A cross-sectional study of the role of epithelial cell injury in kidney transplant outcomes

Abstract

Background: Expression of acute kidney injury-associated (AKI-associated) transcripts in kidney transplants may reflect recent injury and accumulation of epithelial cells in "failed repair" states. We hypothesized that the phenomenon of failed repair could be associated with deterioration and failure in kidney transplants.

Methods: We defined injury-induced transcriptome states in 4,502 kidney transplant biopsies injury-induced gene sets and classifiers previously developed in transplants.

Results: In principal component analysis (PCA), PC1 correlated with both acute and chronic kidney injury and related inflammation and PC2 with time posttransplant. Positive PC3 was a dimension that correlated with epithelial remodeling pathways and anticorrelated with inflammation. Both PC1 and PC3 correlated with reduced survival, with PC1 effects strongly increasing over time whereas PC3 effects were independent of time. In this model, we studied the expression of 12 "new" gene sets annotated in single-nucleus RNA-sequencing studies of epithelial cells with failed repair in native kidneys. The new gene sets reflecting epithelial-mesenchymal transition correlated with injury PC1 and PC3, lower estimated glomerular filtration rate, higher donor age, and future failure as strongly as any gene sets previously derived in transplants and were independent of nephron segment of origin and graft rejection.

Conclusion: These results suggest 2 dimensions in the kidney transplant response to injury: PC1, AKI-induced changes, failed repair, and inflammation; and PC3, a response involving epithelial remodeling without inflammation. Increasing kidney age amplifies PC1 and PC3.

Trial registration: INTERCOMEX (ClinicalTrials.gov NCT01299168); Trifecta-Kidney (ClinicalTrials.gov NCT04239703).

Funding: Genome Canada; Natera, Inc.; and Thermo Fisher Scientific.

Keywords: Molecular diagnosis; Nephrology; Organ transplantation; Transplantation.

Figure 1. CONSORT diagram and study design.

(A) CONSORT diagram showing biopsy inclusion for the INTERCOMEX and Trifecta-Kidney Study (N = 4,502) combined dataset. (B) Study design. ATAGC, Alberta Transplant Applied Genomics Centre; CV, cross-validation; MMDx, Molecular Microscope Diagnostic System.

Figure 2. Factor maps in the N = 4,502 kidney transplant biopsy population showing the correlations between the input variables (red circles) and the principal components in the N = 4,502 kidney transplant biopsy population.

AA groups are shown in green. The correlations between the PCA input variables and the PC scores are shown as factor maps in (A) PC2 vs. PC1 and (B) PC2 vs. PC3.

Figure 3. PCAs for biopsies.

(A) PC2 vs. PC1 in all 4,502 biopsies, (B) PC2 vs. PC3 in all 4,502 biopsies, (D) PC2 vs. PC1 in 2,479 no rejection biopsies, (E) PC2 vs. PC3 in 2,479 no rejection biopsies, (C) UMAP visualization of the 4,502 population with variation compressed into 2 dimensions only, and (F) UMAP visualization of the 2,479 population of no rejection biopsies. PCAs and UMAP panels are colored by the 5-archetype injury model cluster assignments (normal, mild CKD, AKI1, AKI2, and CKDAKI). D–F show the no rejection samples within the original plots, rather than generating new PCAs/UMAPs using only the no rejection samples. MCAT, mast cell transcripts; IGT, immunoglobulin transcripts; ci>1_Prob, ci lesion classifier; ct>1_Prob, ct lesion classifier; Prot_Prob, proteinuria classifier; IRITD5, injury and repair induced transcripts day 5; IRITD3, injury and repair induced transcripts day 3; IRRAT, injury and repair associated transcripts; DAMP, damage-associated molecular pattern transcripts; lowGFR_Prob, probability of low GFR ≤ 30 cc/min/M².

Figure 4. Relationships between injury principal component scores, injury archetypes, eGFR, and time posttransplant in all 4,502 biopsies with available data.

(A) Restricted cubic splines showing the relationship between DAMP, IRITD3, IRITD5, IRRAT, and lowGFR_Prob and time posttransplant. (B) Restricted cubic splines showing the relationship between MCAT, IGT, ct>1_Prob, ci>1_Prob, ci lesion, and Prot_Prob. (C) Restricted cubic splines showing the relationship between injury PC1, PC2, PC3 and time posttransplant in the N = 4,502 biopsy population. (D) Restricted cubic splines showing the relationship between injury archetypes and time posttransplant. Scores were standardized before analysis so that they could be shown on the same scale. (E) Restricted cubic splines showing the relationship between injury PC1, PC2, PC3, and eGFR. (F) Restricted cubic splines showing the relationship between injury archetypes and eGFR.

Figure 5. Visualizing the relationship between injury and 3-year postbiopsy death-censored survival (1 random biopsy per kidney) in all 4,502 biopsies.

(A) Kaplan-Meier plots for the injury archetype groups in all 4,502 biopsies. (B) Kaplan-Meier plots for the injury archetype groups in biopsies ≤42 days. (C) Kaplan-Meier plots for the injury archetype groups in biopsies >1 year. (D–F) Relative variable importance plots from random survival forest analyses using injury PC1, PC2, and PC3; rejection PC1, PC2, and PC3; and time of biopsy posttransplant (TxBx) as predictors (D) in all 4,502 biopsies, (E) in biopsies ≤42 days posttransplant, (F) and in biopsies >1 year posttransplant. (G) Time-varying effects of injury PC1 and PC3 on the hazard of graft loss, estimated from a Cox proportional hazards model with interaction terms between time posttransplant and the injury PCs.

Figure 6. Heatmap showing pairwise Spearman correlations between the PBT scores of the 12 new gene sets in the N = 4,502 data set.

Figure 7. Factor maps in the N = 4,502 kidney transplant biopsy population showing the correlations between the input variables (red circles) and the principal components.

Additional variables of interest (injury archetype scores in green and gene sets from ref. as blue circles) were not used for the analysis but were projected as supplementary variables into the original PCA in (A) PC2 vs. PC1 and (B) PC2 vs. PC3. (C) Restricted cubic splines showing the relationship between the PBT scores for the 12 new gene sets and log(time posttransplant). N = 4,301 samples with TxBx available. MCAT, mast cell transcripts; IGT, immunoglobulin transcripts; ci>1_Prob, ci lesion classifier; ct>1_Prob, ct lesion classifier; Prot_Prob, proteinuria classifier; IRITD5, injury and repair induced transcripts day 5; IRITD3, injury and repair induced transcripts day 3; IRRAT, injury and repair associated transcripts; DAMP, damage-associated molecular pattern transcripts; GFR, probability of low GFR ≤ 30 cc/min/M²; New1, oxidative stress; New2, hypoxia; New3, inflammation; New4, EMT; PT, proximal tubule; TAL, thick ascending limb; tL, thin limb; DCT, distal convoluted tubule.

Figure 8. Relative variable importance plots for predicting 3-year survival after biopsy, using injury pathogenesis-based transcript set scores and injury principal component scores (using 1 randomly selected biopsy per transplant).

(A) All 4,502 biopsies. (B) Biopsies >1 year. (C) Biopsies ≤42 days. Error rates are 1.0 minus the C statistic (which is the survival analysis equivalent of the AUC).