Cellular RNA levels are determined by the interplay of tightly regulated processes for RNA production (transcription), processing (e.g
., polyadenylation, splicing, transport, localization), and depletion (degradation). In addition to transcriptional regulation1,2
, changes in RNA degradation can also significantly affect differential gene expression3,4
, particularly in mammalian cells where RNA half-lives are typically longer3
The response of immune dendritic cells (DCs) to pathogens provides a compelling model of a temporal transcriptional sensory response in mammalian cells5,6
. DCs initiate and regulate immune responses. Upon stimulation with pathogen components, DCs activate a regulatory program, which unfolds over ~24h and involves the activation of ~1700 genes and repression of ~2000 genes7
, some peaking as early as 30 minutes, whereas others peak after 6 hours or more. In a recent study, we identified over a hundred transcription factors, and at least a dozen RNA binding proteins controlling this response7
There are two key questions on the roles of transcription and degradation in regulating RNA levels in this response. (1) Which of the two processes contributes most to shaping changes in RNA levels over time? (2) Do such changes primarily result from variation of constant rates between genes or from variability of the rates for each gene over time?
The extent to which RNA stability contributes to dynamic changes in RNA levels is still unclear and debated. Most works focus on transcriptional mechanisms8-12
, tacitly assuming that degradation rates (per gene) are constant over time13
(‘constant degradation hypothesis’). However, recent studies suggest that changes in a gene’s mRNA level following stimulation are strongly affected by corresponding changes in its RNA degradation rate13,14
, modeled either by a single change3,4,15
or as a continuous shift16,17
over time (‘varying degradation hypothesis’). Indeed, it was proposed that such changes in degradation rates may determine up to half of the temporal changes in RNA levels in mammalian cells16
. Distinguishing between these two hypotheses is hampered by the shortcomings of the indirect methods used for determining transcription and degradation rates, which may limit their relevance in vivo
. For example, nuclear run-on assays for measuring transcription rates are conducted ex-vivo
in isolated nuclei15,16,18
, and methods for estimating degradation rates by transcriptional inhibition, either with antibiotics or temperature sensitive mutants3,4,13,17,19
, are not well adapted to dynamic settings and severely affect cell growth and survival20
Improved direct measurement of RNA production rates may allow us to address these questions. Recent studies used metabolic labeling of RNA with 4-thiouridine (4sU), a naturally occurring modified Uridine, to distinguish recently-transcribed RNA from the overall RNA population, with minimal interference to normal cell growth21-26
. The modified base is incorporated into the growing RNA chain in place of Uridine, marking it, and serving as an attachment point for a biotin tag for easy separation of newly transcribed RNA from the total RNA population (Supplementary Fig. 1
). In previous work, labeled RNA was hybridized to standard microarrays, requiring relatively large quantities of RNA and hence lengthier 4sU labeling times (1-2h). Thus, most existing studies focused on variation between genes during steady state conditions22,25,26
, and a single 4 time points microarray study23
, though promising, lacks a systematic dynamic analysis.
Here, we use metabolic labeling coupled with advanced RNA quantification assays and computational modeling to study RNA regulation in the response of mouse DCs to Lipopolysaccharide (LPS). Leveraging the Nanostring nCounter technology for accurate multiplex measurement of RNA27
and massively parallel sequencing28
, we significantly reduce metabolic labeling time to directly measure RNA transcription rates at high temporal resolution for a selected set of signature genes, and at a lower temporal resolution on a genome-scale. We develop new computational models to decompose RNA levels into the separate contributions of RNA production and degradation, and estimate changes in degradation rates between genes and over time. We leverage the reduced abundance of rRNA and other stable RNA populations in recently transcribed RNA, to sequence a broad representation of the labeled RNA transcriptome, and determine the processing rates of precursor mRNA (pre-mRNA).
We discover key principles of temporal RNA regulation in mammalian cells. We find that changes in transcription rate highly correlate with changes in RNA level, preceding them by ~15-30 min, with about twice as long a delay in down-regulated than up-regulated genes. In contrast to recent works16,17
, we find that dynamic changes in degradation rates have minimal effect on most RNA profiles, but that they do play a unique role in genes with sharp ‘peaked’ responses. Genome-wide analysis shows substantial variation in both degradation and processing rates between
genes, rather than over time, consistent with their regulatory and functional differences. Our method is a new and effective tool for studying key processes controlling cellular RNA levels.