NuRD and CAF-1-mediated silencing of the D4Z4 array is modulated by DUX4-induced MBD3L proteins

Thank you for submitting your article "MBD3L2 relief of NuRD and CAF-1 silencing of the D4Z4 array amplifies DUX4 expression" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Jessica Tyler as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

The macrosatellite repeat locus D4Z4 harbors the retrogene encoding the DUX4 transcription factor. The expression of DUX4 in early embryos activates a cleavage-specific transcriptional program. DUX4 is normally repressed in somatic tissues but D4Z4 repeat contraction or mutations in chromatin regulators of D4Z4 result in facioscapulohumeral muscular dystrophy (FSHD) due to mis-expression of DUX4 in skeletal muscle. In this manuscript the authors have applied a powerful proteomics approach using LC-MS to identify the proteins that are bound to the D4Z4 macrosatellite locus in myogenic cells. They validate these factors and explore their regulatory potential through siRNA knockdowns. In this way they identify components of the NURD complex and of the chromatin assembly factor (CAF-1), bound to the D4Z4 locus. Down-regulation of HDAC1 and HDAC2 induces Dux4 and ZSCAN4 expression, similarly to CHD4. Knock down of CAF-1 p150 or p60 subunits induces Dux4 and ZSCAN4 expression in FSHD mutant myoblasts. In the last figure, they show that siRNA for Mbd2, Chd4, CAF-1 subunits and HDAC1 and 2, induce DUX4 expression in human iPS cells. The authors go on to propose a new mechanism, via MBD3L2 induction and competition, for amplification and spread of the DUX4 signal after an initial myonuclear burst.

Overall this study is impressive both technically, for the approach the authors developed to identify proteins at the D4Z4 locus and for the therapeutic perspectives as the proteins discovered could provide future therapeutic targets for FSHD. The mechanistic insights as to how the factors identified influence D4Z4 are still quite limited. Nevertheless, this manuscript presents interesting new data and opens up the field for FSHD research, and is also of potential interest for macrosatellite control more generally.

There are however a number of issues that need to be addressed, listed below. In particular, the number of cell lines tested is rather limited – the claims would be more convincing if further FSHD1 and FSHD2 lines were tested. Also enthusiasm was limited for the MBD3L2 feed-forward mechanism and reviewers felt that it occupied too central of a position in the title and Abstract. Focusing on the MBD3L2 mechanism detracts from the impact of the impressive work of identifying the D4Z4 proteome, testing its components, and showing that NURD and CAF-1 are essential for repression of the locus, even in FSHD cells. Indeed as the manuscript stands, the title, impact statement and Abstract overstate the importance of MBD3L: "The early embryonic transcription factor DUX4 induces the expression of MBD3L genes that inhibit the NuRD complex revealing a mechanism of amplifying DUX4 expression in facioscapulohumeral dystrophy". The role of MBD3L would need to be developed further in order to support such claims, as the size of the effects seen was not major. At the very least, the title, Abstract and impact text need to be modified, and should better reflect the main topic of this paper, which is the identification of factors associated with D4Z4.

Essential revisions:

1) D4Z4 underlies FSHD and the authors explore disease-relevant cell lines FSHD1 and FSHD2. In FSHD, it is common to see variability from patient to patient, not just clinically but also in patient-derived cells. Most of the analysis is performed on a single FSHD and a single control myoblast line. By using single FSHD1 and FSHD2 lines, it does not seem that the authors can distinguish between individual patient differences or differences specific to the disease lesions in FSHD1 and FSHD2 (e.g., in the MBD1 and MBD2 experiments). If possible, more individual cell lines (both FSHD1 and FSHD2) should be used for the validations and in order to support the claims about MBD3L2 more convincingly in particular.

2) For the knockdown analyses, the authors are commended for providing comprehensive control data for all of their knockdowns in each experiment. However the study uses n=3 throughout, and performs t-tests to demonstrate significance. As sample size decreases, the t-test becomes extremely sensitive to the assumption of normality, and although a sample of 3 can be consistent with a normal distribution of the population, it cannot provide high confidence that the population is in fact normally distributed. Therefore, the authors should either increase sample size, or use a nonparametric test to calculate significance.

3) Some of the knockdowns were much less effective compared to others, for example CHD3, MBD3, SETDB1 – it seems that this establishes caveats that should be mentioned by the authors.

4) Could the authors please provide the provenance of the eMHF2 iPS line? If this is the control cell line – please make this point clearly. Also, please address the inconsistency between the statement "we have previously shown that DUX4 is expressed in iPS cells" and the data that it is actually quite repressed in iPS cells until NuRD and CAF-1 are disabled.

5) Similar to point 2, above, the effect of NuRD and CAF-1 inhibition should be evaluated in at least two additional iPS cell lines.

6) In the mass spec summaries (Table 1 and Supplementary file 1), the authors indicate the number of peptides detected for each protein. This metric biases in favor of larger proteins, as they break into more peptides. Could the authors please add% coverage as an additional column alongside number of reads?

7) The authors should improve their interpretation (and discussion) of the function of CAF-1 and Nurd. CAF-1 regulates the early embryo cleavage expression programme, including ZSCAN4 and Dux, in mouse ES cells (Hendrickson et al., 2017 and Ishiuchi et al., 2015). Likewise, Trim28, Kdm1a/Lsd1 and HDAC activity (TSA, VPA) have been shown to regulate ZSCAN4 and the 'early cleavage programme' in the mouse (Macfarlan et al. Nature 2014). Neither the Macfarlan, nor the Ishiuchi references are included in the text – for example in the discussion on CAF-1 (Discussion). Furthermore, there are conceptual mistakes in the current text concerning CAF1, concerning its known biochemical and cellular activities (see reviewer 3's comments for more detailed information).

8) Chromatin remodelers and other positive transcriptional regulators might be expected to be present at the contracted D4Z4 array in FSHD cells. The association with repressed D4Z4s in healthy cells is therefore surprising. It would be interesting if the authors commented on the enrichment of SMARCA5, BRD3, and BRD4 at the D4Z4 arrays in healthy myoblasts.

9) There are several overstatements and over interpretations in this paper, particularly the role of MBDL3 as mentioned above and also:

- The authors cannot claim to have "uncovered a mechanism for relieving D4Z4 epigenetic repression in the early embryo" – please change.

- Interpretation in the last paragraph of the subsection “Components shared by the NuRD and CAF-1 complexes mediate D4Z4 repeat repression”, on CAF-1 involved in multiple transcriptionally inhibitory factors needs revision.

- Subsection “Proteins that repress the D4Z4 array in myoblasts also silence DUX4 in iPS cells”. There is no evidence in Hendrickson and Whiddon references that "DUX4 […] activates a cleavage-stage specific transcriptional program in human 4-cell embryos". Please rephrase.

- The conclusion “These results indicate that NuRD and CAF-1 complexes that silence D4Z4 array in muscle cells also contribute to the regulation of this locus during early development” is not founded. Please remove.

10) The impact and value of the study is not conveyed by the title neither by the impact statement and Abstract. The first 5 figures have nothing to do with MBD3L2 relieving silencing. This notion comes up at the end, in Figure 6, and is not developed, certainly not to the point of focusing the title on it. In view of the importance of the purification method to the success of the study, this should be better highlighted in the title and Abstract.

11) Similarly, the schematic describing the purification could be included in Figure 1 proper, rather than in the supplemental data for Figure 1.

Reviewer #1:

This work has two novel and significant components: 1) Using a clever strategy for unbiased biochemical enrichment, the authors identified, and then verified, many of the factors bound at the FSHD-associated D4Z4 macrosatellite in myogenic cells in vivo; 2) identification of a new mechanism, via MBD3L2 induction and competition, for amplification and spread of the DUX4 signal after an initial myonuclear burst. This study represents overall solid work with minor concerns. In FSHD, it is common to see variability from patient to patient, not just clinically but also in patient-derived cells. By using single FSHD1 and FSHD2 lines, it does not seem that the authors can distinguish between individual patient differences or differences specific to the disease lesions in FSHD1 and FSHD2 (e.g., in the MBD1 and MBD2 experiments). Similarly, the crux of the paper and rationale for the title is Figure 6, and the key experiment is done only in FSHD2 and not in FSHD1 cells. The level of knockdown is great, but the effect on DUX4-fl expression is slight (~25%) in what is basically the most relevant therapeutic target cell (myoblasts); is this because they are using FSHD2 cells? For potential therapeutic targeting, FSHD1 cells would seem to be the better choice (and possibly the cells in which the endogenous mechanism is most important). Alternatively, it seems worth looking at DUX4-fl transcripts at a later timepoint (perhaps 48 hours is too early to see much of an effect). Overall, the data does support the claims, which are significant, and the biochemical screening stands on its own. Some of the differences found in validation and the MBD3L claims would be more convincing if performed in more individual cell lines (both FSHD1 and FSHD2), yet that likely would not change the big picture end results, just some of the details.

Reviewer #2:

Enthusiasm is very high for this paper, which identifies proteins bound to D4Z4 and evaluates their regulatory potential through knockdowns. The discovery of the important role of the CAF/1 and NuRD repressive complexes will be of great interest to the FSHD field.

There are a few significant concerns, however:

1) The impact and value of the study is not conveyed by the title, and to some degree, neither by the Abstract. The first 5 figures have nothing to do with MBD3L2 relieving silencing. This notion comes up at the end, in Figure 6, and is really not very developed, certainly not to the point of focusing the title on it. The beautiful magnum opus of purifying D4Z4 chromatin, identifying and then testing a large number of factors is not even mentioned in the Abstract. The long first half of the Abstract does not actually address the contents of the manuscript, but rather serves as an introduction – it could be condensed by half.

2) In view of the importance of the purification method to the success of the study, I suggest including the schematic describing the purification in Figure 1 proper, rather than in the supplemental data for Figure 1.

3) Rigor and reproducibility. First, the authors are commended for providing comprehensive control data for all of their knockdowns in each experiment. Nevertheless, there is a concern with rigor. In the current environment, it is becoming more difficult to justify extremely low sample size experiments. The study uses n=3 throughout, and performs t-tests to demonstrate significance. As sample size decreases, the t-test becomes extremely sensitive to the assumption of normality, and although a sample of 3 can be consistent with a normal distribution of the population, it cannot provide high confidence that the population is in fact normally distributed. Therefore, the authors should either increase sample size, or use a nonparametric test to calculate significance.

3) Some of the knockdowns were much less effective compared to others, for example CHD3, MBD3, SETDB1 – it seems that this establishes caveats that should be mentioned by the authors.

4) Most of the analysis is performed on a single FSHD and a single control myoblast line. Could the authors please take whatever in their view is the most significant knockdown (or combination) culminating from this study and demonstrate its effectiveness in a small panel of FSHD and control cell lines? This will give confidence that the results will be broadly relevant.

5) Could the authors please provide the provenance of the eMHF2 iPS line. I assumed that this is a control cell line – please make this point clearly. Also, please address the inconsistency between the statement "we have previously shown that DUX4 is expressed in iPS cells" and the data that it is actually quite repressed in iPS cells until NuRD and CAF/1 are disabled.

6) Similar to point 4, above, please evaluate the effect of NuRD and CAF/1 inhibition in at least two additional iPS or ES cell lines.

7) In the mass spec summaries (Table 1 and Supplementary file 1), the authors indicate the number of peptides detected for each protein. This metric biases in favor of larger proteins, as they break into more peptides. Could the authors please add% coverage as an additional column alongside number of reads?

Reviewer #3:

The authors report a locus proteomics approach using LC-MS that identifies components of the NURD complex and of the chromatin assembly factor (CAF1), bound to the D4Z4 locus in myoblasts. They validate their targets by performing siRNA for several subunits of NURD, alone or in combination. Based on this, they conclude that combined down-regulation of HDAC1 and HDAC2 induces Dux4 (and ZSCAN4) expression, similarly to CHD4. Likewise, they show that siRNA for p150 or p60 subunits of CAF-1 induces Dux4 and ZSCAN4 expression in FSHD mutant myoblasts, as measured by RT-QPCR. Finally, in the last figure, they show that siRNA for Mbd2, Chd4, CAF-1 subunits and HDAC1 and 2, induce DUX4 expression in human iPS cells.

The work is properly controlled, and the method by itself is very interesting. While I do not have specific technical comments on the data, without any further characterisation (for example, biochemical) on how Nurd or CAF-1 act to regulate D4Z4, I find the manuscript not very novel and a little preliminary (in short, MassSpec of the D4Z4 locus and many siRNA coupled to RT-QPCR of the subunits of two complexes to validate their MassSPec). The question that remains is how does any of these complexes actually regulate D4Z4 chromatin structure and DUX4 expression?

CAF-1 has already been shown to regulate the 'early cleavage expression' programme, including ZSCAN 4 and Dux, in mouse ES cells (Hendrickson et al., 2017 and Ishiuchi et al., 2015). Likewise, Trim28, Kdm1a/Lsd1 and HDAC activity (TSA, VPA) have been shown to regulate ZSCAN4 and the 'early cleavage programme' in the mouse (Macfarlan et al. Nature 2014). Neither the Macfarlan, nor the Ishiuchi references are included in the text. I find this particularly surprising, especially considering the discussion on CAF1 (Discussion).

At the very least, the authors should improve their interpretation (and discussion) of the function of CAF-1 and Nurd, the former in which I find important conceptual mistakes, based on the known biochemical and cellular activities of CAF-1. I also find several of their conclusions/interpretations overstated, which will unnecessarily confuse the readers, and which the authors should correct, as per my comments below.

CAF-1 is not a transcriptionally repressive complex (subsection “Silencing the D4Z4 array requires components of the MBD1/CAF-1 complex”, first paragraph; subsection “Silencing the D4Z4 array requires components of the MBD1/CAF-1 complex”, last paragraph; subsection “Components shared by the NuRD and CAF-1 complexes mediate D4Z4 repeat repression”, first paragraph), but a chromatin assembly complex which functions only in parallel to DNA synthesis of the whole genome, mainly in S-phase, and under some circumstances (e.g. DNA repair), during G1. In fact, its expression is regulated throughout the cell cycle (see works by Paul Kauffman, Torsten Krude and Genevieve Almouzni). Along these lines, the conceptual interpretation of CAF-1 function is misleading. Also, the authors do not really show how CAF-1 (or NuRD) regulate chromatin function: for example, do myoblasts proliferate, and hence CAF1 function may be related to its chromatin assembly activity?

I find the following sentences overstated, and there is no data in the manuscript that supports such conclusions, please rephrase/remove:

1) The authors have not "uncovered a mechanism for relieving D4Z4 'epigenetic repression in the early embryo".

2) Interpretation in the last paragraph of the subsection “Components shared by the NuRD and CAF-1 complexes mediate D4Z4 repeat repression”, on CAF-1 involved in multiple transcriptionally inhibitory factors needs revision.

3) Subsection “Proteins that repress the D4Z4 array in myoblasts also silence DUX4 in iPS cells”. There is no evidence in Hendrickson and Whiddon references that "DUX4 […] activates a cleavage-stage specific transcriptional program in human 4-cell embryos". Please rephrase.

4) The conclusion “These results indicate that NuRD and CAF-1 complexes that silence D4Z4 array in muscle cells also contribute to the regulation of this locus during early development” is not founded. Please remove.