Quantitative Phosphotyrosine Profiling of Patient-Derived Xenografts Identifies Therapeutic Targets in Pediatric Leukemia

Acute lymphoblastic leukemia (ALL) is the commonest cancer in childhood and adolescent age groups, accounting for a third of all pediatric malignancies and 80% of childhood leukemia (1, 2). The disease results from a clonal proliferation of malignant, immature white blood cells of B- and T-cell lineages in the bone marrow. The B-cell precursor (BCP) - and T-ALL are further divided into a number of subtypes based on clinical features, chromosomal translocations, and genetic characteristics (3). Despite an overall survival rate approaching 90%, the heterogeneous nature of ALL confers a differential response to treatment for the diverse patient groups (4). Given this heterogeneity in prognosis, modern therapeutic approaches seek to modify treatment regimens based on a patient's risk of relapse at diagnosis using clinical, cytogenetic, and biologic criteria, in risk-adapted therapy (5, 6). More recently, attempts have been made to improve risk stratification by classifying these high-risk patients into distinct subtypes at a molecular level, via gene expression analysis and immunophenotyping (7, 8). Whole-genome and transcriptome sequencing of certain high-risk patient subsets such as early T-cell precursor ALL (ETP-ALL; refs. 9, 10) and Philadelphia chromosome–like (Ph-like; refs.11, 12) ALL have also revealed several activating gene fusions, alterations, and mutations that could result in constitutively activated tyrosine kinases (TK). This further leads to unconstrained phosphorylation of downstream substrates by TKs, impacting several key signaling pathways and resulting in increased cell survival and proliferation.

Recent advances in mass spectrometry (MS) and the use of anti-phosphotyrosine antibodies for the enrichment of tyrosine phosphorylated peptides have greatly facilitated characterization of kinase signaling networks and identified several activated TKs and their phosphorylated substrates in cancer cells (13–15). Furthermore, quantitative tools such as stable isotope labeling by amino acids in cell culture (SILAC) have also been efficiently integrated into this phosphoproteomic workflow to globally map and quantify changes in phosphorylation events (16, 17). Using such an approach, activated oncogenic kinases were identified in non–small cell lung cancer patient samples (15). Phosphotyrosine profiling also revealed specific signaling networks in basal breast cancer cells and highlighted multiple kinases and substrates for therapeutic evaluation (18). These studies also indicate that patient stratification is feasible based on activated kinase profiles, which could suggest specific kinase-targeted drugs to be used either alone or in combination with established chemotherapeutics. Although this quantitative approach has been successfully extended to patient samples in solid tumors, studies in hematologic disorders have been confined to cell line models.

In this study, we have used an MS-based phosphotyrosine profiling approach to characterize activated TK signaling in patient-derived xenografts (PDX) of high-risk pediatric ALL patients. Integrating a “spike-in” SILAC approach, we mapped close to 1900 class I phosphosites with >0.75 localization probability and 99% confidence in 16 PDXs, of which 1394 tyrosine phosphorylated sites had a heavy SILAC partner that allowed quantification. In particular, individual PDXs with ABL1, FLT3, and JAK mutations with aberrant kinase signaling were targeted with commercially available TK inhibitors (TKI) both in vitro and in vivo. Aberrant ABL1 kinase signaling observed in a Ph⁺-ALL and a PDX with high phospho-ABL1 (harboring a NUP214-ABL1 translocation) indicated dasatinib treatment and a significant delay in disease progression was achieved in these PDXs. Similarly, the uniquely activated FLT3 in one PDX correlated with an objective response to the multikinase inhibitor sunitinib. Thus, this study demonstrates an important step forward in the preclinical utility of an unbiased and quantitative tool to identify aberrant TK signaling in high-risk ALL PDXs and highlights its potential to identify tractable drug targets.

All PDX studies had received prior approval from the respective institute's Human Research Ethics Committees and Animal Care and Ethics Committees. Continuous xenografts from childhood ALL biopsies (demographics detailed in Table 1) were established in immunodeficient mice as described previously (19, 20). Spleens with >95% infiltration of human leukemic cells were harvested and mononuclear cells were used for sample preparation. Methods for PDX engraftment and proteomic sample preparation are detailed in Supplementary Materials and Methods.

The JURKAT (T-ALL), NALM-6 (BCP-ALL), MUTZ-5, and MHH-CALL-4 (JAK mutated/CRLF2 high BCP-ALL) cell lines were used in this study to represent a mixed population of T-, BCP-, and JAK-mutated samples, and had been validated by STR analysis no more than 3 months before use. Detailed culture conditions and SILAC MIX preparation are described in Supplementary Materials and Methods.

Lysates from the PDX samples were spiked with SILAC MIX and were reduced, alkylated, and trypsin digested as previously described (21, 22). Purified and lyophilized peptides were immunoenriched using P-Tyr-100 antibody as described previously (14, 18, 23). Enriched phosphotyrosine peptides were subjected to TiO2-based phosphopeptide enrichment using TiO2 microcolumns (GL Sciences) as described by Larsen and colleagues (24) following the manufacturer's protocol to facilitate removal of antibody fragments and nonphosphorylated peptides that coelute in the elution step. Detailed methods of phosphopeptide immunoprecipitation with SILAC-spiked PDX samples and TiO2 microcolumn clean-up are described in Supplementary Materials and Methods.

Mass spectra of phosphotyrosine immuno-purified peptides were generated in an Orbitrap Velos (Thermo Electron) mass spectrometer. Raw mass spectrometric files were analyzed in MaxQuant (25) version 1.3.0.5 for feature detection, protein identification, and quantification using the integrated Andromeda search engine for database searching (26). Methods for liquid chromatography, MS, MaxQuant search criteria, protein identification, and quantification are detailed in the Supplementary Materials and Methods.

Cryopreserved PDX cells were thawed at 37°C and washed in RPMI1640 media (Invitrogen) supplemented with 10% FCS, 100 U/mL penicillin, 100 μg/mL streptomycin, and 2 mmol/L l-glutamine. After centrifugation (500 × g, 5 minutes) and a second wash, cell density, and viability were calculated following an exclusion assay using 0.4% Trypan blue (Sigma-Aldrich). Cells were resuspended at the appropriate cell density (4–5 × 10⁵ cells/well) in QBSF-60/F [containing QBSF-60 media (Quality Biological) supplemented with 20 ng/mL Fms-like tyrosine kinase 3 ligand (Flt-3L; ProSpec), 100 U/mL penicillin, 100 μg/mL streptomycin, and 2 mmol/L l-glutamine]. For ETP-1 and ETP-8 PDXs, the media were additionally supplemented with 10 ng/mL recombinant human IL7 (Jomar Biosciences).

All drugs were obtained from Selleck Chemical, serially diluted in QBSF-60/F, and 20 μL added to wells in triplicate to give final drug concentrations of 1 pmol/L to 10 μmol/L. Following a 48-hour incubation at 37°C/5% CO₂, 15 μL of Alamar Blue [containing 75 mg resazurin, 12.5 mg methylene blue, 164.5 mg potassium hexocyanoferrate (III) and 211 mg potassium hexacyanoferrate (II) trihydrate (Sigma Aldrich) in 500 mL PBS] was added to each well (27). Fluorescence was measured (560 nm excitation, 590 nm emission) on a Victor X3 Multilabel Plate Reader (PerkinElmer) immediately following the addition of Alamar blue and at 6-hour postaddition. The 0-hour read was subtracted from the 6-hour read and all readings were normalized as a percentage of the corresponding control well. The experiment was repeated as above for three biologic replicates for each PDX or on three passages of the cells lines to give a biologic triplicate result. Dose–response curves were graphed and IC₅₀ values calculated by GraphPad Prism 6 software (http://www.graphpad.com/scientific-software/prism/). Dephosphorylation of phosphorylated kinase targets post inhibitor treatment was validated using immunoblotting (detailed in the Supplementary Materials and Methods).

For ruxolitinib efficacy studies, mice were randomized to treatment or vehicle (4–7 mice per arm) once xenografts had engrafted with sufficient disease burden to detect >1% peripheral human cCD3⁺ blasts. Ruxolitinib (INCB018424) was administered for the entire 3-week treatment period in chow formulation provided by Incyte. Studies were terminated and mice were euthanized after 3 weeks of treatment, when vehicle-treated mice became ill according to standard parameters set on the Institutional Animal Care and Use Committee protocol. Spleen was harvested at the time of euthanasia. Disease burden was assessed every 7 to 14 days by flow cytometric measurement of human cCD3⁺ blast count in peripheral blood, using CountBright absolute counting beads (Invitrogen) as described in the manufacturer's instructions, and at euthanasia, by measuring absolute splenic blasts (total splenic cell count × %human cCD3+/CD45⁺ cells) as we have described previously (28).

RNA samples were used to prepare cRNA with Illumina Total Prep RNA Amplification Kit (Life Technologies). cRNA was then hybridized to Illumina Human Beadchip HT12 Arrays. Gene expression datasets were analyzed using GenePattern v3.2.3 as we have described previously (29). Gene expression heatmaps were generated using GenePattern, whereby the range of color coding extends from minimum to maximum values per gene (per row). In each case, red indicates high, and blue low, level of expression. Unsupervised hierarchical clustering was performed using the Hierarchical Clustering module in GenePattern using the entire 47,323 probes representative of 34,694 genes present in the Illumina Human Beadchip HT12 Arrays. Gene expression datasets can be accessed at www.ncbi.nlm.nih.gov/geo (accession no. GSE52991, GSE57795, GSE74460).

To characterize the global tyrosine phosphorylation status in pediatric ALL PDXs, we adopted an immunoaffinity profiling of phosphotyrosine peptides protocol as described previously (14, 15, 18). Patient demographics of the 16 PDXs, along with their subtype classification and key molecular lesions, are listed in Table 1. 

To initially establish the immunoaffinity profiling in PDX models, two T-ALL (ALL-8, ALL-31) and two BCP-ALL (ALL-17, ALL-19) PDXs were selected. Using the phosphotyrosine immunoaffinity profiling method, we identified 540 Class I phosphotyrosine sites along with 61 phosphoserine and 49 phosphothreonine sites (Supplementary Table S1). These sites correlated to 424 phosphoproteins, of which 17% were kinases (Supplementary Fig. S1A). This method enabled us to identify phosphorylation on all the key TKs that have been reported in T-cell signaling pathways in the two T-ALL PDXs, ALL-8, and ALL-31 (30). For example, ZAP70 and LCK (Supplementary Fig. S1B) were identified along with other T-cell–associated signaling molecules CD3, CD247, LAT, GRAP2, and SH2D2A (highlighted in yellow in Supplementary Table S1). ZAP70 is an essential kinase in T-cell signaling and its phosphorylation is indicative of recruitment of other T-cell signaling proteins (LCK, LAT; ref. 16).

As expected, phosphorylation at the T-cell–specific sites was not observed in the BCP-ALL PDXs, in which the identified phosphotyrosine sites were primarily B-cell signaling molecules. The TKs SYK, BTK, and LYN were highly phosphorylated in the BCP-ALL PDXs (Supplementary Fig. S1B; Supplementary Table S1; red highlight). ALL-19 was the only BCP-ALL PDX with markedly elevated tyrosine phosphorylation of ABL1 kinase. We then plotted the combined intensity of all phosphorylated tyrosine sites in a TK against the total intensity of all the identified TKs in that PDX (Supplementary Fig. S1B). The pie charts clearly show the distinction between the T- and BCP-ALL PDXs based on the intensity of TK-specific tyrosine phosphorylated peptides. The phosphotyrosine sites of the TKs identified either in T-ALL or in the BCP-ALL PDXs are listed in Supplementary Table S2.

To compare phosphotyrosine profiles across multiple PDXs, we undertook developing a more quantitative analysis method. The immunoaffinity protocol optimized for the PDX models was combined with a “spike in” SILAC (21, 31) approach to generate a quantitative phosphotyrosine profile of high-risk pediatric ALL. Figure 1 shows a schematic of the quantitative phosphotyrosine profiling approach in PDX models of ALL. As profiling of phosphotyrosine sites in cellular signaling is compounded with artifacts when clinical sample processing is delayed or when samples are frozen (32), extra care was taken for extraction of proteins under optimal conditions from freshly harvested human mononuclear cells from mice spleens. In fact, we compared phosphotyrosine enrichment of freshly processed ALL-2 PDX cells with that of frozen ALL-2 PDX cells that were stored in liquid nitrogen in a FBS+10% DMSO medium. Both the frozen and fresh ALL-2 PDX samples were the same in all respects (generated from the same patient biopsy and harvested from mice at the same passage number) with freezing being the only variable. As expected, the number of class I phosphotyrosine sites was reduced by 80% in frozen samples compared with freshly harvested ALL-2 spleens (Supplementary Fig. S2A). Furthermore, as a novel heterozygous FLT3 mutation (c.1715A>C; p.Y572S) was observed in the ALL-2 PDX, we tracked the tyrosine phosphorylated sites on FLT3. In the fresh ALL-2 samples, two autophosphorylated FLT3 sites (Y842 and Y969) were identified compared with only one (Y842) in the frozen ALL-2 samples. Moreover the intensity of the Y842 site was a third in frozen samples compared with freshly lysed samples (Supplementary Fig. S2B). All the PDX samples used in this study were freshly harvested and lysed under optimal conditions using appropriate phosphatase and protease inhibitors.

As all PDX samples were spiked with equal concentration of the heavy labeled internal standard, the light (PDX) to heavy (SILAC) ratio generated for each of the enriched phosphopeptides enabled an accurate indication of their relative abundance. This allowed a direct comparison of phosphopeptide abundance between different ALL PDXs relative to the SILAC standard. To ensure the amount of heavy labeled phosphopeptide enriched during the immunoaffinity process was sufficient to provide quantitative information on the light phosphopeptides enriched from the PDX sample, different ratios of starting material and “spike-in” standard were initially tested (Supplementary Fig. S3) and the optimized 20:1(PDX:SILAC) ratio was then used to generate quantitative phosphotyrosine profiling data on a panel of 16 PDXs, including 4 BCP-ALL, 5 T-ALL, 4 ETP-ALL, and 3 Ph-like ALL (Table 1).

Enriched phosphopeptides from at least two technical replicates for each PDX (up to 4 replicates for some) were identified by MS. Using this quantitative phosphotyrosine method, we mapped in total 1,912 class I phosphosites (1,375 phosphotyrosine, 376 phosphoserine, and 161 phosphothreonine) with >0.75 localization probability and 99% confidence (Supplementary Table S3). Of these, 1,394 class I phosphosites (73%) had a heavy partner intensity that allowed quantification (Supplementary Table S4). The number of phosphosites identified and quantified in each of the PDXs is represented in Supplementary Fig. S4. Marked differences were observed in the number of class I phosphosites that were identified in each of the different subgroups. Although the highly aggressive ETP-ALL and the Ph-like ALL PDXs had higher number of phosphosites ranging from 400 to 850, the other T-ALL and BCP-ALL PDXs had phosphosites ranging from 250 to 400 (Supplementary Fig. S4).

Hierarchical clustering based on the average normalized ratios of all quantified phosphotyrosine sites (Supplementary Table S4) for each PDX revealed clustering into two distinct groups broadly based on B and T lineage (Fig. 2A), with the stem cell-like ETP-ALL clustering either with the B lineage (ETP-12 and ETP-1) or with the T lineage (ETP-13 and ETP-8). In the BCP-ALL group, the Ph⁺-ALL, ALL-4 clustered independently of the remaining PDXs. Notably, segregating the molecularly defined ETP, Ph-like, and Ph⁺-ALL PDXs and reclustering them resulted in grouping according to their subtype (Supplementary Fig. S5). Furthermore, the distinct clustering into the B and T lineages was also validated by gene expression profiling. Unsupervised hierarchical clustering of basal gene expression profiles of 14 PDXs revealed two broad branches reflecting each leukemia subtype (Supplementary Fig. S6A).

Unsupervised hierarchical clustering was performed to identify PDX clustering patterns based on the identified phosphotyrosine sites on TKs. As shown in Fig. 2B, the 16 PDXs clustered into two broad groups, separating the BCP-ALL from the T-ALL and a distinct group consisting of the highly aggressive ETP-ALL and Ph-like ALLs. The only exception was ETP-8, which clustered with the T-ALLs. Unsupervised hierarchical clustering of TKs by microarray analysis of gene expression in 14 PDXs also showed two broad groups, separating the T-ALL PDXs from the BCP-ALL and Ph-like ALLs, with one ETP-ALL clustering each with the T-ALL and BCP-ALL clusters (Supplementary Fig. S6B).

Analysis of individual PDXs revealed that in the Ph⁺-ALL, ALL-4 the TKs ABL1 (Y469, Y393), DDR1 (Y792, Y796) and PTK2B (Y579, Y580) showed the highest levels of tyrosine phosphorylation, along with BTK, CSK, FYN, and EPHB4 (Fig. 2B; Supplementary Table S5). Similarly, the receptor TK (RTK) ZAP70 phosphosites (Y493, Y292 and Y248) were uniquely present in the ETP group with Y248 having the highest level of phosphorylation in ETP-12, ETP-13, and ETP-1 (Supplementary Table S5). Y493 and Y292 have previously been reported as autophosphorylation sites responsible for the complete activation of ZAP70 by displacing the activation loop from the catalytic site (33). Furthermore, in the ETP group, ETP-8 had the highest levels of phosphorylation of the JAK1 autophosphorylation phosphosites Y1034/35, which could be a post translational effect of the activating JAK1 mutations described by Zhang and colleagues in ETP-ALL (10).

In the Ph-like ALL group, the TKs with the highest levels of tyrosine phosphorylation included INSR, JAK1, JAK2, LYN, FYN, SYK, and LCK (Fig. 2B; Supplementary Table S5). The RTK INSR has previously been reported to be involved in pre-B leukemia (34) and this kinase had the highest level of phosphorylation in this group with phosphorylation at five different sites (Y1190, Y1189, Y1185, Y1355, and Y1361). Conversely, Y842 FLT3 was unique to PAMDRM across the entire panel, whereas the Y570 site in JAK2 and the Y702 site in AXL were unique to PALLSD. PALLSD also had Y209 HCK, Y344 BTK, Y145 DYRK1A, and Y188 BLK phosphosites unique to itself. A complete list of all the TK phosphotyrosine sites that were identified in the ETP-ALL, Ph-like, and Ph⁺-ALL PDXs is shown in Fig. 2B and Supplementary Table S5.

As anticipated, the highest levels of TK phosphotyrosine sites in T-ALL PDXs included ZAP70, LCK, and FYN, with none of these identified in the BCP-ALL PDXs (Fig. 2B). In fact, the Y420 site of FYN that regulates T-cell receptor signaling by autophosphorylation was present at very high levels in all five T-ALL PDXs compared with the BCP-ALLs. Within the BCP-ALLs, ALL-19 had the highest level of phosphotyrosine sites on ABL1 (Y393), LYN (Y194, Y193), INSR (Y1189, Y1185), and JAK2 (Y1007). In particular, the autophosphorylation site Y1007 of JAK2 was uniquely observed in ALL-19, whereas the RTK FLT3 and the nonreceptor TK (nRTK) TYK2 were uniquely identified in ALL-2. Both of the identified FLT3 phosphosites in ALL-2 (Y842 and Y969) are key autophosphorylation sites for kinase signaling activation (35). The profound differences in levels of pTyr site phosphorylation of TKs (ABL1, FLT3 and JAK1) and key signaling substrates could not be explained by mere differences in gene expression levels (Supplementary Fig. S7).

We further investigated altered tyrosine phosphorylation in pediatric ALL PDXs to assess the therapeutic potential of specific TKIs. Immunoblot analyses were performed to validate activated sites and their dephosphorylation upon TKI treatment, and in vitro cytotoxicity assays and/or in vivo efficacy experiments were performed to determine drug efficacy.

The autophosphorylation site of the ABL1 kinase Y393 was notably upregulated in ALL-4 (Fig. 3A). The upregulation of this site has also been confirmed by others studying BCR-ABL1 signaling in chronic myeloid leukemia (13, 23). Imatinib-sensitive tyrosine phosphorylation sites in GAB1, SHC, BCR, ABL, and STAT5A were also identified (highlighted in yellow, ALL-4 sheet; Supplementary Table S6). Furthermore, upregulation of tyrosine phosphorylation sites in the SRC kinase family members (SRC, FYN, FGR, LCK, HCK, BLK, and LYN; highlighted in yellow, ALL-4 sheet; Supplementary Table S6) prompted us to test the TKI dasatinib against ALL-4. Exposure of ALL-4 cells to dasatinib resulted in complete dephosphorylation of pCRKL and pSTAT5 (Fig. 3B). The BCP-ALL PDX ALL-19 also exhibited very high intensity of ABL1-pY393, despite the absence of the BCR-ABL1 translocation (Fig. 3A), and downstream targets are highlighted in yellow, ALL-19 sheet; Supplementary Table S6. Upon dasatinib treatment, ALL-19 showed similar dephosphorylation of pCRKL and pSTAT5 as ALL-4 (Fig. 3B), although ALL-4 and ALL-19 exhibited disparate in vitro sensitivity to dasatinib (Fig. 3C). ALL-4 and ALL-19 were also evaluated for their in vivo sensitivity to dasatinib monotherapy (36). Although dasatinib significantly decreased the proportion of both ALL-4 and ALL-19 cells in the murine peripheral blood compared with vehicle-treated control mice (Fig. 3D), the sensitivity of ALL-4 (complete response) was greater than ALL-19 (significant progression delay). Subsequent analysis to investigate the aberrant ABL1 in ALL-19 using RT-PCR of fusion partner panels revealed a rare NUP214-ABL1 translocation. This translocation in the ALL-19 patient was further confirmed by Sanger sequencing (data not shown).

Phosphorylated FLT3 (Y969, Y842; Fig. 4A and B) was markedly upregulated only in ALL-2 along with phosphorylation of key downstream tyrosine phosphorylation targets (GAB1, GAB2, STAT5A, STAT3, and PLCG2) (highlighted in ALL-2 sheet; Supplementary Table S6). Exposure of ALL-2 cells to the multitargeted TKI sunitinib in vitro resulted in a marked dephosphorylation of pFLT3 (Fig. 4C). Subsequent exome sequencing of ALL-2 revealed a novel heterozygous mutation in FLT3 (c.1715A>C; p.Y572S). Although ALL-2 demonstrated limited in vitro sensitivity to sunitinib (Supplementary Fig. S8), it was the only PDX out of eight tested to achieve an objective response (complete response) to sunitinib monotherapy when tested in vivo (Fig. 4D; ref. 37).

The unbiased quantitative phosphotyrosine profiling of high-risk pediatric ALL PDXs also revealed elevated levels of phosphorylation of JAK1 phosphosites in all of the ETP-ALL PDXs examined when compared with the T-ALL PDX panel. The light (PDX) to heavy (SILAC) ratio of the JAK1 site pY1034 was the highest in ETP-8 (15-fold) followed by ETP-1 (3-fold; Fig. 5A). ETP-12 and ETP-13 also had an increased pJAK1 ratio in comparison with the other non-Ph–like PDXs. ETP-ALLs have been reported with mutations in the JAK kinases (10). To confirm that these activating JAK mutations translated from the genomic to the posttranslational level, the phosphorylation levels of STAT family proteins were investigated. pSTAT5 and pSTAT3 were clearly increased in ETP-ALL PDXs compared with T-ALLs (Fig. 5B). Overall, on the basis of the phosphotyrosine profiling and analysis of pSTAT proteins, it appears that the ETP-ALLs have a higher basal expression of JAK-STAT pathway activation compared with non-ETP T-ALLs despite the fact that not all of the ETPs profiled harbor JAK mutations (Table 1).

The marked upregulation of the JAK-STAT pathway in the ETP-ALL samples led us to hypothesize that inhibition of this pathway would considerably effect the survival of ETP-ALL PDX cells. We therefore tested the specific JAK1/2 inhibitor ruxolitinib, which is FDA approved for treatment of myeloproliferative diseases (38). Ruxolitinib inhibited JAK signaling by decreasing pSTAT5 and pSTAT3 levels in all ETP-ALL PDXs (Fig. 5B). However in the T-ALL PDXs ALL-8 and ALL-31, elevated levels of pSTAT5 and pSTAT3 were not observed (Fig. 5B). In vitro the ETP-ALL samples were marginally more sensitive to ruxolitinib compared with non-ETP T-ALL samples (Fig. 5C). ETP-8, which had the highest ratio of phosphorylated JAK in the profiling study (Fig. 5A), was also the most sensitive to ruxolitinib (IC₅₀ 0.2 μmol/L), followed by ETP-1 (IC₅₀ 0.8 μmol/L; Fig. 5C). Furthermore, in vivo ruxolitinib treatment of the most sensitive ETP-8 PDX resulted in significant reduction of spleen absolute blast counts (P = 0.0003) and weights (P < 0.0001) compared with the vehicle-treated group (Fig. 5D).

Although MS-based phosphotyrosine profiling is not currently feasible in routine clinical diagnostics, xenografting allows a viable option for this methodology to identify and predict clinically actionable kinase targets for patients who are difficult to treat with conventional therapies. As proof of principle, we xenografted a primary refractory BCP-ALL patient biopsy sample (A6199). Although the patient achieved an adequate morphologic remission with subsequent rounds of chemotherapy and received a donor transplant, the rising minimal residual disease levels were indicative of incipient ALL relapse. Low-resolution genomic analysis was unable to predict any targetable candidates. However, phosphotyrosine profiling of xenografted samples (ALL-77) of this patient revealed elevated levels of LYN kinase phosphorylation (Fig. 6A and Supplementary Table S7). We further validated the upregulation of LYN phosphorylation using immunoblot analyses and compared it with the patient sample along with a SRC kinase-activated positive control ALL-4 (Fig. 6B). Although phospho-LYN was detected in the frozen patient bone marrow samples, the levels were considerably higher in the freshly harvested xenografted samples. Furthermore, the exposure of the xenografted cells to dasatinib (1 μmol/L, 1 hour) resulted in complete dephosphorylation of Lyn kinase along with its downstream target phospho ERK (Fig. 6B).

Whole-genome sequencing and gene expression profiles of aggressive and chemoresistant high-risk subtypes of pediatric ALL have revealed several activating gene fusions, alterations and mutations that could result in constitutively activated TKs (11, 12, 39, 40). In this study, we have developed a quantitative MS-based phosphotyrosine profiling approach compatible with batch mode analysis of proteins obtained from pediatric ALL PDXs to track the global phosphorylation status of TKs. By targeting the activated kinase pathways with commercially available inhibitors, we have also shown that significant in vivo responses can be achieved. Overall, this study demonstrates the preclinical utility of our approach and in particular its potential to identify novel, individualized treatments for aggressive and chemorefractory pediatric ALL.

The success of characterizing the tyrosine phosphoproteome in cancer is exemplified by a number of MS-based studies (14, 15, 18, 41). These studies have provided insights into the role of tyrosine phosphorylation in leukemia progression and identified altered TK phosphorylation and novel targets for therapeutic opportunities (42–45). For example, a common phosphotyrosine signature for the BCR-ABL kinase in chronic myeloid leukemia cell lines along with novel fusion kinase signaling and potential drug-responsive biomarkers was identified in one study (23), whereas in another, the changes in T-cell receptor signaling pathways following cisplatin-induced apoptosis were analyzed (43). However, phosphotyrosine profiling studies in leukemias have used cell line models rather than primary patient samples due to the large amounts of protein extract required per experiment.

Even though in vitro model systems provide in-depth kinase signaling information, it is very difficult to predict the true effect that any signaling regulation observed would have within a disease microenvironment. To fully understand the relevance and regulation of tyrosine phosphorylation and improve prediction of clinical response after inhibitor treatment, it becomes imperative to study tyrosine phosphorylation in in vivo models or directly in patient samples. We recognize that there are significant technical hurdles and challenges for the profiling approach to become a diagnostic clinical tool. One of the key limiting factors being the procurement of large amounts of starting material (peripheral blood/bone marrow) required to profile the relatively low abundance (0.05%) phosphotyrosine molecules (46). However, in the interim, PDX models offer a relevant and feasible alternative to direct patient samples. We have previously shown that childhood ALL PDXs accurately reflect systemic disease and retain fundamental biologic characteristics of the original disease (including responsiveness to treatment; refs.19, 20). The establishment of a quantitative phosphotyrosine profiling method to track the TK signaling changes in a clinically relevant model of ALL is an important step forward.

The regulation of cellular signaling events at a molecular scale by TKs and phosphatases is not only affected by the phosphorylation at selected sites on a given protein, but also by their level of phosphorylation. By adopting a decoupled “spike-in” SILAC quantitative methodology (21) in PDX models of pediatric leukemia, we were able to successfully integrate this with the phosphotyrosine enrichment protocol. However, we initially observed a preferential capture of heavy SILAC-labeled phosphotyrosine peptides when equal amounts of PDX sample and “spike-in” SILAC standard were used. The likely explanation for this difference is that the xenografted samples had to go through extended processing steps before the cleared mononuclear cells could be lysed or frozen in comparison with cell lines that were directly lysed after a brief wash. It has been well established that processing times in patient samples have an impact on the phosphorylation status due to hypoxia and stress (17). Another possible explanation is that the cell lines receive a pervanadate treatment before cell lysis, in order to retain as many phosphorylated peptides as possible, whereas the PDX samples did not. The percentages of PDX phosphopeptides with a labeled SILAC partner and vice versa are shown in Supplementary Fig. S9.

One crucial finding in this analysis was a very high intensity of the Y393 site of TK ABL1 in the ALL-19 (BCP-ALL) PDX along with the classic B-cell signaling non receptor TKs. Y393 is an autophosphorylation site for the ABL1 kinase and is often constitutively activated in BCR-ABL1⁺ ALL patients (23, 30). In this study, we reconfirmed the phosphorylation of Y393 in ALL-4, which has a BCR-ABL1 translocation. The identification of active ABL1 in the BCP-ALL PDX (ALL-19) was a unique discovery as there was no previously reported BCR-ABL translocation in this leukemia. Upon further genetic analysis, we were able to discover a NUP214-ABL1 translocation in ALL-19 (unpublished observations). Episomal amplification of NUP214-ABL1 has been reported in T-cell malignancies and this cell type has also shown sensitivity to dasatinib monotherapy (47, 48). Although the NUP214-ABL1 translocation is rare in B-cell malignancies, it has been previously reported in Ph-like ALL (12). Moreover, we observed similar sensitivity to dasatinib in ALL-19 in vivo indicating the potential of phosphotyrosine profiling to identify activated protein kinases without prior knowledge of kinase translocations and the therapeutic value of the targets. Similarly, the unique upregulation of phosphorylated FLT3 in ALL-2 and the subsequent exome sequencing identification of a novel Y572S mutation highlights the potential of phosphotyrosine profiling to track functional aspects of kinase dysregulation due to point mutations. A similar mutation at site Y572 (Y572 >C) has been reported as a gain-of-function mutation that induced constitutive FLT3 activation. Y572, the first amino acid of the juxtamembrane domain of FLT3, plays a key role in kinase auto inhibition. The crystal structure of the auto inhibited form also reveals that any mutation at this site would result in disruption of its extensive interactions with the surrounding FLT3 subdomains and consequent destabilization of the inactive kinase conformation (49). The increase in phosphorylated FLT3 at known autophosphorylation sites (Y969, Y842) is consistent with high activity of FLT3 (50).

Protein kinase inhibitors represent a possible new treatment option for the chemoresistant ETP-ALL subtype. This high-risk subtype has been reported with high frequency of activating mutations within cytokine receptor and JAK/STAT signaling pathways (10, 51). In agreement with the genetic studies, we observed activation of JAK family proteins (JAK1, JAK2, and TYK2) as well as tyrosine phosphorylation of downstream targets STAT5, STAM1, and STAM2A previously reported to be phosphorylated by JAK1-JAK3 (Supplementary Table S6; refs. 10, 52). Furthermore, inhibiting the JAK-STAT pathway with ruxolitinib, a specific JAK1/2 inhibitor, clearly demonstrated the reliance of the ETP cells on the JAK/STAT pathway for survival. Taken together, these findings and the previous genetic studies suggest that JAK/STAT pathway hyperactivation may be a hallmark of ETP-ALL and that specific inhibition of this pathway has clear therapeutic relevance, even for those ETP-ALLs without JAK mutation. More recently, we have extended in vivo efficacy studies of ruxolitinib to a panel of 6 ETP-ALL PDXs and observed that ruxolitinib demonstrates robust activity irrespective of their JAK mutational status (28).

Deregulation of signaling pathways, whether as a result of fusion, deletion, mutation, or amplification of component gene products is a hallmark of high-risk ALL. Targeted inhibition of perturbed kinases, the key regulators of these signaling pathways, opens novel therapeutic options for high-risk pediatric ALL. Using an unbiased and quantitative tool to identify, the tyrosine phosphorylation state of activated kinases in xenograft models of leukemia, we have demonstrated the practical application of this approach in a clinically relevant environment. This study therefore highlights the potential of profiling relapsed and refractory pediatric ALL patients to identify and validate clinically relevant kinases for targeted inhibition with FDA-approved drugs in future clinical trials.

R. Sutton is a Conjoint Associate Professor in the University of New South Wales. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S. Dolai, K.C.S. Sia, P.J. Houghton, M.A. Smith, R.J. Daly, R.B. Lock

Development of methodology: S. Dolai, K.C.S. Sia, F. Hochgräfe, P.J. Houghton, R.J. Daly, M.J. Raftery, R.B. Lock

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Dolai, K.C.S. Sia, L. Zhong, A.K. Robbins, R. Sutton, T. Revesz, T. Revesz, D.T. Teachey, M.J. Raftery, R.B. Lock

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Dolai, K.C.S. Sia, T. Revesz, P.J. Houghton, R.J. Daly, M.J. Raftery

Writing, review, and/or revision of the manuscript: S. Dolai, K.C.S. Sia, L. Zhong, A.K. Robbins, T.L. Vincent, R. Sutton, T. Revesz, T. Revesz, P.J. Houghton, M.A. Smith, D.T. Teachey, R.J. Daly, R.B. Lock

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Dolai, T.L. Vincent, R.T. Kurmasheva, P.J. Houghton

Study supervision: R.T. Kurmasheva, P.J. Houghton, R.B. Lock

Other (Chief Investigator of grant funding this research): R. Sutton

The authors thank the Tissue Resources Core Facility of St. Jude Children's Research Hospital (Memphis, TN) for the provision of primary ETP-ALL samples and the Children's Oncology Group for the provision of the P9906 JAK-mutated primary patient samples. The authors also thank Agnes M. Mendomo for her assistance in development of the phosphotyrosine enrichment protocol in xenograft models of leukemia. Children's Cancer Institute Australia is affiliated with UNSW Australia and the Sydney Children's Hospitals Network.

This work was funded by Project Grant 1024232 from the Australian National Health and Medical Research Council. R.B. Lock was supported by a Senior Research Fellowship (1059804) from the Australian National Health and Medical Research Council.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.