Transcriptional trajectories of human kidney injury progression

Abstract

Background: The molecular understanding of the progression from acute to chronic organ injury is limited. Ischemia/reperfusion injury (IRI) triggered during kidney transplantation can contribute to progressive allograft dysfunction.

Methods: Protocol biopsies (n = 163) were obtained from 42 kidney allografts at 4 time points after transplantation. RNA sequencing-mediated (RNA-seq-mediated) transcriptional profiling and machine learning computational approaches were employed to analyze the molecular responses to IRI and to identify shared and divergent transcriptional trajectories associated with distinct clinical outcomes. The data were compared with the response to IRI in a mouse model of the acute to chronic kidney injury transition.

Results: In the first hours after reperfusion, all patients exhibited a similar transcriptional program under the control of immediate-early response genes. In the following months, we identified 2 main transcriptional trajectories leading to kidney recovery or to sustained injury with associated fibrosis and renal dysfunction. The molecular map generated by this computational approach highlighted early markers of kidney disease progression and delineated transcriptional programs associated with the transition to chronic injury. The characterization of a similar process in a mouse IRI model extended the relevance of our findings beyond transplantation.

Conclusions: The integration of multiple transcriptomes from serial biopsies with advanced computational algorithms overcame the analytical hurdles related to variability between individuals and identified shared transcriptional elements of kidney disease progression in humans, which may prove as useful predictors of disease progression following kidney transplantation and kidney injury. This generally applicable approach opens the way for an unbiased analysis of human disease progression.

Funding: The study was supported by the California Institute for Regenerative Medicine and by the Swiss National Science Foundation.

Keywords: Chronic kidney disease; Nephrology; Transcription; Transplantation.

Conflict of interest statement

Conflict of interest: The authors have declared that no conflict of interest exists.

Figures

Figure 1. The kidney allograft transcriptome across individuals and time.

(A) Gene expression correlation analysis including the 500 most variable genes in RNA-seq data (RPKM values) from 163 protocol biopsies obtained from 42 kidney allografts at 4 time points. Clusters of interest are highlighted in yellow (#1 renal physiology, #2 response to kidney injury, #3 fibrosis, #4 adaptive immunity; gene ontology analysis is presented in Supplemental Figure 1A). (B) T-distributed stochastic neighbor embedding (t-SNE) analysis on RNA-seq data, including all samples and showing the separation of the transcriptomes in 2 major clusters: early phase (green; PRE, before implantation; POST, after implantation), late phase (blue; 3M, 3 months after transplantation; 12M, 12 months after transplantation). (C) Gene expression variance decomposition analysis in linear mixed models showing the contribution of individual and time to gene expression variation. Genes showing an individual-driven variance are shown in red; genes with a time-drive variance are shown in blue, and some relevant examples are specified.

Figure 2. Early transcriptional response to ischemia/reperfusion.

(A and B) Pseudotime analysis including samples collected before (PRE, n = 38) and after implantation (POST, n = 39). (A) Sample state ordering in the reduced dimensional space, as determined by the Monocle algorithm. PRE samples were classified in 2 groups and are shown in cyan; POST samples ordered along a pseudotime line from right to left and are shown in red. Among the POST samples, the circles mark samples from living donors (LD), and the black squares mark samples from donors after cardiac death (DCD). (B) Cluster analysis of representative genes differentially expressed along the pseudotime: samples are aligned from left to right according to the order shown in A. Genes are vertically aligned and classified in 2 clusters. The colors indicate the relative expression of the genes (log10 scale). The complete list of genes is presented in Supplemental Figure 2 and Supplemental Table 1. (C) Influence score for the top 14 transcription factors as determined in the network analysis based on a modified Mogrify algorithm. (D) Venn diagram including genes differentially expressed in POST compared with PRE (human) and homologous mouse genes differentially expressed 2 hours after IRI compared with control. Significance of enrichment was determined by hypergeometric test. (E and F) Reads per kilobase per million mapped reads (RPKM) values along the early time-course analysis after IRI in mice (n = 3 for each time point). (G) Venn diagram including genes differentially expressed in POST compared with PRE in the human kidney and in the liver. Significance of enrichment was determined by hypergeometric test. (H) Cluster analysis of representative genes differentially expressed along the pseudotime in the liver.

Figure 3. Transcriptional trajectory of transition from acute to chronic kidney injury.

(A) Minimum spanning tree of kidney biopsy transcriptomes at 3 (3M, dark blue, n = 38) and 12 (12M, light blue, n = 39) months after transplantation, and 10 samples collected after implantation (POST, green). Community A (marked in blue) separated from the rest of the study population (community B, in red). (B) Violin plots showing mRNA per sample (RPS) values of representative genes selected as markers for kidney injury (HAVCR1, VCAN), fibrosis (COL1A1), and chronic inflammation (CCL19). Adjusted P values are reported (Benjamini-Hochberg). (C and D) Pseudotime analysis including samples collected after implantation (POST) and 3 and 12 months after transplantation. Sample state ordering in the reduced dimensional space is shown. The colors indicate the time point of biopsy collection in C and the classification in communities A or B in D. (E) Similar analysis including only 3- and 12-month samples based on genes differentially expressed in communities A and B. The color of the dots indicate the progression along the pseudotime, as indicated. (F) Cluster analysis of the top 2,000 genes differentially expressed along the pseudotime shown in E. Genes are vertically aligned and classified in 4 clusters (the complete list of genes is reported in the Supplemental Table 8). (G) Representative example of 1 gene expressed early (HAVCR1) and late (MMP2) in the transition to chronic injury. The colors indicate the pseudotime, as indicated. The numbers indicate the cluster, according to F. (H) Glomerular filtration rate at 12 months, estimated by CKD-EPI equation (eGFR) in communities A and B. P value was calculated by Mann-Whitney U test. (I) Histogram of the degree of fibrosis 12 months after transplantation, quantified by ci-score on conventional histology. The percentage of patients in each fibrosis category in communities A and B is reported. The groups were compared by χ2 test. (J) Venn diagram including genes differentially expressed in community A compared with community B (human) and homologous mouse genes differentially expressed 12 months after IRI compared with control. Significance of enrichment was determined by hypergeometric test. (K and L) RPKM values along the late time-course analysis after IRI in mice (n = 3 for each time point).

Figure 4. Early markers of transition to kidney fibrosis.

(A) Schematic representation of the definition of groups along the divergent branches of the pseudotime analysis presented in Figure 3C. (B–E) Box plots of RPKM values of the indicated genes in the group defined in A and PRE biopsies obtained from living donor (LD), as a surrogate of normal renal tissue. The groups were compared by Mann-Whitney U test. (F) Expression profile of EP300 along the pseudotime presented in Figure 3E in comparison with other genes, reflecting the very early upregulation of EP300 along the transition to chronic kidney injury. (G) Samples cluster analysis based on the 33 genes highly correlated with EP300 in groups 1 and 2 (Pearson’s correlation > 0.88). The division of the samples in 2 clusters is indicated by the blue (group 1) and green (group 2) bars on the left. (H) Conceptual summary of the study highlighting the common early injury response after ischemia/reperfusion followed by a multifactorial transition phase with divergent long-term outcomes: recovery versus the initiation of a chronic injury signature. Some of the critical genes involved in each phase are shown.