The Dynamics of Cytoplasmic mRNA Metabolism

Abstract

For all but a few mRNAs, the dynamics of metabolism are unknown. Here, we developed an experimental and analytical framework for examining these dynamics for mRNAs from thousands of genes. mRNAs of mouse fibroblasts exit the nucleus with diverse intragenic and intergenic poly(A)-tail lengths. Once in the cytoplasm, they have a broad (1000-fold) range of deadenylation rate constants, which correspond to cytoplasmic lifetimes. Indeed, with few exceptions, degradation appears to occur primarily through deadenylation-linked mechanisms, with little contribution from either endonucleolytic cleavage or deadenylation-independent decapping. Most mRNA molecules degrade only after their tail lengths fall below 25 nt. Decay rate constants of short-tailed mRNAs vary broadly (1000-fold) and are larger for short-tailed mRNAs that have previously undergone more rapid deadenylation. This coupling helps clear rapidly deadenylated mRNAs, enabling the large range in deadenylation rate constants to impart a similarly large range in stabilities.

Keywords: 5-ethynyl uridine; PAL- seq; TAIL-seq; deadenylation rates; decapping rates; mRNA decay rates; mRNA uridylation; metabolic labeling; poly(A)-tail lengths.

Figure 1.. Global Tail-Length Dynamics of Mammalian mRNAs

(A) Schematic of 5EU metabolic labeling. Experiments were performed with two 3T3 cell lines designed to induce expression of either miR-155 or miR-1 (cell lines 1 and 2, respectively) but cultured without microRNA induction. The 8 is in parentheses because an 8-h labeling period was included for only one line (cell line 1). For simplicity, all subsequent figures show the results for cell line 1, unless stated otherwise. (B) Tail-length distributions of mRNA molecules isolated after each period of 5EU labeling (key). Left: distributions were normalized to each have the same area. Right: distributions were scaled to the abundance of labeled RNAs in each sample and then normalized such that the steady-state sample had an area of 1. The steady-state sample was prepared with unselected RNA from the 40-min time interval. Each bin is 2 nt; results for the bin with tail lengths ≥ 250 nt are not shown. (C) Distributions of mean poly(A)-tail lengths for mRNAs of each gene after the indicated duration of 5EU labeling. Values for all genes that passed the tag cutoffs for tail-length measurement at all time intervals were included (n = 3,048). Each bin is 2 nt. Genes with mean mRNA tail-length values greater than ≥ 250 nt were assigned to the 250-nt bin. (D) Tail lengths over time. Mean tail lengths for mRNAs from each gene (n = 3,048) are plotted along with box-and-whiskers overlays (line, median; box, 25^th–75^th percentiles; whiskers, 5^th–95^th percentiles). Ss, steady state. See also Figures S1 and S2.

Figure 2.. Correspondence Between mRNA Half-life and Deadenylation Rate

(A) Relationship between half-life and mean steady-state tail length of mRNAs in 3T3 cells. For mRNAs of each gene, standard PAL-seq data were used to determine the length distribution of tails ≥ 50 nt, and data generated from a protocol that used single-stranded ligation to the mRNA 3′ termini (rather than a splinted ligation to the tail) were used to determine both the length distribution of tails < 50 nt and the fraction of tails < 50 nt. Compared to the tail-length distribution generated by only standard PAL-seq data, this composite distribution better accounted for very short and highly modified tails. Nonetheless, using the standard PAL-seq data without this adjustment produced a similar result (Figure S3G). Results for mRNAs of ribosomal protein genes (RPGs) and immediate-early genes (IEGs) (Tullai et al., 2007) are indicated (blue and red, respectively). (B) Relationship between mRNA half-life and mean tail length of metabolically labeled mRNAs isolated after 2 h of labeling. Otherwise as in (A). See also Figures S3A–S3D and S3G.

Figure 3.. Tail-Length Dynamics of mRNAs with Different Half-Lives

Tail-length distributions for mRNAs from individual genes. For each time interval (key), the distribution is scaled to the abundance of labeled RNA in the sample (top), and the distribution is also represented as a heatmap (bottom), with the range of coloration corresponding to the 5^th–95^th percentiles of the histogram density. Each bin is 5 nt. Bins for tails < 10 nt are not shown because the splinted ligation to the tail used in the standard PAL-seq protocol depletes measurements for tails < 8 nt. Bins for tails ≥ 250 nt are also not shown. See also Figures S3F and S3H–S3J.

Figure 4.. Computational Model of mRNA Deadenylation and Decay Dynamics

(A) Schematic of the computational model. k₀, k₁, and k₂ are terms for mRNA production, deadenylation, and decay, respectively, and ∅ represents the loss of the mRNA molecule. The curves (right) indicate the distributions used to model probabilities of production and decay as functions of tail length. They are schematized using the globally fitted parameters (v_p, m_d, and v_d) that defined each distribution (Table S2). The parameter m_p controls the mean (μ) of the negative binomial distribution (top curve), whereas the decay rate constant, β, scales the decay distribution (bottom curve) (Table S2). (B) Correspondence between the model and the experimental data. Results for mRNAs of these four genes are shown as representative examples because their fits fell closest to the 10^th, 25^th, 75^th, and 90^th percentiles of the distribution of R² values for all genes that passed expression cutoffs in the PAL-seq datasets (Figure S4F; n = 2,778). For each time interval, the blue line shows the fit to the model, and the red line shows the distribution of observed tail-length species, plotted in 2-nt bins and scaled to standards as in Figure 1B, right. Ss, steady state. (C) Correspondence between mean tail lengths generated from the model simulation and tail lengths measured in the metabolic labeling experiment. Shown for each gene are mean tail lengths for mRNAs at each time interval (key) from the simulation plotted with respect to the values observed experimentally. The discrepancy observed for some mRNAs at early time intervals was attributable to low signal for long-lived mRNAs at early times. The dashed line indicates y = x. See also Figures S4, S5A, and S5B and Tables S1 and S2.

Figure 5.. Dynamics of Cytoplasmic mRNA Metabolism

(A) Distribution of deadenylation rate constants (k₁ values), as determined by fitting the model to data for mRNAs from each gene (n = 2,778). (B) Tail lengths at which mRNAs decay, as inferred by the model. The model rate constants were used to simulate a steady-state tail-length distribution for each gene. The abundance of each mRNA intermediate was then multiplied by the decay rate constant k₂ to yield a distribution of decay events over all tail lengths. Plotted is the combined distribution for all mRNA molecules of all 2,778 genes. Results were indistinguishable when the distribution from each gene was weighted equally. Values for tails < 20 nt are shown as a dashed line because the model fit steady-state tail lengths < 20 nt as an average of the total abundance of tails in this region and, thus, did not provide single-nucleotide resolution for decay rates of these species. (C) Mean tail lengths at which mRNAs from each gene (n = 2,778) decayed, as inferred by the model. Otherwise, as in (B). (D) Distribution of decay rate constants (k₂ values) for mRNAs with 20-nt tail lengths, as determined by fitting the model to data for mRNAs from each gene (n = 2,778). (E) Correlation between the deadenylation rate constant (k₁) and the decay rate constant (k₂) at a tail length of 20 nt. The dashed line indicates y = x. See also Figure S4.

Figure 6.. A Modest Buildup of Short-Tailed Isoforms of Short-Lived mRNAs

(A) Relationship between the steady-state fraction of tails < 20 nt and mRNA half-life. For mRNAs of each gene, the fraction of tails < 20 nt was calculated from a composite distribution generated as in Figure 2A, which accounted for very short and highly modified tails. (B) Metatranscript distributions of steady-state tail lengths of short- and long-lived mRNAs (red and blue, respectively), with mRNAs from each gene contributing density according to their abundance. Results were almost identical when mRNAs were weighted such that each gene contributed equally. This analysis used the composite distributions as in (A). (C) Uridylation of short-lived mRNAs with short poly(A) tails. For mRNAs with half-lives < 20 min, the fraction of molecules with the indicated poly(A)-tail length at steady state is plotted, indicating for each tail length the proportion of tails appended with 0 through 10 U nucleotides (key). For mRNAs with poly(A)-tail length of 0, U residues were counted only if they could not have been genomically encoded. As poly(A) tails approached 20 nt, the ability to map reads with ≥ 3 terminal U residues diminished, but the ability to map reads with 1–2 terminal U residues was retained for poly(A) tails of each length. (D) Uridylation of long-lived mRNAs (half-lives > 10 h) with short poly(A) tails. Otherwise as in (C). (E) Distribution of tailless tags (regardless of mRNA half-life) as a function of their distance from the annotated 3′ end of the UTR. Tags with a terminal A (or with a terminal A followed by one or more untemplated U) were excluded, even if the A might have been genomically encoded. The proportion of tails appended with 0 through 10 U nucleotides is shown (key). (F) Relationship between the standard deviation of steady-state tail length and mRNA half-life. Otherwise as in (A). See also Figures S5C–S5J.

Figure 7.. Deadenylation and Decay Dynamics of Synchronous mRNA Populations.

(A) Schematic of 5EU metabolic-labeling and actD treatments used to analyze synchronized cellular mRNAs. Cells from cell line 2 were treated for 1 h with 5EU, then treated with actD continuously over a time course spanning 15 h. (B) Tail-length distributions of labeled mRNA molecules observed at the indicated times after stopping transcription (key). Left: distributions were normalized to all have the same area. Right: distributions were scaled to the abundance of labeled RNAs in each sample and then normalized such that the 0-h time interval had an area of 1. Each bin is 2 nt; results for the bins with tail lengths < 8 nt and ≥ 250 nt are not shown. At 0 h, 7% of the tails were still ≥ 250 nt, which helps explain why the density for the remainder of the tails fell below that observed at 1 h. (C) Distributions of mean poly(A)-tail lengths for labeled mRNAs of each gene after the indicated duration of transcriptional shutoff. Values for all mRNAs that passed the cutoffs for tail-length measurement at all time points were included (n = 2,155). Each bin is 2 nt. (D) Relationship between half-life and mean tail length of labeled mRNAs from each gene after 1 h of actD treatment. (E) Labeled mRNA abundance as a function of mean tail length over time. Results are shown for mRNAs grouped by half-life quantiles (95%, 75%, 50%, 25%, and 5%, left to right, with mRNAs in the 5% bin having the shortest half-lives). Each half-life bin contains 100 genes. mRNA abundance was determined from paired RNA-seq data. Each line connects values for mRNA from a single gene. (F) Simulation of mRNA abundance as a function of mean tail length over time. For each gene in (E), model parameters fit from the continuous-labeling experiment were used to simulate the initial production of mRNA and its mean tail length from each gene, as well as the fates of these mRNAs and mean tail lengths after production rates were set to 0. Results are plotted as in (E), but using a shorter time course (key) to accommodate the faster dynamics observed without actD. See also Figure S3E.