Not all reference samples are equal in single-cell transcriptomics of human kidney tissue
Rajasree Menon, Paul L. Kimmel, Edgar A. Otto, Lalita Subramanian, Christopher L. O’Connor, Bradley Godfrey, Cathy Smith, Fadhl Alakwaa, Celine C. Berthier, Minnie M. Sarwal, E. Steve Woodle, Laura Pyle, Ye Ji Choi, Patricia Ladd, John R. Sedor, Sylvia E. Rosas, Sushrut S. Waikar, Abhijit S. Naik, Ricardo Melo Ferreira, Michael T. Eadon, Markus Bitzer, Petter Bjornstad, Jeffrey B. Hodgin, Matthias Kretzler, for the Kidney Precision Medicine Project (KPMP)
Rajasree Menon, Paul L. Kimmel, Edgar A. Otto, Lalita Subramanian, Christopher L. O’Connor, Bradley Godfrey, Cathy Smith, Fadhl Alakwaa, Celine C. Berthier, Minnie M. Sarwal, E. Steve Woodle, Laura Pyle, Ye Ji Choi, Patricia Ladd, John R. Sedor, Sylvia E. Rosas, Sushrut S. Waikar, Abhijit S. Naik, Ricardo Melo Ferreira, Michael T. Eadon, Markus Bitzer, Petter Bjornstad, Jeffrey B. Hodgin, Matthias Kretzler, for the Kidney Precision Medicine Project (KPMP)
View: Text | PDF
Research Article Endocrinology Nephrology

Not all reference samples are equal in single-cell transcriptomics of human kidney tissue

Abstract

Identifying mechanisms of kidney disease commonly involves comparing diseased samples with healthy reference tissues; however, the effects of variability in tissue procurement, storage, and donor characteristics remain underexplored. In this study, we systematically evaluated 3 reference tissue types — tumor nephrectomy (TN), pretransplant biopsies from living donors (LD), and percutaneous biopsies from healthy control volunteers (HC) — to determine their impact on differential gene expression across 3 diabetic kidney disease states. We observed distinct injury markers, cell state proportions, and gene signatures associated with procurement method, sex, and donor age. Adjustment for these confounding factors significantly influenced pathway analysis results. Specifically, correcting for age and sex eliminated significant enrichment of IFN-γ response when comparing the diabetes mellitus–resilient group and HC group. Processes related to biological aging were enriched in older reference tissues, potentially confounding disease-specific interpretations. Importantly, TNF signaling via NF-κB remained enriched in LD and TN samples relative to HC, even after accounting for confounders. These results underscore the critical importance of selecting appropriate control tissues and rigorously adjusting for confounding variables to reliably discern the molecular mechanisms underlying kidney diseases.

Authors

Rajasree Menon, Paul L. Kimmel, Edgar A. Otto, Lalita Subramanian, Christopher L. O’Connor, Bradley Godfrey, Cathy Smith, Fadhl Alakwaa, Celine C. Berthier, Minnie M. Sarwal, E. Steve Woodle, Laura Pyle, Ye Ji Choi, Patricia Ladd, John R. Sedor, Sylvia E. Rosas, Sushrut S. Waikar, Abhijit S. Naik, Ricardo Melo Ferreira, Michael T. Eadon, Markus Bitzer, Petter Bjornstad, Jeffrey B. Hodgin, Matthias Kretzler, for the Kidney Precision Medicine Project (KPMP)

×

Figure 7

Analysis of the top pathways.

Options: View larger image (or click on image) Download as PowerPoint
Analysis of the top pathways. (A) Hierarchical clustering on the right g...
(A) Hierarchical clustering on the right generated using aggregate read counts at the sample type level. Hierarchical clustering on the left utilized aggregate read counts with batch effect correction applied via the edgeR package. For HC, early DKD, DM-R, and DKD, age and sex were used as the covariate in batch effect correction; for TN and LD, age, process effect score, and sex served as covariates. (B) Plot displaying the top significant MsigDB Hallmark pathways (adjusted P < 0.05) associated with differentially expressed genes among the 3 reference groups. (C) DKD versus LD and DKD versus TN. DKD versus HC comparison was excluded due to nonoverlapping age distributions between these groups. Differential expression analysis was performed using DESeq2, both with and without adjustment for confounding variables: age, preprocess effect, and sex were included for DKD versus LD and DKD versus TN analyses. (D) DKD (n = 13) and LD (n = 15) in Visium data. The observed trends were consistent with those from single-cell analyses. (E) DM-R versus each of the 3 reference groups. Age and sex were used as the confounders for the DM-R versus HC comparison; age, sex, and preprocess effect were included for DM-R versus LD and DM-R versus TN. (F) DM-R (n = 15) versus LD (n = 15) in Visium data. (G) Early DKD versus HC. Comparisons of early DKD with other reference groups were excluded due to the absence of overlapping age distribution in these groups. Age and sex were used as the confounders for the early DKD versus HC comparison. (H) Violin plots depict pathway scores for selected MsigDB pathways, calculated from genes differentially expressed (P < 0.05) between early DKD (n = 4) and HC (n = 5) samples in Visium transcriptomic data.