Phosphoproteomic profiling of T cell acute lymphoblastic leukemia reveals targetable kinases and combination treatment strategies

Protein kinase inhibitors are amongst the most successful cancer treatments, but targetable kinases activated by genomic abnormalities are rare in T cell acute lymphoblastic leukemia. Nevertheless, kinases can be activated in the absence of genetic defects. Thus, phosphoproteomics can provide information on pathway activation and signaling networks that offer opportunities for targeted therapy. Here, we describe a mass spectrometry-based global phosphoproteomic profiling of 11 T cell acute lymphoblastic leukemia cell lines to identify targetable kinases. We report a comprehensive dataset consisting of 21,000 phosphosites on 4,896 phosphoproteins, including 217 kinases. We identify active Src-family kinases signaling as well as active cyclin-dependent kinases. We validate putative targets for therapy ex vivo and identify potential combination treatments, such as the inhibition of the INSR/IGF-1R axis to increase the sensitivity to dasatinib treatment. Ex vivo validation of selected drug combinations using patient-derived xenografts provides a proof-of-concept for phosphoproteomics-guided design of personalized treatments.

3) One of the limitations of the PDX models is that their ex vivo growth is limited, and often depend on supplemented cytokines (IL7?) or a feeder layer. If I am correct, the ex vivo culture conditions have not been included in the manuscript. The drug-response of each PDX may be highly variable because of 'sub-optimal' growth conditions and may skew the conclusion. Is this the reason why the effect of milciclib was not evaluated in the PDX models? The monotherapy effect of Dasatinib and INSR inhibitor presented in supplemental figure 8 should be included in Figure 6 as this is crucial info for the interpretation of these results. Because of the limitations (ex vivo culture conditions), further evaluation of the therapeutic effects of milciclib and dasatinib+INSR inhibition in an in vivo setting (on T-ALL PDX-02) certainly would strengthen the manuscript. Alternatively, authors should mention the limitations of their study. Figure 4: the correlation between the INKA score and the response to BMS-754807 is not obvious. Not only two cell lines with high INKA score, SUP-T1 and Jurkat, are not responsive; also two other cell lines that have low INKA score, PEER and CCRF-CEM, are responsive. How specific is this inhibitor? Also the synergistic effect with the dasatinib+IGF1R in supplemental Figure 7 is not convincing. These data should be confirmed with a second IGF1R/INSR inhibitor or via (druginducible) genetic depletion.

4)
Reviewer #2 (Remarks to the Author): The authors of "Phosphoproteomic profiling of T Cell acute lymphoblastic leukemia reveals targetable kinases and combination treatment strategies" have employed mass spectrometry-based global phosphoproteomic profiling using a panel T-ALL cell lines to identify targetable kinases. Herein they identify an impressive 21,000 phosphosites on 4,896 phosphoproteins, including 217 kinases. The application of phosphoproteomics using clinically relevant patient samples to aid in clinical decision making, is the holy grail of clinical proteomics for cancer patients. The authors have established bioinformatic pipelines to help deconvolute these complex phosphoproteomes and employ them to reveal SRC family kinases and cyclin-dependent kinase show ubiquitous activity in T-ALL cell lines.
The article is well written however, I have several key issues with the manuscript. The authors omit key T-ALL references such as references to mutations in the tyrosine-protein kinase JAK3 seen 16% of T-ALL cases PMID: 26206799 and conclude that no other global, MS-based phosphoproteomic study have been published to predict drug sensitivity, however, miss this study PMID: 28852199. The authors should list known genomic drivers of these cell lines in Supplementary Table S1, so correlation between selection of effective therapies based on phosphoproteomic signatures differ from those that would be encouraged based on genomics. There are no statistics in Supplementary Figure 1? It is not overly clear to me how the enriched phosphopeptides were quantitated across samples? It is also unclear whether biological or technical replicates were performed except in LOUCY and CCRF-CEM performed in duplicate, hence statistical analysis comparing these cell lines is impossible. I also find it troubling that there are no controls cell lines or tissue. This is highlighted by the rapamycin results which are highly effective across cell lines but known to be highly immunosuppressive in any case. I find it challenging to use the data in Fig 1 to come to my own conclusions. Given the priorities of assessment of the tyrosine phosphoproteome, I am unclear as to why the authors start their results by dissection of the serine/threonine phosphoproteome, which yields little in the way of a therapeutic vulnerabilities. All of Figure 2 requires corroborative statistical analysis. Is the difference (statistical) in viability (A/V+) HSB-2 cells treated with 100 nM Milciclib? It is an interesting choice to assess substrate-level phosphorylation after 3-day treatment with Dasatinib in Figure 3? Authors need to include an assessment of cell viability markers to show ontarget effects or whether this is general cell death. Previous studies also report the benefit of combined SFK and IGF-1R inhibition in vitro and in vivo. Indeed, these studies also show upon IGF-1R inhibition, increased activity of SFK, highlighting the potential of the combination. Although the manuscript tests the efficacy of therapies following a repeat of INKA analysis using isolated blasts ex vivo, what is missing are studies to assess the preclinical utility of the pipeline/results.
We thank the reviewers for the careful and detailed evaluation of our manuscript and the appreciation of our work. We have addressed the points raised by the reviewers as outlined in the point-by-point response below. Additionally, we have highlighted in yellow the alterations in the text, figures, and legends of the original manuscript. Moreover, we provide here additional data to answer some of the points raised by the reviewers (" Figures for reviewers").

REVIEWER COMMENTS Reviewer #1 (Remarks to the Author):
In this elegant study, Valentina Cordo and colleagues have performed a phosphoproteomics profiling of T-ALL to identify targetable kinases. The authors started with an exploratory unbiased profiling of a panel of 11 T-ALL cell lines. To translate their findings to clinical applications, they supplemented their profiling with drug screenings and validation in a cohort of patient-derived xenograft models. This work is excellent, technically well performed with inclusion of necessary replicates and correct data interpretations. Their approach highlights the need for more proteomics-based approaches to identify leukemia vulnerabilities and exemplifies that it can be used to guide combination treatment strategies. Few remarks that need to be addressed: 1) The intro/discussion should be extended with reported druggable kinases in T-ALL, like TYK2, PIM1 and Polo-like kinases (PLK) and Aurora kinases; are these also identified in the phosphoproteomics profiling? The font of the identified kinases in Figure 1B is too small.
We thank the reviewer for the appreciation of our elaborate work and for pointing out some interesting kinases that were missing in our introduction/discussion. Therefore, we added text and references indicating the role of TYK2, PIM1, PLKs and AURK in T-ALL and their possible targeting in the introduction section (lines 62-66 and ref #8-13).
In our phosphoproteome profiling, we identify a low TYK2 kinase activity in the cell lines (HSB-2, PEER, ALL-SIL, HPB-ALL, and CCRF-CEM) and in the four PDXs as reported in Figure 1B (ranking 14 th -20 th in the Top20-tyrosine INKA plots) and Figure 6B (ranking 12 th -20 th in the Top20-tyrosine INKA plots), respectively. Very low PIM1, PLK1 and AURKA activity was inferred (ranking beyond the Top20 active kinases in each sample and were therefore not depicted in the plots).
To improve the readability of Figures 1B and 1C, we increased the font of the text in the INKA plots.
2) The use of the 4 PDX models makes the study highly relevant. Due to the limited number of T-ALL patients samples it is logic that authors did their first exploratory phase using T-ALL cell lines. I fully agree with their discussion that in future a second more extended profiling will be highly informative by using more patients from different molecular T-ALL subtypes. At this point, I am missing details on the used PDXs in supplemental table 1, which makes it impossible to link the results with the genetic profile of these patients.
We agree with the reviewer that the missing genomic information is important for the correct interpretation and analysis of our results. Therefore, we added an additional supplementary table (Supplementary Table 3) listing all the somatic mutations identified in the 4 PDXs by whole-exome sequencing. Moreover, we highlight in the results and discussion sections the relation between the lack of somatic mutations in INSR and IGF1R and the detection of active INSR/IGF-1R in lines 252-254 and lines 308-311. In addition, the presence of JAK mutations in PDX-04 and the high JAK activity inferred, the sensitivity to ruxolitinib, and the synergy between ruxolitinib and dasatinib are discussed in lines 259-260, lines 291-293 and lines 324-325.
3) One of the limitations of the PDX models is that their ex vivo growth is limited, and often depend on supplemented cytokines (IL7?) or a feeder layer. If I am correct, the ex vivo culture conditions have not been included in the manuscript. The drug-response of each PDX may be highly variable because of 'sub-optimal' growth conditions and may skew the conclusion. Is this the reason why the effect of milciclib was not evaluated in the PDX models?
We thank the reviewer for the detailed analysis of our PDX-based drug screening results. Viably frozen purified blasts from T-ALL PDXs were thawed and cultured in RPMI1640+Glutamax supplemented with 20% fetal calf serum and antibiotics for 3 days, in the absence of cytokines and feeder layer to reduce the complexity of our experimental settings. In fact, the addition of cytokines such as IL7 could alter the intracellular signaling and promote therapy resistance as previously described (Delgado-Martin et al., Leukemia 2017;van der Zwet et al., Leukemia 2021). We now added a more detailed description of our experimental settings in the Material and Methods section at lines 437-439.
We initially did not include the evaluation of milciclib sensitivity in PDXs since we wanted to focus on the validation of our strategy for the identification of INKA-guided effective treatment combinations from phospho-tyrosine data (the dataset that provides the LCK, SRC, INSR/IGF-1R activities already investigated in the cell lines panel). Nevertheless, we tested the sensitivity of the 4 PDXs to milciclib treatment ex vivo and we also evaluated the cell survival/proliferation, as illustrated below and in Supplementary Figures 5G-H. The 4 PDXs had a different survival and proliferation rate during 72 hours of culture in vitro in the absence of cytokines and feeder layer. While untreated PDX-01 and PDX-04 cells had a decrease in survival (decrease in mean luminescence signal at day 3 compared to day 0), PDX-02 cells survived well in vitro and PDX-03 showed even a clear proliferation during 72 hours of culture in vitro (panel G). Panel H illustrates the dose-response curves for milciclib treatment. All the PDXs showed sensitivity to milciclib with IC50 values lower than 200nM and included in the clinical range (lower than the maximum plasma concentration achieved in patients, green box). It is important to notice that PDX-03, which showed active proliferation in vitro seems to be the most sensitive to milciclib treatment. We agree with the reviewer that sub-optimal culturing condition could affect the results of the drug screening (in particular when testing chemotherapeutics or drugs that interfere with the cell cycle). Nevertheless, this seems not to be the case for milciclib in our limited PDX cohort since we did not see resistance to treatment in the absence of cell proliferation.  Figure 6 as this is crucial info for the interpretation of these results.

The monotherapy effect of Dasatinib and INSR inhibitor presented in supplemental figure 8 should be included in
In agreement with the reviewer, we now moved the monotherapy dose-response curves in panels C and E of Figure 6, respectively.

Because of the limitations (ex vivo culture conditions), further evaluation of the therapeutic effects of milciclib and dasatinib+INSR inhibition in an in vivo setting (on T-ALL PDX-02) certainly would strengthen the manuscript. Alternatively, authors should mention the limitations of their study.
We fully agree with the reviewer on the limitations of our drug screening approach. We also agree that in vivo data would certainly strengthen the validity of our findings. However, the main aim of this study was to propose phosphoproteomics as useful tool to identify non-genomic targets and to guide the testing of novel treatment strategies, rather than validating a single treatment option. Therefore, now we list and discuss the limitations of our study (such as the in vitro drug screening in the absence of active cell proliferation that might affect sensitivity to some drugs, the need to expand the limited PDXs cohort, the need for in vivo validations of effective treatment strategies) in the discussion (lines 332-338). Figure 4: the correlation between the INKA score and the response to BMS-754807 is not obvious. Not only two cell lines with high INKA score, SUP-T1 and Jurkat, are not responsive; also two other cell lines that have low INKA score, PEER and CCRF-CEM, are responsive. How specific is this inhibitor?

4)
We agree with the reviewer that the correlation between the INKA score and the BMS-754807 response might not be immediate. First, the absolute INKA score for INSR/IGF-1R should be considered within a single sample (therefore in the context of the Top20 active kinases for each cell line). Figure 4A is used to show that most of the T-ALL cell lines have active INSR/IGF-1R. As illustrated in the Figure 1 for Reviewers below, we took Figure 4B from the manuscript and highlighted in orange the lines with low/no INKA score for INSR/IGR-1R such as PEER, CCRF-CEM and HSB-2 (panel A). Differently from the most sensitive lines highlighted in blue (ALL-SIL, MOLT-16 and HPB-ALL, IC50 values < 250nM), PEER, CCRF-CEM and HSB-2 have IC50 values around 1µM. We therefore checked the selectivity of BMS-754807 and all the possible targets inhibited by the drug at different concentrations (dataset from Klaeger et al., The target landscape of clinical kinase drugs, Science 2017) and we report in the figure below (panel B) the list of BMS-754807 targets and their activity inhibition at 100nM, 1µM, and 2µM. As reported by Klaeger and colleagues, kinase inhibitors lose their specificity when increasing the drug concentration. In fact, BMS-754807 has only 3 predicted targets (inhibition > 50%) at 100nM while the number of kinases inhibited (inhibition > 50%) goes up to 11 (1µM) and 15 (2µM) at increasing concentrations. Therefore, we cannot exclude that the response obtained with BMS-754807 concentrations above 1 µM are caused by off-target inhibition of additional kinases rather than INSR and IGF-1R only. In our study, we generally consider cell lines sensitive to a kinase inhibitor when the IC50 values are in the nanomolar range (< 1µM) to avoid misinterpreting an off-target kinase inhibition/general cytotoxicity with an actual on-target effect. Indeed, we consider only ALL-SIL, HPB-ALL and MOLT-16 as sensitive cell lines as indicated in the text at lines 203-205. For the same reason we decided to use only the IC20 values of BMS-754807 (30-300nM) for the drug combination screenings in the cell lines.  Figure 7 is not convincing. These data should be confirmed with a second IGF1R/INSR inhibitor or via (drug-inducible) genetic depletion.

Also the synergistic effect with the dasatinib+IGF1R in supplemental
We agree with the reviewer regarding the need of further validation of the synergistic effect of dasatinib and an INSR/IGF-1R inhibitor. Therefore, we tested two other INSR/IGF-1R inhibitors, linsitinib (OSI-906) and GSK1904529A (GSK-4529A), respectively. We confirmed the synergistic effect of combined SRC/LCK and INSR/IGF-1R inhibition when treating cells with dasatinib and a fixed concentration (IC20) of linsitinib (500nM) and GSK-4529A (60nM). We therefore removed the graph in Supplementary Figure 7 (which now illustrates the dose-response curves to the single INSR/IGF-1R inhibitors and their corresponding IC20 values) and added the two additional dasatinib + INSR/IGF-1Ri combinations in Figure 4E of the manuscript, demonstrating synergistic effects of all 3 INSR/IGF-1R inhibitors when combined with dasatinib. The text at lines 213-218 in the results section was accordingly changed.

Reviewer #2 (Remarks to the Author):
The authors of "Phosphoproteomic profiling of T Cell acute lymphoblastic leukemia reveals targetable kinases and combination treatment strategies" have employed mass spectrometrybased global phosphoproteomic profiling using a panel T-ALL cell lines to identify targetable kinases. Herein they identify an impressive 21,000 phosphosites on 4,896 phosphoproteins, including 217 kinases. The application of phosphoproteomics using clinically relevant patient samples to aid in clinical decision making, is the holy grail of clinical proteomics for cancer patients. The authors have established bioinformatic pipelines to help deconvolute these complex phosphoproteomes and employ them to reveal SRC family kinases and cyclin-dependent kinase show ubiquitous activity in T-ALL cell lines. The article is well written however, I have several key issues with the manuscript.
The authors omit key T-ALL references such as references to mutations in the tyrosine-protein kinase JAK3 seen 16% of T-ALL cases PMID: 26206799 and conclude that no other global, MS-based phosphoproteomic study have been published to predict drug sensitivity, however, miss this study PMID: 28852199.
We thank the reviewer for the insightful suggestions. We added the recommended references in the introduction section of the manuscript at lines 60 (ref #6) and lines 83-84 (ref #25). Table S1, so correlation between selection of effective therapies based on phosphoproteomic signatures differ from those that would be encouraged based on genomics.

The authors should list known genomic drivers of these cell lines in Supplementary
We agree with the reviewer that genomic information of the cell lines used is necessary to highlight the differences of putative targets suggested by the phosphoproteomic analyses. Therefore, we now added the subtype, the genomic drivers and known mutated oncogenes of the 11 cell lines in Supplementary Table 1.
There are no statistics in Supplementary Figure 1?
Regarding to the lack of statistics in Supplementary Figure 1A, we would like to clarify that the graph only shows the number of phosphorylated peptides recovered and identified for each cell line in a single mass spectrometry measurement. To exclude that the visible higher recovery of phosphorylated peptides in HSB-2 could be due to a possible error, a higher total lysate input or a different efficiency of purification during the phospho-tyrosine immunoprecipitation, we checked the total phosphorylation level prior to any phospho-enrichment via western blotting (antiphosphotyrosine antibody) as shown in Supplementary Figure 1C. Despite an equal protein loading (blue Coomassie staining in Supplementary Figure 1C-right), HSB-2 cells show a higher phosphotyrosine signal compared to any other cell line already before any phospho-enrichment. Therefore, we concluded that HSB-2 cells present an enhanced phospho-tyrosine signaling (possibly due to the TCRβ-LCK translocation that results in higher LCK expression) as intrinsic characteristic. We did not perform any statistical analysis to check whether the higher recovery of phosphorylated peptides was significantly different in comparison to the other lines.

It is not overly clear to me how the enriched phosphopeptides were quantitated across samples?
Label-free mass spectrometry (MS) measurements with ion intensity-based relative quantification were performed as previously described (Piersma SR et al., Journal of Proteomics, 2015;Beekhof R. et al., Molecular Systems Biology, 2019).
We added the following details to the supplementary methods section (page 1-2, line 15-31): Massspectrometry measurements were performed using data-dependent acquisition MS. Every scan cycle of ~1 sec starts with measurement of intact peptide masses (MS1) and subsequently the top15 highest peptide signals are sequentially isolated in the quadrupole and fragmentated in the HCD collision cell to acquire MS/MS spectra. For peptide and protein identification, MS1 and MS/MS spectra were searched against the Swissprot complete human proteome FASTA file (release January 2018, 42,258 entries) using the MaxQuant 1.6.0.16 software and applying a false discovery rate (FDR) cut-off of 1% at the peptide, phosphosite and protein level.
Phosphopeptides and phosphosites were quantified from the MS1 Intensities of the eluting phosphopeptides. Phosphopeptide intensities were used by MaxQuant to calculate phosphosite intensities. The relevant information can be extracted from the phospho(STY)sites.txt and modificationSpecificPeptides.txt files produced by MaxQuant. The quantification of proteins was performed using the MaxLFQ algorithm (Cox J. et al., Mol Cell Proteomics 2014).