Structure and functional mapping of the KRAB‐KAP1 repressor complex

Abstract Transposable elements are a genetic reservoir from which new genes and regulatory elements can emerge. However, expression of transposable elements can be pathogenic and is therefore tightly controlled. KRAB domain‐containing zinc finger proteins (KRAB‐ZFPs) recruit the co‐repressor KRAB‐associated protein 1 (KAP1/TRIM28) to regulate many transposable elements, but how KRAB‐ZFPs and KAP1 interact remains unclear. Here, we report the crystal structure of the KAP1 tripartite motif (TRIM) in complex with the KRAB domain from a human KRAB‐ZFP, ZNF93. Structure‐guided mutations in the KAP1‐KRAB binding interface abolished repressive activity in an epigenetic transcriptional silencing assay. Deposition of H3K9me3 over thousands of loci is lost genome‐wide in cells expressing a KAP1 variant with mutations that abolish KRAB binding. Our work identifies and functionally validates the KRAB‐KAP1 molecular interface, which is critical for a central transcriptional control axis in vertebrates. In addition, the structure‐based prediction of KAP1 recruitment efficiency will enable optimization of KRABs used in CRISPRi.

I need to state that this is not really my area, so my review is limited to my structural biology expertise and interest as a general reader. I am not familiar enough with the field to have a strong opinion on the novelty and significance.
The paper describes the structural determination of a complex between two truncated proteins (and the title should really reflect that these are not full-length proteins) that allows full biochemical characterisation of the interface between a KRAB domain (of ZNF93) and the RBCC domain of TRIM28. The key residues identified in the interface are then validated through mutagenesis and a silencing assay. The structural data are high quality, with 5 crystal structures reported, including 4 SeMet substitutions to aid sequence assignments in regions of ambiguous density. This is an important point, as in an almost entirely helical structure, the register of residues is important to get right. I'm not totally sure I understood what the alphafold derived models add to the manuscript. Despite not being in the area, I found the manuscript to be interesting and well-written, and the structure was not previously known, indeed the suggestion that there is likely to be significant disorder prior to complex formation is an interesting one, and one that could be worth following up experimentally. I don't have any concerns with the structural data.

Referee #3:
This paper is of interest to readers in the fields of structural biology, chromatin and epigenetics, KRAB zinc finger proteins and transposable elements/endogenous retroviruses. The manuscript reveals the structure of the interaction of the KRAB domain of the Zinc finger protein (KZFP) ZNF93 with the RBCC domain of the common KZFP co-repressor Kap1, made possible by 1) adding the T4 lysozyme to the N terminus of the RBCC domain 2) deletion of the flexible B-box 1 domain of KAP1, and 3) adding the CUE1 domain of the chromatin remodeler SMARCAD1, which binds to the KAP1 coiled coil domain. The authors make use of recently developed Alphafold2 and Alphafold-Multimer tools to validate and compare their structural model, which make the structure convincing. In addition, they carefully assess a panel of previously established mutations that abolish silencing both in the KRAB domain (initially identified for the related KZFP ZNF10) and in KAP1's RBCC domain (in particular V293S/K296A/M297A/L300S, individually or combined (when combined called CCmut, which the authors themselves had previously introduced and shown to abolish KRAB interactions). To demonstrate the biological importance of the interaction, the authors use KAP1 KO HEK293T cells complemented with WT and e.g. CCmut KAP1 in luciferase reporter assays to show loss of repressive activity upon mutation of the interacting residues. They also assess these complemented cells with H3K9me3 cut&run to investigate if loss of the interaction affects heterochromatin formation via the KAP1 interacting H3K9me3 histone methyltransferase SETDB1. These experiments convincingly show that KZFP interactions are indispensable for repressing transposable elements in HEK293T cells. The presented data sufficiently support the claims of the paper, are well controlled and well presented. The results present a significant advance in our understanding of KAP1 and its interactions with KZFPs, the largest transcription factor family in human. These results will allow researchers in other fields where KAP1 has essential roles (e.g. stem cell or developmental or cancer biology) to assess the importance of KZFP interactions.
Major concerns • There have been many reports of direct interactions between components of the KRAB/KAP1/SETDB1 complexes, and this multivalency might suggest that KAP1 could be recruited independently of KRAB proteins to some chromatin domains. DO mutations that alter KRAB /KAP1 binding also prevent KAP1 binding to chromatin at ERVs/TEs? This is certainly suggested by the reduction in H3K9me3, but would be more convincing if the authors showed that KAP1/SETDB1 fails to be recruited to these sites in their "rescued cell" lines. There is an available KAP1 antibody that has been used successfully for ChIP-seq (e.g. with ab10484 used by Frank Jacobs lab) Minor concerns: • Page 17: KRAB-KAP1 complexes recruit multiple effectors that modify the and epigenetic conformational landscape of chromatin target loci including HP1, SETDB1, NuRD and SMARCAD1. Remove 'and': '...modify the and epigenetic conformational landscape...' • Page 12/ Fig 4C+D: ' ...V293S [,] was not expressed at detectable levels in KAP1 KO HEK293T cells'.
If feasible it may be worth repeating the overexpression experiment, given this mutation is part of the 4 mutations used in the functional characterization (CCmut) in the remainder of the manuscript.

Editor's Summary
As you will see, the reviewers are overall positive and appreciate the findings and their interest to the field. However, they do raise a number of points which should be addressed in the revised version of the manuscript. In particular, the KAP1-KRAB/ SETDB1 complex and its recruitment should be assessed in further detail (ref#1-point 5; ref#3-major point 1), and the analysis and interpretation of the AlphaFold results discussed in more detail (ref#1-point 1; ref#2-point 2). Please also carefully consider all other referee comments and revise the manuscript and figures as needed, as well as providing a detailed response to each comment.

Dear Editor, Dear Referees,
We thank all three referees for their attentive reading of this manuscript, and for their positive, constructive comments and useful suggestions.
We have carefully and extensively revised our manuscript in response to the referees' comments. Revisions include new experimental data shown in the expanded view figures, along with text edits throughout the manuscript. The most important additions and changes are the following: • To address the only "major concern", from Referee 3, regarding whether recruitment of KAP1 complex to chromatin is KRAB-dependent, we performed genome-wide profiling of KAP1 distribution by ChIP-seq, as suggested by Referee 3. Analysis of the ChIP-seq data showed that KAP1 enrichment in cells expressing WT KAP1 occurred at H3K9me3-positive regions and gene promoters (new figure panels Fig EV4A,B). No enrichment was observed over H3K9me3-positive regions in cells expressing a KRAB binding-deficient KAP1 variant, indicating that KAP1 recruitment to these sites depends on KRAB binding. Interestingly, WT and KRAB binding-deficient KAP1 both remained enriched over gene promoters, suggesting that KAP1 recruitment to promoters is independent of KRAB binding.
• We have added new immunoprecipitation data demonstrating that the KAP1-SETDB1 interaction is unaffected by mutations inhibiting KRAB binding (new figure panel Fig EV4E). This supports our hypothesis that the loss of H3K9me3 in the KRAB binding-deficient variant is due to KAP1 not being recruited to H3K9me3 loci (rather than disruption of SETDB1 binding). Moreover, we show that the KAP1-SETDB1 interaction is dependent on SUMOylation.
• We have added a paragraph in the Materials and Methods describing how the AlphaFold2 models were generated, specifying which installations of AlphaFold used and the URL of the server website.
• We have added additional new figure panels (Figs EV5A,D,E), new text in the Results (p. 7-9) and Methods (pp. 13-17), and minor edits throughout the manuscript to address the remaining minor concerns of the referees.
We attach a copy of the manuscript with all substantive changes to the text highlighted (please note that text deleted since the previous version is not shown).

Yorgo Modis (on behalf of the authors)
28th Sep 2022 1st Authors' Response to Reviewers

Referee #1:
General summary and opinion about the principal significance of the study, its questions and findings The KRAB domain defines the largest transcription factor (TF) family in the human genome. KRAB-domain containing zinc-finger proteins (KZFPs) control the expression of transposable elements (TEs) as well as that of host genes and thereby mediate multiple key biological processes. For its versatile repressive function, the KRAB domain has been adapted to confer repression at specific genomic loci using its fusion protein with a catalytically deficient Cas9 (dCas9). This dCas9-KRAB has become a popular genetic tool known as CRISPR interference (CRISPRi) to dissect gene regulatory systems through targeting promoter or enhancer regions.
Although it has long been known that the KRAB domain deposits H3K9me3 by recruiting a scaffold protein, KAP1, and the key domains and residues important for the KRAB-KAP1 interactions have been identified, its interaction has not been solved structurally, which hampered our deeper understanding of it.
Stoll et al. (hereafter, the authors) solved the structure of the KRAB-KAP1 complex, revealing residues involved in the binding from both proteins. The authors further validated the involvement of the identified KAP1 residues in its functions by reporter assays and by H3K9me3 CUT&RUN.
This manuscript presents enough evidence to support the authors' claims. Given that the KRAB-KAP1 interaction is important for not only understanding the biology of KZFPs but also designing CRISPRi with a higher potency, we believe this work is of interest for wide research areas, although we think some minor concerns deserve to be addressed before publication.
Specific major concerns essential to be addressed to support the conclusions We do not have any major concerns.
Minor concerns that should be addressed 1. The authors should add detailed information about the AlphaFold2 and AlphaFold-Multimer analyses so that the readers can reproduce them.
We have added a paragraph in the Materials and Methods describing how the AlphaFold2 models were generated including the specific installations of AlphaFold used and the URL of the server website.
2. The Tycko et al paper is a great resource that complements this study by a highthroughput analysis of human KRAB domains and we believe the authors should take full advantage of it. By a deep mutational scan, they identified 12 KRAB A-BOX residues that significantly contribute to its silencing. Although several of them are mentioned in the manuscript for their direct involvement in the binding, the authors should summarise if any explanation could be given for the rest of the residues based on the structure they obtained Of the 12 KRAB A-Box residues identified by Tycko et al (2020), 10 were already discussed in the context of our crystal structure (specifically residues Asp8, Val9, Ile11, Phe13, Glu17, Leu21, Tyr29, Val32, Met33 and Glu35 in ZNF93 residue numbering; see pages 6 and 7). The two remaining residues are Trp18 and Asn36. Asn36 contributes to the KAP1 binding interface, forming a hydrogen bond with the side chain of Asn376. Asn36 also stabilizes the first helix of the KRAB A-box via an intramolecular hydrogen bond. Trp18 is unusual in that although it is well conserved and was reported to be required for silencing by Tycko et al (2020), it does not contribute to the KAP1 binding interface or stabilize the KRAB domain fold based on the crystal structure. Hence our structure cannot explain why Trp18 is required for silencing. We have added a few lines in the corresponding Results section (p. 7) summarizing these observations. We have also modified Fig 3C to show Trp18 and Asn36 in the appropriate closeup views of the structure. Hence all 12 residues identified by Tycko et al are now shown in Fig 3 and discussed in the text. 3. It is known that a phylogenetic tree of the human KRAB domains forms two major clades: one closer to the consensus (standard KRAB) and the other away from it (variant KRAB). The standard KRABs often recruit KAP1, while the variant KRABs do not; however, there are several KRABs that do not follow this trend as shown in the Tycko et al paper. Could the author explain these based on residues involved in KRAB-KAP1 binding?
We thank the referee for this question. To address it, we produced a multiple sequence alignment of 39 variant KRAB domains and two unusual standard KRABs (new figure panel, Fig EV2B). We highlight residues that form contacts with KAP1. For residues that differ from the ZNF93 sequence we use a green, orange, and red color code to indicate whether they are expected to be tolerated, detrimental, or inhibitory for KAP1 binding. Based on this analysis, and consistent with the literature, most variant KRABs are not expected to bind KAP1, but a few (e.g. ZNF75D, ZNF26) are predicted to bind tightly. Our analysis extends to some KRABs for which there is no data, offering prospective predictive value. Fig 4C and D, the authors should clarify what comparisons were made to assess statistical significance.

In
Upon reexamining these figure panels, we opted to remove the statistical significance asterisks as the difference between WT KAP1 and other KAP1 variants is self-evident without statistical tests. All three data points are shown for each sample. Fig 4C and D, it looks like some KAP1 mutants de-silence the SVA reporter more weakly than the LINE1 reporter. Is this statistically significant? If so, could this be explained by the difference in the KRAB domains of ZNF91 and ZNF93?

In
The fold repression observed for WT KAP1 and the KAP1 mutants are indeed different for the LINE1 and SVA reporter. This could be due to differences in the KRAB domains of the cognate ZNFs (ZNF93 and ZFN91, respectively), as the referee suggests. Differences could include differences in KAP1 binding affinity, DNA binding affinity, ZNF expression level, or the epigenetic background of the reporter. Because the relative (and absolute) repression observed with WT KAP1 is different in the two assays, direct statistical comparisons of the KAP1 mutants across the two assays (e.g. with a t-test or Anova test) would not be appropriate. In this context, although the KAP1 M297A mutant may be less derepressed than the other mutants in the SVA reporter, we refrain from speculating on the basis of this difference.
6. The authors show that mutations in KAP1-KRAB interacting residues abolished reporter silencing as well as genome-wide H3K9me3 deposition. This is indeed likely due to failure in KAP1-KRAB interaction, but an alternative explanation is that such mutation(s) interferes with KAP1-SETDB1 interaction. I believe that showing the mutant KAP1 still binds similar levels of SETDB1 will make the authors' claim complete.

The referee raises a valid point. We have addressed this by performing coimmunoprecipitation experiments of KAP1 and SETDB1 from cell lysates. These pulldowns showed that WT KAP1 and a KRAB binding-deficient variant (CCmut) recruit SETDB1 equally well (see new figure panel Fig EV4E), demonstrating that SETDB1 recruitment by KAP1 is independent of KRAB binding. Moreover, the SETDB1-KAP1 interaction was only observed in the presence of the SUMO protease inhibitor N-ethylmaleimide (NEM), which confirms that KAP1 recruits SETDB1 in a SUMO-dependent manner.
7. In Fig 5C, the authors show that CCmut completely failed to restore the H3K9me3 levels in retrotransposons. The authors should also assess it genome-wide as it is known that some KZFPs bind other classes of TEs as well as genic regions. Fig 5B heatmaps Fig EV5).

Thank you for spotting this typo-it has been corrected (note that Fig EV4 from the original submission is now
9. Also, the authors should show quantitative data to support further their claim, such as the difference in H3K9me3 signal between the HUSH-bound loci and the rest. figure panel (Fig EV5D) shows that there is significantly smaller average reduction in H3K9me3 levels over the HUSH-bound loci. That there is measurable reduction in H3K9me3 levels over HUSH-bound loci in the KAP1 KO may reflect the cooperation between the two complexes at certain loci, as detailed in Robbez-Masson et al (2018; cited in the manuscript). We have added a sentence to the text to mention this conclusion (p. 9).

We assessed changes in H3K9me3 in control and KAP1 KO cells over two sets of H3K9me3 rich loci: KAP1-dependent loci and those bound by HUSH (based on Douse et al, 2020 and Seczynska et al., 2022). A new
10. In EV4, the authors should show what proportion of the KAP1-independent H3K9me3 loci can be explained by HUSH-binding or centromeric regions by a bar plot. a new figure panel, Fig EV5E. 11. In Fig 5D, the heatmap annotation for THE1 says "center" unlike the other two TEs. The authors should explain what this means. We believe that it would be more informative to show a position-aligned heatmap just like the other ones.

Fig 5D has been amended as requested to show a position-aligned (+/-5kb) heatmap for all three TE types (based on the THE1 consensus sequence).
Additional non-essential suggestions for improving the study (which will be at the author's/editor's discretion) 12. We believe testing if ZIM3 S46F indeed strengthens its potency as speculated by the authors would be a significant improvement to this manuscript. This would not only further support the authors' claim but also contribute to rational design of a KRAB domain with higher potency, which greatly benefits its application in CRISPRi.
We agree it will be useful to test if ZIM3 S46F indeed strengthens its transcriptional repression potency for CRISPRi applications. We plan to do this in future studies.
13. Ideally, one would like to predict if a KRAB domain binds KAP1 or not solely based on its sequence. Could the authors come up with a scoring system that roughly estimates how likely a KRAB binds KAP1 based on a given KRAB sequence?
A scoring system to rank and predict KAP1 binding by KRAB domains would indeed by useful. To do this effectively we believe we would need to increase the size of our dataset and examine additional KRAB domains as even with 40 or so KRABs it remains unclear what the rules should be for a scoring algorithm. We hope to incorporate this in a future study.

Referee #2:
I need to state that this is not really my area, so my review is limited to my structural biology expertise and interest as a general reader. I am not familiar enough with the field to have a strong opinion on the novelty and significance.
The paper describes the structural determination of a complex between two truncated proteins (and the title should really reflect that these are not full-length proteins) that allows full biochemical characterisation of the interface between a KRAB domain (of ZNF93) and the RBCC domain of TRIM28. The key residues identified in the interface are then validated through mutagenesis and a silencing assay. The structural data are high quality, with 5 crystal structures reported, including 4 SeMet substitutions to aid sequence assignments in regions of ambiguous density. This is an important point, as in an almost entirely helical structure, the register of residues is important to get right.
We thank the Referee for their positive overall review.
I'm not totally sure I understood what the alphafold derived models add to the manuscript. Despite not being in the area, I found the manuscript to be interesting and well-written, and the structure was not previously known, indeed the suggestion that there is likely to be significant disorder prior to complex formation is an interesting one, and one that could be worth following up experimentally.
We have added a paragraph in the Materials and Methods describing how the AlphaFold2 models were generated including the specific installations of AlphaFold used and the URL of the server website (as also requested by Referee 1).
I don't have any concerns with the structural data.

Referee #3:
This paper is of interest to readers in the fields of structural biology, chromatin and epigenetics, KRAB zinc finger proteins and transposable elements/endogenous retroviruses. The manuscript reveals the structure of the interaction of the KRAB domain of the Zinc finger protein (KZFP) ZNF93 with the RBCC domain of the common KZFP co-repressor Kap1, made possible by 1) adding the T4 lysozyme to the N terminus of the RBCC domain 2) deletion of the flexible B-box 1 domain of KAP1, and 3) adding the CUE1 domain of the chromatin remodeler SMARCAD1, which binds to the KAP1 coiled coil domain. The authors make use of recently developed Alphafold2 and Alphafold-Multimer tools to validate and compare their structural model, which make the structure convincing. In addition, they carefully assess a panel of previously established mutations that abolish silencing both in the KRAB domain (initially identified for the related KZFP ZNF10) and in KAP1's RBCC domain (in particular V293S/K296A/M297A/L300S, individually or combined (when combined called CCmut, which the authors themselves had previously introduced and shown to abolish KRAB interactions). To demonstrate the biological importance of the interaction, the authors use KAP1 KO HEK293T cells complemented with WT and e.g. CCmut KAP1 in luciferase reporter assays to show loss of repressive activity upon mutation of the interacting residues. They also assess these complemented cells with H3K9me3 cut&run to investigate if loss of the interaction affects heterochromatin formation via the KAP1 interacting H3K9me3 histone methyltransferase SETDB1. These experiments convincingly show that KZFP interactions are indispensable for repressing transposable elements in HEK293T cells. The presented data sufficiently support the claims of the paper, are well controlled and well presented. The results present a significant advance in our understanding of KAP1 and its interactions with KZFPs, the largest transcription factor family in human. These results will allow researchers in other fields where KAP1 has essential roles (e.g. stem cell or developmental or cancer biology) to assess the importance of KZFP interactions.
Major concerns • There have been many reports of direct interactions between components of the KRAB/KAP1/SETDB1 complexes, and this multivalency might suggest that KAP1 could be recruited independently of KRAB proteins to some chromatin domains. DO mutations that alter KRAB /KAP1 binding also prevent KAP1 binding to chromatin at ERVs/TEs? This is certainly suggested by the reduction in H3K9me3, but would be more convincing if the authors showed that KAP1/SETDB1 fails to be recruited to these sites in their "rescued cell" lines. There is an available KAP1 antibody that has been used successfully for ChIP-seq (e.g. with ab10484 used by Frank Jacobs lab) We agree with the reviewer and wanted from the beginning to probe KAP1-chromatin interactions in addition to using the indirect H3K9me3 readout. Indeed, to directly test our model that the CCmut KAP1 fails to be recruited to chromatin via the KRAB domain, we made extensive attempts to get KAP1 CUT&RUN and CUT&Tag to work at an early stage of this project (and since). All these attempts have so far have been unsuccessful, perhaps due to KAP1-chromatin interactions being indirect.
We therefore reverted to crosslinking approaches and assessed KAP1 chromatin association using KAP1 ChIP-seq, as suggested by the Referee, in KAP1-KO (negative control), WT-complemented and CCmut-complemented lines (new figure panels Fig EV4A,B).
The ChIP-seq experiment showed relatively weak KAP1 enrichment over background compared to the cleaner H3K9me3 readout. Because of this we restricted ourselves to a relatively simple analysis.
We found that KAP1 enrichment in the WT-complemented cells occurred at two types of loci: H3K9me3-positive regions and gene promoters (Fig EV4A,B). No KAP1 enrichment was observed over H3K9me3-positive regions in the CCmutcomplemented cells, indicating that KAP1 recruitment to these sites depends on KRAB binding. By contrast, KAP1 remained enriched over gene promoters in the CCmut-complemented cells, suggesting that KAP1 recruitment to promoters does not depend on KRAB binding (Fig EV4A). We are unsure of the significance of the promoter peaks, but we note that KAP1-promoter interactions involving the PHDbromodomain region of KAP1 have been reported previously (e.g. Bacon et al, 2020, now cited in the manuscript). We cannot, however, discount the possibility that the promoter peaks are an artifact of overexpression and crosslinking of the recombinant protein to open chromatin (relative expression levels in WT cells and KAP1complemented KAP1 KO cells are shown in Fig EV4C,D). It is outside the scope of this manuscript to follow up on this, but we report the finding in the Results section with appropriate caveats (p. 8). Nonetheless, the ChIP-seq data provide some additional support for the model that KRAB binding is necessary for KAP1 recruitment at H3K9me3-positive regions. Moreover, we now show with new immunoprecipitation experiments that the KAP1-SETDB1 interaction is unaffected by CCmut (see new figure panel Fig EV4E), which further supports our hypothesis that the loss of H3K9me3 in the CCmut variant is due to KAP1 not being recruited to those loci.
• Page 12/ Fig 4C+D: ' ...V293S [,] was not expressed at detectable levels in KAP1 KO HEK293T cells'. If feasible it may be worth repeating the overexpression experiment, given this mutation is part of the 4 mutations used in the functional characterization (CCmut) in the remainder of the manuscript.
We thank the reviewer for this comment. In attempting to repeat the overexpression experiment for the V293S mutant we discovered a frameshift mutation upon resequencing the original plasmid. This spontaneous frameshift mutation was probably responsible for the lack of expression of the original V293S construct in HEK293 cells. Due to time and staffing limitations, we were not able to complete a new set of luciferase reporter assays with the V293S single mutant, but this data would not have changed any of the conclusions in this paper. We have deleted the phrase stating that V293S single mutant was not expressed (p. 8).