Global analyses of UPF1 binding and function reveal expanded scope of nonsense-mediated mRNA decay

Abstract

UPF1 is a DNA/RNA helicase with essential roles in nonsense-mediated mRNA decay (NMD) and embryonic development. How UPF1 regulates target abundance and the relationship between NMD and embryogenesis are not well understood. To explore how NMD shapes the embryonic transcriptome, we integrated genome-wide analyses of UPF1 binding locations, NMD-regulated gene expression, and translation in murine embryonic stem cells (mESCs). We identified over 200 direct UPF1 binding targets using crosslinking/immunoprecipitation-sequencing (CLIP-seq) and revealed a repression pathway that involves 3' UTR binding by UPF1 and translation but is independent of canonical targeting features involving 3' UTR length and stop codon placement. Interestingly, NMD targeting of this set of mRNAs occurs in other mouse tissues and is conserved in human. We also show, using ribosome footprint profiling, that actively translated upstream open reading frames (uORFs) are enriched in transcription factor mRNAs and predict mRNA repression by NMD, while poorly translated mRNAs escape repression. Together, our results identify novel NMD determinants and targets and provide context for understanding the impact of UPF1 and NMD on the mESC transcriptome.

Figure 1.

Consistent derepression of hundreds of mRNAs with and without canonical NMD features occurs following UPF1 depletion and translational inhibition. (A) Overlap of mRNAs that changed expression by more than 1.1-fold in the same direction in each of three NMD inhibition experiments (shRNA Upf1-1, shRNA Upf1-2, and CHX treatment). (B) Cumulative distribution functions (CDFs) of changes in mRNA abundance following UPF1 depletion (shRNA Upf1-1) for all dEJ mRNAs (dashed red line), consistently changing dEJ mRNAs (solid red line), or mRNAs without an annotated dEJ (black line). P-value determined by Wilcoxon rank sum test. (C) Ratios of median fold expression change following NMD inhibition of dEJ_cons to non-dEJ isoforms. Error bar represents standard error of the two populations compared. P-values determined as in B. (D) As in B for isoforms behaving consistently with different annotated 3′ UTR lengths (different green lines). Median expression changes and standard error are shown at right. See also Supplemental Figure S1F. (E) As in C for mRNAs with long (1500–10k nt) versus short (50–350 nt) 3′ UTRs. (F) Interaction between dEJ and 3′ UTR length. Median fold change in expression following UPF1 depletion of consistent mRNAs with different 3′ UTR lengths without (left) and with (middle) an annotated dEJ. Ratios of median expression change following UPF1 depletion between mRNAs with and without an annotated dEJ of a given 3′ UTR length (right). Significance of differences between different 3′ UTR length bins was determined by permutation test (n = 2000). This trend was also observed when comparing expression changes between each dEJ isoform and non-dEJ isoforms with equivalent 3′ UTR lengths (±10%) following NMD inhibition (Supplemental Fig. S1H). Results were similar for other NMD inhibitory treatments. All fold change values and ratios are plotted on a log₂ scale. P-values: (*) P < 0.05, (**) P < 0.01, (***) P < 0.001, (****) P < 0.0001.

Figure 2.

Translation of uORFs is associated with UPF1-mediated repression. (A) mRNA-seq (blue) and ribosome footprint (orange) reads from UPF1 depleted (shRNA Upf1-1) cells mapping to Dmtf1 and Armc1 mRNAs. Dmtf1 (top) contains a tuORF (outlined with dark gray lines), while Armc1 contains a single ntuORF (dashed lines). (B) CDFs of changes in mRNA abundance following UPF1 depletion (shRNA Upf1-1) for all genes with a tuORF (dashed yellow line), consistent genes with a tuORF (orange line), genes with a uORF (dashed gray line), and genes with an ntuORF (black line). P-values determined by Wilcoxon rank sum test. (C) Ratios of median fold expression change following NMD inhibition of consistent genes with a tuORF to genes with an ntuORF. Error bars represent standard error of the two populations compared. P-values determined as in B. (D) Fraction of genes harboring a tuORF for all expressed genes, all expressed transcriptional regulators (GO:0045449), and transcriptional regulators derepressed by more than 1.1- or 1.2-fold in at least two out of three NMD inhibition experiments. P-value of enrichment determined by hypergeometric test. Numbers of genes in each category are indicated. Error bars indicate binomial standard deviations. Fold change values and ratios shown in B and C are plotted on log₂ scale. Asterisks as in Figure 1.

Figure 3.

UPF1 interacts predominantly with 3′ UTRs of mature mRNAs. (A) Distribution of CLIP-seq reads mapping to 5′ UTRs, coding sequences (CDS), and 3′ UTRs using RNase A (“A”) or RNase I (“I”) and RNA-seq reads, from control and CHX+ treatments. (B) Metagene plot of average UPF1 CLIP tag density per gene in 500-nt regions flanking the TC (red line) under normal (top) and CHX (bottom) conditions. Density was smoothed using a Gaussian with standard deviation of 10 nt. (C,D) Correlation of UPF1 CLIP samples binding in 3′ UTRs of genes with minimum FPKM (fragments per kilobase per million mapped reads) of 50. Correlations of MBNL1 CLIP data in mouse C2C12 cells and two mouse brain samples (Wang et al. 2012) and of AGO2 CLIP data in wild-type and Dicer null mESCs (Leung et al. 2011) are shown for comparison. In C, correlation coefficients were calculated between densities of CLIP binding over all UTRs. In D, correlation coefficients were calculated between densities over 100-nt windows across all UTRs. Grayscale emphasizes higher values in C. (E) Mean nucleotide content flanking UPF1 binding sites within 3′ UTRs. Lines indicate moving average of values for each nucleotide using a 3-bp sliding window. (F) UPF1 binding to the Esrrb mRNA in three CLIP experiments in untreated cells. Schematic of Esrrb mature mRNA displayed below. Width of gray bar indicates CDS and UTR regions, and vertical black bars indicate exon-exon junctions.

Figure 4.

3′ UTR binding by UPF1 is associated with mRNA repression. (A) CDFs of gene expression changes following UPF1 depletion (shRNA Upf1-1) for consistently behaving genes bound by UPF1 in the 3′ UTR (blue line), all genes bound by UPF1 in the 3′ UTR (cyan line), and unbound genes (black line). (B) Ratios of median fold expression change following NMD inhibition between consistently behaving genes bound by UPF1 in the 3′ UTR and unbound genes in mESCs (dark gray), as well as for Smg1 KO in MEFs (McIlwain et al. 2010), and Upf2 KO in mouse liver (Weischenfeldt et al. 2012) (light gray), and for human homologs of these genes following UPF1 depletion in HeLa (Cho et al. 2012) and U2OS cells (Wang et al. 2011) (pink). Error bars represent standard error of the two populations compared. (C) As in A, except mRNA abundance measurements were made in wild-type and Upf2 KO mouse liver (Weischenfeldt et al. 2012). (D) Fraction of all genes and genes bound by UPF1 in the 3′ UTR that have an annotated isoform harboring a dEJ. Error bars indicate binomial standard deviation. (E) Distribution of 3′ UTR lengths of genes bound by UPF1 in their 3′ UTRs. Lengths were assigned based on the best-annotated isoform for each gene. (F) As in A for consistently behaving genes bound in the 3′ UTR by UPF1 and genes sampled with replacement to match the distribution of expression levels and 3′ UTR lengths. Significance was calculated by bootstrapping (n = 20,000). Mean and 95% confidence intervals of subsampled populations are shown in black and gray lines. Median fold change in expression between UTR groups shown is 1.16, and results were similar using shRNA Upf1-2 and CHX treatment (median fold changes 1.10 and 1.15, respectively) (data not shown). (G) Mean number of 50-nt regions with high G content (95th percentile) per kb of 3′ UTR for genes with UPF1-bound 3′ UTRs, all genes, and genes not bound by UPF1 (controlled for expression level and 3′ UTR length). Error bars represent standard error of the mean (Bound or All genes) or standard error of the means of sampled populations (Not bound). (H) As in A, except gene names were identified by homology with mouse genes either bound or unbound by UPF1 in their 3′ UTR, and mRNA abundance measurements were made in control- and UPF1-depleted HeLa cells (Cho et al. 2012). All expression fold change values and ratios are plotted on a log₂ scale. Asterisks as in Figure 1. See also Supplemental Figure S4.

Figure 5.

Relationship between TE and NMD-triggering gene features. (A) Median fold change of mRNA abundance following UPF1 depletion (shRNA Upf1-1) for consistently behaving genes grouped by percentile rank of translational efficiency (TE). Results were similar using shRNA Upf1-2 and CHX treatment (not shown). (B) As in A, except for non-dEJ (gray) and dEJ (red) mRNAs. Ratios of median fold change in expression following UPF1 depletion between dEJ and non-dEJ mRNAs with given TE is shown at right (black). For dEJ calculation, expression changes were calculated on an isoform level and TE was assigned based on the TE of the full ORF. (C,D) As in B, except for UPF1 binding in 3′ UTRs (C, blue) and 3′ UTR length (D, green). Long and short 3′ UTRs were defined as 1500–10,000 nt and 50–350 nt, respectively. Significance of differences in expression changes between TE bins of non-feature-containing genes (non-dEJ, not bound, and short 3′ UTR) were similar to those of all genes in A. All expression fold changes and ratios are plotted on a log₂ scale. P-values calculated using Wilcoxon rank sum test. Asterisks as in Figure 1.

Figure 6.

NMD features and models of UPF1-dependent mRNA repression. (A) Predictive capacity of mRNA features for NMD-regulation. Fractions of expressed genes harboring a dEJ (red), long 3′ UTR (green), UPF1 3′ UTR binding (blue), uORF (light orange), and/or tuORF (orange) that were derepressed consistently are shown. Shown for comparison are the fractions of genes derepressed consistently regardless of feature content (medium gray), without any NMD-inducing feature (light gray), and with at least one feature (dark gray). P-values above each feature indicate significance relative to all expressed genes; brackets indicate significant comparisons between features (hypergeometric test). See also Supplemental Figure S5A. Asterisks as in Figure 1. (B) (Left) Canonical dEJ-mediated regulation. EJC components (blue and gray) are deposited as a consequence of splicing in the nucleus ∼24 nt upstream of an exon-exon junction (black bar). Members of the EJC, including UPF2 and UPF3, help to stabilize the transient UPF1-ribosome interaction as well as to stimulate UPF1's phosphorylation and helicase activity, ultimately leading to decay of the message. (Right) 3′ UTR UPF1 binding-mediated regulation. UPF1 binds to mRNA 3′ UTRs independent of the presence of an exon-exon junction. At some frequency, UPF1 is activated by interaction with cytoplasmic EJC components. These factors may either be recently released from mRNAs due to translation or perhaps stably associated with message 3′ UTRs independent of an exon-exon junction.