Activity-based CRISPR scanning uncovers allostery in DNA methylation maintenance machinery

  1. Kevin Chun-Ho Ngan
  2. Samuel M Hoenig
  3. Hui Si Kwok
  4. Nicholas Z Lue
  5. Pallavi M Gosavi
  6. David A Tanner
  7. Emma M Garcia
  8. Ceejay Lee
  9. Brian B Liau

  • Curated by eLife

    eLife logo

    Evaluation Summary:

    This highly interesting manuscript will be relevant to colleagues studying cancer and those developing cancer therapies. The work describes the use of a large-scale CRISPR screen to identify mechanisms underlying resistance to the hypomethylating anti-cancer agent decitabine, which acts by inhibiting the DNA methyltransferase DNMT1. A specific focus is given to allosteric mechanisms of resistance that emerge, including those that appear to act as gain-of-function mutations in both DNMT1 and its interacting partner UHRF1. These findings showcase the power of large-scale genomic editing screens for the discovery of novel drug resistance mechanisms, which may guide the development of next-generation cancer therapies.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. The reviewers remained anonymous to the authors.)

Reviewed by

Read the full article

Listed in

Log in to save this article

Abstract

Allostery enables dynamic control of protein function. A paradigmatic example is the tightly orchestrated process of DNA methylation maintenance. Despite the fundamental importance of allosteric sites, their identification remains highly challenging. Here, we perform CRISPR scanning on the essential maintenance methylation machinery—DNMT1 and its partner UHRF1—with the activity-based inhibitor decitabine to uncover allosteric mechanisms regulating DNMT1. In contrast to non-covalent DNMT1 inhibition, activity-based selection implicates numerous regions outside the catalytic domain in DNMT1 function. Through computational analyses, we identify putative mutational hotspots in DNMT1 distal from the active site that encompass mutations spanning a multi-domain autoinhibitory interface and the uncharacterized BAH2 domain. We biochemically characterize these mutations as gain-of-function, exhibiting increased DNMT1 activity. Extrapolating our analysis to UHRF1, we discern putative gain-of-function mutations in multiple domains, including key residues across the autoinhibitory TTD–PBR interface. Collectively, our study highlights the utility of activity-based CRISPR scanning for nominating candidate allosteric sites, and more broadly, introduces new analytical tools that further refine the CRISPR scanning framework.

Article activity feed

  1. Version published to 10.7554/elife.80640 on eLife
  2. eLife

    Author Response

    Reviewer #1 (Public Review):

    Liau and colleagues have previously reported an approach that uses PAM-saturating CRISPR screens to identify mechanisms of resistance to active site enzyme inhibitors, allosteric inhibitors, and molecular glue degraders. Here, Ngan et al report a PAM-saturating CRISPR screen for resistance to the hypomethylating agent, decitabine, and focus on putatively allosteric regulatory sites. Integrating multiple computational approaches, they validate known - and discover new - mechanisms that increase DNMT1 activity. The work described is of the typical high quality expected from this outstanding group of scientists, but I find several claims to be slightly overreaching.

    Major points:

    The paper is presented as a new method - activity-based CRISPR scanning - to identify allosteric regulatory sites using DNMT1 as a proof-of-concept. Methodologically, the key differentiating feature from past work is that the inhibitor being used is an activity-based substrate analog inhibitor that forms a covalent adduct with the enzyme. I find the argument that this represents a new method for identifying allosteric sites to be relatively unconvincing and I would have preferred more follow-up of the compelling screening hits instead. The basic biology of DNMT1 and the translational relevance of decitabine resistance are undoubtedly of interest to researchers in diverse fields. In contrast, I am unconvinced that there is any qualitative or quantitative difference in the insights that can be derived from "activity-based CRISPR scanning" (using an activity-based inhibitor) compared to their standard "CRISPR suppressor scanning" (not using an activity-based inhibitor). Key to their argument, which is expanded upon at length in the manuscript, is that decitabine - being an activity-based inhibitor that only differs from the substrate by 2 atoms - will enrich for mutations in allosteric sites versus orthosteric sites because it will be more difficult to find mutations that selectively impact analog binding than it is for other active-site inhibitors. However, other work from this group clearly shows that non-activity-based allosteric and orthosteric inhibitors can just as easily identify resistance mutations in allosteric sites distal from the active site of an enzyme (https://www.biorxiv.org/content/10.1101/2022.04.04.486977v1). If the authors had compared their decitabine screen to a reversible DNMT1 inhibitor, such as GSK3685032, and found that decitabine was uniquely able to identify resistance mutations in allosteric sites, then I would be convinced. But with the data currently available, I see no reason to conclude that "activity-based CRISPR scanning" biases for different functional outcomes compared to the "CRISPR suppressor scanning" approach.

    We appreciate the reviewer’s comments and thank them for their constructive feedback. We agree with the reviewer that our claims regarding the utility of activity-based CRISPR scanning would be more strongly supported with a head-to-head comparison against a non-covalent, reversible inhibitor. To address this point, we conducted a CRISPR scanning experiment on DNMT1 and UHRF1 using GSK3484862 (GSKi), which is shown in Fig. 1e–h. We observed that the top enriched sgRNA under GSKi treatment targets H1507, which directly interacts with the drug and contributes to compound binding. (Fig. 1e,h, Supplementary Fig. 1e). Our results are consistent with previous structural and biochemical studies of these inhibitors (reported in Pappalardi, M.B. et al., Nat. Cancer 2021), in which they demonstrate that the H1507Y mutation reduces GSK3685032 (a derivative of GSK3484862) inhibition of DNMT1 by >350-fold compared to wild-type DNMT1. By contrast, the top enriched sgRNA under decitabine (DAC) treatment targets D702 in the autoinhibitory linker region (Fig. 1c). Furthermore, comparison of sgRNA resistance scores across DAC and GSKi treatment conditions reveals highly distinct sgRNA enrichment profiles (Fig. 1g). Taken together, our data suggest that these two mechanistic classes of inhibitors may exert differential selective pressures that lead to unique enrichment profiles.

    While we consider these data to strengthen our claim that activity-based CRISPR scanning can preferentially enrich for mutations in allosteric sites versus orthosteric sites, we also recognize that allosteric site mutations can be identified without the use of activity-based inhibitors, as the reviewer points out. To address this point, we have modified the text to suggest that the use of activity-based inhibitors may exert a greater bias for the enrichment of allosteric site mutations but clarifying that the enrichment of such mutations are not exclusive to the use of activity-based inhibitors.

    How can LOF mutations from cluster 2 be leading to drug resistance? It is speculated in the paper that a change in gene dosage decreases the DNA crosslinks that cause toxicity. However, the immediate question then would be why do the resistance mutations cluster around the catalytic site? If it's just gene dosage from LOF editing outcomes, would you not expect the effect to occur more or less equally across the entire CDS?

    This is an excellent point. As outlined previously above, we recognize that our gene dosage hypothesis regarding the mechanism of cluster 2 sgRNAs may lack sufficient explanation to convey our reasoning clearly, and we have added more text and data to clarify and support our claim.

    Mutations that are highly likely to lead to a nonfunctional protein product (i.e., frameshift, nonsense, splice site disrupting) are annotated as “loss-of-function” (LOF) in the text, with all other protein coding mutations designated as “in-frame.” The key insight underlying our gene dosage hypothesis is that sgRNAs targeting essential protein regions and functional domains generate greater proportions of null (i.e., knockout) mutations and undergo stronger negative selection compared to sgRNAs targeting non-essential protein regions (see Shi, J. et al., Nat. Biotechnol. 2015). This is because in-frame coding mutations in protein regions that are functionally important (e.g., DNMT1 catalytic domain) are more likely to disrupt protein function than those in non-essential protein regions. As a result, sgRNAs targeting functional protein regions are more likely to generate in-frame mutations resulting in a null allele and are thus “effectively LOF.” Importantly, the observation that sgRNAs targeting specific protein regions are more likely to lead to null mutations also implies that 1. not all CDS-targeting sgRNAs are equivalent at inducing LOF effects and 2. sgRNAs that are more effective at generating null mutations may exhibit preferential clustering within functionally important protein regions.

    In this context, we reasoned that cluster 2 sgRNAs, which target the essential catalytic domain, may be more effective at reducing DNMT1 gene dosage than other DNMT1-targeting sgRNAs because in-frame mutations generated by these sgRNAs are more likely to lead to nonfunctional DNMT1 protein. That is, cluster 2 sgRNAs may generate greater proportions of “effectively LOF” in-frame mutations that disrupt DNMT1’s essential function. Consequently, we posited that the observed clustering of these sgRNAs in the catalytic domain is likely a reflection of its functional importance. To test this idea, we transduced WT K562 cells with 6 individual sgRNAs targeting the N-terminus, RFTS domain, and catalytic domain of DNMT1, and performed genotyping on the cellular pools over 28 days (Fig. 4f). We observed that sgRNAs targeting outside of the catalytic domain exhibited increasing frequencies of in-frame mutations over time, consistent with the idea that these sgRNAs generate functional in-frame mutations that are not under strong negative selection. By contrast, catalytic-targeting sgRNAs exhibited significant depletion of inframe mutations over time, supporting the notion that in-frame mutations in essential regions are functional knockouts and thus negatively selected under normal growth conditions. Consequently, the ability of catalytic-targeting sgRNAs to generate greater proportions of null mutations would therefore make them more effective at conferring resistance through gene dosage reduction than other DNMT1-targeting sgRNAs.

    Our hypothesis implies that a large proportion of in-frame mutations generated by cluster 2 sgRNAs are functionally equivalent to LOF mutations (i.e., frameshift, nonsense, splice site disruption), and therefore neither in-frame or LOF mutations should be preferentially selected for under DAC treatment, in contrast to the positive selection of gain-of-function (GOF) in-frame mutations in cluster 1 sgRNAs. Consistent with this idea, our data indicate that the relative proportions of in-frame and LOF mutations in cluster 2 sgRNAs remain comparable across vehicle and DAC treatments (Fig. 4b). Furthermore, since the selective pressure on in-frame and LOF mutations should be similar if they are functionally equivalent, the relative proportions of in-frame versus LOF mutations in cluster 2 sgRNAs should be primarily dictated by their frequencies as editing outcomes. Consistent with this idea, the observed proportions of in-frame versus LOF mutations in cluster 2 sgRNAs under DAC treatment do not deviate significantly from their expected proportions as predicted by inDelphi (Supplementary Fig. 4c). Conversely, cluster 1 sgRNAs exhibit greater ratios of in-frame versus LOF mutations under DAC treatment than their predicted ratios from inDelphi (Supplementary Fig. 4c,d). Altogether, these data are consistent with the notion that cluster 2 sgRNAs may operate through a gene dosage reduction effect.

    In general, I found the screens, and integrative analyses, highly compelling. But the follow-up was rather narrow. For example, how much do these mutations shift the IC50 curves for DAC?

    To address this point, we derived two clonal cell lines from the screen harboring endogenous DNMT1 mutations in either the autoinhibitory linker or the RFTS domain (Supplementary Fig. 3g). We treated these cell lines, in addition to WT K562 cells, with varying concentrations of DAC and observed a partial growth rescue in the mutant cell lines relative to WT K562 cells (Fig. 3i). We also show that these mutant cell lines exhibit DAC-mediated degradation of DNMT1, consistent with our fluorescent reporter results (Supplementary Fig. 3h). To further validate whether these endogenous DNMT1 mutations confer partial resistance to DAC, we transduced WT K562 cells with vectors encoding an shRNA targeting the 3' UTR of the endogenous DNMT1 transcript and a DNMT1 overexpression vector encoding WT and mutant DNMT1 constructs (Supplementary Fig. 3i). Upon treating these knockdown and overexpression cells with varying concentrations of DAC, we again observed a partial growth rescue in the presence of mutant versus WT DNMT1 (Fig. 3j).

    What kinetic parameters have changed to increase catalytic activity?

    We performed enzyme activity assays at various temperatures with recombinant DNMT1 protein for WT and mutant DNMT1 constructs, observing that mutant DNMT1 constructs exhibit varying degrees of overactivity relative to WT DNMT1 at different temperatures (Fig. 3h, Supplementary Fig. 4f). Whereas the autoinhibitory linker mutations display consistently higher levels of activity relative to WT DNMT1 at all temperatures tested, we observed that RFTS and CXXC mutants exhibited decreasing levels of overactivity with increasing temperature (Fig. 3h). Previous studies (see Berkyurek, A.C. et al., J. Biol. Chem. 2014) have observed similar behavior with RFTS mutations, suggesting that these mutations may disrupt critical hydrogen bonds at the autoinhibitory interface that reduce the activation energy required to release DNMT1 from an autoinhibited to active conformation. Our RFTS and CXXC mutations exhibit behavior that are consistent with this hypothesis, which may explain the decreasing levels of overactivity with increasing temperature.

    Do the mutants with increased catalytic activity alter the abundance of methylated DNA (naively or in response to the drug)? It is speculated that several UHRF1 sgRNAs disrupt PPIs and not DNA binding, but this is never tested.

    While we derived clonal cell lines containing DNMT1 mutations, as noted above, it proved too difficult to compare these drug-resistant cells to naïve cells because they were cultured in the presence of DAC for 2 months, leading to large changes in DNA methylation that may confound any conclusions about the effects of the mutations alone. Additionally, the reviewer also brings up valid limitations regarding our studies on UHRF1, which also proved very difficult to biochemically purify and beyond our expertise. After some initial studies, we chose not to pursue these additional experiments further but instead prioritized the GSKi CRISPR-suppressor scan and cluster 2 studies, as suggested by the reviewers. We acknowledge these limitations in the text.

    Reviewer #2 (Public Review):

    In this manuscript, Ngan and coworkers described a CRISPER-based screening approach to identify potential variants of DNMT1 and UHRF1 that can suppress the anti-proliferation role of decitabine. In theory, such an effect can be achieved by at least two types of gain-of-activity DNMT1/UHRF1 mutants by directly boosting the enzymatic activity or by indirectly abolishing the intrinsic inhibitory activity of the DNMT1-UHRF1 axis. Through systematically targeting the DNMT1-UHRF1 reading frames with a rationally designed sgRNA library, the authors identified and characterized a few potential hotspots within multiple autoinhibitory motifs. While the approach has its merits in regard to the unbiased screening of the target proteins in living cells, there are the following serious concerns in terms of how the data were interpreted and the limitation of the approach itself as detailed below.

    (1) Although the authors identified multiple hotspots in the DNMT1-UHRF1 complex with their alterations associated with the resistance to decitabine, it is risky to argue these mutations increase DNMT1 activity simply because they are clustered within known auto-inhibitory regions. There are many alternative explanations for this observation. For instance, some mutants may allosterically alter how DNMT1 recognizes decitabine-containing vs native GpC motifs; others may recruit other proteins as modulators. The key gap here is to associate the decitabine-resistance phenotype to the loss of auto-inhibitory functions because multiple hotspots were in the auto-inhibitory regions.

    In our original manuscript, we supported our claim that gain-of-function DNMT1 mutations enhance DNMT1 activity with experimental data using purified DNMT1 protein constructs in enzyme activity assays (Fig. 3g, Fig. 4g), so our conclusion was not solely inferred from sgRNA clustering at the autoinhibitory interface, but also experimentally validated. In our revised manuscript, we provide additional experimental biochemical characterization to further support the claim that autoinhibition is weakened in the DNMT1 mutants we identified (Fig. 3h, Supplementary Fig. 4f). Moreover, we provide cellular data using clonal cell lines harboring endogenous DNMT1 mutations in addition to knockdown/overexpression experiments, demonstrating that RFTS and autoinhibitory linker mutations confer partial growth rescue to DAC treatment (Fig. 3i,j). We agree that we cannot rule out the possibility that these mutations may exert other effects that independently contribute to the observed resistance phenotype (e.g., altered CpG recognition), and we have added a statement acknowledging this limitation.

    (2) Lack of general biological relevance of the corresponding findings. Through this work, the author identified multiple DNMT1-UHRF1 variants that alter the anti-proliferation role of decitabine. However, the observation that the multiple mutants were clustered in a hotspot doesn't mean that these mutants have to act via the same mechanism. The authors seem to underestimate the complexity of how these mutants can render the same biological readouts and even haven't considered the possibility of transcriptional modulation of antagonists or agonists in the DNMT1-UHRF1. Therefore, the biological relevance of these findings remains unclear.

    We agree that although the cluster 1 mutations share a common property of increased DNMT1 activity, it does not preclude alternative mechanisms. Indeed, it is likely that these mutations have complex and nuanced mechanistic differences in the biochemical alterations underlying their observed increases in DNMT1 activity. Indeed, we have included enzyme activity data suggesting that autoinhibitory linker mutations may exhibit a different biochemical basis for increased DNMT1 activity than RFTS and CXXC mutations. That said, we did not intend to make broader claims regarding biological relevance and were instead focused on conveying that this activity-based methodology can identify gain-of-function mutations, which we directly support with experimental data. To clarify these points, we have adapted the text to more precisely convey our intended claims and have acknowledged that other complex mechanisms may also be involved.

    (3) Collectively for reasons (1) and (2), the mechanistic analysis seems only to associate the current findings with known regulatory pathways. Without detailed in vitro and in-cell characterization of the DNMT1-UHRF1 mutants, the novel regulatory mechanisms, which may exist, could be largely missed.

    We have added some additional characterization of these mutations in the revised manuscript, which have been detailed above, and we would like to note that we identified new sites in DNMT1 and UHRF1 that may be functional based off our allele analysis. However, since this manuscript is intended more as a methodology, we believe that extensively exploring novel regulatory mechanisms and their mechanism is beyond the scope of this report.

    (4) The current CRISPER-based screening approach has the technical limitation of mainly screen deletion with some exceptions for point mutations. As a result, the majority of loss/gain-of-function point mutations will be missed by the CRISPER-based screening method.

    We acknowledge that a technical limitation of this Cas nuclease-based mutational scan is that it is biased toward insertion/deletion mutations versus point mutations. However, we disagree with the reviewer’s claim that this means that the majority of the loss-/gain-of-function mutations will be missed, since insertion/deletions are often larger perturbations than point mutations and thus have stronger effect sizes in many cases. In principle, the selection modalities (e.g., activity-based inhibitors) used here — which are the primary focus of the study — can also be combined with alternative genomic editing approaches to assess distinct mutational perturbations, such as base editing for point mutations (see Lue, N.Z. et al., Nat. Chem. Biol. 2022). To acknowledge the reviewer’s concern, however, we have added additional text explicitly stating that the screen is biased against point mutations and that future integration with base editing and other mutational modalities may be useful to complement our nuclease-based approach.

    (5) The current CRISPER-based screening approach can work only in the context of living cells. As a result, robust cellular readouts are needed. The DNMT1-UHRF1 in combination with decitabine is among few suitable targets for such application.

    While running CRISPR-based screens requires robust cellular assays, the main advantage of CRISPRbased mutational scanning is the ability to mutagenize the endogenous protein target in situ and assess the effect of the perturbation in the native cellular and genomic context. Resistance to activity-based probes — and small molecules more broadly — provides a robust phenotypic readout that has been extensively used by our group and many others. Alternatively, other types of phenotypic readouts that do not rely on cell viability can also be employed with these screens, including those used to assess DNA methylation (see Lue, N.Z. et al., Nat. Chem. Biol. 2022). Given the increasingly large body of literature applying CRISPR-based screens towards a multitude of biological pathways and diverse targets, we disagree with the reviewer’s claim that only a few targets can be evaluated in such a manner.

    (6) Although the authors claim that their mutants are "gain-of-function" for DNMT1/UHRF1, they were indeed due to the loss of inhibitory regulation. It is a little disappointing because the screening outcomes still fall into the conventional expectation of the loss-of-function variants.

    We agree that the mutations are not truly neomorphic, but instead likely hypermorphic due to loss of an autoinhibitory mechanism, resulting in gain-of-function increase in catalytic activity. While discovering neomorphic mutations would be extraordinary, we do not believe that our results are disappointing since the identification of autoinhibitory mechanisms is nevertheless impactful.

    Collectively, the current status of the manuscript is short of merits in terms of the impacts of technology and biological findings.

    We respectfully disagree with the reviewer’s comment as we believe that the experimental and computational methodology may be broadly useful for the field. Indeed, we have already implemented many of the tools developed here in our current ongoing work.

  3. eLife

    Evaluation Summary:

    This highly interesting manuscript will be relevant to colleagues studying cancer and those developing cancer therapies. The work describes the use of a large-scale CRISPR screen to identify mechanisms underlying resistance to the hypomethylating anti-cancer agent decitabine, which acts by inhibiting the DNA methyltransferase DNMT1. A specific focus is given to allosteric mechanisms of resistance that emerge, including those that appear to act as gain-of-function mutations in both DNMT1 and its interacting partner UHRF1. These findings showcase the power of large-scale genomic editing screens for the discovery of novel drug resistance mechanisms, which may guide the development of next-generation cancer therapies.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. The reviewers remained anonymous to the authors.)

  4. eLife

    Reviewer #1 (Public Review):

    Liau and colleagues have previously reported an approach that uses PAM-saturating CRISPR screens to identify mechanisms of resistance to active site enzyme inhibitors, allosteric inhibitors, and molecular glue degraders. Here, Ngan et al report a PAM-saturating CRISPR screen for resistance to the hypomethylating agent, decitabine, and focus on putatively allosteric regulatory sites. Integrating multiple computational approaches, they validate known - and discover new - mechanisms that increase DNMT1 activity. The work described is of the typical high quality expected from this outstanding group of scientists, but I find several claims to be slightly overreaching.

    Major points:

    The paper is presented as a new method - activity-based CRISPR scanning - to identify allosteric regulatory sites using DNMT1 as a proof-of-concept. Methodologically, the key differentiating feature from past work is that the inhibitor being used is an activity-based substrate analog inhibitor that forms a covalent adduct with the enzyme. I find the argument that this represents a new method for identifying allosteric sites to be relatively unconvincing and I would have preferred more follow-up of the compelling screening hits instead. The basic biology of DNMT1 and the translational relevance of decitabine resistance are undoubtedly of interest to researchers in diverse fields.

    In contrast, I am unconvinced that there is any qualitative or quantitative difference in the insights that can be derived from "activity-based CRISPR scanning" (using an activity-based inhibitor) compared to their standard "CRISPR suppressor scanning" (not using an activity-based inhibitor). Key to their argument, which is expanded upon at length in the manuscript, is that decitabine - being an activity-based inhibitor that only differs from the substrate by 2 atoms - will enrich for mutations in allosteric sites versus orthosteric sites because it will be more difficult to find mutations that selectively impact analog binding than it is for other active-site inhibitors. However, other work from this group clearly shows that non-activity-based allosteric and orthosteric inhibitors can just as easily identify resistance mutations in allosteric sites distal from the active site of an enzyme (https://www.biorxiv.org/content/10.1101/2022.04.04.486977v1). If the authors had compared their decitabine screen to a reversible DNMT1 inhibitor, such as GSK3685032, and found that decitabine was uniquely able to identify resistance mutations in allosteric sites, then I would be convinced. But with the data currently available, I see no reason to conclude that "activity-based CRISPR scanning" biases for different functional outcomes compared to the "CRISPR suppressor scanning" approach.

    How can LOF mutations from cluster 2 be leading to drug resistance? It is speculated in the paper that a change in gene dosage decreases the DNA crosslinks that cause toxicity. However, the immediate question then would be why do the resistance mutations cluster around the catalytic site? If it's just gene dosage from LOF editing outcomes, would you not expect the effect to occur more or less equally across the entire CDS?

    In general, I found the screens, and integrative analyses, highly compelling. But the follow-up was rather narrow. For example, how much do these mutations shift the IC50 curves for DAC? What kinetic parameters have changed to increase catalytic activity? Do the mutants with increased catalytic activity alter the abundance of methylated DNA (naively or in response to the drug)? It is speculated that several UHRF1 sgRNAs disrupt PPIs and not DNA binding, but this is never tested.

  5. eLife

    Reviewer #2 (Public Review):

    In this manuscript, Ngan and coworkers described a CRISPER-based screening approach to identify potential variants of DNMT1 and UHRF1 that can suppress the anti-proliferation role of decitabine. In theory, such an effect can be achieved by at least two types of gain-of-activity DNMT1/UHRF1 mutants by directly boosting the enzymatic activity or by indirectly abolishing the intrinsic inhibitory activity of the DNMT1-UHRF1 axis. Through systematically targeting the DNMT1-UHRF1 reading frames with a rationally designed sgRNA library, the authors identified and characterized a few potential hotspots within multiple autoinhibitory motifs. While the approach has its merits in regard to the unbiased screening of the target proteins in living cells, there are the following serious concerns in terms of how the data were interpreted and the limitation of the approach itself as detailed below.

    (1) Although the authors identified multiple hotspots in the DNMT1-UHRF1 complex with their alterations associated with the resistance to decitabine, it is risky to argue these mutations increase DNMT1 activity simply because they are clustered within known auto-inhibitory regions. There are many alternative explanations for this observation. For instance, some mutants may allosterically alter how DNMT1 recognizes decitabine-containing vs native GpC motifs; others may recruit other proteins as modulators. The key gap here is to associate the decitabine-resistance phenotype to the loss of auto-inhibitory functions because multiple hotspots were in the auto-inhibitory regions.

    (2) Lack of general biological relevance of the corresponding findings. Through this work, the author identified multiple DNMT1-UHRF1 variants that alter the anti-proliferation role of decitabine. However, the observation that the multiple mutants were clustered in a hotspot doesn't mean that these mutants have to act via the same mechanism. The authors seem to underestimate the complexity of how these mutants can render the same biological readouts and even haven't considered the possibility of transcriptional modulation of antagonists or agonists in the DNMT1-UHRF1. Therefore, the biological relevance of these findings remains unclear.

    (3) Collectively for reasons (1) and (2), the mechanistic analysis seems only to associate the current findings with known regulatory pathways. Without detailed in vitro and in-cell characterization of the DNMT1-UHRF1 mutants, the novel regulatory mechanisms, which may exist, could be largely missed.

    (4) The current CRISPER-based screening approach has the technical limitation of mainly screen deletion with some exceptions for point mutations. As a result, the majority of loss/gain-of-function point mutations will be missed by the CRISPER-based screening method.

    (5) The current CRISPER-based screening approach can work only in the context of living cells. As a result, robust cellular readouts are needed. The DNMT1-UHRF1 in combination with decitabine is among few suitable targets for such application.

    (6) Although the authors claim that their mutants are "gain-of-function" for DNMT1/UHRF1, they were indeed due to the loss of inhibitory regulation. It is a little disappointing because the screening outcomes still fall into the conventional expectation of the loss-of-function variants.

    Collectively, the current status of the manuscript is short of merits in terms of the impacts of technology and biological findings.

  9. Version published to 10.1101/2022.05.14.491958v1 on bioRxiv