Two cohorts of patients and healthy donors were recruited at the Rheumatology Department of Hospital Universitario La Princesa (HUP, Madrid, Spain): a main cohort including 51 SLE (35 iSLE and 16 aSLE), 52 SSc patients (38 lSSc and 14 dSSc), and 66 healthy donors (HD), of which 52 were age and sex matched to SLE patients and 54 to SSc patients, and a second cohort including 13 SLE, six SSc, and 15 HD gender-matched. SLEDAI index was used to categorize SLE patients according to disease activity, considering those patients with SLEDAI value ≤ 4 as iSLE and those patients with SLEDAI values > 4 as aSLE (39). SSc patients were classified according to EULAR/ACR classification criteria for SSc as patients with lSSc or dSSc (10). All patients were previously diagnosed and had established diseases.

Serum levels of PSGL-1, ADAM8, and P-, E-, and L-selectins, as well as ADAM8 cell membrane expression, were analyzed in the main cohort, and L-selectin cell membrane expression was analyzed in the second cohort.

This cross-sectional study was realized following the “Strengthening the Reporting of Observational Studies in Epidemiology” (STROBE) recommendations and was approved by the Ethics Committee for Drug Research of HUP (reference numbers: No. PI758: acta 14/14, approved date 24 July 2014; No. 3106: acta 11/17, approved date 08 June 2017; and No. 4033: acta CEIm 05/20, approved date 3 December 2020). This study has been carried out in accordance with the Code of Ethics of the World Medical Association (Declaration of Helsinki) for experiments involving humans (https://www.wma.net/policies-post/wma-declaration-of-helsinki-ethical-principles-for-medical-research-involving-human-subjects). Informed consent was obtained and signed by all subjects involved in the study.

Clinical symptoms, presence of autoantibodies, treatment, and other clinical characteristics are summarized in Tables 1 , 2 .

Serum levels of P-, E-, and L-selectin, PSGL-1, and ADAM8 were determined by commercially available ELISA kits: the Human PSGL-1/CD162 ELISA Kit (Novus Biologicals, Centennial CO, USA); the Human CD62E ELISA Kit; the Human L-SELECTIN ELISA kit and Human CD62P ELISA Kit (Diaclone SAS Company, Besancon Cedex, France); and the Human A Disintegrin And Metalloprotease 8 (ADAM8) ELISA Kit (Elabscience, Houston, TX, USA). Samples were thawed at room temperature and homogenized before assay. ELISA assays were performed following manufacturer recommendations and read in a microplate reader and luminometer, GloMax Discovery (Promega, Madison, WI, USA).

To analyze the membrane expression of ADAM8 and L-selectin, 50 µl of whole blood was blocked with human γ-globulin (Sigma-Aldrich, St Louis, MO, USA) for 10 min at 4°C. Next, cells were labeled with monoclonal antibodies for specific molecular targets of monocytes (mouse antihuman CD14-APC, BioLegend, San Diego, CA, USA), neutrophils (mouse antihuman CD16-APCH7, BD Biosciences, San Diego, CA, USA), B cells (mouse antihuman CD19-APC/cyanine7, Biolegend), and T cells (mouse antihuman CD3-PE/Cyanine7, BD Biosciences, San Diego, CA, USA) combined with purified goat antihuman ADAM8 (R&D Systems, Minneapolis, MN, USA) and mouse antihuman CD62L (L-selectin)-FITC (BioLegend) for 20 min at 4°C in the dark. Cells were then washed with PBS + EDTA 5 mM + BSA 0.5% and labeled with donkey antigoat Alexa Fluor 488 (Life-Technologies-Invitrogen, Eugene, OR, USA) for 15 min at 4°C in the dark. After labeling, cells were washed and erythrocytes were lysed with 2 ml of FACS lysis solution diluted 1:10 in distilled water (BD Biosciences) for 15 min in the dark at room temperature (RT). Finally, cells were washed, acquired in a FACSCanto II cytometer, and analyzed with DIVA Software (BD Biosciences). Positivity was established using corresponding isotype-matched control antibodies. The gating strategy was established by selecting cell populations by FSC/SSC, then monocytes (CD14+), granulocytes (CD16+), T cells (CD3+), and B cells (CD19+) were identified and analyzed for L-selectin and ADAM8 expression.

Data were analyzed using SPSS 15.0 (IBM Corporation, New York, NY, USA) and R Statistical Software (v4.3.3, R Core Team 2024). The normality condition was assessed using either Kolmogorov–Smirnov’s or Shapiro–Wilk’s tests, depending on the sample sizes, whereas the homoscedasticity condition among groups was assessed using Levene’s test. Based on the results from these tests, the following analyses were performed as appropriate. When comparing two unpaired samples, a two-tailed Student’s t-test or Mann–Whitney’s U test was used, whereas one-way ANOVA or Kruskal–Wallis’ tests followed by Holm’s post-hoc correction were used when comparing more than two independent groups. Finally, for bivariate correlation analysis, Pearson’s r or Spearman’s rho coefficients were calculated according to whether the data were normally distributed or not. Statistical significance was defined at p-value < 0.05.

The association of the variables with different clinical outcomes was assessed by binary logistic regression (BLR) to discriminate HD from SLE patients, HD from SSc patients, and SLE from SSc patients. For all BLR models, odds ratio (OR), sensitivity, specificity, area under the ROC curve, and overall percentage of correctly classified individuals were calculated. To compare SLE and SSc, SSc patients were considered “positive” subjects when computing sensitivity and specificity metrics.

The variable selection process for BLR was carried out as follows: First, univariant BLR for each independent variable was performed, and corresponding receiver operating characteristic (ROC) curves were assessed. Afterward, to select the optimal cut-off value, the Youden index (YI = [{sensitivity + specificity}/100] − 1) was computed for each point of the ROC curve, so that the point with the highest YI value was chosen and used to binarize its corresponding variable. Binarized independent variables were then included in a multivariable BLR model, and backward stepwise elimination based on likelihood ratio was performed to conserve the most influential variables.

These variables identified by backward elimination were then individually included in a different multivariable BLR, together with age and sex as confounder factors. Additionally, when discriminating SLE from SSc patients, years from diagnosis, glucocorticoids, and other immunosuppressant agents were also included as confounding variables in the multivariable BLR. For BLR model evaluation, five times fourfold repeated cross-validation was carried out using R’s package caret v6.0–94 (40), splitting the global cohort each time into different training and testing subcohorts. Sensitivity, specificity, ROC area under the curve (AUC), and overall percentage of correctly classified individuals from the testing subcohort at each of the 20 subjects’ resamplings were computed as model performance estimators.

In the first approach, we characterized the serum levels of PSGL-1 and its ligands in SLE and SSc patients. SLE patients showed increased levels of sPSGL-1, sE-selectin, and sADAM8, regardless of disease activity ( Figures 1A, C, E ), and decreased sL-selectin levels ( Figure 1D ) compared to HD, while no differences were observed in sP-selectin levels ( Figure 1B ). In SSc patients, sPSGL-1, sP-selectin, and sADAM8 levels were similar to those of HD ( Figures 1F, G, J ), whereas sE-selectin was higher and sL-selectin lower than in HD ( Figures 1H, I ). We also found higher serum levels of PSGL-1, E-selectin, and ADAM8 in patients with diffuse SSc compared to those with limited SSc ( Figures 1F, H, J ). Comparative analysis between the two autoimmune disorders revealed that SLE patients exhibited higher sADAM8 levels ( Figure 1O ) and lower sL-selectin levels ( Figure 1N ) than SSc patients, while sPSGL-1, sP-selectin, and sE-selectin levels were similar in both diseases ( Figures 1K–M ).

The metalloproteinase ADAM8 binds to and cleaves PSGL-1, as well as E- and L-selectins, which play an important role in the regulation of leukocyte extravasation (22–24). Hence, we analyzed the correlations between serum levels of PSGL-1 and selectins with sADAM8 levels. HD showed a negative correlation of sP-selectin (r = −0.250) and sE-selectin (r = −0.310) with sADAM8 ( Figures 2A–D , left panels). SLE patients showed a negative correlation of sPSGL-1 (r = −0.246) and sP-selectin (r = −0.418) and a positive correlation of sL-selectin (r = 0.326) with sADAM8 ( Figures 2A–D , middle panels). SSc patients showed a positive correlation of sE-selectin (r = 0.270) and a negative correlation of sL-selectin (r = −0.275) with sADAM8 ( Figures 2A–D , right panels).

Previous work showed increased ADAM8 membrane expression in circulating monocytes and lymphocytes of SSc patients compared to HD (33). In this work, we did not find differences in ADAM8 membrane expression levels between SLE patients and HD in neutrophils, monocytes, and lymphocytes ( Figures 3B, D, F, H ; left panels). However, the percentage of ADAM8(+) monocytes and T lymphocytes was higher in SLE patients than in HD ( Figures 3C, E ; left panels), while no difference was found in B lymphocytes ( Figure 3G ). Importantly, the percentage of ADAM8(+) neutrophils and ADAM8 membrane expression level in neutrophils were higher in SSc than in SLE patients ( Figures 3A, B ; right panel). Regarding L-selectin, membrane levels on all leukocyte subsets ( Figures 4B, D, F, H ) and the percentage of L-selectin(+) neutrophils ( Figure 4A ) and monocytes ( Figure 4C ) were reduced in SLE and SSc patients compared to HD. Additionally, although no changes were observed in the percentage of T lymphocytes expressing L-selectin ( Figure 4E ), the percentage of L-selectin(+) B lymphocytes was also lower in SLE patients compared to SSc patients and HD ( Figure 4G ).

We explored potential associations between serum protein levels and clinical characteristics, including clinical manifestations and the administration of various treatments in both SLE and SSc patients. In SLE patients, we observed higher sADAM8 levels associated with the presence of anti-double-stranded DNA (anti-dsDNA) and anti-Sjögren’s syndrome-related antigen A (SSA) antibodies ( Figure 5A ) and with the presence of cutaneous manifestations ( Figure 5B ). Regarding treatments, we found lower sPSGL-1 levels associated with taking azathioprine (AZA) ( Figure 5C ) and lower sP-selectin levels associated with taking hydroxychloroquine (HCQ) ( Figure 5D ).

In SSc patients, lower sPSGL-1 levels were associated with the presence of anti-centromere autoantibodies, while higher levels were associated with the presence of anti-topoisomerase antibodies (anti-Scl70) ( Figure 5E ). Regarding clinical manifestations, we found higher levels of sPSGL-1 ( Figure 5F ) and sE-selectin ( Figure 5G ) were associated with the presence of ILD; higher levels of sE-selectin were also associated with the presence of PAH ( Figure 5G ). In addition, patients with ILD showed a significant reduction in the forced vital capacity (FVC)-predicted percentage ( Figure 5H ). Furthermore, sE-selectin levels negatively correlated with FVC predicted percentage (r = −0.264) in SSc patients ( Figure 5I ) and positively correlated with sPSGL-1 levels (r = 0.440) in SSc patients with ILD ( Figure 5J ).

Following the variable binarization process described in “Materials and methods”, we included categorized serum levels of PSGL-1, ADAM8, and P-, E-, and L-selectins of HD (n = 50) and SLE patients (n = 46) in a multivariable BLR analysis, adjusting by gender and age, with binary outcome for HD or SLE. As a result of backward stepwise elimination performed on this BLR model, categorized sL-selectin (sL-selectin_cat) and sADAM8 (sADAM8_cat) were identified as relevant variables for SLE outcome. Thus, we separately included each of these variables in new multivariable BLR models, adjusting by gender and age. The odds ratio (OR), sensitivity (S), specificity (Sp), AUC, and overall percentage (OP) of correctly classified individuals for these models are shown in Supplementary Table S1 .

Afterward, we computed the sL-selectin/sADAM8 ratio in HD and SLE patients and included this new variable in another multivariable BLR model adjusted by age and gender (identified as model 1 in Supplementary Table S1 ). Interestingly, we found that almost all SLE patients showed a sL-selectin/sADAM8 ratio below 2.06, while 52% of HD had a ratio over this value ( Figure 6A ). After binarizing sL-selectin/sADAM8 ratio according to this cut-off value, we obtained one last multivariable BLR model (model 2 in Supplementary Table S1 ), in which a low sL-selectin/sADAM8 ratio was identified as a significant SLE risk factor (OR = 65.65 [6.47–666.42], p = 0.0004). Most importantly, this model showed the best sensitivity value (98%) compared to our previously obtained models and correctly classified 71.9% of individuals in the cohort ( Figure 6B ; Supplementary Table S1 ).

In the absence of a distinct cohort of HD and SLE patients on which we could externally validate our results, we decided to perform a cross-validation-based evaluation of model 2. S, Sp, AUC, and OP results from five times fourfold repeated cross-validation for this model are presented in Figure 6C . Moreover, we wondered how our model would perform when compared to other well-established SLE biomarkers, such as the presence of anti-dsDNA and anti-Smith (Sm) antibodies or low levels of C3 and C4 complement proteins. Thus, we included these variables (alone or all three of them together), as well as sex and age, in multivariable BLR models and cross-validated them in the same manner. According to these results, our sL-selectin/sADAM8-based model significantly showed much better sensitivity values than all the other biomarker-based models ( Figure 6D ). Conversely, the sL-selectin/sADAM8 model was outperformed in terms of Sp by the rest of the models ( Figure 6E ). Moreover, AUC and OP values from our sL-selectin/sADAM8-based model were higher than those from anti-Sm and low complement-based models ( Figures 6F, G ).

Next, to study our discrimination capacity of SSc from HD, we similarly included binarized serum levels of PSGL-1, ADAM8, and P-, E-, and L-selectins of SSc patients (n = 52) and HD (n = 53) in a multivariable BLR analysis, together with gender and age. Backward elimination performed on this model showed that categorized sL-selectin (sL-selectin_cat) and sE-selectin (sE-selectin_cat) had a relevant contribution to the SSc outcome. Thus, we separately used each of these variables in new multivariable BLR models, adjusting by gender and age. OR, S, Sp, AUC, and OP values for these models are presented in Supplementary Table S2 .

With the aim of improving the performance of these first models, we also computed the sL-selectin/sE-selectin ratio for this cohort and calculated another BLR model, adjusting by gender and age. Results for the model based on this ratio showed high Sp but low S values. Thus, we investigated other variable ratios and found that sE-selectin/sPSGL-1 ratio-based BLR had better S ( Supplementary Table S2 ). Consequently, we obtained the best cut-off values for both ratios ( Figures 7A, B ) and used them to categorize both sL-selectin/sE-selectin and sE-selectin/sPSGL-1 ratios. Including these three categorized variables adjusted by gender and age, we obtained the models identified as model 1 (categorized sL-selectin/sE-selectin ratio), model 2 (categorized sE-selectin/sPSGL-1 ratio), and model 3 (both categorized ratios) in Supplementary Table S2 . According to the results of this last model, both the low sL-selectin/sE-selectin ratio (OR = 6.77 [2.04–22.46], p = 0.002) and the high sE-selectin/sPSGL-1 ratio (OR = 2.62 [1.09–6.28], p = 0.031) were significant risk factors for SSc compared to HD. Globally, model 3 showed the best S (67%), AUC (0.786), and OP values (71.4%) of all explored models for HD and SSc ( Figure 7C ; Supplementary Table S2 ).

Next, we assessed models 1, 2, and 3 performances by means of five times fourfold repeated cross-validation ( Figure 7D ) and compared their classification capacity with other SSc-related biomarkers, such as the presence of anti-centromere and anti-Scl70 antibodies. To do this, we cross-validated in the same way new BLR models, including the presence of anti-centromere, anti-Scl70, or both antibodies as predictor variables, adjusted by gender and age. Cross-validation results showed that model 3 had a significantly higher S than the models relying only on either anti-centromere or anti-Scl70 antibodies ( Figure 7E ), but lower Sp values ( Figure 7F ). Finally, the AUC and OP values of Model 3 were significantly higher than those of the anti-Scl70-based model ( Figures 7G, H ).

Finally, since there are currently no good biomarkers that distinguish SLE from SSc, we addressed whether our data could also help discriminate between these two diseases. To this aim, we included binarized serum levels of PSGL-1, ADAM8, and P-, E-, and L-selectins of SLE (n = 27) and SSc patients (n = 28) in a multivariable BLR analysis, together with gender, age, disease duration, and treatment (glucocorticoids and immunosuppressants) as confounder factors. Backward elimination on this model identified sADAM8 (sADAM8 _cat) as a relevant variable for the discrimination of SLE and SSc. Thus, this variable was included in a new multivariable BLR model, adjusted by all the confounding factors listed above (model 1 in Supplementary Table S3 ). However, the contribution of sADAM8 did not reach statistical significance (probably due to the influence of corticosteroid treatment).

Since SLE and SSc patients showed differences in the percentage of ADAM8(+) neutrophils ( Figure 3A ), we then computed sADAM8/% ADAM8(+) neutrophil ratio for each patient and found lower levels in SSc when compared to SLE patients, with an optimal cut-off value of 21.1 ( Figure 8A ). Accordingly, when this ratio was included alongside the confounding variables in a second BLR model, identified as model 2 in Supplementary Table S3 , its contribution to the discrimination of SLE and SSc patients was statistically significant (OR = 0.91 [0.84–1], p = 0.044). In addition, this ratio was also categorized using its identified cut-off value and individually introduced in a BLR analysis, including all the confounder factors (model 3, Supplementary Table S3 ). Model 3 improved OR (7.84 [1.4–43.89]), S (86%), and OP values (83.6%) compared to model 2 ( Figure 8B ; Supplementary Table S3 ).

Subsequently, we assessed model 3 performance by means of five times fourfold repeated cross-validation ( Figure 8C ) and compared their classification capacity with the same SSc-related and SLE-related biomarkers that we considered before (anti-centromere, anti-Scl70, anti-dsDNA, anti-Sm, and low complement levels). To carry out this comparison, we cross-validated in the same way new BLR models including these disease-specific biomarkers adjusted four all confounder factors. Cross-validation results showed that model 3 had similar S, AUC, and OP ( Figures 8D–G ) than the models relying on disease-specific biomarkers.