Cell-Type Specific Single-Cell Signatures Reveal Nephrotoxic Drug Affects

Abstract

Drug-induced acute kidney injury (AKI) affects about 20% of hospitalized AKI patients, a significant contributor to morbidity and mortality. The lack of understanding of the kidney system and functioning of nephrotoxic drugs contributes to hospital-acquired AKI cases. AKI is difficult to predict because of its complex injury mechanism and the numerous pathways through which it manifests. Traditional toxicity biomarkers, like elevated creatinine levels, detect AKI only after significant kidney injury has occurred. Concurrently, advancements in single cell RNA sequencing (scRNAseq) have improved our ability to map cellular heterogeneity within tissues, potentially enabling the study of drug effects at a single cell level. We hypothesized that only particular subtypes of kidney cells may be responsible for observed nephrotoxicity and explain prediction challenges. To test this, we generated cellular response scores for 32 kidney cell types from the Human Cell Atlas and estimated drug effects. We identified significant expression differences in 6 cell types (e.g. Indistinct intercalated cell p = 0.009, Epithelial Progenitor cell, p = 0.04). We also developed an ensemble model that achieved an AUROC of 0.6 across different kidney cell populations - a significant improvement over using traditional bulk RNA sequencing alone. The single-cell transcriptomic signatures we identified potentially reveal unexplained molecular mechanisms of nephrotoxicity.

Keywords: Single cell RNAseq; nephrotoxicity; single cell signatures.

Figure 1.

Workflow diagram – Drug target information is collected from ChEMBL and Drugbank and consolidated. Ryan reference, ACS FDA label data and DIRIL datasets are combined to extract toxicity labels for drugs. Logical OR is used to label them to maximize the number of nephrotoxic drugs. To create a unified Target-Toxicity database the targets and toxicity labels are combined. This database is used in all downstream analyses. For the single-cell analysis, data is collected from HCA and GEO, selecting only healthy kidney samples. We use CellRanger to process FASTQ files, Seurat to analyze the data, QC, filter and integrate (see Methods). We annotate the clusters using SingleR, with the Kidney Cell Atlas as reference at two levels of abstraction (4 cell types, called abstract, and 32 cell types called detailed.) We analyze both levels of abstraction at all phases downstream. We then use Drug2Cell package to extract drug scores based on the target genes in the single cell matrix. We log these scores to ensure normal distribution. Our bulk RNAseq analysis extracts kidney-specific LINCS data to identify differentially expressed genes and extractable toxicity signatures. Independently, we simulated drug-affected single-cell expression data and pseudo-bulked it to compare data type power for toxicity predictions. We then used both single-cell and bulk RNAseq toxicity matrices to train machine learning methods for identifying drug toxicity signatures.

Figure 2.

4 cell-type results. In figures a), b), c) are power plots for different response rates. Response rates are defined as the percentage of cells within each sub-type that respond to a drug. Power is defined as the number of differentially expressed genes divided by total number of genes, represented on the y axis of the plots. On the x axis is the effect size. Figure d) shows a UMAP of the clustered single cell data coded by color. Many of the cells are Nephrons (82.4%). Boxplots showing the logged drug scores are in figure e). There is a marked difference in the drug scores between nephrotoxic (dark) and non-nephrotoxic (light) drugs. The nephrotoxic drug scores trend higher.

Figure 3.

Plots showing single cell data having higher nephrotoxicity detection power compared to pseudo-bulked data. Box plots showing the increased drug scores of nephrotoxic drugs (dark) in 32 cell types.