Background/Purpose: Lupus nephritis (LN) causes substantial morbidity and mortality. LN is histopathologically divided into six classes, which currently serves as the basis for making treatment decisions. Pathogenesis underlying different classes of LN is unclear. To identify the molecular differences, we studied the quantitative protein changes across all six LN classes using tandem mass spectrometry proteomics analyses of kidney biopsies from patients.

Methods: Kidney biopsies from 48 subjects, including 10 normal donor kidneys and 38 LN kidneys, were obtained from UCLA Pathology. All biopsies were reviewed independently by two pathologists. Protein was extracted from biopsy tissues and subjected to tandem mass spectrometry proteomics analyses. We measured the peptides expression quantitatively using Orbitrap LC-MS/MS system. Peptides were annotated and the abundance of peptides was normalized. The data were presented as mean ± SD and median with range (minimum–maximum), and categorical data were presented as frequencies and proportions. Wilcoxon rank-sum test with two-tailed distribution was used in the statistical comparisons between groups unless otherwise indicated. The p-values were adjusted for multiple testing with p.adjust in R using false discovery rate. The principal component analysis based on Spearman’s rank correlation coefficients between samples was performed using R. We employed a machine learning analysis with random forest classification to build a probabilistic-based prediction model of LN disease vs. healthy controls. Pathway analyses of differentially expressed peptides with absolute log2 fold change greater than one was completed with Ingenuity Pathway Analysis.

Results: Proteomics analysis identified 2190 peptides quantifiable in all 48 kidney biopsies. Of these, 655 peptides were significantly differentially expressed, including 304 upregulated peptides and 351 downregulated peptides (p < 0.05). In principal component analyses, all class VI biopsies clustered with the control specimens, and when class VI biopsies were excluded controls neatly separated from LN (classes I-V). Through random forest classification, we built a probabilistic-based prediction model that can discriminate LN disease (class I-V) vs. healthy controls utilizing 273 of the 655 peptides differentially expressed between the groups, which maintained a receiver operating characteristic area under the curve accuracy of 87.5% with 95% CI (0.7131, 0.9985), and an out-of-bag error rate of 3.7%. Of these 273, peptides representing VIM, ETFB, SERPINA1, BHMT, IGHG1, and MDH2 had the highest mean decrease accuracy and Gini. Next, we utilized Ingenuity pathway analysis (IPA) to seek differentially expressed proteins and pathways in individual LN classes compared to controls. While a set of proteins and pathways were significantly differentially expressed across LN classes, certain proteins and pathways discriminated individual LN classes.

Conclusion: Our data indicate the unique molecular signatures that differentiate LN classes and pave the way for defining the unique molecular pathogenesis of individual LN classes, thus introducing the basis for designing class-specific treatment in LN.

ACR Meeting Abstracts - https://acrabstracts.org/abstract/molecular-heterogeneity-between-different-classes-of-lupus-nephritis-as-revealed-by-kidney-biopsy-proteomics/