Orthogonal proteogenomic analysis identifies the druggable PA2G4-MYC axis in 3q26 AML

The overexpression of the ecotropic viral integration site-1 gene (EVI1/MECOM) marks the most lethal acute myeloid leukemia (AML) subgroup carrying chromosome 3q26 abnormalities. By taking advantage of the intersectionality of high-throughput cell-based and gene expression screens selective and pan-histone deacetylase inhibitors (HDACis) emerge as potent repressors of EVI1. To understand the mechanism driving on-target anti-leukemia activity of this compound class, here we dissect the expression dynamics of the bone marrow leukemia cells of patients treated with HDACi and reconstitute the EVI1 chromatin-associated co-transcriptional complex merging on the role of proliferation-associated 2G4 (PA2G4) protein. PA2G4 overexpression rescues AML cells from the inhibitory effects of HDACis, while genetic and small molecule inhibition of PA2G4 abrogates EVI1 in 3q26 AML cells, including in patient-derived leukemia xenografts. This study positions PA2G4 at the crosstalk of the EVI1 leukemogenic signal for developing new therapeutics and urges the use of HDACis-based combination therapies in patients with 3q26 AML.

(3) The variability in RIME assays is such that typically at least 4 biological replicates are recommended to obtain statistically significant (p-values) fold changes between pull-downs.As the number of replicates for EVI1 and the corresponding IgG control was 2 in each pull down it is difficult to know the statistical confidence of the enriched proteins.Could the authors reanalyse the data using, for example, label free quantitative analysis or if possible isobaric labelling (TMT) .Ideally this would be done with a minimum of 3 biological replicates.
(4)The PDXL samples were established by tail injection of patient (AML) mononucleated cells (PR#003).More detail is required on the procedure to isolate human cells from mice for RIME analysis.This is important as there is considerable sequence identity between human and mouse proteins and contamination from mouse cells will compromise the data.For PR#003 samples are unique peptides unique for the protein or unique for human in these samples?Could the authors please detail how this sample was analysed.
(5) It is confusing what the figures in tables for each pull down refer to (eg What do the figures in columns D, H & L in table HNT34 Enriched Protein List for Evi-1 RIME Final Assay 35210 (HNT34) refer to ?
Reviewer #2: Remarks to the Author: The manuscript by Marchesini et al. describes the identification of HDAC inhibitors as drugs that suppress 3q26 AMLs cell viability, presumably by lowering high levels of EVI1 in these cells, and through a mechanism that it is mediated by the PA2G4-MYC axis.I am not convinced that the work as presented deserves publication in Nature Comm.
• HDAC inhibitors have already been explored as potential treatments for AML, even reaching clinical stage.The novelty in this work might be that HDAC inh might be selective and more efficacious for a subset of AML, 3q26 AMLs, because the some HDAC inh lower EVI1, a driver of cell proliferation for these cancer cells.However, although the data indicates that HDAC inhibitors indeed lower EVI1 levels in EVI1 high AML cells, the evidence is quite underwhelming that the effect seen in cell viability by HDAC inh is dependent on reducing EVI1 levels, because there is no difference in the potency of the HDAC inh tested between EVI1 high and low AML cells: Fig 2A and B show that cells TF1 and MUTZ-3 have much lower levels of EVI1 than the other AML cell lines tested, however the IC50s for cell proliferation shown in Fig 2G are essentially the same, suggesting that EVI1 cellular levels do not matter for the potency of HDAC inh in these AML cells.In addition, Fig 3E shows AUC data for the effect of AR-42, belinostat, entinostat for 5 EVI1High and 7 EVI1Low AML cell lines and the actual differences are very small, although somewhat statistically significant.Same for patient derived cells in Fig 4C .In addition, Fig S1 D shows that, as a class, HDAC inh are not selective for EVI1 high cells.In general, the authors do not show convincing data that the pharmacological effects of HDAC inhibitors on cell viability are larger for AVI1 high vs low cells, and draw a clear conclusion that the effects seen in cell growth are due to AVI1 downregulation.
• The discovery that a PA2G4 inhibitor downregulates AVI1 appears to be novel, and in fact, the pharmacological effects on cell viability shown for a PA2G4 inhibitor appears to be more dependent on AVI1 levels (Fig S9 A) than HDAC inh.Puzzling that the authors never tested the combination of HDAC inh and the PA2G4 inhibitor to test whether there might be some additive or synergistic effects.S1-1 headers refer to; the screen was done in duplicate.What groups are compared to calculate p-values from the screening data?The POC index includes a differences in % cell viability to control and an Adj.P. How is this Adj.P calculated?

• It is unclear what the p-values in Table
• Line 100: "The HDACis AR-42, belinostat, trichostatin A, and entinostat preferentially affect the proliferation of EVI1High AML cells compared to EVI1low".What selectivity criteria was used to identify selective EVI1 high vs low compounds?Did the criteria include both potency (IC50) and efficacy (% cell viability).
• Line 110: "Most (56.2%) of the HDACis sampled in the CMap libraries scored among the top 250 hits (P = 0.0006)".I am assuming this statement refers to the cell proliferation screen, so I am then confused as to what is plotted in Fig 1E .In addition, there is no mention in the text to the panels for the other compound classes.Shouldn't the y-axis be % cell viability in Fig 1E?.The legend seems to suggest that the x-axis is number of perturbagenes in the connectivity map.It is not clear how many total number of perturbagenes in the connectivity map: 2836 total or 2836 small molecules?From the x-axis, it might seem that there are >4000 perturbagenes total.
• Line 158: The IC50 values have a high number of significant digits.Are their measurements really that precise?What are the errors for the IC50 values?.In this regard, the authors should include the number of biological and technical replicates whenever they show error bars, eg.Fig 2G .• I don't understand Fig 1F .What is the p-value?Is it an enrichment score for the % of compounds in each target class that were considered active (top 90% by POC index) in the cell viability screen?• In Fig5G and H, there is no comparison of MYC levels in AVI1 high vs low cells.Is there a correlation between the levels of both proteins in AML cancer cells?that would help demonstrate that the pharmacological effect of HDAC inh on MYC levels in these AVI1 high cells is through downregulation of AVI1 and not directly on transcription regulation of MYC bu HDAC inh, which has been shown before.
• Also, there is not data shown for the effects of WS6 inhibited cell viability in AML EVI1 High vs low cell lines and clinical samples to demonstrate that the pharmacological effects of PA2G4 inh on 3q26 AMLs cell viability is indeed EVI1 dependent.
Reviewer #3: Remarks to the Author: In this study Dr. Marchesini and co-authors used integrated high-throughput drug screening and expression profile analysis to identify selective inhibitors of EV1/MECOM signaling in AML with 3q26 abnormalities, one of the AML subtypes with the most dismal outcome and lacking targeted and effective therapies.HDACIs were demonstrated potent repressors of EVI1 and further examined in downstream analyses.
Major comments Page 5, line 87: the authors should specify that the MOLM1 cell line used for the initial screen, although carrying an inv(3) leading to MECOM (EVI1) overexpression, is a chronic myeloid leukemia and not acute myeloid leukemia.
In order to find candidates for effective inhibition in AML cells with EVI1 deregulation, the authors intersected the results from the high-throughput drug screening with those from the gene expression differential analysis of TF-1 cells with siRNA targeting EVI1.However, only one cell line, TF-1, not included in the drug screening, was analyzed.An alternative approach may be to knock-down EVI1 in MOLM1 or UCSD-AML1 cells, perform transcriptome sequencing and then compare the gene expression data with the results from the drug screening and from the experiment in TF1 cells.Moreover, in order to strengthen their results, the authors could also analyze the public available data from the Cancer Dependency Map in AML cell lines with EVI1 overexpression and define a common list of targets.
The authors hypothesized that HDACIs can impair the function of the EVI1 co-transcriptional complex and profiled by RNAseq two AML cell lines with high expression of EVI1 post 16 hours treatment with AR-42 and entinostat and found a strong association with MYC signature genes.One key experiment that is missing to demonstrate the deregulation of the EVI1 transcriptional complex is the analysis of EV1 chromatin-binding together with H3K27ac profiling by ChIP-seq (or CUT&RUN) in cells pre and post treatment.
The scRNAseq analysis on follow-up samples provides interesting findings on how the combination azacitidine and entinostat targets stem and progenitor cells, however, as the authors are aware, it does not distinguish the effect of the single drugs.Can the authors perform cell viability assay and single cell RNAseq on normal CD34+ overexpressing EVI1 (for example by viral transduction) or in PDX cells before and after treatment with the combination and single azacitidine and entinostat?
We would like to express our appreciation for the thoughtful comments provided by the Reviewers.Their feedback has influenced our study, guiding us in addressing important aspects and refining our methods and interpretations.In response to their feedback, we have prepared a point-by-point rebuttal, which will be presented below.We would like to acknowledge that we have created n=35 figures panels, intended to clarify specific points raised by the Reviewers.However, upon careful consideration, we decided to include only part of these data in the revised version of the manuscript.The remaining data are presented here as part of our rebuttal.

Reviewer #1:
The editors have specifically requested I review only the mass spectrometry (RIME) related aspects of this study.
The authors have used RIME (rapid immunoprecipitation and mass spectrometry of endogenous proteins) to interrogate protein complexes in which EVI1 was the bait protein and the negative control unrelated IgG.A number of proteins were identified from three samples (2 cell lines HNT34 & UCSD/AML1 and 1 PDLX model) that were enriched over the negative controls.There are some concerns that should be addressed to improve this section of the manuscript.
(1) The raw MS (thermo QE .rawfile) data is not available and without access to the PEAKS software it is difficult to evaluate the quality of the data.Could the authors please provide access to the Thermo .rawfiles?
We are pleased to inform you that we can provide access to the Thermo.rawfiles at: ProteomeXchange title: RIME of EVI1 in HNT34 and UCSD/AML1 cell lines and one primary sample to identify interactors ProteomeXchange accession: PXD042440 PubMed ID: Not applicable Publication DOI: Not applicable Project Webpage: http://www.ebi.ac.uk/pride/archive/projects/PXD042440FTP Download: ftp://ftp.pride.ebi.ac.uk/pride/data/archive/2023/05/PXD042440 (2) Could the authors detail if the RIME assays are biological replicates (at least different passages of the cell lines/different mice).
In this work we wanted to explore the native interactors of EVI1 by leveraging the RIME assay.For this purpose, we used the HNT34 and UCSD/AML1 cell lines, as well as the clinical PR#003 sample.Hence, for this experiment, the biological replicates representative of a 3q26 AML model are equal to three (n=3).All these models exhibited the t(3:3)(q21;q26) genomic rearrangement, and each experiment was carried out in technical replicates (n=2).
(3) The variability in RIME assays is such that typically at least 4 biological replicates are recommended to obtain statistically significant (p-values) fold changes between pull-downs.As the number of replicates for EVI1 and the corresponding IgG control was 2 in each pull down it is difficult to know the statistical confidence of the enriched proteins.Could the authors reanalyse the data using, for example, label free quantitative analysis or if possible isobaric labelling (TMT).Ideally this would be done with a minimum of 3 biological replicates.
We thank the Reviewer for bringing attention to this aspect as it gives us the opportunity to better clarify our data analysis workflow.RIME assays have been successfully used in various research fields to study protein interactions, functions, and disease mechanisms 1,2,3,4,5,6,7,8,9,10 .The data obtained from these assays, even with a limited number of replicates, have helped to depict the protein interactome and potential biological pathways involved in different cancer models 11,12,13 .
In our analysis, we focused on proteins that were present in the pull-down sample but absent in the IgG control.This is consistent with the literature in this field.In Supplementary Table S4-1, we report the lists of uniquely identified proteins for all samples after removing proteins present in the IgG control and other proteins considered as background (below the spectral count of 5).This list was generated by averaging the spectral counts for common elements as identified in replicate R1 and R2 pull-downs for each model.The averaged spectral counts between biological experiments are also reported in the table (column E).Excluding the identification of the bait protein EVI1, a total of 155, 107, and 117 proteins were identified in PR#003, HNT34, and UCSD/AML1 samples, respectively (Figure R1A).Then, only proteins that were common to all three biological replicates were considered for further investigation.This resulted in 68 common targets, excluding EVI1, as reported in Table S4 Hence, in our original draft, we did not perform a fold change analysis, and it was not our intent to do so.The goal was consistently to identify a set of EVI1 interactors using methodological approaches that have been successfully explored in the literature.However, as suggested by the reviewer, we have reanalyzed our data using a label-free quantification (LFQ) method.In Figure R1B, we report the correlation analyses based on the protein LFQ intensities between biological and technical replicates.We observed a robust correlation of LFQ signals among both technical (r = 0.9493 PR#003; r = 0.9668 UCSD/AML1 and r = 0.9533 HNT34) and biological (r = 0.8981 HNT34-UCSD/AML1; r = 0.8548 HNT34-PR#003 and r = 0.8276 UCSD/AML1-PR#003) replicates, indicating a highly reproducible, relative LFQ between samples.
For a quantitative analysis, we included our technical and biological replicates (n = 6) in a group named "3q26 AML" and compared them to the group including all the correlative IgG controls (n=6), named "Control".The Thermo raw files were analyzed using MaxQuant (MQ) version 2.4.2.0.The LFQ intensities of proteins from MQ analysis were imported and filtered for reverse identifications (false positives), contaminants, and proteins "only identified by site".Data were transformed to log2 scale.Then, we imputed missing values and replaced them from a normal distribution.The protein quantification and calculation of statistical significance were performed with a two-sample t-test and with a permutation-based correction controlled by using an FDR threshold of 0.05.A protein was considered as EVI1's interactor if the difference between the "3q26 AML" and "Control" groups was statistically significant (P < 0.05), the fold change was 4, and it was identified with a minimum of two peptides.Using this approach, we identified 147 proteins that were enriched in 3q26 AML samples with a fold change greater than 4 (Table S4-4).We then asked whether the group of EVI1-MYC interactors identified by combining RIME and ssGSEA (as described in the manuscript and presented in Figure S9D and Table S4) were also identified using this alternative MS data analysis.As shown in Figure R1C and presented in this new version of the manuscript in Figure S9E, we found that 10 out of 11 interactors scored in the LFQ analysis (RACK1, scored as an EVI1 interactor by RIME, was not confirmed by this analysis).Additionally, we observed that, as in the case of the RIME analysis, EVI1 coregulators identified previously using isotope labeling-based MS 14 were represented among the 3q26 AML enriched proteins (Figure R1C).
To further validate the results obtained both by RIME and LFQ analysis, we performed biochemical co-immune precipitation (co-IP) experiments to confirm EVI1 and PA2G4 binding.We chose the HNT34 cell line as a 3q26 AML model based on the relatively higher expression level of EVI1 (Figure 2A and 4D) in the effort to compensate for the difference in sensitivity between co-IP and RIME.We observed a reciprocal co-immunoprecipitation of EVI1 and PA2G4.We repeated a co-IP in 293T cells after an ectopic overexpression of both EVI1 (NM_005241.3) and PA2G4 (Gene BankBC001951.1)open reading frames and confirmed, also in this setting, a reciprocal co-IP between EVI1 and PA2G4 (Figure R1D, included in the manuscript, Figure S9G).
Despite we are aware that RIME is an exploratory experimental approach, the selection of PA2G4 was based on multiple orthogonal evidences, including its potential significance in the context of 3q26 leukemia and its modulation by HDAC inhibitors.
• First, PA2G4 was identified as a common protein across three 3q26 leukemia models, including primary-derived cells.We provided evidence of a biochemical interaction between EVI1 and PA2G4 interaction that corroborates co-localization studies (Figure S9F).In addition, as reported in Table S4-2 of our manuscript PA2G4 was part of a long list of EVI1-interactors in different disease models (n=236) of wild type EVI1 transduced 293T 15 .• Second, PA2G4 is part of MYC-dependent signatures, and both the Myc pathway and PA2G4 were modulated by HDAC inhibitors in 3q26 AML models.• Third, overexpression of PA2G4 rescued 3q26 AML from the HDACi-induced phenotype, suggesting its involvement in the mechanism of action of these molecules.
In conclusion, we believe that RIME assay, despite its innate limitations that are similar to other approaches, provides valuable insights into the protein interactome and potential biological mechanisms given the intersectionality of this approach with the rest of the study.We will continue to refine our methods and explore opportunities to incorporate alternative analyses in future studies.
(4) The PDXL samples were established by tail injection of patient (AML) mononucleated cells (PR#003).More detail is required on the procedure to isolate human cells from mice for RIME analysis.This is important as there is considerable sequence identity between human and mouse proteins and contamination from mouse cells will compromise the data.
For PR#003 samples are unique peptides unique for the protein or unique for humans in these samples?Could the authors please detail how this sample was analysed.
We thank the Reviewer for this comment and concern.RIME analysis was carried out referring to human protein database.In fact, PDXL samples (PR#003) used in this specific context were established by tail injection of patient leukemic cells into mice.Subsequently, the cells were transplanted subcutaneously to ensure an adequate supply of protein for analysis.These steps were taken to maximize the representation of human proteins in the samples and to allow for sufficient protein production.The transplanted tumor maintained the original translocation, as shown by FISH analysis.
We apologize if this information was not explicitly stated in the text, but it was presented in Figure S7A and S7B of original version of the manuscript (now in Figure S9A-B) where the experimental procedure and analysis workflow are described in more detail.
( We apologize for the missing information.The file mentioned by the Reviewer, sent during the revision, contains the lists of uniquely identified proteins in each RIME experiment (HNT34 sample in this case).Proteins considered as background (spectral count < 5) and proteins found in the negative control (IgG) were previously filtered.Lists going from columns B to D and J to L were obtained by overlapping candidate peptides identified in the replicates R1 (B to D) and R2 (J to L) and their correlative IgG negative controls.The list going from F to H contains proteins that were identified in both replicates.Gene names (columns B, F and J) and protein names (columns C, G and K) are indicated.Columns D and L accommodate the spectral counts of uniquely identified proteins (with a spectral count > 5) of the replicates R1 and R2 respectively.The column H contains instead the averaged spectral count for common elements as identified in R1 and R2 pull-downs.

Reviewer #2:
The manuscript by Marchesini et al. describes the identification of HDAC inhibitors as drugs that suppress 3q26 AMLs cell viability, presumably by lowering high levels of EVI1 in these cells, and through a mechanism that it is mediated by the PA2G4-MYC axis.
HDAC inhibitors have already been explored as potential treatments for AML, even reaching clinical stage.The novelty in this work might be that HDAC inh might be selective and more efficacious for a subset of AML, 3q26 AMLs, because the some HDAC inh lower EVI1, a driver of cell proliferation for these cancer cells.However, although the data indicates that HDAC inhibitors indeed lower EVI1 levels in EVI1 high AML cells, the evidence is quite underwhelming that the effect seen in cell viability by HDAC inh is dependent on reducing EVI1 levels, because there is no difference in the potency of the HDAC inh tested between EVI1 high and low AML cells: We understand the concern of the Reviewer, however we would like to point out that clinical and preclinical studies have demonstrated that different subtypes of AML may respond differently to specific HDAC inhibitors (HDACis) based on the presence, for example, of leukemia-associated fusion proteins (LAFPs) or specific genetic alterations rather than solely relying on the expression level of the specific target 16,17 .
For instance, studies have shown that certain LAFPs, such as PLZF::RARα, PML::RARα, and RUNX1::RUNXT1 16,17,18 can sensitize AML cells to HDAC inhibition, leading to improved response rates with specific HDACis.This suggests that the recruitment of HDACs to specific promoters through the interaction of fusion proteins can enhance the susceptibility of cells to HDAC inhibition.Additionally, the response to different HDACis can be influenced by specific molecular pathways.Belinostat promotes an anti-leukemia effect in several AML cell lines, especially in acute promyelocytic leukemia and, in combination with all-trans retinoic acid, accelerates granulocytic differentiation 19 .However, in this case, the mechanism that explains why the chromosomal translocation t(15;17) bearing PML::RARα fusion differs in susceptibility to HDACis has not been investigated.Entinostat instead degrades FLT3 by inhibiting the chaperone protein 90 in AML cells 20 and reduces MYC transcription in PICALM::MLLT10 or PML::RARα rearranged AML cell lines 21 .Unfortunately, in phase I 22 or phase II study of entinostat 22,23 , with and without azacitidine, patients were not segregated by cytogenetic mutational status, limiting any potential conclusion of entinostat activity on specific AML subgroups.This appears to be a more general limitation for AML trials.
It is important to note that the understanding of the precise mechanisms underlying the differential responses to HDACis in various AML subgroups, including those with specific chromosomal translocations or genetic alterations like EVI1, is still limited.
That said, is absolutely possible that other mechanisms are involved in cell-line specific response to HDACis.However, in this version of manuscript we further provided evidence that the expression of EVI1 sensitize cells to HDAC inhibition (Figure R2 A-H).
These new data are presented in Figures 3E-I, S4E and S8C-D in the new version of the manuscript.Furthermore, we included addition 17 primary cases to our study (Figure 4C) that confirmed our original results.
In addition, Fig 3E shows AUC data for the effect of AR-42, belinostat, entinostat for 5 EVI1High and 7 EVI1Low AML cell lines and the actual differences are very small, although somewhat statistically significant.Same for patient derived cells in Fig 4C.
The small differences between EVI1 Low and EVI1 High are somewhat expected, considering the potency of HDACis in several tumor models including AML.We acknowledge that the observed differences in drug response may be influenced by multiple mechanisms driven by HDACs, potentially resulting in varying responses even within the EVI1 Low group.For example, we have demonstrated the presence of two distinct populations in MOLM1 cells (we excluded contamination by repeating this observation in multiple STR profiled batches): one that is EVI1-positive, characterized by large nuclei and prominent nucleoli, and another consisting of EVI1-negative cells, which are smaller in size and exhibit weak or no EVI1 protein expression (Figures 2B, 3G-H).We sorted MOLM1 cells based on this characteristic and subsequently tested HDAC inhibitors.In this isogenic model, HDAC inhibitors displayed greater efficacy in MOLM1-EVI1 High cells compared to MOLM1-EVI1 Low cells, which, at certain concentrations, still exhibited sensitivity to HDAC inhibition (Figure 3I).
The inclusion of additional data, further described below, in Figures 3E-I, S4E, S8C-D and Figure 4C) and the demonstration of differences in drug response between EVI1 High and EVI1 Low populations further support our findings.These results suggest that there may be distinct molecular characteristics and sensitivities to HDACis in the EVI1 High subgroup, which can contribute to the observed differences in drug response.
In addition, Fig S1 D shows that, as a class, HDAC inh are not selective for EVI1 high cells.
In general, the authors do not show convincing data that the pharmacological effects of HDAC inhibitors on cell viability are larger for AVI1 high vs low cells and draw a clear conclusion that the effects seen in cell growth are due to AVI1 downregulation.
We appreciate the Reviewer's input but respectfully disagree with the statement that our data does not demonstrate the differential effects of HDAC inhibitors on EVI1 High vs. EVI1 Low cells or support the conclusion that the observed effects on cell growth are due to EVI1 downregulation.First, we have shown that the absence of EVI1 leads to a significant delay in cell growth in 3q26 AML, indicating its role in promoting cell proliferation.In this version of the manuscript we included a second EVI1 silenced 3q26 AML cell lines (Figure R3A Secondly, HDAC inhibitors are known to have diverse effects 24 and their response is not expected to be a simple binary "yes" or "no" outcome, as seen with certain targeted tyrosine kinase inhibitors.We understand that the data presented in Figure S1D may not show a perfect segregation between EVI1 High and EVI1 Low cells in terms of their response to HDAC inhibitors.In Figure S1D, we have indeed tested both selective and pan-HDAC inhibitors of different classes (e.g selective HDAC inhibitors pyroxamide, tacedinaline, ISOX, WT-161, entinostat, trichostatin-a, Merk60, mocetinostat, vorinostat, NCH-51, apicidin, droxinostat and others; pan-HDAC inhibitors panabinostat, HC toxin, givinostat, belinostat, phenylbutyrate, dacinostat and others).The variability in selectivity among HDAC inhibitors can indeed contribute to the observed differences in the degree of the effect.The diverse selectivity profiles of HDAC inhibitors can result in differential modulation of histone and nonhistone proteins, leading to distinct cellular responses.Some inhibitors may primarily target specific HDAC isoforms associated with critical pathways in the context of EVI1 expression and 3q26 status, thereby resulting in a more pronounced effect.Nevertheless, it is important to consider that the responses observed are relative and clearly demonstrate a trend towards enhanced sensitivity in EVI1 High cells.We discussed this aspect more explicitly in the revised manuscript and in the answers above.
To meet the Reviewer' expectations, we have indeed included two new experimental models to investigate the role of EVI1 and the activity of HDAC inhibitors.As described above, we isolated MOLM1-EVI1 High cells from MOLM1 cells and showed their greater dependency on the HDAC machinery compared to MOLM1-EVI1 Low cells (Figure 3G-I).We overexpressed EVI1 in HL60 cells and demonstrated that EVI1-cells are more sensitive to HDAC inhibitors compared to non-expressing cells (Figure 3E-F and S4E).Finally, we have included additional new primary cases that collectively recapitulate the data presented in the original submission (Figure 4C and Table S2).
In summary, we conducted five different types of comparisons, involving AML cell lines, inducible models (U937T), stable overexpression (HL60), isogenic models (MOLM1 EVI-high vs. MOLM1-EVI1 Low and primary samples.These comparisons consistently indicate that the 3q26 status/EVI1 expression sensitizes cells to HDAC inhibition, resulting in a reduction in EVI1 levels.Furthermore, this outcome is, in part, mediated by PA2G4 (Figure 7B).Importantly, we observed the same preferential activity toward EVI1 High cells with a PA2G4 inhibitor, suggesting a potential avenue for drug development.
The discovery that a PA2G4 inhibitor downregulates AVI1 appears to be novel, and in fact, the pharmacological effects on cell viability shown for a PA2G4 inhibitor appears to be more dependent on AVI1 levels (Fig S9 A) than HDAC inh.Puzzling that the authors never tested the combination of HDAC inh and the PA2G4 inhibitor to test whether there might be some additive or synergistic effects.
We appreciate the reviewer's recognition of the novelty in our discovery of a PA2G4 inhibitor downregulating EVI1.We have indeed conducted the suggested experiment and evaluated the combination of HDAC inhibitors and the PA2G4 inhibitor.The results demonstrate an additive/synergistic effect on cell viability when combining HDAC inhibitors with the PA2G4 inhibitor WS6.These findings support the idea that PA2G4 may indeed act as a mediator of the HDAC response.These findings are shown in Figure R3D, and included as Figure S11B in the new version of the manuscript.S1-1 headers refer to; the screen was done in duplicate.What groups are compared to calculate p-values from the screening data?The ΔPOC index includes a differences in % cell viability to control and an Adj.P. How is this Adj.P calculated?

It is unclear what the p-values in Table
We apologize to the Reviewer for not being clearer about the screening procedure in our methods section and Table S1 Line 100: "The HDACis AR-42, belinostat, trichostatin A, and entinostat preferentially affect the proliferation of EVI1High AML cells compared to EVI1low".What selectivity criteria was used to identify selective EVI1 high vs low compounds?Did the criteria include both potency (IC50) and efficacy (% cell viability).
To compare the effect of single agents between EVI1 High and EVI1 Low models (Figure S1D) we choose the Area Under the fitted dose response Curve (AUC) criterion.AUC improves the predictive accuracy for classifying samples as sensitive or resistant compared to more traditional metrics such as IC50, or % cell viability.A substantial body of literature has concluded that AUC, a parameter that combines potency and efficacy into a single measure 25, 26 , is robust approach and surpasses response metrics when the goal is to compare a single drug across cell lines exposed to identical dose ranges.For example, IC50 presumes a typical sigmoidal shape in dose-response curves, with there's no growth inhibition in the absence of the compound and complete growth inhibition at high compound doses.This assumption doesn't distinguish between samples that reach 50% growth inhibition at the same dose, even if one of the samples achieves significantly greater growth inhibition at higher doses.Consequently, we consistently used AUC in experiments requiring the categorization compounds sensitive or resistant in EVI1 High and EVI1 Low in cell lines (Figure 3J • Line 110: "Most (56.2%) of the HDACis sampled in the CMap libraries scored among the top 250 hits (P = 0.0006)".I am assuming this statement refers to the cell proliferation screen, so I am then confused as to what is plotted in Fig 1E .In addition, there is no mention in the text to the panels for the other compound classes.Shouldn't the y-axis be % cell viability in Fig 1E?.The legend seems to suggest that the x-axis is number of perturbagenes in the connectivity map.It is not clear how many total number of perturbagenes in the connectivity map: 2836 total or 2836 small molecules?From the x-axis, it might seem that there are >4000 perturbagenes total.
A connectivity mapping approach was used to prioritize small molecules as previously described 27 .Top 100 up-and down-regulated probes based by the signal-to-noise ratio, defining an EVI1 "On" versus an "Off" change from the GSE16238 dataset were submitted to the cMap-Touchstone query version 1.1 (https://clue.io/query) of 8864 perturbagens, of these 2836 are compounds.In CMap, the summary score is a perturbagen-centric measure of connectivity that summarizes the results observed in individual cell types, while the connectivity score (in our model EVI1 "Off" score) is a quality control measurement comparing an observed enrichment score from EVI1 gene set to reference gene sets, ranging from -100 to 100 and, for perturbagenes with a positive connectivity score > 0 (n= 4187), presented as Log10 (2= highest correlation with an EVI "Off" state).In this regard, the y axis is correct since it refers to a compound's probability of mimicking an EVI1 "Off" state.The x axis indicates n= perturbagenes.We clarified this point in the method section and in the figure legends.
• Line 158: The IC50 values have a high number of significant digits.Are their measurements really that precise?What are the errors for the IC50 values?In this regard, the authors should include the number of biological and technical replicates whenever they show error bars, eg.Fig 2G.
We used a digital dispenser TECAN D300e for small molecules dispensing, and a peristaltic cells dispenser for in vitro assays ensuring an accurate measurement of the IC50 and AUC values.The accuracy of the D300e is specified as ±0.2% of the dispense volume, with a dead volume of 2 µL.This means that the D300e can dispense volumes of liquid with an accuracy of ±0.02 µL.This information was mentioned in the methods section of the manuscript.Additionally, in this version of the manuscript, we have incorporated the standard deviation to indicate the range of small molecule activity, providing a measure of variability in our measurements.The number of biological and technical replicates was present in the figure legends of the original version of the manuscript.We have included the error for IC50 value in the manuscript.

I don't understand Fig 1F.
What is the p-value?Is it an enrichment score for the % of compounds in each target class that were considered active (top 90% by ΔPOC index) in the cell viability screen?
The analysis presented in Figure 1F of the submitted version of our manuscript is now the Figure S2B panel and derives from the results obtained from the in silico small molecule screening.The goal was to determine if the cMap enrichment of HDACi or histamine, glucocorticoid, and acetylcholine receptor antagonists (as examples of groups of molecules in the library with comparable or higher size than HDACi).To check if the rank of these classes of molecules was specific or not, we asked if it could be statistically significant with respect to random sets.To address this challenge, we created random sets of the same size of the class of molecule of interest (n = 32 in the case of HDACi, n = 52 for histamine receptor antagonists, n = 47 for glucocorticoid receptor antagonists and n = 66 for acetylcholine receptor antagonists).Based on the fact that the generated artificial family of molecules had undefined distributions, we leveraged a statistical approach based on the parametric bootstrap method 28 applied to gene-set analysis as described in the study of Coombes et al. 29 obviously referred in our case to molecule-sets.This method allows us to estimate the uncertainty of a p-value.We obtained rank values derived from our random sets, which we compared to the one of interest, using the Wilcoxon test.By following the bootstrap approach, we averaged these results to obtain a p-value whose stability was checked by repeating this procedure 100 times.We did not adjust the p-values for multiple comparisons for the following reasons: • The results could have been influenced by the high number of tests, and thus selfdefeating to the aim of evaluating p-values robustness and stability.• The multiple comparisons had a common data set (rank of molecule of interest), and thus the analysis was not based on independent measurements, a mandatory requirement of the p-value adjustment procedure.
Therefore, Figure S2B shows the distribution of p-values obtained by independent runs of the method described above for the indicated groups of molecules.
In Fig5G and H, there is no comparison of MYC levels in AVI1 high vs low cells.Is there a correlation between the levels of both proteins in AML cancer cells?that would help demonstrate that the pharmacological effect of HDAC inh on MYC levels in these AVI1 high cells is through downregulation of AVI1 and not directly on transcription regulation of MYC bu HDAC inh, which has been shown before.
We performed the experiment suggested by the reviewer and quantified the levels of MYC in the AML cell line available in the laboratory.This data (Figure R3E) is now presented in Figure S8A.Overall, the data show different MYC levels in 3q26 AML, with some cell lines representing clear exceptions.If MYC was the primary target of the HDAC dependency, one would have expected greater activity in MYC High cells.However, this is not the case, Figure S8B (Figure R3F) but, as expected, it is for JQ1 an inhibitor of MYC superenhancers (Figure R3G or S8F in the new version of the manuscript).Instead, when we distinguish the response to HDACis based on the 3q26 status/EVI1 High groups, there is indeed statistical significance as shown in Figure 3J and Figure 4C.Furthermore, conditional expression of EVI1 leads to an enhanced anti-proliferative effect of HDACis, which is not the case when MYC is overexpressed.Moreover, genetic suppression of EVI1 leads to a reduction in MYC levels, suggesting that the loss of EVI1 contributes to the decrease in MYC in the 3q26 model.
Also, there is not data shown for the effects of WS6 inhibited cell viability in AML EVI1 High vs low cell lines and clinical samples to demonstrate that the pharmacological effects of PA2G4 inh on 3q26 AMLs cell viability is indeed EVI1 dependent.
We appreciate the Reviewer's feedback on this topic.In response, we assessed the effects of WS6 on cell viability in AML primary samples with varying levels of EVI1 expression (EVI1 High n= 8 or EVI1 Low , n= 15) as shown in the R3H, included as Figure 7B of the new version of the manuscript.
The results obtained from these experiments support the idea of a dependence of 3q26 AML on the PA2G4 protein.These results also indicate that further optimization of the PA2G4 protein could potentially lead to the development of drug-like molecules.
Reviewer #4: In this study Dr. Marchesini and co-authors used integrated high-throughput drug screening and expression profile analysis to identify selective inhibitors of EV1/MECOM signaling in AML with 3q26 abnormalities, one of the AML subtypes with the most dismal outcome and lacking targeted and effective therapies.HDACIs were demonstrated potent repressors of EVI1 and further examined in downstream analyses.

Major comments
Page 5, line 87: the authors should specify that the MOLM1 cell line used for the initial screen, although carrying an inv(3) leading to MECOM (EVI1) overexpression, is a chronic myeloid leukemia and not acute myeloid leukemia.
In this version of the manuscript, we have accurately defined the MOLM1 cell line as being established from a blastic transformation of chronic myeloid leukemia (CML).We have further substantiated this characterization by conducting a karyotype analysis (Figure R3I), which confirmed the presence of a complex karyotype, as previously reported in 30 ." In order to find candidates for effective inhibition in AML cells with EVI1 deregulation, the authors intersected the results from the high-throughput drug screening with those from the gene expression differential analysis of TF-1 cells with siRNA targeting EVI1.However, only one cell line, TF-1, not included in the drug screening, was analyzed.An alternative approach may be to knock-down EVI1 in MOLM1 or UCSD-AML1 cells, perform transcriptome sequencing and then compare the gene expression data with the results from the drug screening and from the experiment in TF1 cells.Moreover, in order to strengthen their results, the authors could also analyze the public available data from the Cancer Dependency Map in AML cell lines with EVI1 overexpression and define a common list of targets.
We appreciate the reviewers' valuable suggestions.One of the reasons we chose TF1 cells is their high transduction efficiency compared to UCSD1/AML1 and MOLM1 cell lines.We encountered significant resistance to lentiviral transduction in 3q26 AML cell lines, including unsuccessful attempts with MOLM1 cells.This difficulty in transducing 3q26 AML cell lines likely explains the absence of such cell lines in the Cancer Dependency Map, except for TF1 cells.Consequently, generating a comprehensive list of putative targets from the Cancer Dependency Map becomes unrealistic due to the limited representation of EVI1 High cases.To overcome this limitation, we used a transcriptional dataset (E-MTAB-2225) 31 . of AML and characterized EVI1 High and EVI1 Low profiles.Through the analysis of transcriptional profilings, we established a signature that distinguishes between 3q26-positive and 3q26negative cell lines as shown in Figure R4A (Figure S2F of the new version of the manuscript).Subsequently, as shown in Figure R4D-E we used the LINCS database as a complementary approach to cMAP and identified HDAC inhibitors (HDACi) as a compound class associated with the transition from EVI1 "On" to EVI1 "Off" state (data included in Figure 1F and S2G of the manuscript).Furthermore, we successfully sequenced the HNT34 cell line in which EVI1 was lentivirally silenced, as demonstrated in Figure S3A-C, and revealed a significant overlap between EVI1-regulated gene signatures in HNT34 and EVI1dependent genes in TF1 (Figure R4C, included in the manuscript as Figure S2E) indicating a consistent signature derived from EVI1 modulation in these models.Strikingly, the projection of the HNT34 signature onto the LINCS L1000 dataset space demonstrated that transcriptional changes associated with an EVI1 "Off" status mimic and match HDACis signatures as shown in Figure R4D-E included in the new version of the manuscript (Figure 1F and S2G).Furthermore, to validate the results of the in silico small molecule screening in TF-1 cells (Figure 1D-E and S2B), we performed a tertiary small molecule screen of 4942 small drugs that confirmed the potent antiproliferative activity of HDACi compared to non the EVI1 "Off" inducers histamine, glucocorticoid, acetylcholine receptor antagonists or to other antineoplastic compounds contained in Selleck and Sigma-Aldrich collection screened as shown in Figure R4F.We presented this new results in Figure S2C and Table S1 of this new version of the manuscript.
In Figure 2F, we indeed analyzed the immunophenotype in the three conditions (NT, EVI1 sh#16, and EVI1 sh#87).Although this experiment was not the primary focus of our investigation and not included in the manuscript, we made an interesting observation that CD64 (FCGR1A) was overexpressed in the knockdown cells, and confirmed by flow cytometry.This finding was consistent with the single-cell RNA sequencing (scRNA) experiment presented in Figure 5 where CD64 increases upon treatment.While the identification of CD64 is intriguing, a comprehensive characterization of the role of this protein would extend beyond the scope of the current manuscript.Therefore, we have decided to reserve the idea for further investigation in a future study, where we can delve deeper into the role of CD64 in the context of EVI1 regulation.
We apologize for the mistake.
The authors hypothesized that HDACIs can impair the function of the EVI1 co-transcriptional complex and profiled by RNAseq two AML cell lines with high expression of EVI1 post 16 hours treatment with AR-42 and entinostat and found a strong association with MYC signature genes.One key experiment that is missing to demonstrate the deregulation of the EVI1 transcriptional complex is the analysis of EV1 chromatin-binding together with H3K27ac profiling by ChIP-seq (or CUT&RUN) in cells pre and post treatment.
We have performed the experiment suggested by the Reviewer to analyze the deregulation of the EVI1 transcriptional complex.We conducted EVI1 chromatin-binding analysis along with H3K27ac profiling using ChIP-seq in HNT34 cell line before and after treatment with AR-42 0.8 µM and entinostat 4 µM.This experiment is presented in Figure R5A (or Figure S7H of the new version of the manuscript).Analysis of the ChIP-Seq data revealed a quadruple EVI1 binding sites located 1.0 Mb 3′ of the Myc promoter within a region with high levels of H3K27, a histone mark associated with active transcriptions.The first two peaks cover a region of 1 Kb proximal to the MYC first exon, aligning with distribution pattern compatible with the P0, P1 and P2 MYC promoter regions 32 .A third peak identifies a single discrete peak mapping to the 3' region of the first MYC intron, coinciding with the mapping of MYC P3 promoter region, an alternative starting site 33,34 .The last peak is closer to the 3' region of exon 2 of the MYC gene.Furthermore, EVI1 binds the promoter regions of all the three genes PA2G4, FBL, DDX21 identified from the intersection of RIME and the transcriptional analysis of HDACis treated 3q26 AML cells (Figure 6C).Importantly, HDACis remove EVI1 and H3K27 binding from these regions suggesting the requirement of EVI1 for the modulation of a Myc signaling related genes in 3q26 models.These findings not only validate our working hypothesis but also provide an explanation for the data presented in the manuscript.
The scRNAseq analysis on follow-up samples provides interesting findings on how the combination azacitidine and entinostat targets stem and progenitor cells, however, as the authors are aware, it does not distinguish the effect of the single drugs.Can the authors perform cell viability assay and single cell RNAseq on normal CD34+ overexpressing EVI1 (for example by viral transduction) or in PDX cells before and after treatment with the combination and single azacitidine and entinostat?
We have conducted the suggested experiment using patient-derived leukemia xenograft (PLDX) cells derived from PDLX_PR#008.In this study, we evaluated the effects of different treatments, including a vehicle control, azacitidine at 1 mg/kg, entinostat at 10 mg/kg, and the combination of both.Our results demonstrate that entinostat effectively suppresses 3q26 leukemic infiltration (Figure R5B-C and R5F-G that we included in the manuscript as Figure 5G-H and 5K-L) and Myc signaling (Figure R5D-E or 5I-J of the new version of the manuscript).The addition of the hypomethylating agent (azacitidine) did not result in an additional desired synergistic effect.
The findings are further supported by the pharmacodynamic analysis, as depicted in Figure S6E, which reveals the suppression of EVI1 expression in AML upon entinostat treatment but not with azacitidine.This suggests that in the context of our study, the combined treatment did not exhibit an augmented effect on EVI1 suppression beyond what was achieved with entinostat alone.These results provide valuable insights into the potential limitations of hypomethylating agents in 3q26 AML.
Fig 2A and B show that cells TF1 and MUTZ-3 have much lower levels of EVI1 than the other AML cell lines tested, however the IC50s for cell proliferation shown in Fig 2G are essentially the same, suggesting that EVI1 cellular levels do not matter for the potency of HDAC inh in these AML cells.
-C and Figure S3A-C of the manuscript).
, S8E) or primary samples (Figure 4C and R3H, included as 7B in the manuscript) and in MYC High and MYC Low Figure R3F-G (S8B, S8F of the new version of the manuscript).

FigureFigure
Figure R1.(A) Venn diagram showing the overlap of proteins identified by RIME in EVI1 and IgG samples for each technical (R1 and R2) and biological (HNT34, UCSD/AML1, and PR#003) replicates.(B) Pearson correlation analysis of LFQ intensities.Scatter plot showing the correlation of LFQ signals between technical (R1 and R2, top panel) and biological (PR#003, UCSD/AML1, and HNT34, bottom panel) replicates of the RIME experimental setting.(C) Volcano plots showing the enriched proteins in 3q26 AML samples (n = 6) compared to control samples (n = 6), as identified by LFQ analysis.Protein hits with a fold change greater than 4 and a p-value less than 0.01 are shown in dark yellow.In the left panel, blue-labeled proteins are the confirmed MYC-related factors identified by intersecting RIME and ssGSEA data, as described in the manuscript.In the right panel, red-labeled proteins are EVI1 interactors identified using isotope labeling-based MS, as described in the supporting literature.(D) Biochemical validation of EVI1 and PA2G4 interaction.Western blot showing EVI1 and PA2G4 in protein lysates extracted from HNT34 cells (top) or 293T cells (bottom) transduced with a Lenti-hEF1a-ORF-P2A-eGFP-IRES-Puro vector (Transomic, Huntsville, AL, USA, #TLO2015.1)containing the EVI1 (NM_005241.3) and the PA2G4 (GeneBankBC001951.1)ORFs.Cell lysates were immunoprecipitated with an anti-EVI1 or anti-PA2G4 antibody.An anti-IgG antibody was used as a control of immunoprecipitation.The anti-EVI1 (Cell Signaling Technology, Danvers, MA, USA, C50E12, #2593) and the anti-IgG (Rabbit, Merk Darmstadt, Germany, #12-370) antibodies were the same used in RIME.

FoldFigure*Figure
Figure R4.(A) Volcano plot of RNASeq expression in EVI1 High or EVI1 low AML cell lines segregated based on the 3q26 status derived from the E-MTAB-2225 dataset.DEG are depicted in violet if repressed (log 2 fold change ≤ -2, Adj.P ≤ 0.01) or in dark yellow if upregulated (log 2 fold change ≥ + 2, Adj.P ≤ 0.01).(B) Volcano plot derived from RNASeq gene expression data of HNT34 cells transduced with a non-targeting shRNA (Control) or after EVI1-directed shRNAs (sh#16 and sh#87) three days after selection (see also Figure S3A).DEG are depicted in violet if upregulated in Control (log 2 fold change ≤ -2, Adj.P ≤ 0.05) or in dark yellow if upregulated in EVI1 shRNA (log 2 fold change ≥ + 2, Adj.P ≤ 0.05).(C) GSEA analysis of Affymetrix data from Figure S2A using the oncogenic gene sets from HNT34 transduced with shRNAs targeting EVI1.Enrichment plot of the EVI1 overlapping genes is shown.NES, normalized enrichment score; FDR, false discovery rate.(D) L1000FWD fireworks visualization of drug-induced signatures mimicking and reversing the differential gene expression signature generated from EVI1 High (n=5) or EVI1 low (n=2) AML cell lines contained in the E-MTAB-2225 dataset (left) and EVI1silenced HNT34 (right).Aquamarine circles indicate small molecule causing a reverse signature EVI1 "Off".Light violet circles indicate small molecule mimicking a signature EVI1 "On".HDACis drug-induced signatures are indicated in green overlapping with aquamarine circles.(E) Bar plot displaying top drugs inducing an EVI1 "Off' status identified by the L1000CDS2 query of E-MTAB-2225 (from Figure R4A, left) and HNT34 (from Figure R4B, right) datasets.HDACi are highlighted in green.(F) Violin plots comparing ∆POC activity between HDACis and the indicated drug classes."n"indicates number of small molecules within each class.Statistical significance among groups (***P ≤ 0.001, ****P ≤ 0.0001) was determined by one-way ANOVA (F) using Tukey's correction for multiple comparison testing.

-1 and Figure S9C. The description of Table S4-1 that
has been partially revised in this new version of the manuscript.
5) It is confusing what the figures in tables for each pull down refer to (eg What do the figures in columns D, H & L in table HNT34 Enriched Protein List for Evi-1 RIME Final Assay 35210 (HNT34) refer to?
-1.We have revised this section in the new version of the manuscript, as suggested by the Reviewer.The screening was performed in duplicate, and the values in column C (POC_molecules) of TableS1-1 refer to the average POCs derived from technical replicates.The POC_molecules values for each compound were compared to the DMSO control values (n = 24) using an ANOVA test applied to a linear regression model.The resulting p-values were adjusted for multiple comparisons using the Benjamini & Hochberg (BH or FDR) method.We also ran a one-sample t-test to corroborate this analysis.This test is more appropriate for comparing a single value (POC_molecules) to a series of data (DMSO n = 24).As in the case of the ANOVA test, we adjusted the resulting p-values using BH correction.As reported in TableR1, the results confirmed what observed with the previous statistical test, thus we decided to not include the one sample t-test in the final version of our manuscript.