RNA-Seq identifies condition-specific biological signatures of ischemia-reperfusion injury in the human kidney

Five male patients scheduled for total nephrectomy owing to renal cell carcinoma or transitional cell carcinoma were enrolled in the study. Their average age was 64.8 years and their kidney functions before surgery were near normal state (mean creatinine: 0.89 mg/dL, mean estimated glomerular filtration rate [eGFR]: 88.1 mL·min·^− 11.73 m^− 2, mean hemoglobin: 14.2 g/dL) (Supplementary Table S1). The kidney cortical tissue was obtained by gun biopsy at three time points as follows: pre-ischemia (considered as a normal condition), ischemia (after 15 min of hypoxia), and after 10 min of reperfusion (Fig. 1a). The tissue samples were immediately transferred to individual cryotubes prefilled with 0.5 mL RNAlater® (QIAGEN Inc., Hilden, Germany) and stored at room temperature for 24 h. RNAlater® was removed following incubation, and tissues was stored at − 80 °C according to the manufacturer’s instructions until analysis. Our study protocol was approved by the Pusan National University Hospital Ethics Committee (IRB number H-17020002-051). All participants provided written informed consent as requested by our Ethics Committee, and all procedures were implemented in accordance with the Helsinki Declaration.

To identify the changes induced by ischemia and reperfusion at the transcriptomic level, we analyzed the steady-state gene expression pattern during pre-ischemia (normal), ischemia, and reperfusion using RNA-seq (Fig. 1b). Total RNA was isolated from the kidney cortex of five male patients at each condition using mirVana™ miRNA Isolation Kit (ThermoFisher, Inc., Seoul, Korea). The RNA quality was assessed using 2100 Expert Bioanalyzer with RNA 6000 Nano Kit (Agilent, Inc., Santa Clara, CA, USA). Samples with RNA integrity number > 7 were prepared using Illumina TruSeq Standard mRNA Prep kit (Catalog #RS-122-2103; Illumina, San Diego, CA, USA). After quantitative polymerase chain reaction (qPCR) using SYBR Green PCR Master Mix (Applied Biosystems), the libraries were combined such that the indexed sample was present at equimolar concentrations in the pool. Cluster generation was carried out in the flow cell on the cBot automated cluster generation system (Illumina). The flow cell was loaded on HiSeq 2500 sequencing system (Illumina), and sequencing was performed with 2× 100 bp read length. RNA-seq was carried out by DNA Link, Inc., Seoul, Korea (http://​www.​dnalink.​com/​).

FastQC (http://​www.​bioinformatics.​babraham.​ac.​uk/​projects/​fastqc/​) was conducted to assess the quality of RNA-seq data. The RNA-seq data were quantified using an alignment-free tool, Salmon, developed by Patro et. al in 2017 that estimates the relative abundance of all transcripts [17]. GRCh38.p13 was used as the reference transcriptome to quantify read counts. The general information of RNA-seq is shown in Supplementary Table S2, and RNA-seq data are available in Gene Expression Omnibus (GEO) database under the accession number GSE142077.

We employed tximport [18] with Bioconductor differential gene expression package using R (version 3.5.1) to assemble count values from each sample. The assembled count values were used as the input of DESeq2 Bioconductor package [19]. The significantly differentially expressed genes (DEGs) between groups were defined at cut-off criteria of |log₂ fold-change| ≥ 1 and p-value < 0.05. Significantly enriched pathways were examined for the identified DEG sets.

To identify the trajectory patterns of gene expression across all three time points, pre-ischemia, ischemia, and reperfusion, SOM clustering [15] analysis was performed. A 7 × 7 grid panel was selected for the SOM output structure to intuitively interpret the results. Transcripts showing similar expression patterns across three conditions were gathered in a module. Selecting rules were applied to 49 modules to obtain modules of interest. Selected modules with similar expression patterns across three groups were combined as a ‘cluster.’ SOM algorithm implemented in MATLAB 2018b software (http://​www.​mathworks.​com) was used. Genes in each cluster were further analyzed to identify significantly enriched canonical pathways.

The Ingenuity Pathway Analysis (IPA) software (www.​qiagen.​com/​ingenuity, Spring 2019, QIAGEN, CA, USA) was used to identify enriched biological pathways. A p-value or Benjamini-Hochberg adjusted p-value was calculated using Fisher’s exact test, and a cut-off value of less than 0.05 was used to identify significantly enriched canonical pathways based on the Ingenuity Knowledge Base.

We aimed to identify the key genes and pathways through the evaluation of gene expression changes under pre-ischemia (normal), ischemia, and reperfusion conditions in human kidney samples using RNA-seq (Fig. 1b). RNA-seq was performed for downstream bioinformatic analyses, namely, differential gene expression and SOM clustering analyses. Differential expression analysis was performed to identify the significant DEGs between two conditions. SOM clustering was carried out to group the whole gene expression data and identify specific patterns of interest across all three groups without prior knowledge. Finally, the selected genes of interest from both analyses were used as inputs for pathway analysis and to identify the most significantly affected biological pathways separately under ischemia and reperfusion during IRI.

Differential expression analysis is a conventional method for the identification of quantitative changes in the expression levels of genes between two groups. We evaluated the DEGs by performing three comparisons as follows: (1) ischemia versus pre-ischemia to identify the genes affected by ischemia; (2) ischemia versus reperfusion to detect the genes affected by reperfusion after ischemia; (3) reperfusion versus pre-ischemia to analyze the genes affected by the complete IRI process. We used |log₂ fold-change| ≥ 1 and p-value < 0.05 as criteria to identify significant DEGs between the groups (Fig. 2).

As a result, 603 DEGs (upregulated: 402 genes; downregulated: 201 genes) were significantly dysregulated between ischemia and pre-ischemia samples and 135 DEGs (upregulated: 67 genes; downregulated: 68 genes) were found to be significantly dysregulated between reperfusion and ischemia samples. However, 1389 DEGs (upregulated: 1119 genes; downregulated: 270 genes) were dysregulated between reperfusion and pre-ischemia conditions. The top 20 DEGs in each comparison group are shown in Supplementary Tables S3-S5.

To determine the pathways associated with ischemia and reperfusion, the IPA was performed to investigate the biological pathways associated with the DEG sets. The top pathways related to the DEGs between ischemia versus pre-ischemia conditions were mainly metabolic pathways and included genes encoding cytochrome P450 enzymes (CYP1A2, CYP2C8, CYP2C9, CYP2J2, CYP3A4, CYP3A7). Uridine diphosphate glucuronosyltransferase (UGT1A1, UGT1A6, UGT1A7, UCT2A3, UGT2B11, UGT3A1) is involved in the metabolism of various molecules, including steroids, hormones, and drugs (Fig. 3a and Supplementary Table 6) [20, 21]. Ischemia induced changes in metabolites such as nicotine, melatonin, serotonin, and thyroid hormone. Melatonin and serotonin are known to exert antioxidant proprieties under oxidative stress [22], and may protect the function of the kidney during early ischemic injury. In addition, drug metabolism pathways, including bupropion, acetone degradation and estrogen biosynthesis, and pregnane X receptor/retinoic X receptor (PXR/RXR) pathway were related to ischemic process. PXR is a nuclear receptor activated by endogenous compounds and clinical drugs. Activated PXR in conjunction with RXR plays a central role in drug metabolism by inducing the expression of the cytochrome P450 family. This receptor was recently reported to be involved in inflammation, proliferation, and apoptosis [23].

The DEGs between reperfusion and ischemia conditions were found to be enriched in cellular assembly, development and migration, energy production, and inflammasome-related signaling pathways (Fig. 3b and Supplementary Table 7). Fibroblast growth factor (FGF) performs diverse functions through the activation of several pathways by binding to FGF receptor (FGFR). FGF signaling pathway is involved in cellular assembly and development and migration, including hypertrophy, regulation of epithelial-mesenchymal transition, and actin cytoskeleton signaling, which are affected in response to renal ischemia/reperfusion [24, 25]. AMP-activated protein kinase (AMPK) involved in metabolic processes generating ATP and liver X receptor (LXR)/RXR signaling associated with lipid metabolism were also detected. In addition, inflammasome pathway was enriched, indicating that reperfusion may induce an immune response.

The comparison between reperfusion and pre-ischemia (reperfusion versus pre-ischemia) conditions showed that reperfusion induced changes in various pathways related to lipid and drug metabolism such as nicotine, serotonin, and melatonin degradation, FXR/RXR, and PXR/RXR signaling pathway, consistent with the results between ischemia and pre-ischemia conditions (Fig. 3c and Supplementary Table 8). Different pathways are involved in tryptophan, valine, ethanol, and histamine degradation as well as fatty acid oxidation. These results suggest that reperfusion affected not only lipid and drug metabolism but also energy metabolism. In summary, the pathway analysis for DEG sets showed that metabolic pathways were affected under both ischemia and reperfusion conditions. In particular, reperfusion affected hypertrophy, cellular assembly and development, and inflammatory response-related signaling pathways.

Among the clustering algorithms such as hierarchical clustering or k-means clustering, SOM has been extensively used for the analysis of large-scale gene expression data [26‐31]. We performed SOM clustering analysis to examine gene expression patterns across all three conditions. Transcripts per million (TPM) is a commonly used normalization method as previously described in [32]. The log2-transformed TPM values of transcripts for each group were used as SOM input data. Given the small sample size, we calculated median values of log2-transformed TPM for each group. However, as technical and biological biases often generate inexplicable results, we removed zero expression values and filtered the very low signal values (< log₂3) for further analyses. As a result, 38,014 transcripts were used for SOM input. The 7 × 7 grid structure was chosen for SOM output to enhance biological interpretability. The results were visualized as a colored grid panel with blue hexagons and a yellow-black similarity color scheme between the hexagons (Fig. 4a). We defined hexagon as a ‘module’ such that the genes included in each module showed a similar expression pattern across all three conditions. The similarity between adjacent modules was represented with different colors; a color close to yellow was indicative of the similar expression pattern between adjacent modules. On the contrary, a color close to black was suggestive of the distinct patterns of the adjacent modules. The total number of transcripts in each module after clustering is shown in Fig. 4b.

As we aimed to determine the change in the expression pattern of genes under each condition, we focused on the specific patterns of gene expression under ischemia (genes upregulated or downregulated in ischemia versus pre-ischemia) and reperfusion (genes upregulated or downregulated in reperfusion versus ischemia). Modules of interest among 49 modules were selected based on the criteria rules (Table 1). As a result, module 42, 44, 45, 46, 48, and 49 including the genes upregulated in ischemia were defined as ‘Cluster 1.’ Module 1, 8, 11, 17, and 19 carrying the genes downregulated during ischemia were defined as ‘Cluster 2.’ Similarly, the genes in modules 36 and 43 upregulated by reperfusion as compared to ischemia and pre-ischemia were defined as ‘Cluster 3.’ Lastly, the genes in module 15, 16, and 23 that were downregulated in reperfusion were defined as ‘Cluster 4’ (Fig. 5). In total, 3035 and 1917 genes were affected in ischemia and reperfusion, respectively (Table 2).

We performed pathway analysis using IPA for the selected clusters to identify significantly enriched pathways. The genes dysregulated by ischemia (Cluster 1 and 2) were mainly enriched in apoptosis-related pathways, including aryl hydrocarbon receptor and death receptor signaling pathway (Table 3). During ischemia, FAS cell surface death receptor (FAS) and tumor necrosis factor (TNF) related to complex signaling pathways for cell death were down-regulated, suggestive of the presence of a protective mechanism against cell death. In addition, intracellular and secondary messenger signaling pathways such as protein ubiquitination, adipogenesis, and apelin adipocyte signaling pathway were identified. Lipid accumulation and deposition is known to induce lipotoxicity, thereby leading to ischemia-mediated kidney injury [27]. Further, metabolism-related pathways such as PXR/RXR signaling pathway and xenobiotic metabolism signaling were enriched. PXR/RXR signaling pathway was also detected in DEG analysis.

The genes dysregulated by reperfusion (Cluster 3 and 4) were mainly related to cellular functions of growth, proliferation, and migration (semaphorin signaling, mTOR signaling, E74-like factor 2 [elF2] signaling, integrin signaling, apelin muscle signaling, and actin-based motility by Rho-related signaling) (Table 4). Semaphorins, a family of growth cone guidance molecules during neurodevelopment, interact with the members of the Rho family [33]. In particular, semaphorins can be synthesized in podocytes and tubular epithelial cells within the kidney and are implicated in cell migration, growth, and immune response in AKI [34, 35]. Integrin and actin signaling pathways linked to Rho signaling related to cytoskeletal remodeling process during cell growth and wound healing were enriched in the reperfusion effect clusters. In addition, immune response-related pathways such as antigen presentation pathway and sphingosine-1-phosphate signaling as well as intracellular and second messenger pathways were enriched.