Introduction

Pulmonary arterial hypertension (PAH) is a potential complication of connective tissue diseases like scleroderma, systemic lupus erythematosus (SLE), and mixed connective tissue disorder (MCTD). It is one of the causes of death in SLE patients. Due to the heterogeneity and complexity of the underlying disease, pathogenesis, prevalence, risk factors, and the natural history of PAH in SLE remain poorly understood.1 As a result, there are no definite guidelines for screening, managing, and monitoring pulmonary hypertension in SLE patients.

Several autoantibodies are associated with PAH in SLE patients. Associations with antiribonucleoprotein (anti-RNP)2 and anticardiolipin antibodies3 have been reported in Chinese SLE cohorts. At the same time, Prabu et al4 found an association of PAH with lupus anticoagulant in an SLE cohort from the United Kingdom. Anti-Ro and anti-La have been associated with PAH in a French SLE cohort.5 Till now, the literature on antibody association in Indian lupus PAH patients is limited. A recently published study from the Hopkins Lupus Cohort where lupus phenotype associated with PAH was defined using multiple autoantibodies.1 Here clustering analysis was used for the first time to define five autoantibody clusters in lupus-PAH. In the present study, clustering analysis has been used to test the hypothesis that this technique would bring out unique phenotypic groups of PAH in Indian lupus patients. This method can help drive future investigations into risk factors and mediators of disease, enabling better targeted therapeutic approaches.

In another study by Sun et al,6 two distinct phenotypic clusters of SLE-PAH were identified based on disease activity. They identified two clusters—the one with high disease activity had serositis, nephritis, rash, and other disease manifestations, while the one with low disease activity had the purer form of PAH. Our study has combined autoantibodies with disease activity to generate distinct phenotypes in Indian lupus PAH patients.

Objectives

Our study had the following objectives:

• To identify distinct autoantibody-based clusters in SLE-PAH patients using cluster analysis.

• To compare demographic characters, clinical phenotypes, and therapeutic strategies across the clusters.

Statistical Analysis

K-means cluster analysis (nonhierarchical clustering or quick cluster; SPSS version 10 software [BM SPSS Statistics 21 License Authorization Wizard]) was used to identify clustering in SLE-PAH patients based on autoantibody. Briefly, this first involved defining a disease metric to quantify the degree of similarity between autoantibody patterns in two patients. Euclidian distance (the square root of the sums of squared differences between patients for each autoantibody) was used. The clusters’ initial centers were chosen in the first pass of data, and patients were assigned to the closest center. Next, the cluster centers were recalculated on the patients in the cluster. The patient process continued till the clusters’ means did not shift more than a given cutoff value or until the iteration limit was reached. Because it was not known in advance how many autoantibody clusters would be observed, 3, 4, and then 5 clusters were specified, respectively, in the K-means analysis. The outputs from the analyses with 3, 4, and 5 clusters of patients were then compared with each other regarding the prevalence of the individual autoantibody not shifting more than a given cutoff value or until the iteration limit is reached. Finally, by clustering antibodies into 3 clusters, three distinct groups of patients with very different autoantibody profiles were obtained. After clustering, associations between final clusters and patient characteristics were assessed by one-way analysis of variance (ANOVA) with post hoc analysis for continuous variables. Fisher’s exact Hamilton test for categorical variables was done; p-value < 0.05 was considered significant. All statistical analyses were performed using SPSS for Windows version 10.0.

Results

1. Baseline Characteristics

A total of 71 patients with pulmonary hypertension were identified and included in the analysis. The demographic characteristics, the frequency of specific autoantibodies, and various clinical phenotypes are shown in Table 1; 12.7% of the patients had associated skin thickening and 15.5% of patients were found to have the Raynaud phenomenon.

1. Clustering of Patients

Three distinctive antibody clusters were identified—cluster 1 (anti–Sm-RNP, anti-Sm, anti-SSA), cluster 2 (no definitive autoantibody association), and cluster 3 (anti-dsDNA, nucleosome, histone, ribosomal P protein). Twenty-six patients were assigned to cluster 1, 36 patients were assigned to cluster 2, and 9 patients were assigned to cluster 3. Low complements were found in all three clusters, while anticardiolipin and anti-SSB did not show specific association with any cluster. In cluster 2, the anticardiolipin antibody was present in 10 patients, which was not statistically significant. Table 2 depicts the three clusters with their antibody association. The frequency of the specific autoantibodies in each cluster is shown in Table 3.

1. Demographic and Clinical Characteristics across Clusters

The demographic characteristics and clinical manifestations of SLE patients in the three clusters is shown in Table 4. A significant association with cytopenia was seen in cluster 2 (77.8%) as compared to clusters 1 (42.3%) and 3 (55.6%) (p 0.01, Fisher value 8.25). Cluster 3 patients had significantly higher disease activity (p 0.049, Fisher value 5.93) and dsDNA association (p 0.001, Fisher 14.33) (Table 4).

1. Mean RVSP in Different Clusters

Mean (± standard deviation [SD]) RVSP in clusters 1, 2, and 3 were 52.9 ± 15.9, 64.7 ± 18.6, and 46 ±11.4 mm Hg, respectively (Table 5) with significant high value in cluster 2 (p 0.003) (by ANOVA testing) . Post hoc analysis showed that cluster 2 had significantly higher mean RVSP than clusters 1 and 3 (p 0.023 and 0.011) (Table 6).

Discussion

In this study, using clustering techniques for autoantibodies, three clusters were identified. The largest cluster of patients had no definite antibody association. The next large cluster was associated with anti–Sm-RNP, anti-Smith, and anti-Ro antibodies, while the smallest cluster included anti-dsDNA, antinucleosome, antihistone, and antiribosomal P protein antibodies. In patients with no definite autoantibodies, significantly more frequent cytopenia, skin thickening, and higher RVSP were noted.

There are two published studies on clustering analysis in lupus patients with associated PAH. Sun et al6 used derivation and validation cohorts, each having approximately 100 patients of SLE-PAH. They reported that patients clustered into high and low disease activity groups, with the former having less time to diagnose. They adjusted for immunosuppressive and vasodilator treatments and noted that survival differences existed between the two groups. This study did not use antibodies as independent variables.

Mizus1 studied 10 serological variables in 154 patients with SLE-PAH. One of their five clusters was defined by antiphospholipid antibodies and another by anti-Ro and anti-La. The third, as in our study, had a low frequency of all antibodies. Of note, this work was derived from the Hopkins Lupus Cohort. Though the numbers in our study are slightly less than this study, the composition of antibodies in our cluster is different from that in the Hopkins Lupus Cohort. Anticardiolipin antibodies did not figure in any cluster in our study. The cluster with significant numbers and definite antibody representation included anti-Smith and anti–Sm-RNP. This difference may highlight the genetic differences which shape autoantibody specificity among Indian and Hopkins Lupus Cohort of SLE-PAH.

There have been many cluster analysis publications in 2005, 2012, 2015, and 2019 on SLE patients. The first publication7 of cluster analysis done on 1,257 patients tested on seven autoantibodies identified three clusters with an almost similar number of patients in the groups. The cluster with dsDNA, aCL, and LAC had the highest incidence of thrombosis and stroke damage. The cluster, which included anti-Ro, anti-La antibodies, had the highest presence of Sjogren syndrome. The cluster with the Sm-RNP antibody had the lowest frequency of cytopenia. Li8 from Hongkong reported on 1,928 patients, identified the dsDNA cluster with nephritis; the Sm-RNP, Sm, aCL cluster with arthritis and serositis; and the anti-Ro/anti-La cluster with photosensitivity. In a Chinese cohort9 of 917 patients, three clusters including no ENA were identified, and differences in age of onset of lupus and other clinical features were reported. Another cluster analysis published from Johns Hopkins cohort analyzing only seven autoantigens, including Ro and RNP with different molecular weights, found two clusters. The Sm cluster was associated with serositis. From Turkey,10 852 lupus patients clustered into five clusters, one that did not have a high frequency of any antibodies but a high organ involvement. Their Sm-RNP cluster is associated with Raynaud’s phenomenon and PAH.

The concept of two different pathophysiological mechanisms of SLE-PAH in groups with different disease activities, as proposed by Sun et al, can be extrapolated to suggest the possibility of different pathological mechanisms at play in the cluster which lacked a definite antibody association in our study.

The use of immunosuppressive therapy has been reported to predict short-term survival in CTD-PAH.11 In an open-labeled observational study, we have also previously reported that immunosuppressant therapy may have a modest benefit in improving PAH in SLE patients.12 Hence at the bedside, the choice of whether to use potentially toxic immunosuppressive therapy in SLE-PAH is guided by active lupus elsewhere and serological activity in a patient. Any further study would need to account for these confounders while analyzing autoantibodies as independent predictors of survival in SLE-PAH patients.

Our study’s limitations were that pulmonary artery pressure was not documented by right heart catheterization, and other antiphospholipid antibodies (LAC, anti-B2 glycoprotein 1) were not measured.

We hope to repeat this exercise on a larger SLE-PAH and SLE group to reproduce the ethnic variation in autoantibodies in lupus and its subgroups. We are surprised to report that patients with Sm-RNP did not have a higher incidence of skin thickening or Raynaud’s phenomenon, suggesting underlying scleroderma as a PAH cause. This reiterates that the patients included in our study have no underlying scleroderma and constitute an SLE-PAH population.

Conclusion

This study confirms that in SLE-PAH, autoantibodies do exist as clusters, and this association has a bearing on phenotypic manifestations. The largest cluster had no definite antibody association but the most severe PAH. The underlying differences in different pathophysiologic autoantibody clusters will need further studies.