A feedback loop between nonsense-mediated decay and the retrogene DUX4 in facioscapulohumeral muscular dystrophy

1) The effects were overall modest, in particular in experiments with the 3' UTR intron and in the assessment of the fraction of nuclei expressing DUX4 protein in FHSD myotubes.

We agree that the experiments with the chimeric Gl-DUX4-Intron2 constructs generally yielded modest effect sizes, such as increases of 2- and 1.4-fold following UPF1 KD or ectopic DUX4 expression (Figure 3G, 4C). These effects were substantially smaller than the corresponding 4- and 5-fold increases in endogenous DUX4 mRNA levels that we observed following UPF1 KD or ectopic DUX4 expression. We are unsure why the chimeric construct exhibited more modest effect sizes than did the endogenous DUX4 mRNA, although it may be due to a combination of its plasmid-based nature and the absence of sequence features of the DUX4 mRNA other than its 3' UTR intron that contribute to its degradation by NMD. As the Gl-DUX4-Intron2 constructs are chimeric, we feel that the most realistic estimates of effect sizes stem from our measurements of the endogenous DUX4 mRNA rather than the Gl-DUX4-Intron2 constructs.

Outside of the chimeric construct, however, many of the other changes that we observed were quite dramatic. For example, the reviewers mention the fraction of DUX4+ nuclei in a FSHD myotube culture following control or UPF1 knock down. DUX4 exhibits a highly variegated expression pattern, with DUX4 detectable in only 0.1-1% of FSHD muscle cells (Snider et al., PLoS Genetics, 2010; Figure 3 from this manuscript). Following UPF1 knock down, we found that the fraction of DUX4+ nuclei increased from 0.3% to 2.1%, a change of an order of magnitude. As the extremely low fraction of DUX4+ cells in FSHD muscle is a defining characteristic of the disease, we feel that an order-of-magnitude increase in the fraction of DUX4+ cells is a substantial effect size, consistent with NMD playing an important role in the regulation of DUX4 levels. We have emphasized this effect size in the revised manuscript, and also updated Figure 3 to show additional microscopy fields.

To place these results in context, it is useful to compare our observations with previous studies where NMD was directly inhibited. For example, a recent study (Hurt et al., Genome Research, 2013), inhibited NMD with shRNAs against UPF1 or cycloheximide treatment, which prevents protein translation and thereby NMD. Hurt et al. then used RNA-seq to quantify global mRNA levels. Hurt et al. identified only 8, 14, or 40 predicted NMD substrates out of >20,000 queried mRNAs that were up-regulated by at least 4-fold in cells treated with one of two distinct shRNAs against UPF1 (8 and 14 mRNAs identified) or cycloheximide (40 mRNAs identified). In the context of these results, we feel that the 4- and 5-fold increases in endogenous DUX4 mRNA levels that we observe following UPF1 knock down or ectopic DUX4 expression are quite substantial.

As discussed further below in specific point #1, we also note that the degree of NMD inhibition that is caused by DUX4 expression is quite remarkable, matching the degree of NMD inhibition caused by genetic knock out of essential NMD factors.

2) Lack of connection between observations in primary myoblast cell lines with overexpressed DUX4 and with authentic patient samples—perhaps too difficult to address, but perhaps it would be possible to ask whether authentic FSHD muscle cells contain increased proportions of predicted NMD substrates.

We agree that our conclusions would be even further strengthened by a direct demonstration that DUX4 causes NMD inhibition in primary patient materials, in the absence of ectopic DUX4 expression. However, DUX4’s variegated expression pattern—with DUX4 detectable in only 0.1-1% of FSHD muscle cells (Snider et al., PLoS Genetics, 2010; Figure 3 from this manuscript)—typically renders such experiments infeasible. For this reason, most studies of DUX4’s molecular function have relied upon ectopic expression to ensure that the majority of assayed cells are DUX4+.

The challenge of detecting DUX4-induced NMD inhibition given DUX4’s variegated expression pattern can be readily seen with a statistical power estimation. For example, we used a poison exon-containing isoform of SRSF3 as an endogenous marker of NMD. This isoform increased dramatically in abundance (from ∼10% to 65% of SRSF3’s mRNA population) following ectopic DUX4 expression. It is therefore reasonable to expect that DUX4+ cells in a FSHD muscle cell culture may similarly express abnormally high levels of this SRSF3 NMD substrate in the absence of ectopic DUX4 expression. However, the variegated nature of DUX4 expression prevents the direct detection of such signals of NMD inhibition simply due to statistical power limitations. DUX4 expression is detectable in 0.3% of FSHD nuclei by immunostaining (Figure 3H). Assume that the SRSF3 NMD substrate is present at isoform ratios of 10% and 65% in DUX4- and DUX4+ cells, respectively. Measurements of the SRSF3 NMD substrate in a bulk FSHD cell population would therefore yield an isoform ratio of (0.003 x 65%) + (0.997 x 10%) = 10.165%. The difference between an isoform ratio of 10% for DUX4- cells and 10.165% for a mixture of DUX4+ and DUX4- cells is not detectable with RNA-seq or quantitative PCR.

In the future, we may be able to overcome the challenges presented by DUX4’s variegated expression pattern by studying single cells with assays such as single-cell RNA-seq or live cell imaging approaches. However, we feel that such assays are beyond the scope of this manuscript. We have added additional text to the discussion in which we briefly describe the challenges with detecting NMD inhibition in a bulk population of FSHD cells.

3) Lack of a potential molecular mechanism to explain how overexpression of a transcription factor leads to selective degradation of a set of factors involved in NMD. One possibility would be to more thoroughly evaluate the levels of other NMD factor (UPF2 etc.) or components of the proteasome itself. Without elucidation of some mechanistic link, this part of the story remains purely correlative and therefore less impactful.

We agree that this aspect of the manuscript is correlative. Given the close temporal connection between decreased UPF1 protein levels and increased levels of NMD substrates, we do feel that it is reasonable to speculate that UPF1 degradation is a likely contributor to NMD inhibition. In the revised manuscript, we are careful to state clearly that the data is suggestive but purely correlative. Nonetheless, as rapid UPF1 protein degradation has not been previously described to our knowledge, we feel that this section of the manuscript is a significant addition to the field, as it suggests a potential new mechanism for the cell to alter RNA surveillance efficiency.

In a previous study (Geng et al., Developmental Cell, 2012), we noticed that many genes that were differentially expressed following DUX4 expression were involved in protein ubiquitination, including numerous E3 ubiquitin ligases. It is possible, for example, that UPF1 is a substrate of one of the many E3 ligases that are up-regulated in DUX4-expressing muscle cells. While we do hope to eventually determine how DUX4 expression triggers UPF1 degradation, we feel that such assays are beyond the scope of this manuscript given the many different ways in which a transcription factor could indirectly influence UFP1 proteolysis. We briefly discuss this in the revised manuscript.

Other more specific points follow:

1) Why is only a small fraction (13%) of predicted NMD targets affected by this mechanism?

We agree that it is interesting that only a minority of premature stop codon-containing mRNAs exhibit increased levels following DUX4 expression and concomitant NMD inhibition. This could be due to imperfect annotation of NMD substrates, differential sensitivity of different NMD substrates to reduced UPF1 levels, or other factors.

It is important to note that this 13% is quite high when compared to other studies where NMD has been inhibited genetically or with RNAi. For example, a recent study (McIlwain et al., PNAS, 2010) used a gene trap to generate mouse embryonic fibroblasts lacking Smg1, which is responsible for UPF1 phosphorylation, a required step in NMD. McIlwain et al. then used RNA-seq to characterize the transcriptomes of Smg1-deficient cells, finding that “in the absence of Smg1, approximately 9% of predicted PTC-containing AS events displayed changes of more than 10%, and approximately 2% showed changes of more than 20%” (PTC = premature termination codon; AS = alternative splicing). Other studies conducted in worms, flies, and mammalian cells found that similarly small fractions of predicted NMD substrates were affected by depletion of core NMD factors.

Therefore, we agree that it is interesting that only 13% of predicted NMD substrates are up-regulated in DUX4-expressing cells. However, as this degree of up-regulation is comparable to that observed in previous studies where NMD was directly inhibited genetically, we believe that this 13% is a remarkably high fraction. We comment on this in the revised manuscript.

2) Figure 2E: Can the authors explain the transient nature of the upregulation of the PTC-containing beta-globin NMD reporter? In the light of the proposed feedback loop, one would anticipate robust and enduring inhibition of NMD (as is observed for SRSF2 and SRSF3 transcripts; Figure 2F) and thus the beta-globin PTC+ levels to remain elevated after 20 hours post transduction with the DUX4 expressing lentivirus.

As the reviewers point out, levels of the NMD(+) β-globin reporter increase at the same time (∼12 h following infection) as the endogenous SRSF2 and SRSF3 markers, but the NMD(+) β-globin reporter up-regulation does not last for the entire time course. This is surprising because (1) SRSF2 and SRSF3 NMD substrates exhibit increased levels throughout the time course, (2) UPF1 levels decrease monotonically throughout the time course, and (3) our RNA-seq data from MB135 and 54-1 cells was obtained at 24 hours and 48 hours time points, respectively, and both samples exhibited obvious global increases in NMD substrate levels. Therefore, while we are confident that NMD inhibition occurs throughout the time course, we are unsure why this enduring behavior is not reflected in the NMD(+) β-globin reporter. We are unaware of a previous study that has used this reporter to study the dynamics of NMD efficiency along a time course. For unclear reasons, such a plasmid-based reporter may be a more useful tool for measuring NMD efficiency in a static, rather than dynamic, context.

3) Figure 2E: The data points for the half-life measurement of DUX4 mRNA under siControl conditions scatter quite a lot; the authors should describe how they did the curve fitting. The conventional way would be to plot the RNA levels logarithmically and calculate the linear regression line through these data points. Furthermore, by limiting the time course to only 2 hours, the extrapolated half-life of 7.86 h in UPF1-depleted cells is most likely highly inaccurate. The time course should be expanded to 6 or 8 hours to get a more precise estimate of this half-life.

To estimate the half-life of DUX4 mRNA, we computed the best-fit exponential model using untransformed data, which is mathematically identical to the reviewers’ suggestion of computing a best-fit linear model for log-transformed data. We apologize for not explaining this half-life estimation procedure more clearly in the initial submission.

We agree that the data points exhibit substantial variation, particularly for the siControl treatment. Unfortunately, this high level of variability is an unavoidable consequence of DUX4’s variegated expression. Since DUX4 is detectable in only 0.3% of nuclei by immunostaining for the siControl treatment (Figure 3), DUX4 mRNA levels are very low even in large cultures of FSHD cells. Estimates of DUX4 mRNA levels therefore are noisy and become increasingly so following transcription shutoff.

We agree with the reviewers that extrapolating a ∼8 hour half-life from a 2 hour time course is prone to substantial measurement error. We accordingly repeated the DUX4 mRNA half-life measurement multiple times with a longer 8 hour time course. However, due to the aforementioned difficulty of accurately measuring DUX4 mRNA following transcription shutoff even in a large culture of FSHD muscle cells, we were unfortunately unable to obtain reliable data from the longer time courses.

As the DUX4 mRNA half-life measurement was not an essential component of the manuscript—since we provide multiple other sources of evidence that DUX4 mRNA is a NMD substrate, due in part to its spliced 3' UTR—we have decided to remove the half-life data entirely from the revised manuscript. We agree with the reviewers that this data is noisy, and feel that it is a distraction from the otherwise clean data that we present here.