C1 CAGE detects transcription start sites and enhancer activity at single-cell resolution

Abstract

26Single-cell transcriptomic profiling is a powerful tool to explore cellular heterogeneity. However, 27 most of these methods focus on the 3'-end of polyadenylated transcripts and provide only a 28 partial view of the transcriptome. We introduce C1 CAGE, a method for the detection of 29 transcript 5'-ends with an original sample multiplexing strategy in the C1 TM microfluidic system. 30We first quantified the performance of C1 CAGE and found it as accurate and sensitive as other 31 methods in C1 system. We then used it to profile promoter and enhancer activities in the cellular 32 response to TGF-β of lung cancer cells and discovered subpopulations of cells differing in their 33 response. We also describe enhancer RNA dynamics revealing transcriptional bursts in subsets 34 of cells with transcripts arising from either strand within a single-cell in a mutually exclusive 35 manner, which was validated using single molecule fluorescence in-situ hybridization. 37Single-cell transcriptomic profiling can be used to uncover the dynamics of cellular states and 38 gene regulatory networks within a cell population (Trapnell, 2015; Wagner, Regev and Yosef, 39 2016). Most available single-cell methods capture the 3'-end of transcripts and are unable to 40 identify where transcription initiates. Instead, capturing the 5'-end of transcripts allows the 41 identification of transcription start sites (TSS) and thus the inference of the activities of their 42 regulatory elements. Cap analysis gene expression (CAGE), which captures the 5'-end of 43 transcripts, is a powerful tool to identify TSS at single nucleotide resolution (Shiraki et al., 2003; 44 Carninci et al., 2006). Using this technique, the FANTOM consortium has built an atlas of TSS 45 across major human cell-types and tissues (Forrest et al., 2014), analysis of which has led to the 46 identification of promoters as well as enhancers in the human genome (Andersson et al., 2014; 47 Hon et al., 2017). Enhancers have been implicated in a variety of biological processes (Lam et 48 al., 2014; Li, Notani and Rosenfeld, 2016), including the initial activation of responses to 49 stimuli (Arner et al., 2015) and chromatin remodeling for transcriptional activation (Mousavi et al., 50 2013). In addition, over 60% of the fine-mapped causal noncoding variants in autoimmune 51 disease lay within immune-cell enhancers (Farh et al., 2015), suggesting the relevance of 52 enhancers in pathogenesis of complex diseases. Enhancers have been identified by the 53 presence of balanced bidirectional transcription producing enhancer RNAs (eRNAs), which are 54 generally short, unstable and non-polyadenylated (non-polyA) (Andersson et al., 2014). Single 55 molecule fluorescence in situ hybridization (smFISH) studies have suggested that eRNAs are 56 induced with similar kinetics to their target mRNAs but that co-expression at individual alleles 57 was infrequent (Rahman et al., 2016). However, the majority of enhancer studies have been 58 conducted using bulk populations of cells...

Transcriptional control by enhancers and enhancer RNAs

“…Current models posit that enhancers are transcribed bidirectionally in a manner resembling gene promoters [79,[81][82][83][84], with unidirectional transcription occurring at a small subset of these elements [76,78,80,85,86]. Interestingly, a recent single-cell transcriptomic profiling technique has been used to demonstrate that enhancers are primarily transcribed unidirectionally from a single strand on a cell-to-cell basis, challenging the previously accepted belief that both strands are transcribed simultaneously [87]. This study also used single-molecule fluorescence in situ hybridization (smFISH) to show that when both strands are transcribed in the same cell, the eRNAs rarely colocalize.…”

Single-cell RNA cap and tail sequencing (scRCAT-seq) reveals subtype-specific isoforms differing in transcript demarcation

“…1b), with sequencing depth of 4 million reads per sample (Supplementary Table 2). Similarly, looking into ERCC data generated by other methods 19,26 , such as C1 CAGE, C1 STRT, we also found high false-positive rates for peaks identified as TSSs in these datasets ( Supplementary Fig. 1c), and applying the machine learning model increased the accuracy to above 88.9% ( Supplementary Fig.…”

“…Previously, machine learning has been successfully used to predict differential alternative splicing 32,33 , but none of them have been developed for the purpose of identifying authentic demarcations of RNA isoforms to elucidate the transcriptomic complexity of single cells. Furthermore, the model trained in this study also improves the accuracy of other methods to over 90%, as evidenced by the ERCC data from C1 CAGE 19,26 and C1 STRT 26 , indicating that our model can be applied to other datasets that contain previously unrecognized high false-positive signals. As a result, the accuracy of our approach for quantification of alternative isoforms is very high, as the measured abundances are highly concordant with the ground truth, with a Pearson's correlation coefficient of 0.98.…”

Single-cell sequencing techniques from individual to multiomics analyses

“…This platform enables the automatic isolation of 96 cells, cDNA synthesis, and amplification based on Smart-seq through microfluidics. C1-CAGE can also be conducted using the C1 system, which enables the profiling of the 5′ end of transcripts with strand information in a single cell 15 . Microdropletbased systems such as Chromium (10× Genomics), ddSEQ (Bio-Rad/Illumina), Nadia (Dolomite), and inDrop (1CellBio) and microwell-based systems such as Rhapsody (BD) and ICELL8 (Takara) also exist.…”