The protein tyrosine phosphatase SHP2 is crucial for oncogenic transformation of acute myeloid leukemia (AML) cells expressing mutated receptor tyrosine kinases. SHP2 is required for full RAS-ERK activation to promote cell proliferation and survival programs. Allosteric SHP2 inhibitors act by stabilizing SHP2 in its autoinhibited conformation and are currently being tested in clinical trials for tumors with overactivation of the RAS/ERK pathway, alone and in various drug combinations. In this study, we established cells with acquired resistance to the allosteric SHP2 inhibitor SHP099 from two FLT3-ITD (internal tandem duplication)-positive AML cell lines. Label-free and isobaric labeling quantitative mass spectrometry–based phosphoproteomics of these resistant models demonstrated that AML cells can restore phosphorylated ERK (pERK) in the presence of SHP099, thus developing adaptive resistance. Mechanistically, SHP2 inhibition induced tyrosine phosphorylation and feedback-driven activation of the FLT3 receptor, which in turn phosphorylated SHP2 on tyrosine 62. This phosphorylation stabilized SHP2 in its open conformation, preventing SHP099 binding and conferring resistance. Combinatorial inhibition of SHP2 and MEK or FLT3 prevented pERK rebound and resistant cell growth. The same mechanism was observed in a FLT3-mutated B-cell acute lymphoblastic leukemia cell line and in the inv(16)/KitD816Y AML mouse model, but allosteric inhibition of Shp2 did not impair the clonogenic ability of normal bone marrow progenitors. Together, these results support the future use of SHP2 inhibitor combinations for clinical applications.
These findings suggest that combined inhibition of SHP2 and FLT3 effectively treat FLT3-ITD–positive AML, highlighting the need for development of more potent SHP2 inhibitors and combination therapies for clinical applications.
The SH2 domain–containing phosphotyrosine phosphatase 2 (SHP2) is a ubiquitously expressed nonreceptor protein tyrosine phosphatase (PTP), encoded by the PTPN11 gene. SHP2 comprises of two Src-homology-2 (SH2) domains at its N-terminus (N-SH2 and C-SH2), a central catalytic PTP domain and a C-terminal tail with two prominent activating phosphotyrosine sites, Tyr542 and Tyr580 (1). In the absence of a stimulus, SHP2 is kept in a closed, autoinhibited state by interactions between the N-SH2 domain and the PTP domain (2). SHP2 is activated by peptides containing appropriately spaced phosphotyrosine residues that bind the N-SH2 and C-SH2 domains of SHP2 in a bidentate manner, causing a conformational change of SHP2 into an open, active conformation. In this state, the interactions between the N-SH2 and the PTP domain are released, making the catalytic site available for substrate recognition and phosphate hydrolysis (2).
SHP2 is a central cellular signaling hub, typically relaying signals from upstream receptor tyrosine kinases (RTK), such as PDGFR (3) and EGFR (4). Downstream of these receptors, SHP2 is known as a key regulator of the RAS-ERK pathway (5).
Somatic PTPN11 mutations predominantly affect the N-SH2 and PTP domains, rendering SHP2 constitutively active, and are found in acute myeloid leukemia (AML), B-cell acute lymphoblastic leukemia (B-ALL), juvenile myelomonocytic leukemia (JMML), and solid tumors (6, 7). Furthermore, wild type (WT) SHP2 is essential for oncogenic transformation downstream of mutated RTKs (8). Combined, these findings support SHP2 as a potential drug target in the context of cancer treatment.
In 2016, Novartis developed a first-in-class potent and selective allosteric, noncovalent SHP2 inhibitor, SHP099. This molecule stabilizes SHP2 in its inactive conformation by binding to the interface between the N-SH2, C-SH2, and PTP domains (9). The therapeutic potential of SHP2 allosteric inhibitors is currently investigated for solid tumors in several clinical trials (10).
A recent study reported the effectiveness of SHP099 as a single agent in clinically relevant mouse models of AML, reducing leukemogenesis and leukemic blast stemness through downregulation of a Myc-driven gene signature (11). This observation was in contrast with published data showing that allosteric SHP2 inhibition (SHP2i) is only effective as combination treatment with MEK or RAS inhibitors in RTK mutation–driven cancers (12) and in a subset of KRAS-driven cancers (13, 14).
AML is a bone marrow malignancy characterized by a blockage of differentiation and an increased proliferation of myeloid hematopoietic progenitor cells (15). To investigate the sensitivity of human AML to allosteric SHP2i as a single agent, we focused on AML cell lines with internal tandem duplication (ITD) in the juxta-membrane domain of the RTK FLT3. This mutation is present in 25% of patients with AML and it leads to constitutive activation of its tyrosine kinase activity (16).
Here, we show that the initial phospho-ERK (pERK) downregulation observed in FLT3-ITD–positive AML cell lines treated with SHP099 is transient, and rapidly overpowered by adaptive resistance mechanisms. To analyze the global molecular changes induced by resistance to SHP099, we employed quantitative mass spectrometry (MS)-based proteomics and phosphoproteomics. We further investigated the role of SHP2 phosphorylation on Tyr62 using knockdown and single-site mutations to uncover a feedback activation of FLT3 that leads to the emergence of adaptive resistance through reactivation of the SHP2-RAS-ERK axis. Finally, we show that this adaptive resistance can be overcome by cotargeting components of the same pathway, such as the MEK or FLT3 kinases.
Materials and Methods
The MOLM-13 cell line was purchased from DSMZ line (catalog no. ACC-554, RRID: CVCL_2119), while the MV-4-11 and U-2 osteosarcoma (OS) cell lines were purchased from ATCC (catalog no. CRL-9591, RRID: CVCL_0064; catalog no. HTB-96, RRID: CVCL_0042). The HB11;19 cell line (RRID: CVCL_8227) was kindly provided by Dr. Yana Pikman. MOLM-13 and HB11;19 cells were cultured in RPMI1640 with GlutaMAX (Gibco) supplemented with 10% FBS and 1% penicillin-streptomycin (P-S). MV-4-11 cells were cultured in Iscove's Modified Dulbecco's Medium with GlutaMAX (Gibco) supplemented with 25 mmol/L HEPES (Gibco), 10% FBS, and 1% P-S. The human osteosarcoma cell line U-2 OS was cultured in DMEM with GlutaMAX (Gibco) supplemented with 10% FBS and 1% P-S. AML cell lines were starved in respective serum-free media, supplemented with 1% BSA (Sigma-Aldrich). Cell lines were kept in a humidified incubator at 37°C, with 5% CO2. All cell lines were regularly checked for Mycoplasma using the EZ PCR Mycoplasma detection Kit (Biological Industries). To avoid artefacts of long-term culture of immortalized cell lines, cell cultures were started up from an early passage vial and interrupted after 8–12 weeks.
Cell treatment and acquired resistance generation
SHP099, RMC-4550, cobimetinib, gilteritinib, and KW-2449 were purchased from Selleckchem. II-B08 was purchased from Sigma-Aldrich. Recombinant human, animal free EGF was purchased from Peprotech.
For acquired resistance generation, cells were initially treated with SHP099 at their respective IC50, followed by dose increase at each passage for over 3 weeks, stopping the increase at doses corresponding approximately 5× IC50 (2.5 μmol/L for MV-4-11 and 10 μmol/L for MOLM-13), and continuing cultivation at those doses of SHP099 for a maximum of 6 weeks.
Cell survival assay and IC50 determination
Approximately 25,000 cells were plated in a clear bottom 96-well plate and treated with increasing drug doses. Cell viability was measured after 72 hours, using the Cell counting Kit 8 (Tebu-Bio), according to manufacturer's instructions. Absorbance was measured using a FLUOstar Omega microplate reader (BMG Labtech). IC50 values were determined in GraphPad Prism v8–9 (RRID: SCR_002798), using nonlinear regression analysis (log inhibitor vs. response, four parameters).
inv(16)/KitD816Y AML mouse model
The inv(16)/KitD816Y AML mouse model has been described before (17). Animals were housed according to institutional guidelines at the University of Copenhagen (Copenhagen, Denmark) and experiments were approved by the Danish Animal Research Ethical Committee.
Colony-forming cell assay
Colony-forming assay (CFC) assay was performed as described previously (18). A total of 1,500 of MV-4-11 cells or 12,500 of primary murine bone marrow cells were seeded in a 12-well plate in 0.5 mL of MethoCult Methylcellulose-Based Media (StemCell Technologies: H4435 for human cells, M3434 for murine cells) supplemented with 1% P-S. Colonies were manually counted after 7–14 days.
A detailed protocol can be found in the Supplementary Materials and Methods. Information on number of replicates for each experiment is provided in Supplementary Table S1.
Determination of Ras-GTP levels
The active RAS pulldown and detection kit (Thermo Fisher Scientific, catalog no. 16117) was used according to manufacturer's instructions.
A total of 1 × 106 cells per condition were transfected with 20 nmol of siRNA using the Neon Transfection System (Invitrogen) with the following setting: 1,400 V; 20 ms; 2 pulses. All assays 100 μL kit were carried out after 72 hours knockdown. Double-stranded siRNAs targeting human PTPN11 were purchased from Dharmacon as SmartPool. Stealth RNAi siRNA GFP Reporter Control duplex (Invitrogen) was used as negative control.
PTPN11 point mutations were generated from the parental pCMV-SHP2-WT (RRID: Addgene_8381) using site directed mutagenesis (QuikChange II XL, Agilent), according to manufacturer's instructions.
U-2 OS cells were transfected for 24 hours using Lipofectamine 2000 (Invitrogen) with either 1 μg (6-well plate) or 15 μg (15 cm plate) pCDNA 3.1 (Thermo Fisher Scientific, catalog no. V79020), WT SHP2 or mutants.
Cellular thermal shift assay
Cellular thermal shift assay (CETSA) was performed as described previously (19). A detailed protocol can be found in the Supplementary Data.
Evaluation of drug synergy
The synergy of small-molecule inhibitor combinations was calculated utilizing the SynergyFinder web application 2.0 (RRID: SCR_019318; ref. 20).
Multiple sequence alignment
Multiple sequence alignment was performed using the Clustal Omega tool (RRID: SCR_001591).
For the modeling of the SHP2-mutant Y62D, we used the Dynamut web interface (RRID: SCR_021849; ref. 21). We used the deposited crystal structure of closed SHP2, in complex with the SHP099 inhibitor (PDB: 5EHR) and indicated the Y62D mutation on chain A as the mutation.
MS-based quantitative proteomics
Extended methods can be found in Supplementary Data.
Single-shot label-free proteomics and phosphoproteomics
After SDS-based lysis, proteins were on-bead digested through the PAC protocol (22). A total of 750 ng of peptide were loaded on evotips (Evosep) for proteome analysis. The remaining peptides were cleaned-up on Sep-Pak and subjected to automated Ti-IMAC phosphopeptide enrichment (200 μg peptide input). After elution, the phosphopeptides mixtures were loaded on evotips for MS analysis. Samples were analyzed on the Evosep One system (23) coupled to an Orbitrap Exploris 480 (Thermo Fisher Scientific; ref. 24). The mass spectrometer was operated in data-independent acquisition (DIA) mode. Raw MS data were analyzed using the Spectronaut software (25) with the direct DIA workflow (proteome) or with by using a project-specific spectral library (phosphoproteome).
Tandem mass tag phosphoproteomics
After guanidine-based lysis, proteins were digested in solution. Following Sep-Pak cleanup, peptides were tandem mass tag (TMT)-labeled (10-plex kit, Thermo Fisher Scientific), pulled and subjected to TiO2 enrichment (300 μg of peptides per sample: total 3 mg). After elution, phosphopeptide mixtures were separated by offline high-pH reversed-phase fractionation into 12 concatenated fractions. All fractions were acidified, dried, and subsequently solubilized in 5% acetonitrile, 0.1% trifluoroacetic acid prior to MS analysis. Samples were analyzed on an Easy nLC 1200 (Thermo Fisher Scientific) coupled to a Q Exactive HF-X (Thermo Fisher Scientific; ref. 26). The mass spectrometer was operated in data-dependent acquisition (DDA) mode. Raw MS data were analyzed using the MaxQuant software (RRID: SCR_014485; ref. 27).
For all datasets, protein entries identified as potential contaminants were eliminated from the analysis. We kept only the phosphorylation sites with a localization score ≥ 0.75. After log2 transformation, MS intensities were realigned to the median signal. DEqMS (28) was used for the differential expression analysis of the DIA proteome, using the minimum number of peptide spectrum matches per run for variance estimation. Protein groups were considered regulated when presenting a Storey FDR (q-value) ≤ 0.05 between resistant or SHP099 treatment and DMSO (paired analysis). The same FDR strategy was applied to the phosphopeptides using the LIMMA package (RRID: SCR_010943; ref. 29) with a more stringent threshold (q-value ≤ 0.01). When possible, we calculated phosphorylation site occupancy on the mean of replicates as described before (30). For TMT phophoproteomics, the bioinformatics analysis was performed on the Phosphosite(STY).txt table of MaxQuant. Reverse identification were removed. Multiple phosphorylation sites quantified from the same peptide were labeled “protein_site1+site2.” We performed significance analysis using LIMMA for three comparisons (SHP/CTRL, RES/CTRL, RES/SHP), and we considered significantly regulated the phosphorylation sites with a q-value ≤ 0.05. Kinase activity prediction was performed with RoKAI (31) using kinase substrates from PhosphositePlus and Signor.
Data availability statement
All raw MS data were generated by the authors and deposited to the ProteomeXchange Consortium (RRID: SCR_004055), via the PRIDE partner repository (RRID: SCR_003411; ref. 32) with the dataset identifier PXD030338.
FLT3-ITD AML cell lines rapidly develop acquired resistance to SHP099
To characterize the cellular and molecular responses to the allosteric SHP2 inhibitor SHP099, we selected two commercial AML cell lines based on their strong genetic dependency on both PTPN11 and FLT3 in the Cancer Dependency Map (DepMap) database (RRID: SCR_017655; ref. 33): MV-4-11 and MOLM-13 (Fig. 1A; Supplementary Fig. S1A). Both cell lines are known to be sensitive to SHP2 (9) and FLT3 inhibition (34). We confirmed their sensitivity to SHP099, with a calculated half-maximal inhibitory concentration (IC50) of 0.5 μmol/L for MV-4-11 and 1.3 μmol/L for MOLM-13 (Fig. 1B, blue lines). The response observed for SHP099 was mimicked by treatment with RMC-4550 (Supplementary Fig. S1B, blue lines), another potent and selective SHP2 allosteric inhibitor (35).
To investigate whether the two cell lines could develop acquired resistance to SHP099, we treated them for 3 weeks with increasing drug concentration (Fig. 1C, red lines). After chronic drug exposure, both cell lines became drug resistant. We confirmed that these resistant cell lines (MOLM-13/R and MV-4-11/R) showed increased IC50 values in survival assays (Fig. 1B, red lines) as well as cross-resistance to RMC-4550 (Supplementary Fig. S1B, red lines). To test for potential drug resistance reversibility, we removed SHP099 from the culture media of MV-4-11/R and MOLM-13/R cells for 3 weeks. After this washout period, both showed a significant decrease in SHP099 IC50 (Fig. 1D; Supplementary Fig. S1C and S1D), indicating that acquired resistance was reversible.
Proteomics and phosphoproteomics characterization of SHP099 resistance
To elucidate the molecular basis of SHP099 acquired resistance, we analyzed the proteome and phosphoproteome changes induced by SHP099 treatment in the MV-4-11 and MOLM-13 at early (1 and 24 hours) and late (72 hours and 21 days) timepoints (Fig. 2A).
By DIA of biological quadruplicates per treatment condition, 5,495 protein groups were quantified by label-free quantification across all samples (Supplementary Table S2). The principal component analysis (PCA) of the proteome differences is presented in Fig. 2B. SHP099 treatment induced a shift in PC1. Resistant cell proteomes separated from all other conditions in both cell lines but with different patterns: MV-4-11/R separated from the parental cells on the same component as of SHP099-treated cells (PC1), while MOLM-13/R differed from the parental cells on the third principal component (PC3), suggesting that these proteomes presented different features with respect to parental MOLM-13 cells.
In the phosphoproteome, we quantified 17,024 phosphorylation sites across all samples (Supplementary Table S3). Because phosphopeptide upregulation and downregulation can be due to protein-level remodeling, we calculated phosphorylation site occupancy, alongside phosphopeptide relative quantities, for the sites that were located on proteins also detected in the proteome, with quantitation of their cognate nonphosphorylated peptide (30). PCA of phosphorylation sites occupancies (Fig. 2C) shows that SHP099 treatment drove phosphoregulations early in MOLM-13, as SHP099-treated parental cells were separated from parental cells in the PC1. Yet, phosphorylation remodeling was less pronounced in MV-4-11 with separation alongside the PC2 explaining only 7% of variance. For MOLM-13, resistant cells appeared to revert toward the parental cells along the course of SHP099 treatment. To identify the activated kinases responsible for the phosphorylation site occupancies, we performed kinase activity prediction (31), which showed a decrease of predicted MEK/ERK activity at 1 hour that was not maintained over time (Fig. 2D).
We performed the statistical analysis of protein and phosphorylation site remodeling using a LIMMA-based approach (29) where variance was estimated as a function of intensities for the phosphopeptides and number of peptide-spectrum matches for the protein groups (DEqMS; ref. 28). This analysis identified 1,434 and 3,230 protein groups that were differentially regulated upon resistance acquisition in the MOLM-13 and MV-4-11, respectively, whereas 1,522 and 1,013 proteins were regulated after SHP099 treatment in MOLM-13 and MV-4-11 parental cell lines (Fig. 3A). Consistent with the PCA plots (Fig. 2B), all proteins regulated upon short time SHP099 treatment in the parental MV-4-11 cells were found regulated between MV-4-11/R and MV-4-11, whereas in MOLM-13 the set of proteins regulated upon short-term SPH099 treatment differed from long-term MOLM-13 proteome remodeling. Only a partial overlap was observed between the two cell lines, which indicates that cell line–specific mechanisms may be responsible for SHP099 resistance. Nevertheless, many proteins passed the statistical threshold in one of the two cell lines while presenting common expression kinetics over the course of the experiment (Fig. 3B).
Among the phosphorylation sites regulated by SHP099 treatment, we observed a strong dephosphorylation of ERK2/MAPK1-Y187 after 1 hour of SHP099 treatment in both cell lines, while this phosphosite only returned to baseline level of parental cell lines at longer timepoints. The same was observed for ERK1/T202+T207, two phosphothreonines measured on the same peptide that were under the level of detection in the parental cells 1 hour post-SHP099 treatment in both cell lines (Fig. 3C).
To validate the phosphoproteome data discussed above and investigate further the phosphoregulations underlying SHP099 resistance acquisition, we performed a TMT-based quantitative phosphoproteomics experiment (Supplementary Fig. S2A; Supplementary Table S4). TMT labeling allows sample multiplexing with more precise quantification (Supplementary Fig. S2b) and returns fewer missing values than classical label-free strategies (36). We focused on phosphorylation changes induced by SHP099 treatment in both parental and resistant MV-4-11 cells after 1 hour, when pERK rebound was observed in resistant cells (Fig. 3C). Overall, we quantified 18,832 phosphosites. This deepened the phosphoproteome of SHP099-treated MV-4-11 and MV-4-11/R, reaching in total 32,256 phosphorylation sites (Supplementary Fig. S2C). ERK2/MAPK1-Y187 and ERK1/MAPK3-Y204 were both regulated in a similar fashion in the DIA and TMT data, like many other significantly regulated sites (Fig. 4A). More importantly, the TMT-based phosphoproteome dataset included two significantly regulated sites on SHP2: Tyr62 and Tyr542 (Fig. 4B and C; Supplementary Fig. S3A). Tyr542 followed the dynamics of pERK, by being downregulated in SHP099 treated parental cells compared with control and rescued in the acquired resistant cells. Conversely, Tyr62 displayed a specific dynamic behavior characterized by an absence of regulation upon SHP099 treatment in parental cells and an increase only in resistant cells. Importantly, total SHP2 protein level was not regulated in any condition studied, indicating that the observed phosphorylation changes did not result from protein-level regulation (Supplementary Fig. S3B). Phosphorylation on Tyr542 on SHP2 has an adaptor function for the recruitment of GRB2 (37) and subsequent activation of the RAS-RAF-MEK-ERK signaling axis. Moreover, this phosphosite has been shown to interact with the N-SH2 domain, thus relieving basal inhibition of the PTP domain (38). SHP2 Tyr62 phosphorylation, less well studied than Tyr542, was shown to correlate with an activated state of the PTP domain (39).
pERK rebound in acquired resistant cells depends on MEK and SHP2
We further confirmed by immunoblotting that resistant cells reactivated pERK in the presence of the drug within 1 hour of treatment (Fig. 5A; Supplementary Fig. S4A). We noted that pERK rebound happened also after 72 hours of SHP099 treatment in parental cells, confirming the phosphoproteomics data (Fig. 3C). Importantly, in the drug washout condition, SHP099 treatment regained the ability to inhibit ERK phosphorylation to the same extent as parental cells (Fig. 5B; Supplementary S4B), confirming that resistance was reversible. In addition, we tested whether the pERK rebound observed in acquired resistant cells was dependent on MEK. We treated MV-4-11 and MOLM-13 acquired resistant cells with the clinical MEK inhibitor cobimetinib and observed a decrease in the levels of ERK phosphorylation, indicative of its MEK dependency (Fig. 5C; Supplementary Fig. S4C). To assess whether the observed MEK-dependent pERK upregulation was responsible for resistant cell survival, we treated parental and resistant cells with cobimetinib for 72 hours (Fig. 5D; Supplementary Fig. S4D). Cobimetinib treatment completely abolished resistant cell survival advantage, proving that resistant cells depend on the ERK pathway. To validate that SHP099 resistance was caused by RAS-ERK pathway reactivation, we performed RAS-GTP pulldown in the MV-4-11 parental and resistant cell lines (Supplementary Fig. S4E). While SHP099 decreased RAS-GTP loading in parental cells, RAS-GTP levels were fully rescued in acquired resistant cells, verifying RAS activation in presence of the drug.
Next, we analyzed phosphorylation of the Tyr62 and Tyr542 on SHP2 by immunoblotting, confirming its upregulation and rescue in resistant cells, respectively (Fig. 5E; Supplementary S4F). To determine the functional role of the upregulation of SHP2 phosphorylation in resistant cells, we treated parental and resistant MV-4-11 cells with the active-site inhibitor II-B08, which blocks SHP2 phosphatase activity, in combination with SHP099 (Fig. 5F). This experiment showed that II-B08 reverted pERK rebound in resistant cells, indicating that SHP2 regained phosphatase activity was responsible for upregulation of pERK in acquired resistant cells. Interestingly, II-B08 induced pERK upregulation in parental cells, possibly due to concomitant SHP1 inhibition, which is known to activate the ERK pathway (40). To further confirm that SHP2 reactivation was responsible for acquired resistance, we performed SHP2 knockdown for 72 hours in MV-4-11 parental and resistant cells (Fig. 5G). Downregulation of pERK was comparable in the knockdown in both parental and resistant cells. This correlated with a significant reduction in cell survival, both in parental and resistant cells (Fig. 5H), proving that MV-4-11/R cells were still dependent on SHP2.
SHP2 phosphorylation on Tyr62 stabilizes SHP2 in the open conformation
Restored SHP2 phosphatase activity in presence of SHP099 in resistant cells indicated that the drug was not able to inhibit its target, suggesting that SHP2 was in its open, active conformation. To test whether increasing the drug concentration could shift the equilibrium toward the closed, inhibited conformation, we treated the parental and acquired resistant cell lines MV-4-11 and MOLM-13 with increasing concentrations of either SHP099 or RMC-4550 (Fig. 6A; Supplementary Fig. S5A). In MV-4-11/R cells, high doses of SHP099 partially prevented pERK rebound. In MOLM-13/R cells, only the more potent RMC-4550, but not SHP099, was able to partially inhibit ERK phosphorylation.
We investigated which structural, partially reversible, determinant could be responsible for SHP2 conformational change. Tyrosine phosphorylation is a transient posttranslational modification able to regulate protein function through multiple mechanisms, including conformational changes. Tyr62 is highly conserved from zebrafish to humans (Supplementary Fig. S5B) and it is part of the N-SH2 domain of SHP2, which is a known mutational hotspot in multiple diseases, including Noonan syndrome and JMML (6, 7, 41, 42). Mutations in this region generally result in constitutively active SHP2 mutants by stabilizing the open conformation of SHP2, thus antagonizing SHP2i by SHP099 (43). Intriguingly, we noticed a poorly characterized disease-causing mutation involving the Tyr62 on SHP2 (Y62D), which is both a somatic mutation in AML (COSMIC ID: COSV61011774) and a germline mutation in 42 reported cases of Noonan syndrome (dbSNP ID: rs121918460 and rs121918459; ref. 44). We therefore hypothesized that SHP2 phosphorylation on the Tyr62 or the Y62D mutation would stabilize SHP2 in its open conformation, making SHP099 ineffective. To determine changes in SHP2 conformation induced by this mutation, we performed in silico structural homology modeling through the DynaMut web interface (21). This highlighted that most ionic and hydrophobic bonds within the neighboring residues in the three-dimensional (3D) structure are broken upon introduction of aspartic acid in the position 62 (Fig. 6B; Supplementary Fig. S5C). Through the calculation of the free energy changes (ΔΔG), the Dynamut tool predicted that the Y62D mutation was destabilizing (ΔΔG = −1.636 kcal/mol), suggesting that it would lead to opening of the structure.
To experimentally validate this in the context of SHP2i by SHP099, we generated a mammalian SHP2 mutant by site directed mutagenesis, replacing the Tyr62 either by a phosphomimetic glutamic acid (Y62E, gain of function, GoF) or a nonphosphorylatable phenylalanine (Y62F, loss of function, LoF; Supplementary Fig. S5D). We transfected these mutants in U-2 OS cells and monitored pERK upon EGF stimulation as a proxy of SHP2 activation/inhibition in absence or presence of SHP099 (Fig. 6C; Supplementary Fig. S5E). We observed the same trend for pERK downregulation in nontransfected cells and cells transfected with the WT SHP2, with the highest SHP099 concentration (25 μmol/L) not being able to fully inhibit pERK. When we transfected the LoF mutant, 5 μmol/L of SHP099 induced pERK downregulation. Conversely, transfection with the GoF mutant nearly abolished SHP099-driven pERK downregulation.
To investigate the ability of WT and mutant SHP2 to bind SHP099, we performed a CETSA (19) in U-2 OS cells, either nontransfected (NT), transfected with WT SHP2 or with LoF and GoF SHP2 mutants (Fig. 6D). The protein melting temperature of SHP2 in the NT, WT, and Y62F-transfected cells increased upon SHP099 treatment in comparison with DMSO, indicating that nonphosphorylated SHP2 was stabilized, thus bound to SHP099. Conversely, treatment with SHP099 in the Y62E-expressing cells showed no sign of protein stabilization upon drug introduction, implying that phosphomimetic Y62E SHP2 could not bind SHP099.
pERK rebound in acquired resistant cells is caused by feedback activation of the FLT3 receptor
Next, we investigated which kinase could be responsible for SHP2 Tyr62 phosphorylation. Querying the NetworKIN prediction tool (45), we found that this phosphorylation is a predicted target of both EGFR and INSR, suggesting that it depends on an activated RTK (Supplementary Fig. S6A).
As RTKs need an extracellular ligand to be activated, we investigated whether serum deprivation could abolish pERK rebound in acquired resistant cells. Surprisingly, a full serum starvation for 12 hours was not able to prevent pERK rebound in the MV-4-11/R (Fig. 7A). Because MV-4-11 is known to express two FLT3-ITD–mutated alleles, we hypothesized that FLT3 was the RTK responsible for SHP2 reactivation. In support of this, we found that the autophosphorylation site FLT3 Tyr-969 was higher in both MV-4-11 and MOLM-13 acquired resistant cells compared with parental (Fig. 7B). This suggested that chronic SHP099 treatment relieved a negative feedback loop, leading to FLT3 activation in acquired resistant cells. To confirm this, we treated the MV-4-11 and MOLM-13 parental and resistant cells with the clinical second-generation FLT3 inhibitor gilteritinib for 72 hours and calculated its IC50 (Fig. 7C; Supplementary S6B). Both MV-4-11 and MOLM-13 were significantly more sensitive to FLT3 inhibition compared with their SHP099-resistant counterpart, indicating that resistant cell survival is dependent on FLT3. Moreover, we observed a synergistic effect for the combined treatment of RMC-4550 and gilteritinib in a CFC assay, where parental MV-4-11 was treated for 7 days (Fig. 7D). This synergy was validated by using the second-generation FLT3 inhibitor KW-2449, which completely abolished the resistant cell survival advantage (Supplementary Fig. S6C and S6D). In addition, gilteritinib was able to revert pERK rebound both in MV-4-11 (Fig. 7E). These data were confirmed in MV-4-11 and MOLM-13 by using KW-2449 (Supplementary Fig. S6E and S6F).
SHP099-induced pERK rebound is observed in other RTK-driven leukemia models
To validate whether the molecular mechanisms discussed above were limited to FLT3-driven AML, we investigated the B-ALL cell line HB11;19, as it showed the highest genetic dependence on both FLT3 and PTPN11 in the DepMap database (Fig. 1A). In B-ALL, mutations in FLT3 and PTPN11 are common and often present in relapse-fated clones (46). HB11;19 harbors the FLT3 point mutation D835H, located in its tyrosine kinase domain, leading to ligand-independent FLT3 kinase activation. Surprisingly, HB11;19 cells were resistant to SHP099 (Supplementary Fig. S7A). SHP099 resistance was likely due to the fast pERK reactivation and upregulation following SHP099 treatment (Fig. 8A). However, they were sensitive to the FLT3 inhibitor gilteritinib (Supplementary Fig. S7B). Combined treatment with SHP099 and gilteritinib resulted in a synergistic effect on cell survival (Fig. 8B; Supplementary Fig. S7C) and it was able to reverse pERK rebound (Fig. 8C).
To assess whether we identified a general RTK-driven Shp2i resistance mechanism in leukemia, we tested allosteric Shp2i in the context of murine inv(16)/KitD816Y AML (17). Kit is a transmembrane RTK, crucial for the development of hematopoietic stem cells. KitD816Y is one of the most common substitutions in AML, which causes constitutive activation of the receptor (47). Ex vivo AML blasts were sensitive to RMC-4550, while normal bone marrow progenitor clonogenic potential was not affected to the same extent (Fig. 8D). Combined treatment with RMC-4550 and the Kit inhibitor BLUE-285 synergistically reduced AML clonogenic potential (Fig. 8E), suggesting that allosteric Shp2i might be associated with signaling adaptation in a variety of RTK-driven bone marrow malignancies.
Currently, resistance to targeted therapy is the major hurdle in personalized cancer treatment. Intrinsic resistance is usually genetic, while acquired resistance typically arises through a number of diverse and often nongenetic mechanisms (48). Here, we show that the allosteric inhibitor of the oncogenic phosphatase SHP2, SHP099, causes development of acquired resistance through nongenetic mechanisms in AML cells. Previous studies have shown the efficacy of SHP099 monotherapy in FLT3-ITD AML mouse models (11). However, in these investigations mice were only treated for a maximum of 35 days, which might not be enough for adaptive resistance to arise. Prior studies support our model, as they have shown that resistance to SHP099 was caused by feedback activation of RTKs in in vitro models of RTK-driven solid tumors, and FGFR was the RTK responsible for resistance (49).
To understand the mechanism of acquired resistance, we utilized a phosphoproteomics approach, which allowed us to identify the Tyr62 as a resistant-specific phosphorylation site on SHP2. Using site-directed mutagenesis and in silico modeling, we suggest that introducing a high negative charge at residue 62 acts by forcing SHP2 to keep an open conformation, corroborating prior findings of a role for this phosphorylation site in maintaining an activated state of the PTP domain (39). Consistent with this finding, we show that the resistant cell lines still depend on SHP2 and that acquired resistance is caused by on-target reactivation. We prove that the phosphomimetic Y62E GoF variant cannot bind SHP099, resulting in resistance toward SHP2i, while nonphosphorylatable Y62F LoF variant displays the same level of affinity for SHP099 as the WT and is strongly inhibited by SHP099.
In this work, we use fractional stoichiometry to assess changes in phosphorylation dynamics. We opt for the approach originally proposed by our lab (30), which allows preserving information about phosphorylation site localization. A possible alternative for occupancy calculation is the one proposed by Gygi and colleagues (50), which in turn has the advantage to analyze the phosphorylated and unphosphorylated counterparts in the same run, thus reducing the number of missing values.
The finding presented here could be applied in the clinical setting in several ways. First, monitoring phosphorylation levels of SHP2 Tyr62 could be used as a prognostic marker for patients developing resistance to the administered drug. It would also be important to monitor patients receiving SHP099 for longer periods to identify whether they develop mutations in that particular site, or in close proximity. Second, because we found that resistance is reversible and can be targeted by rational combination treatment, such regimens could be developed. Finally, our study shows that treatment strategies based on allosteric SHP2 inhibitors still need further development to avoid resistance in the clinics and it suggests that the next-generation SHP2 inhibitors should prevent phosphorylation of Tyr62.
In conclusion, this study determines the path of adaptive evolution to SHP2i resistance in an AML setting and emphasizes the need for further investigation of allosteric SHP2 inhibitors and whether acquired resistance arises in tumors other than AML.
A. Pfeiffer reports grants from Novo Nordisk Foundation during the conduct of the study and personal fees from Scandion Oncology outside the submitted work. A. Pfeiffer reports grants from Copenhagen Bioscience PhD Programme. G. Franciosa reports grants from Novo Nordisk Foundation during the conduct of the study. S.C. Blacklow reports grants and personal fees from ERASCA, Inc.; grants from Novartis, Inc.; personal fees from Scorpion Therapeutics, Odyssey Therapeutics, MPM Capital, Ayala Pharmaceuticals, and Droia Ventures outside the submitted work. L.J. Jensen reports personal fees from Intomics A/S outside the submitted work and is a founder and owner of Intomics A/S. J.V. Olsen reports grants from Novo Nordisk Foundation during the conduct of the study; non-financial support from Acrivon Therapeutics, grants from Thermo Fisher Scientific and Novo Nordisk a/s outside the submitted work. No disclosures were reported by the other authors.
A. Pfeiffer: Conceptualization, data curation, formal analysis, validation, investigation, methodology, writing–original draft, writing–review and editing. G. Franciosa: Conceptualization, formal analysis, supervision, validation, investigation, writing–original draft, writing–review and editing. M. Locard-Paulet: Formal analysis, investigation, visualization, writing–review and editing. I. Piga: Data curation, validation. K. Reckzeh: Formal analysis, validation. V. Vemulapalli: Investigation, writing–review and editing. S.C. Blacklow: Resources, supervision, writing–review and editing. K. Theilgaard-Mønch: Resources, supervision, writing–review and editing. L.J. Jensen: Resources, supervision, writing–review and editing. J.V. Olsen: Conceptualization, resources, supervision, funding acquisition, project administration, writing–review and editing.
The authors would like to acknowledge Dr. Tasian and Dr. Patrick Brown for generating the B-ALL HB19;19 cell line, as well as Dr. Yana Pikman for providing the aforementioned cell line.
Work at the NNF CPR was funded by a donation from the NNF (NNF14CC0001). G. Franciosa was funded by the NNF Exploratory Interdisciplinary Synergy Programme (NNF20OC0064594). A. Pfeiifer was funded by the Copenhagen Bioscience PhD Program from the NNF (NNF16CC0020906).
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.