UPF1_LL contributes to NMD under normal cellular conditions

We expected that only a subset of genes induced by siUPF1_total would be affected by UPF1_LL, a prediction borne out by the 1,342 genes (72%) that were uniquely up‐regulated in the siUPF1_total condition. However, we also found that 1,028 genes were up‐regulated more than 1.4‐fold upon siUPF1_LL but not siUPF1_total treatment (Fig1B). Because the existence of a large class of genes regulated by siUPF1_LL but not siUPF1_total was unexpected, we pursued three strategies to evaluate whether these genes represent genuine siUPF1_LL targets: (i) investigation of NMD autoregulation in siUPF1_LL and siUPF1_total conditions, (ii) comparison to additional published NMD RNA‐seq datasets, and (iii) analysis of mRNA abundance from cells depleted of UPF1_total, UPF1_LL, or SMG6 by RT–qPCR.

The NMD pathway is governed by a conserved autoregulatory program in which depletion or inactivation of NMD pathway components drives elevated expression of several core NMD factors. NMD feedback regulation has been shown to heavily depend on long 3’UTR‐dependent turnover of NMD factor mRNAs (Singh et al, 2008; Huang et al, 2011; Yepiskoposyan et al, 2011). In siUPF1_total RNA‐seq, we observed pathway‐wide induction of NMD factor mRNAs, with expression of mRNAs encoding SMG1, SMG6, SMG7, SMG9, UPF2, UPF3A, and UPF3B all increased at least 1.4‐fold (Fig1C). In contrast, UPF1_LL depletion had a minimal impact on NMD factor mRNA expression, failing to perturb any by more than 1.2‐fold. In an independent experiment, we knocked down UPF1_total or UPF1_LL and evaluated NMD factor mRNAs by RT–qPCR, again observing up‐regulation of NMD factor mRNAs upon siUPF1_total but not siUPF1_LL treatment (FigEV1C and Dataset EV3). The finding that UPF1_total but not UPF1_LL depletion induced compensatory up‐regulation of NMD components provides a mechanism to explain why some mRNAs might be de‐repressed by UPF1_LL knockdown (which does not induce compensatory feedback regulation of NMD) but not by UPF1_total knockdown (which induces up‐regulation of several core NMD factors).

To further evaluate the contribution of UPF1_LL to cellular NMD, we compared our UPF1_LL‐knockdown RNA‐seq dataset with a published catalog of high‐confidence NMD targets (Colombo et al, 2017). Consistent with the overlap between siUPF1_LL and siUPF1_total in our RNA‐seq studies, we observed significant overlaps among the population of genes induced by UPF1_LL depletion and those previously determined to be repressed by UPF1, SMG6, or SMG7 (FigEV1D), with 618 of the putative UPF1_LL targets represented in the published NMD target catalog.

Co‐regulation of UPF1_LL target mRNAs by SMG6

To gain insight into the involvement of other NMD pathway components in UPF1_LL‐dependent regulation, we took advantage of a recently published RNA‐seq dataset from experiments in which the Gehring laboratory combined CRISPR‐mediated SMG7 knockouts with RNAi‐mediated SMG5 or SMG6 knockdowns (Boehm et al, 2021). For these analyses, we categorized genes as up‐regulated by siUPF1_total only, siUPF1_LL only, or both siUPF1_total and siUPF1_LL. All three classes exhibited significantly enhanced expression in SMG7^ko/SMG6^kd cells, relative to a parental cell line treated with control siRNAs (Fig1D). The greatest degree of up‐regulation in SMG7^ko/SMG6^kd cells was observed for genes induced by both siUPF1_total and siUPF1_LL. Genes that responded to only siUPF1_LL were up‐regulated in SMG7^ko/SMG6^kd cells to a very similar extent to those that responded to only siUPF1_total.

Interestingly, genes induced by siUPF1_LL but not siUPF1_total exhibited distinct responses to SMG5 depletion in the SMG7^ko background. In contrast to the systematic up‐regulation of the siUPF1_LL‐only class of genes in SMG7^ko/SMG6^kd cells, this group of genes was not on average induced in SMG7^ko/SMG5^kd cells (FigEV1E). Reciprocally, genes that were up‐regulated in SMG7^ko/SMG6^kd but not SMG7^ko/SMG5^kd cells were most substantially induced by UPF1_LL depletion (FigEV1F). Together, these analyses support the idea that many genes uniquely up‐regulated by siUPF1_LL are genuine NMD pathway targets, as they are responsive to co‐inactivation of NMD factors SMG6 and SMG7. Moreover, these data suggest that these genes, as a class, are particularly dependent on SMG6 for proper regulation.

To corroborate the results of our own and published RNA‐seq datasets, we selected representative genes for evaluation by RT–qPCR of mRNA from cells depleted of UPF1_total, UPF1_LL, or SMG6 (Fig1E and Dataset EV3). We analyzed genes from three major categories: (i) well‐characterized NMD targets, including EJC‐stimulated alternative splice isoforms of SRSF2, SRSF3, and SRSF6 and long 3’UTR decay targets SMG1 and SMG5, (ii) putative UPF1_LL targets regulated by both UPF1_LL and UPF1_total depletion in our RNA‐seq studies, and (iii) putative UPF1_LL targets up‐regulated by UPF1_LL depletion but not UPF1_total depletion. Knockdown of UPF1_LL had no effect on the levels of well‐characterized premature termination codon (PTC)‐containing SRSF2 and SRSF6 transcripts, and increased SRSF3 PTC transcript levels to a much smaller extent (~1.9‐fold) than total UPF1 (~6.2‐fold) or SMG6 (~8.1‐fold) knockdown (Lareau et al, 2007; Ni et al, 2007). Transcriptome‐wide, we found a similar pattern, as depletion of UPF1_total but not UPF1_LL caused systematically elevated expression of PTC‐containing transcript isoforms relative to control PTC‐free isoforms (Appendix Fig S1C). Importantly, all selected UPF1_LL target mRNAs, irrespective of siUPF1_total responsiveness, were significantly up‐regulated by SMG6 depletion (Fig1E). These data further reinforce the conclusion that genes responding to UPF1_LL but not UPF1_total were likely up‐regulated due to UPF1_LL depletion rather than off‐target effects.

Targets of UPF1_LL are enriched for ER‐associated gene products

The set of genes that respond to siUPF1_LL but not siUPF1_total is an interesting class because their regulation can be unambiguously attributed to UPF1_LL and they can be studied in the absence of the overall NMD pathway up‐regulation that results from knockdown of total UPF1 and other NMD factors. However, it is important to note that we do not currently know the mechanisms that determine whether mRNAs regulated by UPF1_LL are responsive to siUPF1_LL alone or both siUPF1_LL and siUPF1_total, as transcripts uniquely affected by siUPF1_LL did not show any significant enrichment for specific NMD‐inducing features like PTCs, uORFs, or long 3’UTRs (Appendix TableS1). Therefore, except where noted, further analyses treat the entire population of siUPF1_LL‐responsive genes as putative UPF1_LL targets, irrespective of the effects of UPF1_total knockdown.

Because genes in functionally related pathways are often coordinately regulated at the posttranscriptional level (Keene, 2007), we performed a gene ontology (GO) enrichment analysis (Eden et al, 2009) to identify commonalities among UPF1_LL targets. This analysis revealed a high degree of enrichment among UPF1_LL targets for genes encoding proteins that rely on the endoplasmic reticulum (ER) for biogenesis (Fig1F and Dataset EV4). In contrast, GO analysis of mRNAs up‐regulated by siUPF1_total treatment yielded no significantly enriched categories. In total, 768 of the 1,621 genes up‐regulated by UPF1_LL depletion are annotated by UniProt as encoding integral membrane, secreted, and/or signal peptide‐containing proteins. We also used a previous survey of ER‐localized translation (Jan et al, 2014) to corroborate the results of the GO analysis, finding that many UPF1_LL target mRNAs were indeed found to be preferentially translated at the ER (FigEV1G).

Affinity purification reveals transcriptome‐wide UPF1_SL and UPF1_LL binding profiles

The observation that specific depletion of UPF1_LL affected a select subpopulation of NMD targets indicated it has distinct cellular functions from those of the major UPF1_SL isoform. To gain insight into how the biochemical properties of the two UPF1 isoforms differ, we performed affinity purification followed by RNA‐seq (RIP‐seq) of each UPF1 variant (Fig2A). For these studies, we engineered HEK‐293 stable cell lines to inducibly express CLIP‐tagged UPF1_LL or UPF1_SL, with a GFP‐expressing stable line as a control, a system that leads to 5‐ to 6‐fold overexpression relative to endogenous UPF1_total levels (FigEV2A). We elected to express CLIP‐tagged UPF1 constructs, as the CLIP tag can be covalently biotinylated for efficient isolation by streptavidin affinity purification (Gautier et al, 2008). We have previously used this system to show that biotinylated CLIP‐UPF1_SL isolated from human cells preferentially associates with NMD‐susceptible mRNA isoforms (Kishor et al, 2020). Analysis of well‐characterized NMD substrate levels following knockdown of total endogenous UPF1 and rescue with the siRNA‐resistant CLIP‐UPF1 constructs confirmed that both CLIP‐tagged UPF1 isoforms were equally able to function in NMD (FigEV2B).

Open in a separate window

Figure 2

UPF1_LL is enriched on NMD‐protected transcripts

1. Scheme for the CLIP‐UPF1 affinity purification (RIP) assay.

2. Scatterplot of CLIP‐UPF1_LL vs. CLIP‐UPF1_SL RIP‐seq efficiency. mRNAs were binned according to 3’UTR length (short, medium, or long).

3. Density plot of recovered mRNAs in CLIP‐UPF1_LL or CLIP‐UPF1_SL affinity purifications. mRNAs were binned according to 3’UTR length. Statistical significance was determined by K–S test.

4. Density plot of recovered mRNAs in CLIP‐UPF1_LL affinity purifications relative to that of CLIP‐UPF1_SL. mRNAs were subdivided by PTBP1 and/or hnRNP L motif density within the 3’UTR, as indicated by the gradient triangle. Statistical significance was determined by K–W test, with Dunn’s correction for multiple comparisons.

5. RT–qPCR analysis of indicated transcripts from CLIP‐UPF1 overexpression RNA‐seq experiments. Relative fold changes are in reference to the GFP‐expressing control line. Significance of CLIP‐UPF1_SL or CLIP‐UPF1_LL overexpression was compared to the GFP‐expressing control line. Asterisk (*) indicates P<0.05, as determined by two‐way ANOVA. Black dots represent individual data points and error bars indicate mean±SD (n=3 biological replicates). Dashed lines indicate log₂ (fold change) of ±0.5. For protected mRNAs, the motif density of PTBP1/hnRNP L within the 3'UTR is indicated. PTC⁺ indicates the use of primers specific to transcript isoforms with validated poison exons (Lareau et al, 2007; Ni et al, 2007). See also Dataset EV3 for P values associated with each statistical comparison.

Source data are available online for this figure.

Open in a separate window

Figure EV2

NMD protection can be overcome by UPF1_LL

1. Western blots of CLIP‐UPF1_SL and CLIP‐UPF1_LL overexpression. Membranes were probed with an anti‐UPF1 antibody that detects both endogenous and CLIP‐tagged UPF1. Wedge indicates serial twofold dilutions of lysate. Mean (±SD) of CLIP‐UPF1 overexpression was determined from two replicate membranes.

2. RT–qPCR analysis of well‐characterized NMD targets following total UPF1 knockdown and rescue with siRNA‐resistant CLIP‐tagged UPF1. Relative fold changes are in reference to the GFP‐expressing control line treated with a NT siRNA. Significance of NMD rescue by CLIP‐UPF1 was compared to the GFP‐expressing control line treated with total UPF1 siRNA. Asterisk (*) indicates P<0.0001, as determined by two‐way ANOVA with multiple comparisons. Black dots represent individual data points and error bars indicate mean±SD (n=3 biological replicates). PTC+ indicates the use of primers specific to transcript isoforms with validated poison exons (Lareau et al, 2007; Ni et al, 2007). See also Dataset EV3 for P‐values associated with each statistical comparison.

3. Density plot of recovered mRNAs in CLIP‐UPF1_LL affinity purifications relative to that of CLIP‐UPF1_SL. Genes were categorized as up‐regulated by siUPF1_total only, siUPF1_LL only, or both siUPF1_total and siUPF1_LL. Statistical significance was determined by K–W test, with Dunn’s correction for multiple comparisons.

4. RT–qPCR analysis of indicated transcripts from UPF1 RIP‐seq experiments. Relative fold enrichment was determined by dividing the recovered mRNA by its corresponding input amount. Significance of differential recovery in CLIP‐UPF1_LL RIP was determined by comparison to that in CLIP‐UPF1_SL. Asterisk (*) indicates P<0.05, as determined by unpaired Student’s t‐test. Black dots represent individual data points, and error bars indicate mean±SD (n=3 biological replicates). For protected mRNAs, the PTBP1/hnRNP L motif density bin of the 3'UTR is indicated. PTC⁺ indicates the use of primers specific to transcript isoforms with validated poison exons (Lareau et al, 2007; Ni et al, 2007). See also Dataset EV3 for P‐values associated with each statistical comparison.

5. Density plots of changes in relative mRNA abundance as determined by RNA‐seq following UPF1_LL (top) or UPF1_SL (bottom) overexpression. mRNAs were binned according to enrichment in the CLIP‐UPF1_LL or CLIP‐UPF1_SL affinity purifications. Statistical significance was determined by K–W test, with Dunn’s correction for multiple comparisons.

6. Density plots of changes in mRNA stability as determined by REMBRANDTS analysis of RNA‐seq following UPF1_LL (top) or UPF1_SL (bottom) overexpression. mRNAs were binned according to enrichment in the CLIP‐UPF1_LL or CLIP‐UPF1_SL affinity purifications. Statistical significance was determined by K–W test, with Dunn’s correction for multiple comparisons.

Source data are available online for this figure.

CLIP‐UPF1 complexes were isolated from whole cell extracts by streptavidin affinity purification using a CLIP‐biotin substrate, with GFP‐expressing cell lines as a negative control for interaction specificity (Appendix Fig S2A). Bound RNAs were then extracted and used for sequencing library preparation. Because recovery of RNA from GFP samples was at least 100‐fold lower than from CLIP‐UPF1 affinity purifications (Dataset EV5), only UPF1 samples were analyzed by RNA‐seq. UPF1 occupancy was assessed by normalizing the abundance of transcripts in RIP‐seq samples to their abundance in total RNA‐seq (hereafter referred to as UPF1 RIP‐seq efficiency).

The UPF1_SL and UPF1_LL isoforms differ only in the domain 1B regulatory loop and therefore share a common RNA binding interface composed of residues from the RecA, 1B, 1C, and stalk domains (Chakrabarti et al, 2011). The majority of UPF1‐RNA contacts involve sugar‐phosphate recognition, enabling high‐affinity, sequence‐nonspecific RNA binding. Consistent with these structural features, CLIP‐UPF1_SL and CLIP‐UPF1_LL exhibited equivalent binding to the vast majority of endogenous mRNAs (mRNAs from 9711 of 10,673 genes were recovered within ±1.4‐fold in the two conditions; Fig2B and Dataset EV5).

UPF1_LL preferentially associates with long 3’UTRs

UPF1 accumulates in a length‐dependent manner on 3’UTRs due to its active displacement from coding regions by translating ribosomes and its nonspecific RNA binding activity (Hogg & Goff, 2010; Hurt et al, 2013; Zünd et al, 2013; Kurosaki et al, 2014; Baker & Hogg, 2017). To evaluate the relationship between 3’UTR length and CLIP‐UPF1 binding, we determined the distribution of RIP‐sequencing efficiencies in three 3’UTR length bins (first tertile: <566nt; second tertile: 566–1,686nt; third tertile: >1,686nt). Consistent with previous findings, the efficiency of mRNA co‐purification with both UPF1_SL and UPF1_LL increased with 3’UTR length (Fig2C).

A potential caveat to the RIP‐seq studies is that the CLIP‐tagged UPF1 proteins are ~5 to 6‐fold overexpressed relative to endogenous UPF1_total (FigEV2A), which may impair the assay’s discriminative power between the two isoforms. In line with this idea, binding of CLIP‐UPF1_LL and CLIP‐UPF1_SL to mRNAs from genes induced upon siUPF1_LL and siUPF1_total treatment was equivalent in these assays (FigEV2C). However, one class of transcript, those with long 3’UTRs, was more efficiently co‐purified with CLIP‐UPF1_LL than CLIP‐UPF1_SL (Fig2B and C). We therefore asked whether this preferential enrichment may give clues to distinct biochemical properties of the two isoforms.

Enhanced UPF1_LL binding to NMD‐resistant transcripts

Transcripts with long 3’UTRs represent a large population of potential NMD targets (Yepiskoposyan et al, 2011; Hurt et al, 2013), only some of which are degraded by the pathway under normal conditions (Toma et al, 2015). Providing a biochemical mechanism to explain evasion of long 3’UTRs from decay, we have identified hundreds to thousands of mRNAs shielded by the protective RNA‐binding proteins (RBPs) PTBP1 and hnRNP L (Ge et al, 2016; Kishor et al, 2019b, 2020). In our previous work, we showed that increased PTBP1 and/or hnRNP L motif binding density within the 3’UTR correlates with reduced UPF1_SL binding and recovery of mRNAs in UPF1_SL RIP‐seq studies (Ge et al, 2016; Kishor et al, 2019b; Fritz etal, 2020). Based on the observation that UPF1_LL more efficiently recovers the longest class of 3’UTRs, we asked whether mRNAs protected by PTBP1 and/or hnRNP L are differentially associated with UPF1_LL versus UPF1_SL.

Subdivision of the transcriptome first by 3’UTR length and then according to the density of PTBP1 and hnRNP L binding sites within the 3’UTR revealed that transcripts with long 3’UTRs and moderate or high densities of protective protein binding sites were more efficiently recovered by CLIP‐UPF1_LL than CLIP‐UPF1_SL (Fig2D and Appendix Fig S2B). This preferential recovery of long 3’UTRs with moderate or high densities of protective protein binding by CLIP‐UPF1_LL was similarly observed when PTBP1 and/or hnRNP L motif densities were restricted to the first 400nt of the 3’UTR (Appendix Fig S2C), which we previously established as a strong feature driving protection and reduced UPF1_SL binding (Ge et al, 2016; Kishor et al, 2019b). Quantitative RT–PCR of select transcripts confirmed these transcriptome‐wide RIP‐seq results (FigEV2D and Dataset EV3).

If UPF1_LL can more efficiently associate with mRNAs normally shielded from NMD by the protective RBPs, then we would expect a correlation between the transcripts affected by protective protein depletion and those enriched for UPF1_LL binding. Indeed, mRNAs preferentially recovered by CLIP‐UPF1_LL were significantly down‐regulated in response to PTBP1 depletion in HEK‐293 cells from our previous work (Ge et al, 2016; Data ref: Ge et al, 2016), a result expected for NMD substrates normally shielded by the protective RBP (Appendix Fig S2D). We observed a similar downregulation of mRNAs enriched for CLIP‐UPF1_LL binding in a publicly available RNA‐seq dataset of mouse neuronal progenitor cells depleted of PTBP1 and its brain‐specific paralogue PTBP2 (Appendix Fig S2E; Linares et al, 2015; Data ref: Linares et al, 2015). Together, our findings indicate that the distinct biochemical properties of UPF1_LL give it the capacity to circumvent PTBP1 and/or hnRNP L to associate with otherwise protected mRNAs.

UPF1_LL overexpression down‐regulates mRNAs normally protected from NMD

We further analyzed the RNA‐seq data from RIP‐seq input samples to ask whether differential transcript recognition by the UPF1 isoforms was reflected in differential regulation upon CLIP‐UPF1_SL or CLIP‐UPF1_LL overexpression. mRNAs preferentially bound by CLIP‐UPF1_LL were systematically down‐regulated upon CLIP‐UPF1_LL overexpression, but not CLIP‐UPF1_SL overexpression (FigEV2E, top). Conversely, a small population of mRNAs preferentially recovered by CLIP‐UPF1_SL were down‐regulated by CLIP‐UPF1_SL but not CLIP‐UPF1_LL overexpression (FigEV2E, bottom). To investigate whether the observed changes in mRNA abundance with UPF1 overexpression were due to enhanced decay, we again used REMBRANDTS software (Alkallas et al, 2017). These analyses found that regulation of gene expression upon UPF1 overexpression was attributable to decreased mRNA stability (FigEV2F and Dataset EV5).

We additionally corroborated the transcriptome‐wide results obtained using RNA‐seq by performing RT–qPCR on select transcripts (Fig2E and Dataset EV3). Notably, validated mRNAs down‐regulated by UPF1_LL overexpression include CSRP1, which we have previously established as a long 3’UTR‐containing mRNA that undergoes decay upon hnRNP L knockdown or mutation of hnRNP L binding sites in its 3’UTR (Kishor et al, 2019b). Reduced exogenous expression of CLIP‐UPF1_LL to levels ~0.7‐fold that of total endogenous UPF1 (Appendix Fig S3A) had only small effects on levels of protected mRNAs (Appendix Fig S3B), indicating that removal of protection requires a more substantial perturbation of UPF1_LL expression. Together, these data support the conclusion that the UPF1_LL isoform is biochemically equipped to overcome the protective proteins to promote decay of mRNAs normally shielded from NMD, but that in cells with normal endogenous UPF1_SL levels, protection is maintained unless UPF1_LL is substantially overexpressed.

SRSF1 is required for expression of the UPF1_LL splice isoform

Knockdown and overexpression of UPF1_LL involve drastic changes in UPF1_LL abundance, both in absolute terms and relative to UPF1_SL. We reasoned that manipulation of a regulator of UPF1 alternative splicing might allow us to manipulate the UPF1_LL:UPF1_SL ratio without changing the total cellular UPF1 expression level. We surveyed publicly available alternative splicing analysis from ENCODE RBP knockdown RNA‐seq data, which reported reduced UPF1_LL splice isoform selection upon knockdown of the serine/arginine‐rich splicing factor 1 (SRSF1) in K562 and HepG2 cells (Yee et al, 2019; Van Nostrand et al, 2020). Consistent with ENCODE data, we observed substantial and specific loss of the UPF1_LL mRNA isoform in semi‐quantitative RT–PCR assays from HEK‐293 cells treated with SRSF1 siRNAs (Fig3A). To test the functional effects of SRSF1‐mediated UPF1 splicing regulation, we depleted SRSF1 from cells overexpressing CLIP‐UPF1_SL or CLIP‐UPF1_LL. We chose to focus on CSRP1 mRNAs for these experiments because we have extensively validated the role of hnRNP L in antagonizing NMD of these transcripts (Kishor et al, 2019b) and, in this study, have shown that they are down‐regulated by CLIP‐UPF1_LL overexpression (Fig2E). Knockdown of SRSF1 caused increased CSRP1 mRNA expression, an effect that was reversed by overexpression of CLIP‐UPF1_LL but not CLIP‐UPF1_SL (Fig3B and Dataset EV3).

Open in a separate window

Figure 3

Splicing regulator SRSF1 is required for UPF1_LL expression

1. Semiquantitative RT–PCR of UPF1_SL or UPF1_LL transcript levels following transfection of HEK‐293 cells with the indicated siRNAs.

2. RT–qPCR analysis of indicated transcripts following transfection of a NT siRNA or SRSF1‐specific siRNA under conditions of CLIP‐UPF1_SL or CLIP‐UPF1_LL overexpression. Relative fold changes are in reference to the GFP‐expressing control line treated with a NT siRNA. Asterisk (*) indicates P<0.05, as determined by unpaired Student’s t‐test. Black dots represent individual data points and error bars indicate mean±SD (n=3 biological replicates). See also Dataset EV3 for Pvalues associated with each statistical comparison.

3. Density plot of changes in relative mRNA abundance as determined by RNA‐seq following SRSF1 overexpression (Data ref: Caputi et al, 2019). Genes were categorized as up‐regulated by siUPF1_total only, siUPF1_LL only, or both siUPF1_total and siUPF1_LL. Statistical significance was determined by K–W test, with Dunn’s correction for multiple comparisons.

4. Density plot as in (C), with genes binned according to enrichment in the CLIP‐UPF1_LL or CLIP‐UPF1_SL affinity purifications.

Source data are available online for this figure.

We next asked whether SRSF1 overexpression would enhance use of the UPF1_LL isoform and, if so, whether an elevated UPF1_LL:UPF1_SL ratio would affect transcripts we have identified as responsive to UPF1_LL knockdown or overexpression. We analyzed a public RNA‐seq dataset from cells in which SRSF1 was overexpressed (Data ref: Caputi et al, 2019), finding an ~1.6‐fold increase in usage of the UPF1_LL mRNA isoform in SRSF1 overexpression relative to vector control cells (SRSF1 overexpression Ѱ=27.8%; vector control Ѱ=17.7%; Appendix Fig S4). This increase in UPF1_LL:UPF1_SL ratio was associated with decreased expression of mRNAs identified as up‐regulated in our siUPF1_LL RNA‐seq dataset, while the expression of mRNAs up‐regulated by only siUPF1_total was decreased (Fig3C). Correspondingly, mRNAs preferentially bound by CLIP‐UPF1_LL versus CLIP‐UPF1_SL were systematically down‐regulated by SRSF1 overexpression (Fig3D). Together, these data establish SRSF1 as a regulator of UPF1 alternative splicing. Moreover, they indicate that relatively subtle changes in UPF1_LL:UPF1_SL isoform ratio are sufficient to significantly favor or impair UPF1_LL activities in cells.

UPF1_LL is less sensitive to PTBP1‐mediated inhibition of translocation

We have proposed that the protective RBPs PTBP1 and hnRNP L exploit the tendency of UPF1 to release RNA upon ATP binding and hydrolysis to promote UPF1 dissociation from potential NMD substrates prior to decay induction (Fritz et al, 2020). In support of this model, deletion of the regulatory loop, which mediates ATPase‐dependent dissociation, rendered UPF1_SL less sensitive to PTBP1 inhibition in vitro (Fritz et al, 2020). Importantly, both the physiological UPF1_LL isoform and the engineered UPF1 variant containing a regulatory loop deletion exhibit a greater affinity for RNA in the presence of ATP than the PTBP1‐sensitive UPF1_SL isoform (Gowravaram et al, 2018). We therefore hypothesized that UPF1_LL can mimic the ability of the loop truncation mutant to overcome negative regulation by PTBP1.

We recently established a real‐time assay to monitor UPF1 translocation activity (Fritz et al, 2020). In this assay, UPF1 translocation and duplex unwinding causes a fluorescently labeled oligonucleotide to be displaced from the assay substrate (Fig4A, left). An excess of complementary oligonucleotide labeled with a dark quencher is provided in the reaction, causing a decrease in fluorescence with increased displacement of the labeled oligonucleotide by UPF1. Inhibition of UPF1 translocation results in sustained fluorescence over time, allowing for the determination of inhibitory effects of PTBP1 on UPF1 unwinding activity. Using this assay in our previous work, we obtained evidence that PTBP1 inhibits UPF1 translocation rather than initial binding (Fritz et al, 2020). This inhibitory effect on UPF1 translocation activity was specific to PTBP1 and was not observed in the presence of the high‐affinity RNA binding Pseudomonas phage 7 coat protein, supporting the conclusions that the protective proteins specifically promote the dissociation of UPF1 and that our assay can robustly assess inhibitors of UPF1 unwinding activity.

Open in a separate window

Figure 4

UPF1_LL overcomes translocation inhibition by PTBP1

1. Scheme of the fluorescence‐based unwinding assay to monitor UPF1 translocation in real‐time (Fritz et al, 2020). An RNA substrate harboring a high‐affinity PTBP1 binding site is incubated with highly purified UPF1ΔCH in the absence and presence of equal molar amounts of highly purified PTBP1. Upon the addition of ATP, UPF1 translocation results in a decrease in fluorescence due to displacement of a labeled, duplexed oligonucleotide and subsequent quenching by a trap strand.

2. UPF1_SLΔCH translocation along an RNA substrate harboring a high‐affinity PTBP1 binding site in the absence and presence of PTBP1. Time to 50% unwound and relative total % unwound at end of assay (1,200s) are indicated. Results of four technical replicates are shown for each dataset and represent at least three independent experiments. Shaded region indicates SD.

3. Results as in (B) but with UPF1_LLΔCH.

Source data are available online for this figure.

We therefore leveraged this system to compare UPF1_LL versus UPF1_SL translocation on a duplexed RNA substrate harboring a high‐affinity PTBP1 binding site (Fig4A, right) (Fritz et al, 2020). For these experiments, we compared the activity of highly purified UPF1 proteins containing the helicase core but lacking the autoinhibitory N‐terminal cysteine‐histidine domain (UPF1ΔCH) (Appendix Fig S5A and B; Chakrabarti et al, 2011; Fiorini et al, 2012; Fritz et al, 2020). UPF1_SLΔCH exhibited robust unwinding activity in the absence of PTBP1, displacing 50% of the duplexed oligonucleotide in 100s (Fig4B). This translocation activity of UPF1_SLΔCH was dependent upon the addition of ATP, as previously demonstrated (Fritz et al, 2020). Addition of PTBP1 substantially impaired UPF1_SLΔCH unwinding activity, reducing both the rate at which the oligonucleotide was displaced (requiring 360s to attain the half‐maximal unwinding value reached by UPF1_SLΔCH alone) and the overall extent of unwinding (73% of the UPF1_SLΔCH total at the end of the assay).

UPF1_LLΔCH also exhibited robust unwinding activity in the absence of PTBP1, displacing 50% of the duplexed oligonucleotide in 90s in an ATP‐dependent manner (Fig4C). The observed enhancement in UPF1_LLΔCH translocation activity over UPF1_SLΔCH is consistent with previous reports of increased catalytic activity of the UPF1_LL isoform relative to UPF1_SL (Gowravaram et al, 2018). In contrast to UPF1_SLΔCH, UPF1_LLΔCH maintained robust unwinding activity in the presence of PTBP1, displacing 50% of the duplexed oligonucleotide by 180s and achieving 94% total duplex unwinding at the end of the assay. These results indicate that UPF1_LL can overcome the translocation inhibition by PTBP1, reinforcing the conclusion that PTBP1‐mediated UPF1 inhibition depends on the clash between the UPF1 regulatory loop and RNA.

Coordinated downregulation of UPF1_LL targets during ER stress and ISR induction

Our in vitro, RIP‐seq, and overexpression studies suggested that UPF1_LL has the biochemical capacity to expand the scope of UPF1‐dependent regulation. Based on these observations, we next investigated whether specific physiological conditions might promote changes in NMD target susceptibility by harnessing endogenous UPF1_LL activity. Multiple lines of evidence led us to examine the regulation of UPF1_LL in the integrated stress response (ISR), which restores homeostasis by repressing translation and inducing expression of a battery of stress response genes (Fig5A; Costa‐Mattioli & Walter, 2020). Activation of the ISR induces hyperphosphorylation of eIF2, driving global downregulation of protein synthesis due to impaired eIF2‐GTP‐Met‐tRNA_i ternary complex recycling and reduced delivery of the initiator Met‐tRNA_i to translating ribosomes (Baird & Wek, 2012; Young & Wek, 2016; Wek, 2018). An established effect of ISR‐mediated translational repression is corresponding stabilization of well‐characterized NMD targets, including several mRNAs encoding factors integral to the activation and resolution of the stress response (Goetz & Wilkinson, 2017).

Open in a separate window

Figure 5

UPF1_LL conditionally remodels NMD target selection during ER stress and induction of the ISR

1. Scheme for activation of the integrated stress response (ISR) and effects on UPF1‐dependent decay.

2. Density plots of changes in relative mRNA abundance as determined by RNA‐seq following treatment of HEK‐293 cells with 1µM thapsigargin for 6h (left) or 9h (right). Genes were categorized as up‐regulated by siUPF1_total only or siUPF1_LL only under basal conditions. Statistical significance was determined by K–W test, with Dunn’s correction for multiple comparisons.

3. RNA‐seq analysis of HEK‐293 cells identifies populations of genes that decreased in abundance with thapsigargin treatment and were rescued by UPF1_LL‐specific knockdown. Indicated are genes that increased in abundance at least 1.4‐fold (FDR<0.05) with UPF1_LL‐specific knockdown under normal conditions.

4. RT–qPCR analysis of indicated transcripts following transfection of HEK‐293 cells with indicated siRNAs and treatment with 1µM thapsigargin for 6h. Relative fold changes are in reference to vehicle‐treated, NT siRNA. Black dots represent individual data points and error bars indicate mean±SD (n=3 biological replicates). Dashed lines indicate log₂ (fold change) of ±0.5. PTC⁺ indicates the use of primers specific to transcript isoforms with validated poison exons (Lareau et al, 2007; Ni et al, 2007). See also Dataset EV3 for P‐values associated with each statistical comparison.

Source data are available online for this figure.

Two of our findings led us to consider the possibility that UPF1_LL activity is regulated by the ISR. First, GO analyses indicated substantial enrichment of ER‐localized mRNAs among those up‐regulated by UPF1_LL knockdown (Fig1F). Second, as an initial test of the hypothesis that certain cellular conditions would promote turnover of mRNAs preferentially bound by CLIP‐UPF1_LL, we analyzed a published RNA‐seq dataset from cells treated with the ISR‐inducing agent tunicamycin (Park et al, 2017; Data ref: Park et al, 2017). This analysis identified systematic downregulation of mRNAs enriched in CLIP‐UPF1_LL RIP‐seq, in contrast to RNAs preferentially bound by CLIP‐UPF1_SL (FigEV3A).

Open in a separate window

Figure EV3

Transcripts targeted by UPF1_LL are coordinately down‐regulated during ER stress and induction of the ISR

1. Volcano plot of relative mRNA abundance as determined from RNA‐seq following treatment of HEK‐293 cells with 1µM tunicamycin for 6h (Data ref: Park et al, 2017). mRNAs were binned by RIP‐seq efficiency in CLIP‐UPF1_LL or CLIP‐UPF1_SL affinity purifications. Statistical significance was determined by K–W test, with Dunn's correction for multiple comparisons. Dashed line indicates the significance threshold P≤0.05 (n=3 biological replicates).

2. Western blot of eIF2α phosphorylation following treatment of HEK‐293 cells with 1µM thapsigargin for 6h.

3. Schematic of the RNA‐seq experimental workflow and conditions for UPF1_LL knockdown and thapsigargin treatment.

4. Density plot of relative mRNA stability as determined by REMBRANDTS analysis of RNA‐seq following treatment of HEK‐293 cells with 1µM thapsigargin for 6h. mRNAs were binned by changes in relative mRNA abundance in thapsigargin. Statistical significance was determined by K–S test.

5. Density plot of relative mRNA stability as determined by REMBRANDTS analysis of RNA‐seq following UPF1_LL knockdown in HEK‐293 cells and treatment with 1µM thapsigargin for 6h (Alkallas et al, 2017). mRNAs were binned by changes in relative mRNA abundance in thapsigargin with UPF1_LL knockdown. Statistical significance was determined by K–W test, with Dunn's correction for multiple comparisons.

6. Quantification of characterized ISR‐target transcript abundance in RNA‐seq of the indicated conditions. Error bars indicate mean±SD (n=3 biological replicates).

7. Quantification of UPF1_LL isoform expression in control and thapsigargin‐treated HEK‐293 cells from rMATS analyses (n=3 biological replicates).

Source data are available online for this figure.

We next directly assessed how UPF1_LL activity contributes to gene expression regulation during ER stress and induction of the ISR by performing RNA‐seq of HEK‐293 cells treated with the ER stress‐inducing agent thapsigargin (Fig5B and Dataset EV6). Western blot analysis showed a 2.5‐fold increase in eIF2 phosphorylation with thapsigargin treatment (FigEV3B), supporting a robust induction of the ISR. Consistent with previous results (Nickless et al, 2014; Li etal, 2017), mRNAs up‐regulated upon total UPF1 knockdown in HEK‐293 cells were on average also up‐regulated following 6 or 9h in 1μM thapsigargin (Fig5B), and the magnitude of the increase correlated with the effects of UPF1 total knockdown. In sharp contrast, genes up‐regulated by UPF1_LL‐specific knockdown exhibited a distinct behavior upon thapsigargin treatment. Rather than increasing, UPF1_LL substrates showed on average a reduction in mRNA levels, and this tendency did not vary according to the magnitude of the effect of UPF1_LL knockdown. These results indicate that UPF1_LL functions distinctly from that of well‐characterized NMD and sustains activity during ER stress and activation of the ISR.

UPF1_LL conditionally remodels NMD target selection during ER stress and ISR induction

To more comprehensively evaluate the role of UPF1_LL in promoting the downregulation of select genes during ISR induction, we transfected HEK‐293 cells with non‐targeting (NT) or UPF1_LL‐specific siRNAs and then treated cells with 1μM thapsigargin for 6 or 9h (FigEV3C). In RNA‐seq analyses, we identified 606 genes that significantly decreased in abundance with thapsigargin treatment, of which 135 (6h) or 143 (9h) were rescued at least 1.4‐fold upon UPF1_LL knockdown (Fig5C, Appendix Fig S6A, and Dataset EV6). In contrast, only 70 (6h) or 62 (9h) of these 606 genes decreased in abundance in response to UPF1_LL knockdown in thapsigargin treatment. These results were highly reproducible between the 6h and 9h thapsigargin RNA‐seq datasets (Appendix Fig S6B), supporting that a unique population of genes are selectively down‐regulated during conditions of ER stress and induction of the ISR in a UPF1_LL‐dependent manner.

Inferred mRNA stability changes using REMBRANDTS software supported that the observed differences in mRNA abundance upon thapsigargin treatment and UPF1_LL knockdown were due to changes in mRNA decay (FigEV3D and E). The changes in gene expression caused by UPF1_LL depletion were not attributable to differential ISR induction, as previously established stress response genes were comparably up‐regulated in response to thapsigargin treatment following NT and UPF1_LL‐specific knockdown (FigEV3F; Ashburner et al, 2000; The Gene Ontology Consortium, 2019). Moreover, thapsigargin treatment did not alter the relative levels of UPF1_LL and UPF1_SL mRNAs (FigEV3G), indicating that UPF1_LL activity in ER stress was likely due to activity of the existing population of protein rather than a consequence of altered UPF1 splicing upon thapsigargin treatment.

Our finding that UPF1_LL has the potential to bind and regulate transcripts protected from NMD by PTBP1 and/or hnRNP L under normal cell growth conditions (Figs2 and EV2) led us to ask whether genes down‐regulated by UPF1_LL during ISR induction included substrates beyond those identified as UPF1_LL targets under normal cellular conditions (Fig1). Of the 135 genes down‐regulated by UPF1_LL upon thapsigargin treatment, 49 genes (36%) were unique to the population of UPF1_LL targets down‐regulated during ISR induction, while 86 genes were identified as UPF1_LL targets under both normal and stress conditions (Fig5C). These data indicate that UPF1_LL activity is maintained or enhanced when cells are subjected to ER stress conditions that inhibit well‐characterized NMD events.

To corroborate the above findings, we performed RT–qPCR on select transcripts identified by RNA‐seq as constitutively or conditionally regulated by UPF1_LL (Fig5D and Dataset EV3). As in the RNA‐seq analyses, RT–qPCR of putative condition‐specific UPF1_LL targets revealed several mRNAs (e.g., TMEM165, LDLR, TIMP2, NXT2, TGOLN2, and COL4A1) that exhibited UPF1_LL‐dependent downregulation upon thapsigargin treatment but were not affected by siUPF1_LL under normal growth conditions. Taken together, these data support a model in which expression of dual UPF1_SL and UPF1_LL isoforms enable conditional remodeling of NMD target selection in response to ISR induction.

UPF1_LL target repertoire is expanded by partial translationalrepression

Because NMD requires detection of in‐frame stop codons, target susceptibility is sensitive to changes in the location and frequency of translation initiation and termination. In addition to modulation of initiation via eIF2 phosphorylation in ER stress (Goetz & Wilkinson, 2017), inhibition of translation elongation (e.g., with cycloheximide and puromycin) inhibits the decay of well‐characterized NMD targets (Carter et al, 1995). We therefore asked whether translational repression would promote UPF1_LL activity outside of the context of the ISR (Fig6A).

Open in a separate window

Figure 6

UPF1_LL activity is enhanced by translational repression

1. Framework for investigation of effects of translation inhibition on UPF1_LL activity.

2. RNA‐seq analysis of HEK‐293 cells identifies populations of genes that decreased in abundance with puromycin treatment (50µg/ml for 4h) and were rescued by UPF1_LL‐specific knockdown. Indicated are genes that increased in abundance at least 1.4‐fold (FDR<0.05) with UPF1_LL‐specific knockdown under normal conditions.

3. RT–qPCR analysis of indicated transcripts following transfection of HEK‐293 cells with indicated siRNAs and treatment with 50µg/ml puromycin for 4h. Relative fold changes are in reference to vehicle‐treated, NT siRNA. Black dots represent individual data points, and error bars indicate mean±SD (n=3 biological replicates). Dashed lines indicate log₂ (fold change) of ±0.5. PTC⁺ indicates the use of primers specific to transcript isoforms with validated poison exons (Lareau et al, 2007; Ni et al, 2007). See also Dataset EV3 for P‐values associated with each statistical comparison.

Source data are available online for this figure.

We again compared published RNA‐seq datasets with our UPF1 RIP‐seq data to provide a preliminary test of this hypothesis. As we observed using published RNA‐seq data from cells undergoing ER stress, mRNAs preferentially bound by UPF1_LL were systematically down‐regulated upon treatment with moderate doses of the initiation inhibitor hippuristanol or the elongation inhibitors emetine and cycloheximide (FigEV4A–C) (Data ref: Martinez‐Nunez & Sanford, 2016; Martinez‐Nunez et al, 2017; Kearse et al, 2019; Data ref: Kearse et al, 2019; Waldron et al, 2019; Data ref: Waldron et al, 2019). In the case of cycloheximide treatment, only 15min of treatment was sufficient to observe significant loss of UPF1_LL‐enriched transcripts, consistent with a decay‐driven process.

Open in a separate window

Figure EV4

UPF1_LL targets decrease in abundance with translational inhibition

1. Volcano plot of relative mRNA abundance as determined from RNA‐seq following treatment of MCF7 cells with 150nM hippuristanol for 1h (Data ref: Waldron et al, 2019). mRNAs were binned by RIP‐seq efficiency in CLIP‐UPF1_LL or CLIP‐UPF1_SL affinity purifications. Statistical significance was determined by K–W test, with Dunn's correction for multiple comparisons. Dashed line indicates the significance threshold P≤0.05 (n=3 biological replicates).

2. Volcano plot as in (A), following treatment of HEK‐293 cells with 50ng/ml emetine for 4h (Martinez‐Nunez et al, 2017). Dashed line indicates the significance threshold P≤0.05 (n=3 biological replicates).

3. Volcano plot as in (A), following treatment of HeLa cells with 100µg/ml cycloheximide for 15min (Data ref: Kearse et al, 2019). Dashed line indicates the significance threshold P≤0.05 (n=3 biological replicates).

We next used siUPF1_LL treatment to directly test whether mRNAs down‐regulated upon moderate translation inhibition required UPF1_LL. We elected to use the translation elongation inhibitor puromycin for these experiments because it is widely used to inhibit canonical NMD events and acts through a completely distinct mechanism from the block to initiation caused by eIF2 phosphorylation. Specifically, the ribosome catalyzes the linkage of puromycin to nascent polypeptides, causing chain termination and peptide release (Nathans, 1964).

To investigate whether UPF1_LL is able to exert sustained or even enhanced post‐transcriptional control in response to translation inhibition via distinct mechanisms, we transfected HEK‐293 cells with NT or UPF1_LL‐specific siRNAs and then treated cells with puromycin (50µg/ml for 4h; FigEV5A). Remarkably, RNA‐seq analyses revealed 2,279 genes that significantly decreased in abundance with puromycin treatment, of which 700 (31%) were rescued at least 1.4‐fold upon UPF1_LL knockdown (Fig6B and Dataset EV7). In contrast, only 124 genes (5%) decreased in abundance in response to UPF1_LL knockdown in puromycin treatment (Appendix Fig S7A). The response to UPF1_LL depletion was highly consistent in cells treated with in 25, 50, and 100µg/ml puromycin (Appendix Fig S7B), supporting a role for UPF1_LL in gene expression regulation during conditions of partial translational repression. Genes identified as destabilized in REMBRANDTS analysis of CLIP‐UPF1_LL versus CLIP‐UPF1_SL overexpression tended to be down‐regulated upon puromycin treatment, while the reverse was true for genes preferentially destabilized by CLIP‐UPF1_SL overexpression (FigEV5B). Consistent with a specific role for UPF1_LL, knockdown of UPF1_LL ameliorated the effects of puromycin treatment on genes regulated by CLIP‐UPF1_LL but not CLIP‐UPF1_SL overexpression.

Open in a separate window

Figure EV5

Reduced translation efficiency promotes UPF1_LL activity

1. Schematic of the RNA‐seq experimental workflow and conditions for UPF1_LL knockdown and puromycin treatment.

2. Density plot of relative mRNA abundance as determined by RNA‐seq following treatment of HEK‐293 cells with 50µg/ml puromycin. mRNAs were binned according to destabilization in CLIP‐UPF1_LL or CLIP‐UPF1_SL overexpression experiments, as determined by REMBRANDTS analysis (Alkallas et al, 2017). Statistical significance was determined by K–S test.

3. Quantification of UPF1_LL isoform expression in control and puromycin‐treated HEK‐293 cells from rMATS analyses (n=3 biological replicates) (Shen et al, 2014).

4. RT–qPCR analysis of indicated transcripts following treatment of HEK‐293 cells with indicated concentrations of puromycin for 4h. Relative fold changes are in reference to vehicle‐treated control. Significance of puromycin treatment on relative transcript abundance was compared to the vehicle‐treated control. Asterisk (*) indicates P<0.05, as determined by two‐way ANOVA. Black dots represent individual data points, and error bars indicate mean±SD (n=3 biological replicates). Dashed lines indicate log₂ (fold change) of ±0.5. PTC⁺ indicates the use of primers specific to the transcript isoform with a validated poison exon (Lareau et al, 2007; Ni et al, 2007). See also Dataset EV3 for P‐values associated with each statistical comparison.

5. Density plot of relative mRNA abundance as determined by RNA‐seq following treatment of HEK‐293 cells with 25µg/ml or 100µg/ml puromycin. mRNAs were binned according to sensitivity to 50µg/ml puromycin and UPF1_LL knockdown. Statistical significance was determined by K–S test.

6. Density plot of relative mRNA stability as determined by REMBRANDTS analysis of RNA‐seq following treatment of HEK‐293 cells with 50µg/ml puromycin for 4h (Alkallas et al, 2017). mRNAs were binned by changes in relative mRNA abundance in puromycin. Statistical significance was determined by K–S test.

7. Density plot of relative mRNA stability as determined by REMBRANDTS analysis of RNA‐seq following UPF1_LL knockdown in HEK‐293 cells and treatment with 50µg/ml puromycin for 4h (Alkallas et al, 2017). mRNAs were binned by changes in relative mRNA abundance in puromycin with UPF1_LL knockdown. Statistical significance was determined by K–W test, with Dunn's correction for multiple comparisons.

Source data are available online for this figure.

We next employed puromycin to identify new conditional UPF1_LL targets, finding that 550 genes (79%) identified as rescued by UPF1_LL from puromycin‐dependent downregulation were uniquely affected during puromycin treatment. The remaining 150 genes (21%) significantly overlapped with those up‐regulated with UPF1_LL depletion under normal cellular conditions (Fig6B). RT–qPCR of select transcripts confirmed the transcriptome‐wide RNA‐seq results (Fig6C and Dataset EV3). Similar to conditions of ER stress, puromycin treatment did not alter UPF1 splicing (FigEV5C), indicating that UPF1_LL activity during conditions of impaired translation was likely due to the existing population of UPF1_LL protein. These data support that translational repression promotes UPF1_LL activity outside of the context of the ISR to conditionally remodel NMD target selection.

UPF1_LL activity requires ongoing translation

Because of the well‐established requirement for translation in NMD, we hypothesized that the UPF1 isoform‐dependent effects of thapsigargin and moderate puromycin treatment were due to changes in the location and/or frequency of translation termination events (Fig6A). To test this hypothesis, we treated cells with a titration of puromycin from 25µg/mL to 400µg/ml. If UPF1_LL activity depends on the infrequent residual translation termination events that occur under puromycin treatment, its activity should be enhanced at low concentrations of puromycin that permit some termination events to persist but be inhibited by high concentrations of puromycin that more efficiently block translation.

In line with these expectations, we observed a dose‐dependent response, in which downregulation of representative UPF1_LL target transcripts FMR1, eIF5A2, ERBB2, and CDK16 was most efficient at lower puromycin concentrations (FigEV5D and Dataset EV3). Treatment with high concentrations of puromycin did not have a significant effect on the levels of these UPF1_LL target mRNAs, consistent with a requirement for translation termination events. Corroborating these findings, we observed globally more efficient downregulation of puromycin‐sensitive UPF1_LL targets with 25µg/ml puromycin than 100µg/ml puromycin in RNA‐seq (FigEV5E and Dataset EV7). Based on these results, we conclude that translation termination is likely required for all UPF1‐dependent decay events but that changes in translation efficiency can drive the downregulation of a novel class of substrates by the UPF1_LL isoform.

Translational repression promotes UPF1_LL‐dependent decay of select mRNAs

Inferred mRNA stability changes using REMBRANDTS software indicated that the observed differences in mRNA abundance upon puromycin treatment and UPF1_LL knockdown from the RNA‐seq studies were due to corresponding changes in mRNA stability (FigEV5F and G and Dataset EV7). To directly evaluate the effect of translational repression on promoting the decay of mRNAs by UPF1_LL, we leveraged the recently established method of Roadblock‐qPCR to assess endogenous mRNA stability (Watson et al, 2021). In this method, 4‐thiouridine (4‐SU) is used to label transcripts produced during a 4‐h timecourse. Isolated RNA is treated with N‐ethylmaleimide, which covalently labels 4‐SU residues, forming a bulky adduct that blocks reverse transcription. RT–qPCR of the remaining unlabeled pool thus allows straightforward quantification of mRNA turnover.

HEK‐293 cells were transfected with NT or UPF1_LL‐specific siRNAs and labeled with 4‐SU in the absence and presence of puromycin. In this analysis, puromycin treatment stabilized the canonical NMD target of ATF4 (Fig7A and Dataset EV3), consistent with previous findings that translational repression inhibits decay of well‐characterized NMD targets. In contrast, representative UPF1_LL targets CDK16 and TNFRSF10D exhibited significantly shorter half‐lives with puromycin treatment, an effect that was dependent upon UPF1_LL expression. Together, these data support the conclusion that translational repression promotes the decay of mRNAs by UPF1_LL.

Open in a separate window

Figure 7

Translational repression promotes the decay of mRNAs by UPF1_LL in the NMD pathway and down‐regulates normally protected transcripts

1. mRNA decay measurement using Roadblock‐qPCR (Watson et al, 2021). RNA was isolated from HEK‐293 cells at indicated timepoints following transfection with indicated siRNAs and treatment with 400µM 4sU and 50µg/ml puromycin. mRNA half‐lives were estimated by fitting the data to a single‐phase exponential decay model. Puromycin treatment was compared to the vehicle control and siUPF1_LL was compared to the siNT in the absence and presence of puromycin treatment using the extra‐sum‐of‐squares F test. Error bars indicate mean±SD (n=4 biological replicates). See also Dataset EV3 for P values associated with each statistical comparison.

2. RT–qPCR analysis of indicated transcripts following transfection of HEK‐293 cells with indicated siRNAs and treatment with 50µg/ml puromycin for 4h. Relative fold changes are in reference to vehicle‐treated, NT siRNA. Black dots represent individual data points and error bars indicate mean±SD (n=3 biological replicates). Dashed lines indicate log₂ (fold change) of ±0.5. PTC⁺ indicates the use of primers specific to transcript isoforms with validated poison exons (Lareau et al, 2007; Ni et al, 2007). See also Dataset EV3 for P values associated with each statistical comparison.

3. Density plot of changes in relative mRNA abundance as determined from RNA‐seq following treatment of HEK‐293 cells with 50µg/ml of puromycin for 4h. mRNAs were subdivided by PTBP1 and/or hnRNP L motif density within the first 400nt of 3'UTR. Statistical significance was determined by K–W test, with Dunn's correction for multiple comparisons.

4. Density plot as in (C), following UPF1_LL‐specific knockdown.

Source data are available online for this figure.

UPF1_LL activity during translational repression requires SMG6

We next asked whether UPF1_LL‐dependent mRNA downregulation during translational repression involved the specialized NMD endonuclease SMG6. To explore this possibility, HEK‐293 cells were transfected with NT or SMG6‐specific siRNAs and then treated with puromycin. Quantitative RT–PCR of select transcripts revealed that knockdown of SMG6 significantly rescued the downregulation of UPF1_LL targets in puromycin treatment (Fig7B and Dataset EV3). This effect was comparable to that observed with UPF1_LL‐specific depletion, supporting the conclusion that select mRNA downregulation during translational repression is due to UPF1_LL activity in the NMD pathway. Furthermore, knockdown of SMG6 under normal conditions increased the abundance of well‐characterized NMD targets and constitutively regulated UPF1_LL substrates but did not significantly affect the levels of conditionally regulated UPF1_LL substrates. These results provide further evidence that UPF1_LL remodels NMD target selection during translational repression to promote the decay of a new class of substrates.

In vitro helicase assays

Unwinding assays were performed as previously described in detail (Fritz et al, 2020). Briefly, 75nM of the pre‐assembled RNA duplex substrate (described below) was combined with 1× unwinding reaction buffer (10mM MES pH 6.0, 50mM KOAc, 0.1mM EDTA), 2mM MgOAc, and 1‐unit RNasin in a well of a half‐area, black, flat‐bottom 96‐well plate (Corning 3993). PTBP1 (80nM) was then added, the sample mixed, and incubated at room temperature for 10min. Then, UPF1 (80nM) was added, the sample mixed, and incubated at room temperature for 10min. Finally, BHQ1 quencher (0.56µM final; GTGTGCGTACAACTAGCT/3BHQ_1) and 2mM ATP were added to initiate the unwinding reaction. An Infinite^R F200 Pro microplate reader and associated i‐control^TM 1.9 software (Tecan) was used to monitor Alexa Fluor 488 fluorescence every 10s for 20min at 37°C. Measured fluorescence intensities were normalized to the zero timepoint for each condition to obtain relative fluorescence values. Four technical replicates for each condition were obtained at the same time. The decrease in fluorescence caused by UPF1 unwinding in the absence of PTBP1 at the end of each 20‐min timecourse was used to calculate the time to 50% maximal unwinding and relative total unwinding for each condition. Because measurements were taken at 10‐s intervals, the earliest time at which 50% maximal unwinding was observed is indicated on each graph.

To generate the RNA duplex, a DNA oligo template (5’ TAATACGACTCACTATAGGGACACAAAACAAAAGACAAAAACACAAAACAAAAGACAAAAACACAAAACAAAAGACAAAAAGCCTCTCCTTCTCTCTGCTTCTCTCTCGCTGTGTGCGTACAACTAGCT 3’) was PCR‐amplified and then in vitro transcribed using the MEGAshortscript^TM T7 Transcription kit (Invitrogen). An 11:7 ratio of the helicase RNA substrate to 5’ Alex Fluor 488 fluorescent oligo strand (Alexa Fluor 488/AGCTAGTTGTACGCACAC) was incubated with 2mM MgOAC and 1× unwinding reaction buffer at 95°C for 3min and 30s and then slowly cooled to 30°C. All DNA oligos were obtained from Integrated DNA Technologies. The 5’ fluorescent oligo probe and 3’ BHQ1 quencher were acquired as RNase‐free, HPLC‐purified.

Cloning, expression, and purification of recombinant UPF1_LLΔCH, UPF1_SLΔCH, and PTBP1 were conducted as previously described (Gowravaram et al, 2018; Fritz et al, 2020). In this study, UPF1_LLΔCH was purified in the same manner as UPF1_SLΔCH (Fritz et al, 2020).

Mammalian cell lines and generation of CLIP‐UPF1 expression lines

HEK‐293 cells used in the endogenous UPF1_LL knockdown and RNA‐seq experiments were received from ATCC (CRL‐3216) and maintained at 37°C and 5% CO₂ in DMEM with 10% FBS (Gibco) and 1% pen/strep. Stable integration of CLIP‐UPF1_SL into human Flp‐In™ T‐Rex™‐293 cells (Invitrogen) and subsequent maintenance of this stable line was previously described (Kishor et al, 2020). CLIP‐UPF1_LL expression lines were generated and maintained in an identical manner. Total UPF1 and SMG6 depletion studies followed by RT–qPCR as well as the Roadblock‐qPCR experiments were performed using HEK‐293 cells maintained as described above. Total UPF1 depletion and RNA‐seq was conducted using parental Flp‐In™ T‐Rex™‐293 cells (Invitrogen) maintained according to manufacturer instructions. To express CLIP‐UPF1 at levels ~0.7‐fold that of the endogenous protein, the CMV promoter and 5’ UTR hairpin from pcDNA HP40GFPTP (Hogg & Goff, 2010) were inserted in place of the tet‐regulated pFRT/TO promoter in the CLIP‐UPF1 expression plasmid (Kishor et al, 2020) via SpeI and HindIII sites. Stable cell lines Flp‐In™ T‐Rex™‐293 in which the resulting UPF1 expression plasmids were integrated were prepared and maintained as above.

Endogenous UPF1, SMG6, or SRSF1 depletion by siRNA

The following siRNAs were used to deplete indicated UPF1 isoforms, SMG6, or SRSF1: total UPF1 (Forward sequence: 5’ CUACCAGUACCAGAACAUA 3’; Reverse sequence: 5’ UAUGUUCUGGUACUGGUAG 3’); UPF1_LL (Forward sequence: 5’ GGUAAUGAGGAUUUAGUCA 3’; Reverse sequence: 5’ UGACUAAAUCCUCAUUACC 3’); SMG6 (Forward sequence: 5’ GCUGCAGGUUACUUACAAG 3’; Reverse sequence: 5’ CUUGUAAGUAACCUGCAGC 3’; Durand et al, 2016); SRSF1 (Forward sequence: 5’ UGAAGCAGGUGAUGUAUGU 3’; Reverse sequence: 5’ ACAUACAUCACCUGCUUCA 3’; (de Miguel et al, 2014)).

CLIP‐UPF1 overexpression RIP‐seq and RNA‐seq

CLIP‐UPF1_SL overexpression RIP‐seq and RNA‐seq sample preparation were previously reported (Kishor et al, 2020); CLIP‐UPF1_LL datasets were generated in parallel. Briefly, CLIP‐UPF1 stable cell lines or a GFP‐expressing control line were seeded in 6×15cm plates and then treated with 200ng/ml doxycycline hyclate (Sigma) for 48h to induce CLIP‐UPF1 expression. Cells were harvested 48h post‐induction at 80–85% confluency and whole cell lysate generated by the freeze/thaw method as previously described (Hogg & Collins, 2007a, 2007b; Hogg & Goff, 2010; Fritz et al, 2018). Equilibrated cell extracts, reserving 1/10^th for downstream input analysis, were then combined with 10µM CLIP‐Biotin (New England Biolabs) and rotated (end‐over‐end) for 1h at 4°C to allow CLIP‐UPF1 to react with the CLIP‐Biotin substrate and yield covalently biotinylated protein. Unbound CLIP‐Biotin was subsequently removed by passing the samples through Zeba™ Spin Desalting Columns, 40K MWCO, 2ml (Thermo Scientific) according to manufacturer instructions. Buffer exchange was performed with HLB‐150 supplemented with 1mM DTT and 0.1% NP‐40. Pre‐washed Dynabeads™ MyOne™ Streptavidin T1 (Invitrogen) were then added, and the samples rotated (end‐over‐end) for 1h at 4°C to immobilize the biotin‐bound CLIP‐UPF1 complexes. The samples were then washed three times with 500 µL of HLB‐150 supplemented with 1mM DTT and 0.1% NP‐40 and 1/100^th reserved for downstream Western blot analysis of CLIP‐UPF1 pull‐down efficiency. The remainder was combined with TRIzol™ (Invitrogen) and RNA was isolated according to manufacturer instructions. DNase treatment was subsequently performed using RQ1 RNase‐Free DNase (Promega), and RNA was isolated by acid phenol chloroform extraction. This resulted in approximately 1µg of total RNA that was then subjected to high‐throughput sequencing. In parallel, the reserved input lysate (approximately 3µg of total RNA) was also sent for high‐throughput sequencing. A total of three biological replicates from each condition were processed. Sequencing libraries were prepared from input and bound RNA using the Illumina TruSeq Stranded Total RNA Human kit and sequenced on an Illumina HiSeq 3000 instrument.

To validate select transcripts by RT–qPCR, an equivalent volume of reserved RNA (3µl) was used as a template for cDNA synthesis using the Maxima First Strand cDNA synthesis kit for RT–qPCR (Thermo Scientific) according to manufacturer instructions. The resulting cDNA was diluted with nuclease‐free water and subsequently analyzed by qPCR using iTaq Universal SYBR Green Supermix (Bio‐Rad) on a Roche LightCycler 96 instrument (Roche). Sequences for gene‐specific primers used for amplification are listed in Appendix TableS2. For input samples, relative fold change was determined by calculating 2^−ΔΔCT values using GAPDH for normalization. For pull‐down samples, relative fold enrichment was determined by dividing the C _q value of the pull‐down by its corresponding input, multiplying by 100 and then normalizing to the relative recovery of the SMG1 transcript.

For assessment of CLIP‐UPF1 expression and pull‐down efficiency by Western blot, reserved input and IP samples were run on a NuPage™ 4–12% Bis‐Tris Protein Gel (Invitrogen) using MOPS buffer according to manufacturer instructions and subsequently transferred to a nitrocellulose membrane according to the NuPage™ manufacturer's protocol (Invitrogen). Membranes were incubated with a blocking buffer for fluorescent Western blotting (Rockland) for 1h at room temperature and then incubated overnight at 4°C with the indicated primary antibody. Primary antibodies used: anti‐RENT1 (goat polyclonal, Bethyl, A300‐038A, 1:1,000) and anti‐β‐actin (mouse monoclonal, Cell Signaling, #3700, 1:1,000). Membranes were subsequently washed three times with 1×TBS supplemented with 0.1% Tween‐20 and then incubated with the appropriate secondary antibody for 1h at room temperature. Secondary antibodies used: anti‐goat IgG (H&L) Antibody DyLight™ 680 Conjugated (Rockland, 605‐744‐002, 1:10,000) and anti‐mouse IgG (H&L) Antibody DyLight™ 680 Conjugated (Rockland, 610‐744‐124, 1:10,000). Membranes were washed three times with 1×TBS supplemented with 0.1% Tween‐20 and then two times with 1×TBS. Western blot images were obtained on an Amersham Typhoon imaging system (GE Healthcare Life Sciences) and quantified using ImageStudio software (LI‐COR Biosciences).

For total UPF1 or SRSF1 depletion and rescue with CLIP‐UPF1, 3×10⁵ cells from the CLIP‐UPF1 stable cell lines, which were engineered as resistant to the described total UPF1 siRNA, or the GFP‐expressing control line were reverse transfected with 40nM of a NT, pan‐UPF1, or SRSF1‐specific siRNA (described above) using Lipofectamine RNAiMAX according to manufacturer instructions. The next day, cells were treated with 200ng/ml doxycycline hyclate (Sigma) for 48h to induce expression of CLIP‐UPF1. Cells were harvested 48h post‐induction, and total RNA was isolated using TRIzol™ (Invitrogen) according to manufacturer instructions. DNase treatment was subsequently performed using RQ1 RNase‐Free DNase (Promega), and RNA was isolated by acid phenol chloroform extraction according to standard protocol. RT–qPCR was performed as described above.

Total UPF1 depletion and RNA‐seq

3×10⁵ HEK‐293 cells (Invitrogen) were reverse transfected with 40nM of a NT or UPF1‐specific siRNA that targets both UPF1 isoforms (described above) using Lipofectamine RNAiMAX according to manufacturer instructions. Seventy‐two hours post‐siRNA transfection, total RNA was isolated using TRIzol™ (Invitrogen) according to manufacturer instructions. DNase‐treatment was subsequently performed using RQ1 RNase‐Free DNase (Promega), and RNA was isolated by acid phenol chloroform extraction according to standard protocol. A total of 2µg RNA was then subjected to high‐throughput sequencing. Three replicates were processed for each condition. Sequencing libraries were prepared using the Illumina TruSeq Stranded Total RNA Human kit and sequenced on an Illumina NovaSeq 6000 instrument.

Total UPF1 and SMG6 depletion with RT–qPCR

3×10⁵ HEK‐293 cells were reverse transfected with 40nM of a NT, total UPF1, or SMG6‐specific siRNA (described above) using Lipofectamine RNAiMAX according to manufacturer instructions. Forty‐eight hours post‐siRNA transfection, cells were replated at a density of 5×10⁵ cells per well of a 6‐well plate. This step was critical to achieve 70–75% confluency on the day of drug treatment (if applicable) and cell harvest. The next day, cells were directly harvested or treated with 50µg/ml puromycin (Sigma) for 4h. A total of three replicates were generated for each condition. Total RNA was then isolated using TRIzol™ (Invitrogen) according to manufacturer instructions. DNase treatment was subsequently performed using RQ1 RNase‐Free DNase (Promega), and RNA was isolated by acid phenol chloroform extraction according to standard protocol. For RT–qPCR, 500ng of total RNA was used as input for cDNA synthesis using the Maxima First Strand cDNA synthesis kit for RT–qPCR (Thermo Scientific) according to manufacturer instructions. The resulting cDNA was diluted with nuclease‐free water and subsequently analyzed by qPCR using iTaq Universal SYBR Green Supermix (Bio‐Rad) on a Roche LightCycler 96 instrument (Roche). Sequences for gene‐specific primers used for amplification are listed in Appendix TableS2. Relative fold changes were determined by calculating 2^−ΔΔCT values using GAPDH for normalization.

Endogenous UPF1_LL knockdown with puromycin or thapsigargin treatment and RNA‐seq

3×10⁵ HEK‐293 cells were reverse transfected with 40nM of a NT or UPF1_LL‐specific siRNA (described above) using Lipofectamine RNAiMAX according to manufacturer instructions. Forty‐eight hours post‐siRNA transfection, cells were replated at a density of 5×10⁵ cells per well of a 6‐well plate. The next day, cells were treated with vehicle control, 25, 50, or 100µg/ml puromycin (Sigma) for 4h, or 1µM thapsigargin for 6 or 9h. A total of three replicates were generated for each condition. Total RNA was then isolated using the RNeasy Plus Mini Kit (QIAGEN). Sequencing libraries were prepared from 2µg total RNA using the Illumina TruSeq Stranded Total RNA Human kit and sequenced on an NovaSeq 6000 instrument. For RT–qPCR validation of changes in target gene expression, 500ng of total RNA was used as input for cDNA synthesis using the Maxima First Strand cDNA synthesis kit for RT–qPCR (Thermo Scientific) according to manufacturer instructions. The resulting cDNA was diluted with nuclease‐free water and subsequently analyzed by qPCR using iTaq Universal SYBR Green Supermix (Bio‐Rad) on a Roche LightCycler 96 instrument (Roche). Sequences for gene‐specific primers used for amplification are listed in Appendix TableS2. Relative fold changes were determined by calculating 2^−ΔΔCT values using GAPDH for normalization.

Roadblock‐qPCR to measure endogenous mRNA stability

mRNA decay measurements were determined using Roadblock‐qPCR as previously described (Watson et al, 2021) but with the following adaptations. 3×10⁵ HEK‐293 cells were reverse transfected with 40nM of a NT or UPF1_LL‐specific siRNA (described above) using Lipofectamine RNAiMAX according to manufacturer instructions. Forty‐eight hours post‐siRNA transfection, cells were replated at a density of 5×10⁵ cells per well of a 6‐well plate. The next day, cells were treated with 400µM 4‐thiouridine (4sU; Cayman Chemical) and vehicle control or 50µg/ml puromycin (Sigma) for a total of 4h. Cells were harvested at indicated timepoints and total RNA was isolated using TRIzol^TM (Invitrogen) according to manufacturer instructions, but with the addition of 1mM (final) DTT to the isopropanol precipitation in order to maintain 4sU in a reduced state (Schofield et al, 2018). Isolated RNA from 4sU‐exposed cells was treated with 48mM N‐ethylmaleimide (NEM; Sigma) as described by Watson et al and then purified using RNAClean XP beads (Beckman Coulter) according to manufacturer instructions. A total of 1µg RNA was used as input for cDNA synthesis with oligo dT₁₈ primers and Protoscript II reverse transcriptase (New England BioLabs) according to manufacturer instructions. The resulting cDNA was diluted with nuclease‐free water and subsequently analyzed by qPCR using iTaq Universal SYBR Green Supermix (Bio‐Rad) on a Roche LightCycler 96 instrument (Roche). Sequences for gene‐specific primers used for amplification are listed in Appendix TableS2. Relative fold changes were determined by calculating 2^−ΔΔCT values using time 0 as the reference and GAPDH for normalization. mRNA half‐lives were estimated by fitting the data to a single‐phase exponential decay model using GraphPad Prism 9.1.0.

Western for phospho‐eIF2

HEK‐293 cells treated with 1µM thapsigargin were lysed in 1X Passive Lysis Buffer (Promega) supplemented with Halt™ Protease and Phosphatase Inhibitor Cocktail (Thermo Scientific) according to manufacturer instructions. A total of 5µg protein was run on a NuPage™ 4–12% Bis‐Tris Protein Gel (Invitrogen) using MOPS buffer according to manufacturer instructions and subsequently transferred to a nitrocellulose membrane according to the NuPage™ manufacturer's protocol (Invitrogen). Detection of phospho and total eIF2 was performed as previously described (Young‐Baird et al, 2020).

Semiquantitative PCR to detect UPF1 isoform ratios

Generated cDNA (1/40^th) from RNA‐seq samples was used as input for PCR amplification with Phusion High‐Fidelity DNA polymerase (New England Biolabs) and UPF1‐specific primers that flank the regulatory loop sequence (Forward: 5’ AACAAGCTGGAGGAGCTGTGGA 3’; Reverse: 5’ ACTTCCACACAAAATCCACCTGGAAGTT 3’). The PCR cycling conditions used were an initial denaturation at 98°C for 30s and then 22 cycles of 98°C for 10s, 63°C for 30s, and 72°C for 15s. PCR products were then run on a 8% Novex™ TBE gel (Invitrogen) according to manufacturer instructions and subsequently stained with SYBR^® Gold Nucleic Acid Stain (Invitrogen). Images were obtained on an Amersham Typhoon imaging system (GE Healthcare Life Sciences) and quantified using ImageStudio software (LI‐COR Biosciences).

Gene‐level differential expression analysis

For analysis of RNA‐seq data from HEK‐293 cells treated with NT, anti‐UPF1_total, or anti‐UPF1_LL siRNAs, raw fastq reads from the NovaSeq 6000 platform were trimmed with fastp (Chen et al, 2018), with the parameters ‐‐detect_adapter_for_pe and ‐‐trim_poly_g. Trimmed reads were aligned with HISAT2 to the hg19/GRCh37 genome and transcriptome index provided by the authors (Kim et al, 2019). For gene‐level differential expression analysis of in‐house and published datasets, reads mapping to Ensembl GRCh37 release 75 gene annotations were quantified with featureCounts (Liao et al, 2014), and differential gene expression was analyzed using limma/voom, as implemented by the Degust server (Powell, 2015; Ritchie et al, 2015). GO analysis was performed with DAVID Bioinformatics Resources, using UniProt keyword classifiers and genes that were represented by >0.5 transcripts per million (TPM) in the RNA‐seq datasets as a background set (Huang et al, 2009a, 2009b; UniProt Consortium, 2021). All comparisons among datasets were performed using genes with average TPM >0.5 in all datasets. Analysis of data from (Boehm et al, 2021) was performed using the authors’ analysis of RNA‐seq of SMG7^ko line 34; similar results were observed with SMG7^ko line 2.

Isoform‐level differential expression analysis

For isoform‐level differential expression analysis, trimmed reads were quantified against a custom HEK‐293 transcriptome index, prepared with Stringtie and TACO as described (Pertea et al, 2015; Niknafs et al, 2017; Kishor et al, 2019b), using kallisto software with parameters ‐‐bias ‐b 1000 ‐t 16 ‐‐single ‐‐rf‐stranded ‐l 200 ‐s 20 (Bray et al, 2016). Differential transcript expression analysis was performed using RUVSeq and edgeR (Robinson et al, 2009; Risso et al, 2014). Normalization and batch correction were performed with the RUVg function, based on transcripts with invariant expression among all samples. The edgeR TMM method was used to obtain normalized differential expression values and to calculate FDRs. Differential isoform usage was calculated for the most abundant PTC and non‐PTC isoforms of each gene using IsoformSwitchAnalyzer and the DEXSeq package (Anders et al, 2012; Vitting‐Seerup & Sandelin, 2019). Alternative splicing was analyzed using rMATS 4.0.1 (Shen et al, 2014), and Sashimi plots were generated using Integrative Genomics Viewer 2.8.2 (Thorvaldsdóttir et al, 2013). ENCODE rMATS data were downloaded from: https://github.com/YeoLab/rbp‐maps (Yee et al, 2019).

Analysis of mRNA features

IsoformSwitchAnalyzeR was used to annotate PTCs in the custom HEK‐293 transcriptome, as described (Kishor et al, 2019b; Vitting‐Seerup & Sandelin, 2019). For PTC analysis, IsoformSwitchAnalyzeR orfMethod “longest” setting was used to predict the longest ORF (min. 100nt) in each annotated transcript, and TCs located >50nt upstream of the final exon junction were designated potential PTCs. Genes represented by at least one transcript predicted to contain a TC within 50nt of the final exon junction or in the last exon and at least one transcript predicted to contain a PTC, defined as a TC more than 50nt upstream of the final exon junction, were selected for analysis of differential isoform usage upon total UPF1 and UPF1_LL‐specific knockdown in Appendix Fig S1C. Putative uORFs were identified using the IsoformSwitchAnalyzeR orfMethod “mostUpstream” setting (min. length 60nt). ORFs predicted by this method that were upstream of the longest predicted ORF were designated potential uORFs. For Appendix TableS1, genes were assigned as PTC‐ or uORF‐containing if at least one transcript with >0.5 TPM in all siUPF1_total, siUPF1_LL, and siNT samples was found to have a putative PTC or uORF.

The most abundant transcript isoform from each gene, as determined by quantification with kallisto as above, was used for analysis of 3’UTR length and PTBP1 and hnRNP L binding motif positions and frequencies. PTBP1 and hnRNP L binding motif position‐specific scoring matrices were downloaded from the RBPmap database and used for motif finding in 3’UTRs derived from the custom HEK‐293 transcriptome with HOMER, as described (Heinz et al, 2010; Paz et al, 2014; Kishor et al, 2019b).

RIP‐seq analysis

Raw fastq reads from CLIP‐UPF1 overexpression RNA‐seq and RIP‐seq data were trimmed with Cutadapt using the following parameters: ‐‐times 2 ‐e 0 ‐O 5 ‐‐quality‐cutoff 6 ‐m 18 ‐a AGATCGGAAGAGCACACGTCTGAACTCCAGTCAC ‐A AGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGTAGATCTCGGTGGTCGCCGTATCATT ‐b AAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA ‐b TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT (Martin, 2011). Trimmed reads were quantified using kallisto software with parameters ‐‐bias ‐b 1000 ‐t 16 ‐single ‐‐rf‐stranded ‐l 200 ‐s 20 (Bray et al, 2016). RIP‐seq enrichment values were obtained by dividing TPM values from IP samples by TPM values from input samples.

GTEx data analysis

GTEx data were downloaded from the GTEx Portal on 11/11/2020. To determine the relative representation of UPF1_LL and UPF1_SL mRNA isoforms, transcript TPM values for transcript ENST00000599848.5 (UPF1_LL) were divided by the total TPM values derived from transcripts ENST00000599848.5 (UPF1_LL) and ENST00000262803.9 (UPF1_SL). GTEx samples were assigned to the indicated tissue types using the sample attributes provided in GTEx Analysis v8.

We thank Sutapa Chakrabarti for the UPF1_LLΔCH recombinant expression plasmid and members of the Hogg laboratory and Nicholas R. Guydosh for critical reading of the manuscript. We are grateful to Sandy Mattijssen and Richard J. Maraia for helpful discussion and to Sara K. Young‐Baird for thapsigargin reagents, helpful discussion, and assistance with phospho‐eIF2 immunoblotting. High‐throughput sequencing was conducted by Yan Luo and Poching Liu in the NHLBI DNA Sequencing and Genomics Core. The Genotype‐Tissue Expression (GTEx) Project was supported by the Common Fund of the Office of the Director of the National Institutes of Health, and by NCI, NHGRI, NHLBI, NIDA, NIMH, and NINDS. This work was supported by the Intramural Research Program, National Institutes of Health, National Heart, Lung, and Blood Institute and utilized the computational resources of the NIH HPC Biowulf cluster (http://hpc.nih.gov).