Disrupting cellular memory to overcome drug resistance

Abstract

Cellular memory describes the length of time a particular transcriptional state exists at the single-cell level. Transcriptional states with memory can underlie important processes in biology, including therapy resistance in cancer. Here we present a new experimental and computational approach for identifying gene expression states with memory at single-cell resolution by combining single-cell RNA sequencing (scRNA-seq) with cellular barcoding. With this technique, we can systematically quantify the full expression profile of these memory states as well as their population dynamics including the relative growth rates of cells within different states and the rates of switching between states. We applied this approach to human melanoma cells and uncovered memory gene expression states that are predictive of which cells will be resistant to combination BRAF and MEK inhibition. While these cells have not been treated with targeted therapies, they already express markers of resistance, and thus are referred to here as “primed” for resistance. From the scRNA-seq data alone, the drug-susceptible and primed cells appear as two distinct and unrelated cell populations. From the lineage barcodes, we found that most cells remain within the same state, demonstrating memory of gene expression. However, we also directly observe state switching as we find that about 18% of lineages contain cells that have switched states. While the molecular drivers of state switching are not immediately apparent from scRNA-seq data alone, we specifically analyzed lineages that undergo state switching and identified TGF-β and PI3K as drivers of state switching at the single-cell level. Through functional validation and single-cell RNA-sequencing, we show mechanistic single-cell manipulation of state switching that ultimately yields changes in cellular phenotype. We leverage these mechanisms of state switching to delay resistance by demonstrating that reducing the number of primed cells with a PI3K inhibitor prior to BRAF and MEK inhibition ultimately leads to fewer resistant colonies. Our results show that modulation of cellular signaling can globally regulate gene expression programs to overcome therapy resistance.

Barcode-free prediction of cell lineages from scRNA-seq datasets

“…We used the LARRY barcoding library 17 to transduce mESCs at low multiplicity of infection, sorted barcoded cells (Extended Data Fig.1a-c) and grew them for 48h before scRNA-seq. In addition, we used public lineage-annotated datasets of other cell types: primary mouse embryonic fibroblasts (MEF) 18 , primary mouse CD8+ T-lymphocytes (CD8), lymphocytic leukemia cells (L1210) (both 29 ), primary mouse hematopoietic stem and progenitor cells (HSPC, more precisely LK and LSK subsets) 17 , hematopoietic stem cells (HSC) 27 , melanoma cells (WM989) 9 , as well as intestinal crypts and organoids 32 . These datasets were generated using different lineage assignment and scRNA-seq platforms and encompassed cells in self-renewal or differentiation conditions grown for 2-14 days (Table 1).…”

“…GEMLI allows to predict cell lineages but also reliably identifies genes displaying gene expression memory, thus opening the door to understanding underlying epigenetic mechanisms in different cell types. The identification of different types of cell divisions (symmetric or asymmetric) by GEMLI can be used to identify cell fate regulators directly from scRNA-seq datasets, alleviating the need for experimental lineage annotation that has been used so far in this context 9,17,19,27 . Notably, GEMLI-predicted lineages can be used as input for trajectory inference and cell fate decision analysis designed for lineage-annotated scRNA-seq data [46][47][48] .…”

“…Heterogeneous mRNA levels for given genes are often transient, but the expression levels of some genes can be preserved through cell division 7 . The existence of such gene expression memory is of major fundamental and biomedical interest, as it is involved in cancer resistance to drugs [8][9][10][11][12][13][14] , disease onset 11,15 , and differentiation trajectories in development, homeostasis and regeneration [16][17][18][19][20][21] . To study gene expression over cell divisions, scRNA-seq has been combined with different lineage tracing techniques.…”

Exploring functional protein covariation across single cells using nPOP

“…2e,f. Recent studies of these melanoma cells identified a transcriptomic signature associated with a cell state that is more likely to resist treatment by the cancer drug vemurafenib 7, 29, 30 . We estimated the differential protein abundance of these markers, which is visualized by color coding the cells by the mean abundance of proteins whose transcripts were identified as markers for each cell state 7 .…”

“…2e,f. Recent studies of these melanoma cells identified a transcriptomic signature associated with a cell state that is more likely to resist treatment by the cancer drug vemurafenib 7,29,30 .…”