Precision-Based Filtering Facilitates Integration of Conventional and Single-Nucleus Transcriptomes to Identify Time- and Temperature-Sensitive Cell Populations

Transcriptome analysis via RNA sequencing (RNAseq) has become a ubiquitous method of molecular characterization from whole organisms, dissected tissues, and single cells. These experiments have provided an extraordinary volume of data describing the molecular states and responses to many conditions. However, standard approaches to RNAseq analysis commonly use expression level filters that eliminate potentially useful data in the service of decreasing noise. Here we describe the implementation of a coefficient of variation-based filter for RNAseq gene expression data. This filter prioritizes consistent data across replicates, allowing lowly-expressed genes with low-variation measurements to be retained for downstream analysis. We show that, in our Arabidopsis RNAseq data set, this filter allows for the inclusion of many more transcription factors than even a low-stringency expression level filter. We find that these lowly-expressed genes mark specific cell clusters in our single-nucleus (sn)RNAseq dataset. We further characterize communities of co-expressed genes, sampled across the day at two growth temperatures, in relation to snRNAseq cell clusters, finding evidence for a highly photosynthetic cell population, and a cell state marked by high cell division and translation. These methods can be expanded to RNAseq analysis in many systems, facilitating the construction of more detailed models of tissue-specific gene regulatory networks.