Background: A considerable subset of patients with systemic lupus erythematosus (SLE) fail to attain low disease activity despite treatment with standard-of-care immunosuppressive or biologic agents, sustaining persistently smoldering or refractory disease and leading to elevated risk for organ damage accrual. However, the molecular determinants sustaining treatment refractoriness remain insufficiently defined. Given SLE interpatient heterogeneity, pathway-level molecular profiling can be leveraged to elucidate the molecular hallmarks of disease refractoriness, enable patient molecular stratification and guide endotype-specific management.

Objectives: Deploying longitudinal transcriptional data, machine learning algorithms and computational drug screening, we sought to pinpoint novel therapeutic agents tailored to counteract individual refractory SLE signatures characterizing unique patient molecular endotypes.

Methods: Whole-blood RNA sequencing was performed in 21 SLE patients who did not achieve Lupus Low Disease Activity State (LLDAS) six months after treatment initiation with cyclophosphamide (n=9), rituximab (n=5), or belimumab (n=7). Functional pathway activity scores were derived using the FAIME algorithm and subjected to unsupervised clustering to define transcriptome-based molecular endotypes. Reproducibility of the clustering was evaluated in an independent belimumab-treated validation cohort (n=23). To facilitate prospective endotype assignment, a multinomial LASSO-derived Molecular Endotype Classification Index (MECI) was constructed and internally validated through 1,000-fold bootstrap resampling. An exploratory analysis of endotype-specific therapeutic modalities was performed through L1000CDS ² -based transcriptome reversal and drug–gene interaction mapping and evaluated based on published transcriptional signatures of CD19 CAR-T cell therapy.

Results: Three mechanistically distinct endotypes of treatment-resistant SLE emerged through our analyses, independent of clinical manifestations, treatment modality, or baseline disease activity. The T cell–dominant endotype was characterized by prominent PD-1 signaling, enrichment of DNA-damage response pathways, and reduced CD28 co-stimulatory signaling—collectively indicative of a senescence-skewed T-cell phenotype. The Cytokine-driven endotype was defined by amplified IL-6, IL-17, TLR, and NF-κB pathway activity with concomitant attenuation of IL-2 signaling, reflecting a disrupted Th17/Treg balance and innate immunity activation. Lastly, the inflammasome-dominant endotype was marked by upregulation of NLRP3/IL-1–related pathways, suggesting a central role of inflammasome activation. Cluster structure was consistently reproduced in the validation cohort (adjusted Rand index 0.826), indicating biological stability of the T cell– and Cytokine-driven endotypes. The 19-gene MECI classifier demonstrated robust discriminative capacity (median AUC-ROC 0.889), enabling accurate transcriptome-based endotype prediction. Endotype-tailored drug repurposing highlighted discrete therapeutic vulnerabilities: glutamine-metabolism and immune checkpoint modulators as candidate therapeutics for the T-cell endotype; JNK-pathway inhibition and IL-2–augmenting approach for the Cytokine endotype; and GSK3β or NLRP3-directed agents for the Inflammasome endotype. Network-based drug–gene interaction analysis additionally identified immunomodulatory agents such as IFN-γ antagonists as potential candidates for the Cytokine endotype. Notably, transcriptome-reversal analysis demonstrated that CD19 CAR-T therapy most effectively counteracted the dysregulated transcriptional milieu of the T-cell endotype, highlighting its added value for this subgroup.

Conclusions: Transcriptome-guided molecular stratification reveals three reproducible patient endotypes underlying treatment-resistant SLE, each marked by a distinct axis of immune dysregulation and corresponding therapeutic targets. The MECI classifier provides an implementable framework for endotype assignment, enabling precision management of refractory SLE patients. These findings offer new insight into the molecular basis of treatment resistance in SLE, elucidating targetable pathways for pathway-driven therapeutics and encouraging biologically-informed clinical trial design in difficult-to-treat SLE.

REFERENCES: NIL.

Acknowledgments: NIL.

Disclosure of Interests: None declared.