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

Abstract

The integration of lineage tracing with scRNA-seq has transformed our understanding of gene expression heritability during development, regeneration, and disease. However, lineage tracing is technically demanding and most existing scRNA-seq datasets are devoid of lineage information. Here we introduce Gene Expression Memory-based Lineage Inference (GEMLI), a computational pipeline allowing to predict cell lineages over several cell divisions solely from scRNA-seq datasets. GEMLI leverages genes displaying conserved expression levels over cell divisions, and allows i.a. identifying cell lineages in a broad range of cultured cell types, in intestinal organoids, and in crypts from adult mice. GEMLI recovers GO-terms enriched for heritable gene expression, allows to discriminate symmetric and asymmetric cell fate decisions and to reconstruct individual cellular structures from pooled scRNA-seq datasets. GEMLI considerably extends the pool of datasets from which lineage information can be obtained, thereby facilitating the study of gene expression heritability in a broad range of contexts. GEMLI is available at (https://github.com/UPSUTER/GEMLI).

Stereotyped cell lineage trees support robust development

“…It is therefore possible to isolate the contribution of transcriptional memory to the expression constraint by contrasting the CV of real CLTs with that of randomized CLTs ( Figure 3A and Methods ), which is hereinafter referred to as the “memory index”. We note that this definition of memory index is similar to that used in previous transcriptional memory-related studies 29,39 .…”

“…which is hereinafter referred to as the "memory index". We note that this definition of memory index is similar to that used in previous transcriptional memory-related studies 29,39 .…”

“…Finally, the memory index was defined as M = (min( CV ) - mean(min( CV Random ))) / min( CV ). Note that the final division by min( CV ) is different from the previously defined memory index 29,39 , but allows comparisons between genes with very different baseline CVs or expression levels.…”

“…We followed previously proposed methods 29,39 to calculate transcriptional memory index. In each cell type and for each gene expressed in >10% of cells of this type, the CV of the expression levels was calculated among all terminal cells of this type within a sub-CLT (containing at least two cells of this type).…”

Toward a systems-level probing of tumor clonality

“… 130 , 131 , 132 , 133 Similarly, computational methodologies relying on transcriptional memory between progenies may also enable barcode-free lineage tracing with single cell RNA-sequencing datasets alone. 134 , 135 However, although clonality in some cancers can be well-described by single cell RNA-sequencing, others may require multi-parameter approaches. 136 Furthermore, it will be critical to perform simultaneous evolutionary measurements of cell-intrinsic factors and extrinsic environments, and identify the chronological order of molecular and cellular events to unravel unifying and cancer-specific systems-level control principles ( Figure 1 ).…”