RESF1 interact with nucleic acids. Perhaps nucleic acids provide a danger signal via toll-like receptors to stimulate the autoimmune response.41 42 Further studies are needed to understand whether these autoantibodies might be pathogenic.
We used innovative technology, a whole human peptidome array, for initial autoantibody identification. Motifs for the 469 peptides bound more in SSA− SjD than healthy controls on the array were associated with SjD-relevant proteins. For example, an identified motif is present in hnRNPs. hnRNPs are a family of RNA-binding proteins that associate with messenger RNA to form protein-RNA complexes that act as substrates for RNA processing43 and are autoantigens in SjD44 along with other systemic rheumatic diseases such as RA, SLE and systemic sclerosis (SSc).45–47 Another motif is present in complement c1q tumour necrosis factor-related protein, CTRP2. CTRP2 regulates insulin tolerance and lipolytic enzymes.48 CTRP2 is similar in structure to adiponectin in the globular domain and can induce phosphorylation of AMP-activated protein kinase (AMPK) and Akt.49 AMPK activation, relevant to SjD, inhibits mammalian target of rapamycin (mTOR) which affects cell growth, survival and proliferation and regulates T cell differentiation.50 We and others have shown that metformin, through mTOR inhibition, might improve SjD.51 52 Thus, motifs bound in SSA− SjD might provide some insight into the functional relevance of our antigenic targets.
Strengths of our study include the innovative whole human peptidome array and our novel statistical approach to identify new peptide targets. We include clinically relevant controls who would typically be referred for an SG biopsy or a possible SjD diagnosis. Finally, we use robust sample sizes to validate our array findings in a large independent population. Limitations of this study include the linear nature of the array and ELISA peptides, which do not have the conformational structure of native proteins. Although in other diseases linear epitopes can be bound, the whole protein binding of anti-DTD2 is unknown.16 53 54 The peptide array does not account for protein modifications, such as glycosylation. Given that the validation cohort was derived from participants referred to the SICCA registry, referral bias cannot be excluded. We tried to limit this bias by including a diverse cohort of participants. Future analyses will include differentiation of anti-DTD2 antibody binding by anti-Ro52/R60 antibody status and study a larger array of diseases such as SSc, hepatitis C infection, sarcoidosis or IgG4-related disease, among others.
In conclusion, we present novel autoantibodies in SSA− SjD compared with AF-controls and sicca-controls that can be used to predict a SjD diagnosis or abnormal FS on labial SG biopsy with good predictive value.
Data availability statement
Data are available on reasonable request. All data relevant to the study are included in the article or uploaded as supplementary information. All data relevant to the study are included in the article or uploaded as supplementary material but data are available on reasonable request.
Ethics statements
Patient consent for publication
Ethics approval
This study was approved by the University of Wisconsin Health Sciences (IRB# 2021-0821 and IRB# 2015-0156). Participants gave informed consent to participate in the study before taking part.
References
Footnotes
Handling editor Josef S Smolen
Contributors Contribution statement: MP, SMcC, MLar and MAS planned experiments. MP, ZZ and AVL carried out the experiments. RAA, MAN and MRL analysed the experiments. MAS, SMcC, MAN, RAA, ZZ and MRL contributed to the interpretation of the results. SMcC and MP wrote the bulk of the manuscript and all authors provided critical feedback and helped shape the research, analysis and manuscript. SMcC is responsible for the overall content as guarantor.
Funding Support for this research was provided by the Sjögren’s Foundation and University of Wisconsin-Madison, Discovery to Product (D2P) with funding from the Wisconsin Alumni Research Foundation (SMcC) and the University of Wisconsin-Madison, Office of the Vice Chancellor for Research and Graduate Education with funding from the Wisconsin Alumni Research Foundation (MAS). Additional support was provided by the Clinical and Translational Science Award programme through the NIH NCATS (1KL2TR002374) (SMcC), US Army Medical Research Acquisition Activity through the Peer Reviewed Medical Research Programme (W81XWH-18-1-0717) (MAS), and UL1TR002373 (MAN). The data reported here have been supplied by the Sjögren's International Collaborative Clinical Alliance (SICCA) Biorepository by the National Institute of Dental and Craniofacial Research (contract HHSN26S201300057C). The interpretation and reporting of these data are the responsibility of the authors and in no way should be seen as an official policy or interpretation of the SICCA investigators or the National Institute of Dental and Craniofacial Research.
Disclaimer The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.
Competing interests SMcC declares consulting for Novartis, Horizon, Target RWE, Otsuka/Visterra, Kiniksa, BMS and iCELL.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.