cPRISM-seq: Cellular Injury Mapping in Kidney Transplantation via Donor-Restricted Methylation Analysis

While donor-derived cell-free DNA (dd-cfDNA) detects renal allograft injury, it lacks cellular resolution(e.g. endothelium vs. epithelium). This is important for classifying rejection according to the Banff criteria. To address this limitation, We developed cPRISM-seq (Cis-Parallel Read Integrated SNP and Methylation Sequencing) to enable cell-type-specific injury tracing by concurrently profiling SNPs and methylation on single DNA molecules.

Target Panel Design: We analyzed whole-genome bisulfite sequencing (WGBS) data from key human renal cell types(glomerular/tubular endothelial cells, tubular epithelial cells, parietal epithelial cells, and podocytes) and clustered them into epithelial and endothelial groups. Differential methylation regions (DMRs) that co-localized with biallelic SNPs and contained ≥4 CpG sites were selected for targeted capture panel design.

cPRISM-seq Assay: Plasma cfDNA was sequenced by targeted capture (~2.18 Gb/sample). Data were used to quantify dd-cfDNA and enrich donor-derived reads via SNP profiling, followed by deconvolution using a renal methylation reference to determine epithelial/endothelial proportions (Figure 1).

In Silico Validation: We performed in silico mixing to assess quantification accuracy. Methylation data from renal endothelial/epithelial cells and recipient leukocytes were mixed across six dd-cfDNA fractions (0.1%, 0.5%, 1%, 2%, 5%, and 10%), with renal endothelial proportions varying 0%-100% (Figure 2).

Clinical Application: cPRISM-seq was applied to plasma from six transplant recipients to quantify renal epithelial/endothelial cfDNA, with results correlated to clinical data.

1. Our targeted panel covered a 109 kb region with 2,489 CpGs and achieved ~1,000X effective sequencing depth.

2. In silico mixing validated cPRISM-seq's strong prediction correlation (R²=0.9967; Figure 3a). By enriching donor-derived reads to remove recipient background noise, the method achieved precise deconvolution (0.0369 MAE at 0.1% dd-cfDNA; Figure 3b) and low renal endo/epi-cfDNA quantification error (0.06% MAE ; Figure 3c), enabling robust cellular tracing under low abundance conditions.

3. In clinical samples, cPRISM-seq resolved cellular injury origins.

1)Renal endothelial cells predominated (72.26% ± 6.26%) over epithelial cells in donor-derived reads in plasma

2)Five low-risk samples exhibited low total dd-cfDNA(0.36%–0.93%) and renal endothelial cfDNA (0.24%–0.63%), consistent with clinical stability.

3)Notably, one sample clinically diagnosed as ABMR-dominant (ddcfDNA=1.61%) showed a specific elevation in renal endothelial contribution (76.63% of dd-cfDNA, corresponding to 1.23% of cfDNA), strongly supporting ABMR and aligning with the clinical diagnosis. 

Current methylation-based deconvolution fails to distinguish donor endothelial from recipient endothelial cells due to highly similar methylation profiles, resulting in high background noise and errors. In contrast, cPRISM-seq overcomes this limitation through simultaneous SNP and methylation analysis on single DNA molecules. Our novel approach enriches donor-derived fragments, removes recipient methylation background, and yields a final quantification error of only 0.06%. When clinically applied, it non-invasively discriminated renal endothelial from epithelial injury, laying the groundwork for larger validation studies. An ongoing effort is underway to integrate podocyte-specific probes to enhance resolution, aiming to detect recurrent diseases such as FSGS or membranous nephropathy by detecting podocyte injury before proteinuria onset, enabling earlier intervention.