If not indicated otherwise, chemicals and reagents were obtained from Sigma-Aldrich (Munich, Germany) or Carl Roth (Karlsruhe, Germany).

K-562 cells (RRID: CVCL_0004), established from the pleural effusion of a 53-year-old woman (14), were obtained from the German Collection of Microorganisms and Cell Cultures (DSMZ, Braunschweig, Germany). Cell maintenance, generation of biological replicates of TKI-resistant sublines, and analyses of cell line authenticity were described elsewhere (15, 16). Cells were resistant against lowIM (0.5 µM imatinib), highIM (2 µM imatinib), lowN (0.05 µM nilotinib) and highN (0.1 µM nilotinib). The concentrations were chosen to reflect the clinically typical range of estimated imatinib plasma concentration, as well as the 20-fold higher potency of nilotinib.

Total RNA was isolated using E.Z.N.A Total RNA kit 1 (Omega bio-tek, Norcross, GA, USA). Cell line DNA was purified using Gentra Puregene Kit (Qiagen, Hilden, Germany).

Exome sequencing was performed using Illumina InView Human Exome Advance sequencing technology, a random-primed cDNA library, 60x coverage, and 2 x 150 bp read length at Eurofins Genomics (Ebersberg, Germany). Raw data was mapped against GRCh38. Exome data was processed similarly to Künstner et al. (17). For the detailed bioinformatic analysis, see supplement .

Exome sequencing data was validated using Next Generation Sequencing (NGS) SBS technology with Illumina MiSeq after PCR amplicon preparation with the Nextera XT Sequencing Kit (Illumina, San Diego USA). For this purpose, amplicons of the respective genes were generated using gene-specific primers, primer-specific annealing temperatures and MyTaq DNA Polymerase (Meridian Bioscience, Memphis, TN, USA). ( Supplementary Table 1 ). Genomic DNA from 2 µM imatinib resistant K-562 cells replicate 2, and 0.1 µM nilotinib resistant K-562 cells replicate 2 served as templates. PCR products were extracted using GeneJet Gel Extraction Kit (Thermo Fisher Scientific, Darmstadt, Germany) according to the manufacturer’s recommendations. MiSeq was performed according to the manufacturer’s protocol, as already described (18).

Microarrays were performed using Clariom S Arrays (Affymetrix; Thermo Fisher Scientific) as previously described (16). Briefly, RNA was isolated using miRVANA microRNA isolation kit (Thermo Fisher), and 100 ng were hybridized onto the arrays according to the manufacturer’s protocol. Further details about data processing and analysis are given in the supplement.

Whole-cell lysates and immunoblotting were performed as described elsewhere (19, 20). Blots were probed with the following antibodies obtained from Santa Cruz or CST (Danvers, MA, USA): phospho-ERK: Cat# sc-7383, RRID AB_627545, 1:1000; ERK: Cat# sc-514302, RRID : AB_2571739, 1:1000; SHP2: Cat# 3397, RRID: AB_2174959, 1:1000; PDGFRβ: Cat# sc-374573, RRID: AB_10990921, 1:100; pan-RAS: Cat# sc-166691, RRID: AB_2154229, 1: 200; GAPDH: Cat# sc-47724, RRID: AB_627678, 1:2000; anti-rabbit: Cat# 926-32211, RRID: AB_621843; Cat# 926-926-68071, RRID: AB_10956166; anti-mouse: Cat# 926-32210, RRID: AB_621842, Cat# 926-680707, RRID: AB_10956588; all 1:10,000, LiCOR (Bad Homburg, Germany). Primary antibodies were diluted in Intercept/TBS blocking solution (LiCOR) supplemented with 0.2% Tween-20, secondary antibodies were diluted in TBS supplemented with 0.1% Tween-20. Total protein staining was performed using Revert 700 Total Protein Stain Solution according to the manufacturer’s protocol (LiCOR). Densitometry was performed using Empiria Studio 1.2 (LiCOR).

PTPN11 phosphatase activity was blocked using the allosteric inhibitor RMC-4550 (ProbeChem, Shanghai, China). For this purpose, 1 x 10⁶ cells per sample were seeded onto 12 well plates and incubated with 1.5 µM RMC-4450 for 3 h in a cell culture incubator. Subsequently, cells were collected, and immunoblotting was performed as described above.

The coding regions of PTPN11 (NM_002834.5) and NRAS (NM_002524.5) were amplified using cDNA from highIM-R2 and highN-R2 cells. PDGFRB coding plasmid was obtained from Sino Biological (NM_002609.3, HG10514-G, Eschborn, Germany). The amplicons were cloned into the pSelect-puromycin-mcs vector (Sigma-Aldrich) using the CloneAmp HiFi PCR premix (Takara) with gene-specific primers and primer-specific annealing temperatures ( Supplementary Table S1 ) including the restriction enzymes BamHI and NcoI/NheI (NEB), cloning enhancer and the In-Fusion HD Kit (Takara). PDGFRB p.(Glu578Gln) was inserted using Q5 site-directed mutagenesis kit (NEB) using the primers PDGFRB_Glu578Gln_F and PDGFRB_Glu578Gln_R at 60°C annealing temperature and 3 min elongation time according to the manufacturer’s protocol ( Supplementary Table S1 ). Sequence identity was confirmed using Sanger sequencing.

Transient transfection was performed using nucleofection and the nucleofector 2 b device (Lonza, Cologne, Germany). 2 x 10⁶ cells were transfected with 5 or 10 µg of the respective plasmid or empty vector control for plasmid transfection or 100 nM Ambion Silencer Select s11524 or negative control #1 for siRNA-mediated knockdown of PTPN11. 24 h after transfection, cells were seeded onto respective cell culture plates to analyze cellular fitness followed by 24-48 h exposure to 2 µM imatinib or 100 nM nilotinib or used for expression analyses as described elsewhere. After incubation time, cells were subducted for subsequent cellular fitness assays as described below. Stably transfected cells were generated by selecting puromycin-resistant cells after 4 weeks of exposure to 1 µg/ml puromycin (Invivogen, Toulouse, France).

Cellular fitness was analyzed as previously described (16, 18, 19). Briefly, cell numbers were obtained by trypan blue staining, WST-1 (Sigma-Aldrich), Caspase Glo 9 Assay (Promega), Bromodeoxyuridine proliferation assay (Merck, Darmstadt, Germany), and MKI ELISA Kit (MyBioSource, San Diego, CA, USA) according to the manufacturers’ recommendations. Data was analyzed by normalizing TKI-treated to non-treated samples, followed by statistical analyses as described below. For analyses of total cell number, proliferation, and apoptosis during the development of imatinib resistance, 0.5 x 10⁶ cells/ml were seeded into cell culture flasks and exposed to 0.1 µM imatinib for 21 days. Cells were counted and cultivated dependent on the cell density. After 21 days, Ki-67 expression and caspase 9 activity were measured as described above. The analyses of 0.2 and 0.3 µM imatinib were performed accordingly.

Unless not stated otherwise, statistical analysis was performed using one-way ANOVA, Dunnett’s test and/or student’s t-test and the GraphPad prism software (Version 8.0 for Windows, San Diego California, USA).

To analyze clonal evolution in TKI resistance, imatinib and nilotinib-resistant sublines derived from TKI-sensitive K-562 cells were established by step-wise exposure to increasing TKI concentrations ( Figure 1A ). Cell lines developing resistance against 0.5 µM imatinib (lowIM) or 2 µM imatinib (highIM), as well as 0.05 µM nilotinib (lowN) or 0.1 µM nilotinib (highN) were obtained generating four biological replicate cell lines of imatinib and two of nilotinib resistance ( Figure 1B ). Subsequently, genetic variants in these twelve TKI-resistant sublines were analyzed by exome sequencing and compared to TKI-sensitive K-562 cells.

First, non-intronic single nucleotide variants (SNVs) exclusively present in TKI-resistant cells were identified by excluding SNVs present in TKI-sensitive K-562 (VAF < 0.05) and applying a ΔVAF > 15% in the TKI-resistant sublines compared to TKI-sensitive cells. The number of variants differed between 103 and 195 in the TKI-resistant sublines ( Figure 1C , Supplementary Table S2 ). For IM-R1 and IM-R4, the majority of SNVs, 128 and 60, respectively, were newly acquired in highIM, whereas for IM-R2 and IM-R3, as well as in N-R1 and N-R2, the majority of SNVs were already present in the respective lowIM or lowN sublines (IM-R2: 61, IM-R3: 87, N-R1: 76, N-R2: 51, Figure 1C ). The total number of SNVs differed between the biological replicates of TKI resistance but increased compared to TKI-sensitive cells in all TKI-resistant cell lines ( Figure 1D ). However, for N-R1, a strong increase in the total SNV number was detected in highN compared to lowN, while in highIM-R2, as well as in highN-R2, the total number of SNVs was lower compared to lowIM-R2 or lowN-R2 cells, respectively ( Figure 1D ). To generate insight into the mutational processes, we determined the mutational signatures (COSMIC, https://doi.org/10.11093/nar/gky1015) of the variants that were acquired in the TKI-resistant sublines (VAF < 5% in TKI-sensitive K-562, △VAF > 15% between TKI-sensitive and -resistant K-562 cells). In all sublines, the signatures of unknown etiology, SBS40, showed the strongest signal ( Supplementary Figure S1 ).

As proteins interact in protein-protein-interaction (PPI) networks, this can be analyzed using network-based approaches, such as network propagation. Following this idea, mutations in single genes (protein) can be viewed as ‘heat sources’ in a PPI network. This heat can diffuse through the rest of the network using an iterative process until a steady state is reached. Proteins close to the mutated protein get higher propagation scores than distant proteins following the biological assumption that proteins underlying similar phenotypes tend to interact with one another (21, 22). Accordingly, the protein-protein interaction network of acquired variants was determined

Three clusters were revealed for the resistant cell lines and 14 clusters for gene sets with highIM-R1, -R3 and -R4 being a distinct cluster separate from the other tested resistant sublines ( Figure 2A ). To compare the network propagation with gene expression data, genome-wide expression analyses of the TKI-resistant cell lines and gene set variation analyses were performed [(16), Figure 2B ]. The resulting pattern of enriched pathways was highly similar to one of the protein-protein interaction network derived from the mutational pattern ( Figure 2A ].

To identify potential driver mutations in the TKI-resistant sublines, acquired SNVs (with the respective AF ≤ 5% in sensitive K-562 cells) were compared to a list of 568 mutational cancer driver genes previously published by Martínez-Jiménez et al. (23) ( Figure 3A , Table 1 ). Between two and five of the detected mutations in each TKI-resistant cell line were mapped to genes from the mutational cancer driver gene list. Among the acquired variants were the well-known pathogenic RAS-family mutation KRAS (KRAS proto-oncogene, GTPase) p.(Ala59Thr) (ClinVar ID: 12581; lowIM-R3: 8.7%, highIM-R3: 66.6%) in IM-R3, KRAS p.(Gly12Asp) in IM-R4 (ClinVar ID: 12582; lowIM-R4: absent; highIM-R4: 29%, Figures 3A, B ), as well as NRAS (NRAS proto-oncogene, GTPase) p.(Gln61Lys) (ClinVar ID: 73058; lowN: 29.2%, highN: 33.3%, Figures 3A, C ) in N-R2. Further, two pathogenic KMT2D (lysine methyltransferase 2D) variants p.(Leu3266Val) and p.(Arg191Trp) (ClinVar ID: 449928) were acquired in IM-R3 (lowIM-R3: 9%, highIM-R3: 37%, Figures 3A, B ). Moreover, PTPN11 (protein tyrosine phosphatase non-receptor 11) p.(Tyr279Cys) was detected in IM-R2 (ClinVar ID: 13328; lowIM-R2: absent; highIM-R2: 69%, Figures 3A, B ). In this cell line, the previously unknown PDGFRB (platelet-derived growth factor receptor beta) variant p.(Glu578Gln) was also detected (lowIM-R2: absent, highIM-R2: 28%, Figures 3A, B , Table 1 ). The gain of these SNVs likely explains the development of TKI resistance in the respective cell lines.

As ABL mutations are frequently the reason for TKI failure, mutations in this gene were also taken into focus showing two variants of unknown significance p.(Leu373Met) in lowN-R1 and p.(Glu208Asp) (VAF: 7%) in highIM-R4, as well as the known pathogenic kinase-domain mutation p.(Glu274Lys), with the latter likely associated with the TKI resistance (VAF: 10%, Figure 3A ).

Presence of variants in NRAS, KRAS as well as PTPN11, PDGFRB, RELA, and KMT2D in the TKI-resistant sublines pointed to recurrent pathway changes, especially in Ras-MAP-kinase signaling ( Figure 2 ; Figure 4A ). As NRAS p.(Gln61Lys) is a well-known driver mutation, described in various cancer types and associated with malignancy and tumor progression, the effect of this mutation in our in vitro-model was analyzed to investigate whether it is solely sufficient for the development of TKI resistance and if this effect is detectable with our in vitro-model ( Supplementary Table S3 ). To address this, TKI sensitive K-562 cells were transfected with either NRAS wild-type or the p.(Gln61Lys) variant. The response to nilotinib was analyzed measuring cell number, metabolic rate activity, apoptosis, and proliferation rates. Successful transfection of K-562 cells led to a 4.4-fold increase in cell number after NRAS WT (p < 0.001) and 6.2-fold after p.(Gln61Lys) transfection compared to the negative control (p < 0.001, Figure 4B ). In addition, metabolic activity was increased in NRAS WT (1.2-fold, p = 0.002) and p.(Gln61Lys)-transfected cells (5.2-fold, p < 0.001; Figure 4B ), while apoptosis, visible on the level of caspase 9 activation, was decreased (WT: 25%; p < 0.001; p.(Gln61Lys): 59%, p< 0.001; Figure 4A ). However, proliferation measured by Ki-67 expression was not significantly altered between the cell lines ( Figure 4B ). A similar effect was also observed under imatinib exposure, as cell number (62%, p = 0.008) and metabolic activity (3.7-fold, p < 0.001) were increased and apoptosis was reduced after NRAS p.(Gln61Lys) transfection (-45%, p = 0.003), while proliferation did not significantly change ( Supplementary Figure S2 ). Overall, our data demonstrate that the presence of NRAS p.(Gln61Lys) is solely sufficient to promote TKI resistance.

Using our established in vitro-analysis pipeline, we focused on the PTPN11 p.(Tyr279Cys) variant (IM-R2: VAF: 48%, Supplementary Table S3 ). Successful transfection of PTPN11 p.(Tyr279Cys) into sensitive K-562 cells led to an increase in cell number (1.7-fold, p = 0.03), accompanied by an increase in metabolic activity (1.7-fold, p = 0.005) and proliferation (2.0-fold, p < 0.001). Nevertheless, a change in apoptosis was not observed ( Figure 4C ).

PTPN11 p.(Tyr279Cys) is a well-known pathogenic germline variant associated with Leopard- and Noonan-syndrome (24, 25). However, this particular variant’s role in cancer and CML is widely unknown. To investigate the effect of PTPN11 p.(Tyr279Cys) in imatinib resistance and on the context-dependent protein function, first, PTPN11 expression was analyzed in the imatinib-resistant sublines showing no expression differences compared to TKI-sensitive K-562 cells ( Supplementary Figure S3 ). As several studies indicated a loss of catalytic function for PTPN11 p.(Tyr279Cys) (26), we hypothesized that the observed effect could be due to altered phosphatase activity. To address this, TKI-sensitive K-562 cells were exposed to the PTPN11 inhibitor RMC-4450, and the effects on Ras-MAP-kinase signaling were analyzed on the level of ERK activation. As expected, PTPN11 inhibition reduced the phosphorylation of ERK ( Supplementary Figure S4A ). To investigate if PTPN11 blockade alters the response to imatinib, an siRNA-mediated knockdown of PTPN11 was performed to mimic reduced protein levels ( Supplementary Figure S4B ). The knockdown cells were subsequently exposed to imatinib resulting in decreased metabolic activity (-32%, p < 0.001) and BrdU incorporation (-33%, p = 0.02), while apoptosis was not altered ( Supplementary Figure S4C ). Moreover, PTPN11 inhibition was performed in the imatinib-resistant cell lines to investigate the extent of pathway addiction in these cells. Interestingly, PTPN11 inhibition only resulted in reduced ERK-phosphorylation in highIM-R2 (-1.3-fold, p = 0.04), but not in the other resistant sublines ( Supplementary Figure S4D ). The siRNA-mediated knockdown of PTPN11 in this cell line showed a slight decrease in imatinib susceptibility, as BrdU incorporation was 1.3-fold increased (p = 0.02), while metabolic activity and apoptosis were not altered ( Supplementary Figures S4E, F ).

As a further candidate variant, we analyzed PDGFRB p.(Glu578Gln), as PDGFRB is a well-known target of imatinib (IM-R2: VAF: 34%, Supplementary Table S3 ). Although overexpression of PDGFRB p.(Glu578Gln) did not lead to a significant increase in cell number, metabolic activity was increased (WT: 1.3-fold, p = 0.01; p.(Glu578Gln): 1.4-fold, p = 0.002) and caspase 9 activity reduced after WT and p.(Glu578Gln) transfection (WT: 25%, p = 0.007; p.(Glu578Gln): 74%, p < 0.001). Analyses of proliferation measured by Ki-67 expression did not reveal significant differences ( Figure 4D ).

Next, stably transfected cell lines expressing either PTPN11 WT or p.(Tyr279Cys), as well as PDGFRB WT or p.(Glu578Gln) were generated. These cell lines were exposed to low dose imatinib (0.1 to 0.3 µM) and the total cell number was analyzed during the development of imatinib resistance in a time-frame of 21 days ( Figure 5 ). An increase in the cell number of PTPN11 p.(Tyr279Cys)-expressing cells were detected in all tested imatinib concentrations, while the other cell lines showed no differences compared to the negative control-transfected cells ( Figure 5A ). As PTPN11 p.(Tyr279Cys) seemed to promote an advantage for the cells during the development of imatinib resistance, proliferation and apoptosis in these cell lines after two weeks of exposure to the respective imatinib concentration was analyzed. Compared to WT and negative control-transfected cells, no significant increase in the proliferation of p.(Y279)-transfected cells was detected ( Figure 5B ). An imatinib dose-dependent effect was observed in WT-expressing cells on apoptosis (0.1 µM: -53%, p < 0.001; 0.2 µM: +40%, p < 0.001; 0.3 µM: -19%, p < 0.001), while cells harboring p.(Tyr279Cys) showed reduced apoptosis in 0.1 µM (74%, p < 0.001) and 0.2 µM imatinib-resistant cells (53%, p < 0.001), but not in resistance to 0.3 µM imatinib ( Figure 5C ).