clonealign: statistical integration of independent single-cell RNA and DNA sequencing data from human cancers

Abstract

Measuring gene expression of tumor clones at single-cell resolution links functional consequences to somatic alterations. Without scalable methods to simultaneously assay DNA and RNA from the same single cell, parallel single-cell DNA and RNA measurements from independent cell populations must be mapped for genome-transcriptome association. We present clonealign, which assigns gene expression states to cancer clones using single-cell RNA and DNA sequencing independently sampled from a heterogeneous population. We apply clonealign to triple-negative breast cancer patient-derived xenografts and high-grade serous ovarian cancer cell lines and discover clone-specific dysregulated biological pathways not visible using either sequencing method alone.

Fig. 1

Assigning single-cell RNA-seq to clone-of-origin using clonealign. a Given independently sampled single-cell DNA- and RNA-seq from the same tumor, the clonealign statistical framework assigns each cell’s gene expression profile to its clone-of-origin, uncovering transcriptional signatures of clonal fitness. b To relate cells as measured in RNA-space to their clones measured in DNA-space, we assume a relationship between gene copy number and gene expression (simulated data). c Simulations demonstrate the robustness of clonealign to the underlying proportion of genes exhibiting a copy number dosage effect. Even if only 30% of genes have a clone-specific copy number effect on expression, clones can still be accurately assigned with an average AUC >0.8. d Simulations demonstrate clonal assignment is accurate even when as few as 10–50 genes lie in regions of differing copy number between clones, allowing clonal assignment from only small-scale genomic rearrangements

Fig. 2

Inferring clone-dependent gene expression in SA501 triple-negative breast cancer xenograft. a Clone-specific copy number for ground truth clones in scDNA-seq (bottom) and clone-specific z-score expression for clonealign inferred clones in scRNA-seq (top) for regions exhibiting inter-clone copy number aberrations. In every copy number segment except one, when the copy number for a given clone is higher than others, then on average the normalized gene expression is also higher. b The mean log expression as a function of copy number across all clones. c Clone assignment probabilities for 1152 single-cell RNA-seq profiles across three clones. clonealign confidently assigns cells to clone A, with some cells exhibiting high assignment uncertainty between clones B and C. d A PCA projection using only genes residing in copy number regions shows the cells clustering by clone along components 2 and 4. ez-score normalized gene expression and copy number profiles for held-out data on chromosomes 8 and 18 as a function of genomic position (gene index along chromosome). In all but one copy number segment, when the copy number profile of a clone is higher, the normalized gene expression in that chromosome is also higher on average. f Differential expression analysis for genes residing in regions whose copy number is identical between clones highlights downregulation of MHC class I proteins

Fig. 3

Clone-specific gene expression in high-grade serous ovarian cancer cell lines. a Single-cell phylogeny for the OV2295R and TOV2295R HGSC cell lines inferred using a Latent Tree Model divided into four clones (TOV2295R_A, TOV2295R_B, OV2295R_C, OV2295R_D). b The scRNA-seq clone assignments for the four clone model (top), then sub-divided into eight clones (bottom). c Expression-CNA relationship on two held out chromosomes for TOV2295R validates the clonealign fit. d Top differentially expressed genes between clones in TOV2295R and e OV2295R