Cell type- and factor-specific nonsense-mediated RNA decay

Abstract

Nonsense-mediated RNA decay (NMD) is a highly conserved RNA turnover pathway that influences several biological processes. Specific features in messenger RNAs (mRNAs) have been found to trigger decay by NMD, leading to the assumption that NMD sensitivity is an intrinsic quality of a given transcript. Here, we provide evidence that, instead, an overriding factor dictating NMD sensitivity is the cell environment. Using several genome-wide techniques to detect NMD-target mRNAs, we find that hundreds of mRNAs are sensitized to NMD as human embryonic stem cells progress to form neural progenitor cells. Another class of mRNAs escape from NMD during this developmental progression. We show that the differential sensitivity to NMD extends to in vivo scenarios, and that the RNA-binding protein, HNRNPL, has a role in cell type-specific NMD. We also addressed another issue in the field-whether NMD factors are core or branch-specific in their action. Surprisingly, we found that UPF3B, an NMD factor critical for the nervous system, shares only 30% of NMD-target transcripts with the core NMD factor UPF2. Together, our findings have implications for how NMD is defined and measured, how NMD acts in different biological contexts, and how different NMD branches influence human diseases.

Graphical Abstract

Figure 1.

Identification of UPF2-regulated genes in human ESCs. (A) The strategy we used to define NMD-target mRNAs in primary stem cells. UPF2-dependent NMD-target mRNAs in ESCs were identified by UPF2 KD with a UPF2 siRNA; see Supplementary Fig. S1) and by using an inducible system to transiently KO UPF2 (UPF2-iKO; described in panel C). Transcripts upregulated in response to these UPF2 disruptions were defined as “high-confidence” NMD-target mRNAs if they harbored a robust NMD-inducing feature. If such transcripts also exhibited high occupancy of pUPF1 (determined by RIPseq analysis), we defined them as “unambiguous” NMD-target mRNAs. UPF3B-dependent NMD-target mRNAs were identified by knocking out UPF3B (UPF3B-KO) in ESCs and then performing the same analysis as described above. UPF2- and UPF3B-dependent NMD-target mRNAs in NPCs were identified by the same approaches as described above in control (Ctrl) and mutant NPCs derived from the ESCs described above. (B) Western blot analysis of ESCs cultured under the indicated conditions. UPF2-iKO generates a truncated version of UPF2. GAPDH is the loading Ctrl. (C) UPF2-iKO scheme. Step 1: LoxP elements were introduced on either side of UPF2 exon 2 in ESCs. Step 2: A Cre-ERT2 construct was introduced into the AAVS1 safe harbor site. 4-OHT treatment triggers Cre-ERT2 nuclear translocation and consequent exon 2 deletion. (D) Unsupervised hierarchical clustering of RNA-seq samples from four groups of ESCs: (i) UPF2^fl/fl ESCs transiently transfected with siUPF2 (“UPF2-KD”), (ii) UPF2^fl/fl ESCs transiently transfected with a scramble siRNA (“Ctrl”), (iii) UPF2^fl/fl Cre-ERT2 ESCs transiently transfected with UPF2 siRNA (“UPF2-iKO & KD”), and (iv) UPF2^fl/fl Cre-ERT2 ESCs transiently transfected with scramble siRNA (“UPF2-iKO”). Two independently derived UPF2^fl/fl or UPF2^fl/fl Cre-ERT2 ESC clones were analyzed, with three experimental replicates for each (e.g. samples 1–3 were from clone 1, and samples 4–6 were from clone 2). All ESCs were incubated with 4-OHT. (E, G, and I) DEGs (q <0.01, fold change >2) from each comparison. (F, H, and J) Biological functions associated with the DEGs defined in the panel to the left. Statistical significance [−Log₁₀ (p-value)] is indicated by the bar. The number of DEGs for a given category is indicated by an “X.”

Figure 2.

Identification of high-confidence NMD-target transcripts in human ESCs. (A–C) Top panel: The total number of DETs, as well as the number of DETs with the indicated NMD-inducing feature, for the UPF2 disruption condition shown (determined from the datasets described in Fig. 1). Bottom panel: Violin plot showing the 3′ UTR length of the up- and down-regulated DETs, as well as all transcripts. ** p <0.01; **** p <0.0001; ns, not statistically significant (unpaired t-test). (D) Overlap amongst high-confidence UPF2-dependent NMD-target RNAs (those with a dEJ and/or an extended 3′ UTR) identified from the datasets in panels (A)–(C). (E–G) Examples of alternatively processed transcripts degraded by NMD in ESCs, as defined by the IsoformSwitchAnalyzeR program. The bar graphs show the expression of the main isoform versus an alternative isoform harboring an NMD-inducing feature that is upregulated by the indicated UPF2 disruption (other isoforms are not shown). The value shown for each isoform is the proportion of the total. Bonferroni-adjusted t-test * q<0.05. (H) Alternatively processed mRNA isoforms upregulated by the indicated UPF2 perturbation condition, as defined by the IsoformSwitchAnalyzeR program. Shown is the total number of upregulated transcripts, as well as upregulated transcripts with the indicted NMD-inducing feature. “Both” refers to upregulated transcripts with both a dEJ and an extended 3′ UTR. (I) Biological functions defined by the Metascape program (p <0.01) associated with the full profile of high-confidence UPF2-dependent NMD-target mRNAs in ESCs [defined from the datasets shown in panels (A)–(C) and (H)].

Figure 3.

Identification of high-confidence UPF3B-dependent NMD-target transcripts in human ESCs. (A) The location of the deletion boundary on either side of UPF3B exon 2 in the two independent UPF3B-KO human ESC clones generated using CRISPR. (B) Western blot analysis of two independent clones of UPF3B-KO and Ctrl ESCs. Shown is a representative result from two independent experiments. (C) Expression of pluripotency genes, based on RPKM (reads per kilobase per million mapped reads) values from RNA-seq analysis. There is no statistical significant difference (Wald test) in the expression of the three pluripotency genes in UPF3B-KO versus Ctrl ESCs. (D) Unsupervised hierarchical clustering of the indicated ESCs assayed by RNA-seq (two independent Ctrl and UPF3B-KO human clones, with three biological replicates from each). (E) Effect of combined UPF3B loss and UPF2 KD on NMD magnitude in ESCs, as quantified by NMD-sensitive/NMD-insensitive mRNA isoform ratio. Expression of these mRNA isoforms from the PTBP1 and TRA2B genes [107] was determined by quantitative PCR (qPCR) analysis of ESCs with the indicated genotype (Ctrl or UPF3B-KO) transiently transfected with a UPF2 siRNA or a negative control scramble siRNA (Ctrl). The values shown are from two independently derived UPF3B-KO ESC clones, with two independent biological replicates. Statistical significance was determined using the Student’s t-test. Different letters (a, b, c, and d) denote statistically significant differences between different groups (p <0.05). (F) DEGs (q <0.01, fold change >2) identified from the RNA-seq analysis in panel (D). (G) Functions associated with the DEGs defined in panel (F). Statistical significance [−Log₁₀ (P-value)] is indicated by the bar. The number of DEGs for a given category is indicated by an “X.” (H) Top panel: The number of DETs, as well as the number of DETs with the indicated NMD-inducing feature. Bottom panel: Violin plot showing the 3′ UTR length of the up- and down-regulated DETs, as well as all transcripts. ** p <0.01; **** p <0.0001 (unpaired t-test). (I) Examples of alternatively processed transcripts degraded by UPF3B-dependent NMD in ESCs, as defined by the IsoformSwitchAnalyzeR program. The bar graph shows the expression of the main isoform versus an alternative isoform harboring an NMD-inducing feature that is upregulated by the UPF3B loss (other isoforms are not shown). The value shown for each isoform is the proportion of the total. Bonferroni-adjusted t-test * q<0.05; ns, not statistically significant. (J) Biological functions defined by the Metascape program (p <0.01) associated with the full profile of high-confidence UPF3B-dependent NMD-target transcripts in ESCs.

Figure 4.

UPF2 and UPF3B predominantly target different transcripts in human ESCs. (A) Overlap of high-confidence UPF2- and UPF3B-dependent NMD-target RNAs (regulated by >2 fold, as defined in Figs 2 and 3, respectively). (B) Overlap of high-confidence UPF2- and UPF3B-dependent NMD-target RNAs exhibiting >1.5 fold regulation, following the same approaches as in Figs 2 and 3. (C) Integrative genomics viewer images showing tracks of normalized pUPF1 occupancy tag counts in ESCs analyzed by RIPseq. pUPF1-1, -2, and -3 are replicates from IP with a pUPF1 rabbit antisera. IgG-1 shows the signal when the IP was performed with normal rabbit antisera. (D) Sequence motifs enriched at pUPF1-occupied 3′ UTR sites, as determined using HOMER. (E) Overlap of UPF2- and UPF3B-dependent NMD-target RNAs regulated by >2-fold [defined in panel (A)] that exhibit pUPF1 occupancy in the 3′ UTR. (F) Overlap of UPF2- and UPF3B-dependent NMD-target RNAs regulated by >1.5-fold [defined in panel (B)] that exhibit pUPF1 occupancy in their 3′ UTR. (G) qPCR analysis of UPF2- and UPF3B-dependent NMD-target mRNAs in ESCs (defined by RNA-seq analysis) with the indicated genotypes and incubated with or without a UPF2 siRNA (UPF2-KD) (as in Fig. 1, above). Three non-NMD-target mRNAs (defined by RNA-seq analysis) were tested in parallel. Ctrl cells were also treated incubated with the small-molecule NMD inhibitor NMDI [113]. The ESC cells used and how they were cultured to deplete UPF2 are described in Fig. 1. Expression in nontreated Ctrl cells was set to “1.” The values shown are from two independently derived ESC clones, with two independent biological replicates. Statistical significance was determined using the Student’s t-test. Different letters (a, b, c, and d) denote statistically significant differences between different groups (p <0.05). (H) qPCR analysis of UPF2-dependent UPF3B-independent NMD-target mRNAs (as defined by RNA-seq analysis in ESCs). The expression of these NMD targets was evaluated in Ctrl or UPF3B-KO cells. UPF3A-KD refers to cells transfected with a UPF3AsiRNA and cultured for 48 h before harvest. Both Ctrl and UPF3B-KO Ctrl cells were transfected with a scramble siRNA. The values shown are from two independently derived ESC clones, with two independent biological replicates. Statistical significance was determined using the Student’s t-test. Different letters (a, b, and c) denote statistically significant differences between different groups (p <0.05).

Figure 5.

Identification of high-confidence NMD-target transcripts in human NPCs. (A–C) Top panel: The number of DETs, as well as the number of DETs with the indicated NMD-inducing feature, for the UPF2 disruption condition shown (determined from the datasets described in Supplementary Fig. S3). Bottom panel: Violin plot showing the 3′ UTR length of the up- and downregulated DETs, as well as all transcripts. * p <0.05; **** p<0.0001 (unpaired t-test). (D) Overlap amongst high-confidence NMD-target RNAs (those with a dEJ or an extended 3′ UTR) identified from the datasets in panels (A)–(C). (E–G) Examples of alternatively processed transcripts degraded by NMD in NPCs, as defined by the IsoformSwitchAnalyzeR program. The bar graph shows the expression of the main isoform versus an alternative isoform harboring an NMD-inducing feature that is upregulated by the indicated UPF2 disruption (other isoforms are not shown). Bonferroni-adjusted t-test * q<0.05; nd, nondetectable; ns, not statistically significant. (H) Alternatively processed mRNA isoforms upregulated by the indicated UPF2 perturbation condition, as defined by the IsoformSwitchAnalyzeR program. Shown is the total number of upregulated transcripts, as well as upregulated transcripts with the indicted NMD-inducing feature. “Both” refers to upregulated transcripts with both a dEJ and an extended 3′ UTR. (I) Biological functions defined by the Metascape program (p<0.01) associated with the full profile of high-confidence NMD-target mRNAs in ESCs [defined from the datasets in panels (A)–(C) and (H)].

Figure 6.

Most NMD-target mRNAs are cell type-specific. (A) Integrative genomics viewer images showing tracks of normalized pUPF1 occupancy tag counts in human NPCs. Three biological replicates, along with one negative Ctrl (IgG), are shown. (B) Sequence motifs enriched at pUPF1-occupancy sites, as determined using HOMER. (C) Overlap of all unambiguous UPF2-dependent NMD-targeted RNAs in human ESCs and NPCs. (D) Overlap of only unambiguous UPF2-dependent NMD-targeted RNAs expressed in both human ESCs and NPCs (expression cutoff: >5 TPM). (E) Identification of transcripts co-expressed in human ESCs and NPCs as defined by module preservation analysis using the WGCNA program (Zsummary >2 was used as the cutoff). Different colors of the dots represent individual modules with different Zsummary statistic. X-axis, gene number in each module. (F) Overlap of only unambiguous UPF2-dependent NMD-target RNAs co-expressed in both ESCs and NPCs as defined by the module preservation analysis in panel (E). (G) qPCR analysis of human ESCs differentiated into endoderm and mesoderm, respectively, using standard protocols [59]. Shown are well-established markers for the indicated germ layers. Data are represented as mean ± SEM (n = 3). (H) NMD-sensitive/NMD-insensitive mRNA isoform ratio (determined by qPCR analysis) of the indicated genes in endoderm and mesoderm treated with the small-molecule NMD inhibitor NMDI [113] or the vehicle, dimethyl sulfoxide (DMSO), alone (Ctrl). The NMD-sensitive mRNA isoform from each of these five genes is an NMD target in ESCs and NPCs (Supplementary Table S5). The expression of the individual NMD-sensitive and -insensitive transcripts is shown in Supplementary Fig. S5. Statistical significance was determined using the Student’s t-test. n = 3; * p <0.05. (I) HNRNPL expression in human ESCs versus NPCs, as detected by RNA-seq. (J) Western blot analysis of HNRNPL protein levels in human ESCs versus NPCs. The right panel shows mean HNRNPL levels relative to the internal Ctrl (GAPDH; n = 3). (K) qPCR analysis of human ESCs transfected with siRNAs against HNRNPL or UPF2. CCNL1-226, DBF4B-205, and METTL6-203 are NMD-target RNAs in NPCs, not ESCs. These transcripts have 70, 55, and 88 predicted HNRNPL-binding sites, respectively, based on the RBPmap program [126]. GADD45B is a negative-Ctrl NMD-target mRNA [36] with only four predicted HNRNPL-binding sites. Statistical significance was determined using the Student’s t-test (n = 3). Different letters (a, b, and c) denote statistically significant differences between different groups (p < 0.05).

Figure 7.

Identification of high-confidence UPF3B-dependent NMD-target RNAs in human NPCs. (A) Unsupervised hierarchical clustering of two independent clones of Ctrl and UPF3B-KO human NPCs assayed by RNA-seq analysis (three experimental replicates for each group). (B) DEGs (q <0.01, fold change >2) identified from the RNA-seq analysis in panel (A). (C) Biological functions associated with the DEGs defined in panel (B). Statistical significance [−Log₁₀ (p-value)] is indicated by the bar. The number of DEGs for a given category is indicated by an “X.” (D) Top panel: The number of DETs, as well as the number of DETs with the indicated NMD-inducing feature. Bottom panel: Violin plot showing the 3′ UTR length of the up- and down-regulated DETs, as well as all transcripts. **** p <0.0001 (unpaired t-test). (E) Examples of alternatively processed transcripts degraded by UPF3B-dependent NMD in NPCs, as defined by the IsoformSwitchAnalyzeR program. The bar graph shows the expression of the main isoform versus an alternative isoform harboring an NMD-inducing feature that is upregulated by UPF3B loss (other isoforms are not shown). Bonferroni-adjusted t-test * q<0.05; ns, not statistically significant. (F) Biological functions defined by the Metascape program (p<0.01) associated with the full profile of high-confidence UPF3B-dependent NMD-target transcripts in NPCs.

Figure 8.

NMD factor-, cell type-, and tissue-specific NMD. (A) Overlap of all high-confidence UPF2- and UPF3B-dependent NMD-target RNAs in NPCs (defined in Figs 5 and 7, respectively). (B) Overlap of all unambiguous UPF2- and UPF3B-dependent NMD-target RNAs [those in panel (A) that are occupied by pUPF1]. (C) Effect of combined UPF3B loss and UPF2 KD on NMD magnitude in NPCs, as quantified by NMD-sensitive/NMD-insensitive mRNA isoform ratio (see Fig. 6H). Expression of the mRNA isoforms was determined by qPCR analysis of NPCs with the indicated genotypes (Ctrl or UPF3B-KO) transiently transfected with a UPF2 siRNA or a negative control scramble siRNA (Ctrl). The expression of the individual NMD-sensitive and -insensitive transcripts is shown in Supplementary Fig. S6. The values shown are from two independently derived UPF3B-KO clones, with two independent biological replicates. Statistical significance was determined using the Student’s t-test. Different letters (a, b, c, and d) denote statistically significant differences between different groups (p <0.05). (D) Overlap of all unambiguous UPF3B-dependent NMD-target RNAs expressed in human ESCs and NPCs. (E) Overlap of only unambiguous UPF3B-dependent NMD-target RNAs co-expressed in both ESCs and NPCs (cutoff: >5 TPM). (F) Overlap of only unambiguous UPF3B-dependent NMD-target RNAs co-expressed in both ESCs and NPCs as defined by module preservation analysis. (G) Unsupervised hierarchical clustering of frontal cortex and cerebellum from four adult Upf3b-null and four littermate control (Ctrl) mice analyzed by RNA-seq analysis. (H and I) Top panel: The number of DETs, as well as the number of DETs with the indicated NMD-inducing feature. Bottom panel: Violin plot showing the 3′ UTR length of the up- and down-regulated DETs, as well as all transcripts. ** p <0.01; **** p <0.0001 (unpaired t-test). (J) Overlap of all high-confidence UPF3B-dependent NMD-target RNAs (those with a dEJ and/or an extended 3′ UTR) that are expressed in mouse frontal cortex or cerebellum [defined from the datasets shown in panels (H) and (I), respectively]. (K) Overlap of only high-confidence UPF3B-dependent NMD-target RNAs co-expressed in both mouse frontal cortex and cerebellum (cutoff: >5 TPM). (L) Overlap of only high-confidence UPF3B-dependent NMD-target RNAs co-expressed in both mouse frontal cortex and cerebellum as defined by module preservation analysis. (M–O) Overlap analyses performed the same as in panels (J)–(L) except that the fold change cutoff used to define UPF3B-dependent NMD-target RNAs in mouse frontal cortex or cerebellum was >1.5 rather than >2.