Systematic elucidation and pharmacologic targeting on non-oncogene dependencies in imatinib-resistant gastrointestinal stromal tumor

Systematic elucidation and pharmacologic targeting on non-oncogene dependencies in imatinib-resistant gastrointestinal stromal tumor | bioRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-M677548'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search New Results Systematic elucidation and pharmacologic targeting on non-oncogene dependencies in imatinib-resistant gastrointestinal stromal tumor View ORCID Profile Prabhjot S. Mundi , Adina Grunn , View ORCID Profile Arsenije Kojadinovic , View ORCID Profile Charles Karan , View ORCID Profile Ronald Realubit , Cristina I. Caescu , View ORCID Profile Hanina Hibshoosh , View ORCID Profile Mahalaxmi Aburi , View ORCID Profile Mariano J. Alvarez , View ORCID Profile Matthew Ingham , View ORCID Profile Denisse Evans , View ORCID Profile Sara Rothschild , View ORCID Profile Gary K. Schwartz , View ORCID Profile Andrea Califano doi: https://doi.org/10.1101/2025.10.12.681609 Prabhjot S. Mundi 1 Department of Systems Biology, Columbia University Irving Medical Center , 1130 Saint Nicholas Ave, New York, NY USA 10032 2 Herbert Irving Comprehensive Cancer Center, Columbia University Irving Medical Center , 1130 Saint Nicholas Ave, New York, NY USA 10032 3 Department of Medicine, Columbia University Irving Medical Center , 630 W 168th Street, New York, NY USA 10032 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Prabhjot S. Mundi For correspondence: psm2134{at}cumc.columbia.edu ac2248{at}columbia.edu Adina Grunn 1 Department of Systems Biology, Columbia University Irving Medical Center , 1130 Saint Nicholas Ave, New York, NY USA 10032 Find this author on Google Scholar Find this author on PubMed Search for this author on this site Arsenije Kojadinovic 4 Department of Internal Medicine, University of Iowa Carver College of Medicine , 375 Newton Rd, Iowa City, IA 52242 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Arsenije Kojadinovic Charles Karan 1 Department of Systems Biology, Columbia University Irving Medical Center , 1130 Saint Nicholas Ave, New York, NY USA 10032 2 Herbert Irving Comprehensive Cancer Center, Columbia University Irving Medical Center , 1130 Saint Nicholas Ave, New York, NY USA 10032 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Charles Karan Ronald Realubit 1 Department of Systems Biology, Columbia University Irving Medical Center , 1130 Saint Nicholas Ave, New York, NY USA 10032 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Ronald Realubit Cristina I. Caescu 1 Department of Systems Biology, Columbia University Irving Medical Center , 1130 Saint Nicholas Ave, New York, NY USA 10032 Find this author on Google Scholar Find this author on PubMed Search for this author on this site Hanina Hibshoosh 2 Herbert Irving Comprehensive Cancer Center, Columbia University Irving Medical Center , 1130 Saint Nicholas Ave, New York, NY USA 10032 5 Department of Pathology and Cell Biology, Columbia University Irving Medical Center , 630 W 168th Street, New York, NY USA 10032 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Hanina Hibshoosh Mahalaxmi Aburi 1 Department of Systems Biology, Columbia University Irving Medical Center , 1130 Saint Nicholas Ave, New York, NY USA 10032 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Mahalaxmi Aburi Mariano J. Alvarez 1 Department of Systems Biology, Columbia University Irving Medical Center , 1130 Saint Nicholas Ave, New York, NY USA 10032 6 DarwinHealth Inc . New York Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Mariano J. Alvarez Matthew Ingham 2 Herbert Irving Comprehensive Cancer Center, Columbia University Irving Medical Center , 1130 Saint Nicholas Ave, New York, NY USA 10032 3 Department of Medicine, Columbia University Irving Medical Center , 630 W 168th Street, New York, NY USA 10032 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Matthew Ingham Denisse Evans 7 The Life Raft Group , 155 US Highway 46, Suite 202, Wayne, NJ USA 07470 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Denisse Evans Sara Rothschild 7 The Life Raft Group , 155 US Highway 46, Suite 202, Wayne, NJ USA 07470 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Sara Rothschild Gary K. Schwartz 2 Herbert Irving Comprehensive Cancer Center, Columbia University Irving Medical Center , 1130 Saint Nicholas Ave, New York, NY USA 10032 3 Department of Medicine, Columbia University Irving Medical Center , 630 W 168th Street, New York, NY USA 10032 8 Case Comprehensive Cancer Center, Case Western Reserve University , 10900 Euclid Ave. Cleveland, OH 44106 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Gary K. Schwartz Andrea Califano 1 Department of Systems Biology, Columbia University Irving Medical Center , 1130 Saint Nicholas Ave, New York, NY USA 10032 2 Herbert Irving Comprehensive Cancer Center, Columbia University Irving Medical Center , 1130 Saint Nicholas Ave, New York, NY USA 10032 3 Department of Medicine, Columbia University Irving Medical Center , 630 W 168th Street, New York, NY USA 10032 9 Department of Biochemistry & Molecular Biophysics, Columbia University Irving Medical Center , 701 W 168th Street, New York, NY USA 10032 10 Department of Biomedical Informatics, Columbia University Irving Medical Center , 622 W 168th Street, New York, NY USA 10032 11 Chan Zuckerberg Biohub New York, New York, NY, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Andrea Califano For correspondence: psm2134{at}cumc.columbia.edu ac2248{at}columbia.edu Abstract Full Text Info/History Metrics Supplementary material Preview PDF Abstract Treatment of gastrointestinal stromal tumor (GIST) with imatinib and other KIT-targeting drugs, has been effective. However, most patients with advanced GIST eventually develop imatinib-resistance and succumb to disease. We have developed mutation-agnostic, network-based methodologies to systematically elucidate and pharmacologically target Master Regulator (MR) proteins representing critical non-oncogene dependencies of cancer cells. Unsupervised, MR-based clustering of 34 GIST patient tumor samples produced two clusters clearly separating imatinib-resistant vs. sensitive tumors. High-throughput profiling of transcriptional responses by two GIST cell lines to FDA approved and late-stage experimental drugs identified six candidate drugs that reversed the MR activity of imatinib-resistant GIST. Predictions were validated in two imatinib-resistant, patient-derived xenograft (PDX) models. The top prediction, linifanib, induced marked tumor growth inhibition in both PDXs across a wide dose range, while selinexor was also effective compared to imatinib. We confirmed in vivo MR activity reversal by these drugs, but not by ineffective drugs. Statement of Significance We leveraged our network-based platforms, OncoTreat and OncoTarget , to characterize Master Regulators of imatinib-resistance in GIST and identify candidate MR-targeting drugs, an unmet clinical need. Top predicted drugs were successfully validated in cognate PDXs, thus providing a path for translation. Introduction The oncogene addiction paradigm has been at the forefront of personalized oncology efforts over the past three decades 1 , with gastrointestinal stromal tumor (GIST) representing one of its earliest successful clinical applications. In this archetype, tumors are highly dependent on the aberrant activity of a singular protein encoded by a recurrently mutated oncogene. GIST is a malignancy of mesenchymal origin arising within the gastrointestinal tract, most frequently the stomach or small intestine, and is one of the most common subtypes of visceral sarcoma 2 . The putative lineage of origin of GIST is the interstitial cells of Cajal, pacemaker cells within the inner circular layer of the muscularis mucosa that propagate intestinal motility, based on the observation of near universal membranous expression of DOG-1 and c-KIT in these tumors 3 – 6 . While some localized GISTs can be cured with surgery alone, higher risk GISTs, characterized by larger tumor size, high mitotic rate, non-stomach location, and intraperitoneal tumor rupture, frequently recur and metastasize, becoming incurable. In the metastatic setting, GISTs are resistant to cytotoxic chemotherapy and radiation 7 . Approximately 70-80% of GISTs harbor activating mutations in the receptor tyrosine kinase (RTK) KIT , most commonly deletions in exon 11 encoding the juxtamembrane domain, with mutually exclusive mutations in the related RTK PDGFRA occurring in an additional 5-10% 3 , 8 . Somatic or germline mutations in BRAF , NF1 , or succinyl dehydrogenase complex subunit genes occur in a sizeable proportion of the remaining KIT/PDGFRA -wildtype GISTs 9 – 11 . Imatinib, a small molecule tyrosine kinase inhibitor (TKI) with selectivity for c-KIT and PDGFR-alpha, significantly reduces or at least delays relapse in the adjuvant setting and is first line therapy for patients with metastatic GIST. Imatinib has demonstrated objective response rates around 54% by RECIST criteria 12 and extended median overall survival for metastatic disease from 18 to 56 months 13 , 14 . The response to imatinib, however, is highly heterogenous; 15% of GISTs present with primary resistance, mainly in KIT -wildtype or KIT exon 9 mutated tumors, while a similar percentage of patients have durable disease control for over five years on imatinib 15 . Most patients fall somewhere in between, with median progression free survival (mPFS) of 18 to 24 months. Unfortunately, acquired resistance to imatinib eventually occurs in the majority of patients with advanced GIST, and is in part attributed to secondary mutations in KIT , including amplification, loss of heterozygosity, and point mutations in the activation loop (exons 17/18) or the ATP-binding pocket (exons 13/14) that reduce imatinib binding 16 , 17 . Other potential mechanisms of resistance include activation of parallel or downstream kinase pathway 18 – 20 , pharmacokinetic changes resulting in decreased drug levels over time mediated by upregulation of ABC transporters in the intestine and increased erythrocyte uptake 21 – 23 , as well as potential epigenetic cell adaptation. Two multi-kinase inhibitors, sunitinib and regorafenib, are approved in the second- and third-line setting, although the duration of clinical benefit from these agents is progressively shorter, with mPFS around 6.3 and 4.8 months, respectively 24 , 25 . More recently, the next generation highly selective c-KIT and PDGFR inhibitors ripretinib and avapritinib were also approved based on their activity against mutations that confer resistance to imatinib including KIT exon 17 and PDGFRA D842V, although objective response rates are low 26 , 27 . We have recently introduced two transcriptome-based, mutation-agnostic methodologies— OncoTarget and OncoTreat —aimed at predicting drugs targeting Master Regulator proteins (hereafter, MRs) representing non-oncogene dependencies and mechanistic determinants of tumor cell state. These have been validated both preclinically 28 , 29 and clinically 30 , 31 . Both methods leverage the VIPER algorithm 32 , which effectively identifies aberrantly activated and inactivated proteins, based on the differential expression of their transcriptional targets. We have shown that the most aberrantly activated and inactivated proteins represent candidate MRs capable of eliciting tumor-specific essentiality 33 , 34 or synthetic lethality 35 , 36 . Furthermore, candidate MRs form tightly autoregulated tumor checkpoints modules (TCMs) that implement highly specific on/off switches controlling cancer cell transcriptional state 37 . VIPER has been extensively tested and has helped elucidate novel, experimentally validated mechanisms of tumorigenesis, progression, and drug resistance in glioma 38 , leukemia 39 , lymphoma 40 , prostate 35 and breast cancer 41 , among several others. Once candidate MR dependencies are identified, OncoTarget prioritizes clinically relevant drugs representing high-affinity MR inhibitors, while OncoTreat predicts drugs capable of inverting the activity of the entire repertoire of TCM MRs 28 , 33 . The latter is accomplished by analyzing high-throughput drug perturbation profiles— representing the transcriptional response of high-fidelity cell lines—to assess differential MR protein activity in drug vs. vehicle control-treated cells 42 , 43 . Herein, we describe the application of these methodologies to identify novel MRs and MR-targeting drugs for the treatment of imatinib-resistant GIST. For this purpose, we collaborated with the Life Raft Group, a patient advocacy organization that has compiled a large GIST tumor bank, with extensive annotation of clinicopathological features and clinical outcomes. We performed transcriptomic profiling of GIST tumors at various stages of disease trajectory, including: initial diagnosis, on-treatment with imatinib in the adjuvant or metastatic setting, and after the development of clinical progression on imatinib and in some cases after resistance to multiple subsequent treatments such as sunitinib and regorafenib. Tumor-specific MRs—identified by unbiased VIPER inference—emerged as highly conserved across imatinib-resistant tumors, representing putative mechanistic determinants of imatinib-resistant vs. sensitive state. We identified six candidate drugs for follow up in vivo validation, based on their highly significant predicted MR targeting of the imatinib-resistant state, two by OncoTarget and four by OncoTreat analysis. The latter were based on drug perturbation profiles generated in two established GIST cell lines presenting differential imatinib sensitivity. We then collaborated with Crown Bioscience, to identify—among 18 available GIST PDX models—the two that most significantly recapitulated the MR activity signature of imatinib-resistant patient tumors, as optimal models for drug efficacy evaluation. The top predicted drug, linifanib, induced significant tumor growth inhibition in both PDXs, across a wide dose range, including tumor regression in both models at the highest dose level. Other predicted drugs also compared favorably to imatinib. We further demonstrate proof-of-concept, by profiling on-treatment biopsy samples, and demonstrate in vivo reversal of MR activity with effective drugs, but not with ineffective drugs. We discuss the implications of our findings to imatinib-resistant GIST, including strategies for further bi-directional translational development—through validation in clinical trials and concurrent single cell RNASeq profiling to characterize intra-tumoral heterogeneity and resistance at the subpopulation level. More broadly, we discuss the potential use of an OncoTreat- based framework to dynamically target evolving drug resistance, applicable across cancer types. Results Characterizing Master Regulators of Imatinib-resistance We collaborated with the Life Raft Group (LRG) patient advocacy organization, the largest private sponsor of GIST research, to profile patient tumor samples obtained at various stages of disease, including at diagnosis, while on treatment with imatinib, in the imatinib-resistant setting, and after progression on multiple drugs. The LRG maintains a GIST patient registry and a companion tissue bank, which links clinical information in the registry to biological samples. RNASeq profiles were generated from 34 GIST samples from the registry that met quality control standards, including 10 samples obtained after progression to imatinib-resistant disease. Differential protein activity was assessed by VIPER, using the centroid of the dataset as reference for differential expression analysis. As discussed, VIPER can assess the differential activity of transcription factors (TFs), co-factors (co-TFs), and signaling proteins (SPs) based on the differential expression of their downstream transcriptional targets, as previously described and extensively validated 32 , 44 . ARACNe, the algorithm that is typically used to generate regulatory networks, requires ≥ 100 samples to produce accurate predictions. Since GIST repositories are much smaller, we leveraged metaVIPER, an extension of VIPER that effectively infers a protein’s activity by integrating information on its transcriptional targets from multiple regulatory networks. In this case, we used metaVIPER to integrate regulatory networks representing the cellular context of 43 distinct cancer types or subtypes 44 , 45 , see Methods . The 34 samples were then stratified by unsupervised, protein activity-based consensus clustering—using the partition around medoids (PAM) approach and 10,000 iterations 46 . The solution with k=2 clusters ( Figure 1A ) was identified as optimal 47 , producing the lowest connectivity and highest silhouette index ( Figure 1B-C ), see Methods . Of the 16 samples in the first cluster ( C 1 ), 10 represented imatinib-resistant cases while all samples in the second cluster ( C 2 ) were imatinib sensitive or naïve to treatment (Fisher’s exact two-tailed p = 0.0001) ( Figure 1A ). While protein activity and cluster analyses were performed agnostic to genotype and clinicopathological features, C 1 also contained all tumors harboring PDGFRA exon 18 (D842V) mutations and all KIT/PDGFRA wildtype tumors (n = 3 and n= 3, respectively), which are known to confer primary resistance to imatinib 48 . Based on this cluster solution ( Figure 1A ), only four of the 34 samples presented with ambiguous cluster placement [i.e., not placed in the same cluster grouping for all 10,000 iterations], with three of the four harboring KIT exon 9 alterations, which are associated with decreased efficacy of imatinib. Download figure Open in new tab Figure 1. Protein activity profiling of Life Raft GIST registry patient tumors. Bulk RNASeq profiling of 34 unique patient samples from the Life Raft registry was completed, including 10 from patients whose cancer had relapsed or progressed while receiving imatinib therapy. MetaVIPER analysis is used to infer differential protein activity in each tumor. (A) Unsupervised consensus clustering of tumors was performed on the protein activity signatures, using a partition around medoids (PAM) approach, with n=10,000 repetitions. The optimal solution to consensus clustering occurred with k=2 clusters. The symmetric consensus matrix heatmap for k=2 demonstrates that most sample pairings either always cluster together (consensus index ∼1, dark brown) or never cluster together (consensus index ∼0, white), with four samples (first column, and three columns in the middle) demonstrating more ambiguous clustering across iterations. Cluster assignment and clinical annotation are provided at the top of the heatmap. All 10 known imatinib-resistant tumors are in cluster 1 [ C 1 ] (Fisher’s exact two-tailed p = 0.0001). (B) Consensus index distribution for different partition numbers (k=2 to 8). The cumulative distribution for k=2 and k=3 contains the fewest ambiguous sample pairings whose consensus index is between 0.2 and 0.8. (C) Additional metrics for optimal partitioning using a PAM approach. The connectivity index (the inverse of the average number of in-cluster samples that are a shorter distance than the closest out-of-cluster sample) and silhouette index (the relative difference in average distance between in-cluster samples and the nearest out-of-cluster sample) are optimized (lowest connectivity and highest silhouette) for k=2 clusters. (D) Heatmap of proteins demonstrating the highest variance in activity between clusters. Analysis considers all proteins (n=7,070) whose activity is inferred by metaVIPER. Columns are organized in the same order as ( A ) . (E) Heatmap of OncoTarget analysis, representing the most differentially activated of a panel of 180 directly druggable proteins. Columns are organized in the same order as ( A ) and color intensity corresponds to the -log10 (FDR p -value) for protein activation. In C 1 , containing all imatinib-resistant samples, XPO1, HDAC10/7, MEK2 ( MAP2K2 ), CDK9, and DOT1L are recurrently activated druggable proteins. Based on these analyses, we proceeded to rank all differentially active proteins by their integrated p -value across all C 1 and C 2 samples, respectively, using Stouffer’s method. The 25 most activated and 25 most inactivated proteins (TF, co-TF, or SP) in C 1 vs. C 2 samples (and vice-versa) are shown in Figure 1D . Consistent with previous analyses, the most activated and inactivated proteins were highly conserved across C 1 and C 2 samples—including MAP2K2 (MEK2), TBL3, ZNF668, and E4F1 (a key post-translational modulator of p53 activity 49 ) as aberrantly activated MRs, and C5orf41 (CREBRF) as an inactivated MR in the majority of imatinib-resistant tumors. This suggests significant conservation of the mechanisms that preside over imatinib sensitivity and resistance across tumors, with imatinib resistance emerging as a transcriptionally distinct cell state, potentially due to additional mutations, cell adaptation, or clonal selection under therapeutic pressure. Pathway enrichment analysis on the integrated protein activity signature of the C 1 cluster, using the MSigDB 50 , 51 gene ontology, cancer hallmarks 52 , immunological signatures, and oncogenic signatures is presented in Figure S1 . There is downregulation in MYC-proliferative signaling, metabolic alterations (e.g., downregulation of oxidative phosphorylation pathway), and complex changes in DNA repair, transcriptional machinery, and immunological signalling 53 in imatinib-resistant tumors, overall suggestive of a complex reprogramming. OncoTarget Drug Prediction for Imatinib-resistant Tumors The OncoTarget algorithm has been developed as an effective approach to identify those MR proteins, representing non-oncogene dependencies, that are also high-affinity targets of clinically relevant drugs, and has been validated both pre-clinically and clinically 28 , 44 . OncoTarget matches the most differentially activated MRs ( p < 10 -5 , FDR-corrected) to a curated target-to-drug list, consisting of 180 high-affinity targets of FDA approved and late-stage experimental (phase 2 and 3) antineoplastic drugs ( Table S1 ). While such MRs may include mutated oncoproteins, they more frequently comprise proteins that are rarely if ever mutated in cancer—e.g., topoisomerases or chromatin remodeling enzymes—thus representing effective non-oncogene dependencies. OncoTarget is CLIA compliant and approved by the NY and CA Dept. of Health. As such, it is readily available as a clinical test to predict drug sensitivity. Figure 1E summarizes the VIPER-assessed activity of the most statistically significant druggable MRs across the 34 GIST samples, ranked by their integrated p -value across each cluster, as well as the activity of c-KIT. The analysis identified MAP2K2 (MEK2), HDAC10, exportin-1 (XPO1), CDK9, and DOT1L as recurrently activated targets (FDR p ≤ 10 -5 ) across a majority of the 10 imatinib-resistant tumors. Interestingly, c-KIT activity was modestly, yet statistically significantly higher in C 2 (imatinib-sensitive) vs. C 1 samples (two sample t-test of VIPER-inferred protein activity scores, p = 2.2 x 10 - 16 ). Only a few recurrently activated, druggable targets were identified by OncoTarget , some of which match to drugs that are difficult to attain for therapeutic study (e.g., pinometostat for DOT1L) or that are considered to be significantly toxic for clinical translation (e.g., the pan-CDK inhibitor dinaciclib for CDK5/9). As such, we leveraged the OncoTreat algorithm to identify additional candidate drugs for imatinib-resistant GIST. OncoTreat Drug Predictions and Generation of Drug Perturbation Profiles OncoTreat is an extensively validated, RNA-based assay to predict tumor sensitivity based on a drug’s ability to invert the activity of MR proteins, see 28 , 33 . Similar to OncoTarget , it is CLIA compliant and approved by the NY and CA Dept. of Health. Distinct from OncoTarget , this algorithm relies on systematic drug perturbation profiles from carefully selected cognate in vitro models. These include cell lines, organoids, or even primary cells that effectively recapitulate the patient-specific MR signature(s) of interest, based on the OncoMatch algorithm 28 , 29 . These assays have been shown to be optimally predictive of in vivo drug sensitivity when performed at 24h using the 48h EC20 of each drug—as assessed from 10-point dose response curves—thus reducing confounding effects arising from engagement of cell death and stress pathways 28 . VIPER analysis of drug vs. vehicle control-treated cells can then be used to rank drugs based on their ability to invert the activity of the top MR proteins—including the 25 most differentially activated (25↑) and 25 most inactivated (25↓) proteins from each tumor sample. Indeed, we have shown that such a number of MRs—also used to define the tumor checkpoint module (TCM)—is sufficient to integrate the effects of >80% of functional mutations in tumor samples 28 , 54 . Similar to OncoTarget , OncoTreat uses a conservative statistical significance threshold ( p ≤ 10 -5 , FDR-corrected, by 1-tailed analytic-rank based enrichment analysis — aREA 32 ) to identify drugs eliciting TCM activity inversion, see Methods . Notably, OncoTreat completely ignores drug sensitivity in vitro (in fact, all drugs are screened at the same EC 20 ), and rather uses in vitro models only as a means for de novo assessment of context-specific drug mechanism of action (i.e., which proteins are significantly activated and inactivated by a drug). Among the limited number of available GIST cell line models, we selected GIST-T1 and GIST430 19 , 55 to generate perturbational profiles. These models have been reported previously as relatively imatinib-sensitive and imatinib-resistant, respectively. Both GIST-T1 and GIST430 were derived from patient tumors with canonical KIT exon 11 deletions—associated with initial imatinib sensitivity, but GIST430 has a secondary, cis-allelic, KIT exon 13 mutation 19 , 55 . We assessed MR-based fidelity of these models to the analyzed samples, based on the enrichment of patient-based TCM MRs ( i.e. , 25↑+25↓) in proteins differentially active in the cell lines (OncoMatch analysis 28 , 29 , see Methods ). Even though these two cell lines do not reflect the full genetic alteration repertoire of GIST, reassuringly and consistent with previous successful validation of OncoTreat predictions 28 , they were found to recapitulate the MR signature of C 1 and C 2 samples at high statistical significance ( p ≤ 10 -5 , FDR corrected, by OncoMatch analysis) ( Figure 2A ). GIST430 provides a modestly stronger match for C 1 (imatinib-resistant enriched) samples ( p = 4.81e−07) compared to GIST-T1 ( p = 3.14e−06). Download figure Open in new tab Figure 2. OncoMatch analysis of Life Raft patient tumor samples to available GIST cell line ( A ) and patient derived xenograft (PDX) models ( C, D ) and OncoTreat drug predictions for patient tumors ( B ) . (A) RNASeq profiling and protein activity inference using metaVIPER was performed on two available GIST cell lines, GIST-T1 and GIST430, with GIST430 previously being reported as less sensitive to imatinib. Integrated protein activity signatures of cluster C 1 and C 2 patient tumors were generated. In OncoMatch analysis, we assess the enrichment of the 25 most activated (red bars) and 25 most inactivated (blue bars) patient tumor master regulator (MR) proteins—i.e., the tumor checkpoint module (TCM) , in the signature of the cell line model. In the plots, proteins are sorted left to right from the most differentially inactivated to the most activated in the cell line model. In general, both cell lines matched somewhat more strongly to the C 2 tumors than C 1 tumors, but nonetheless, they represent a reasonable model match to C 1 tumors (FDR p < 10 -5 ), with GIST430 a slightly better match than GIST-T1 for the imatinib-resistant cluster. (B) OncoTreat drug sensitivity predictions for all GIST patient tumors profiled. High throughput perturbation screens with RNASeq profiling using the PLATESeq platform were performed in GIST430 and GIST-T1, with the analysis subsequently focused on GIST430 given its superior match to the imatinib-resistant cluster. OncoTreat uses de novo drug mechanism information from a cognate model (GIST430) to identify top TCM-inverter drugs. For each patient and drug, we compute the enrichment of TCM proteins in the drug signature, as measured by the aREA algorithm, with negative normalized enrichment scores (NES) indicating reversal, and the associated lower-tail p -value. Color intensity in the heatmap corresponds to the -log10 (FDR p -value), and the predictions are clustered by tumor ( columns— pharmacotype clusters) and drug ( rows ), with cluster reliability indices for cluster assignment shown as barplots on the top and righthand side of the heatmap. A few recurrent OncoTarget predictions are incorporated into the heatmap for simplicity and are denoted as [# target protein], with corresponding OncoTarget -log10 (FDR p -value). Clinical annotation is provided at the bottom of the heatmap. (C) RNASeq and protein activity profiling by metaVIPER of n=18 GIST PDX models developed by Crown Bioscience was performed. The enrichment plots for OncoMatch analysis of the two PDX models most strongly matching to C 1 ( GS5108 and GS5106 ) and C 2 ( GS11354 and GS11351 ) patient tumors is shown. GS5108 and GS5106 are superb matches (FDR p < 10 -80 and p < 10 -30 ) to C 1 tumors, conserving the MRs of the imatinib-resistant state. (D) Heatmap of OncoMatch analysis between individual patient tumors ( columns organized in same order as ( C ) ) and all 18 GIST PDX tumors profiled ( rows ), with color intensity corresponding to the -log10 (FDR p -value) for OncoMatch analysis. GS5108 and GS5106 represent superb matches to all imatinib-resistant tumors. We thus generated PLATE-seq-based drug perturbation profiles for both GIST-T1 and GIST430, using 46 drugs, including 30 FDA approved antineoplastics and 16 late-stage experimental drugs in phase II or III oncology clinical trials ( Table S2 ), see Methods . Cells were harvested at 24-hours following perturbation with each drug at two sublethal concentrations— the 48-hour EC 20 and one tenth of this concentration, as determined by 10-point dose response curves. The lower dose is introduced to assess the potential pharmacologic window of each drug. As discussed, sublethal drug concentrations effectively reveal drug mechanism of action rather than confounding effects associated with cell stress or death pathways 28 , 56 , 57 . We also capped concentrations at each drug’s C Max , defined as the peak serum concentration for the drug’s maximum tolerated dose (MTD), from published pharmacokinetic studies in humans, when available, thus optimizing the translational potential of predicted drugs. Multiplexed, low depth RNASeq profiles (about 2M reads each) were generated by high-throughput microfluidic automation, using the PLATE-Seq technology 58 , which supports either 96 or 384-well plates. VIPER analysis was used to generate a drug-mediated differential protein activity signature from gene expression profiles of drug vs. vehicle control (DMSO) treated cells, thus representing a de novo proteome-wide assessment of drug mechanism of action. We have subsequently completed an expanded screen in GIST430, using 333 drugs, including 122 FDA approved antineoplastics, 192 late-stage experimental drugs in phase II or III trials, and 19 compounds from diversity libraries presenting cell line-specific EC 50 ≤ 2 μM ( Table S3) . Data from the expanded screen were not yet available for the original OncoTreat predictions for in vivo validation, but is shared here, including an unsupervised clustering of drug mechanism based on induced differential protein activity in GIST430 ( Figure S2 ). OncoTreat Predictions Using data from the perturbational screen and the OncoTreat algorithm, we made tumor-specific drug predictions for the 34 LRG GIST samples. Samples optimally segregated into five main clusters, using PAM clustering, based on shared drug predictions ( Figure 2B ). We have previously coined the term pharmacotype to describe tumor clusters presenting shared drug sensitivity predictions. Such a pattern appears to be a recurring theme in most tumor cohorts analyzed to date 28 . Not surprisingly, given the dependency of OncoTreat drug predictions on tumor-specific TCM MRs, and the overlap of these MRs in imatinib-resistant tumors, all ten imatinib-resistant tumors aligned into either pharmacotype I or pharmacotype II ( Figure 2B ). Importantly, we identified five novel candidate drugs, all either FDA approved antineoplastic agents or in late-stage clinical trials, that are predicted to invert the activity of TCM MR proteins in subsets of imatinib-resistant GISTs. Specifically, the VEGFR2 and multi-kinase inhibitor linifanib was predicted as significant by OncoTreat for 8 of 10 imatinib-resistant samples (FDR p ≤ 10 -5 ) and represents the top ranked prediction for four of them. The MEK inhibitor selumetinib was predicted for 5 of 10 imatinib-resistant samples and is consistent with OncoTarget predictions based on aberrant MAP2K2 activity ( Figure 1D ). Finally, the selective estrogen receptor modulator tamoxifen, the WEE1 inhibitor AZD1775 (adavosertib), and the MET/ALK selective inhibitor crizotinib were all predicted for 3 of the 10 imatinib-resistant tumors (FDR p ≤ 10 -5 , by OncoTreat ). In addition to these five OncoTreat -predicted drugs, we also considered two OncoTarget -based predictions for in vivo validation: the selective XPO1 inhibitor selinexor, predicted for 7 of 10 imatinib-resistant tumors, and the pan-HDAC inhibitor panobinostat predicted based on HDAC10 and HDAC7 aberrant activity in 10 of 10 and 7 of 10 imatinib-resistant tumors, respectively (FDR p ≤ 10 -5 , by OncoTarget , Figure 1D ). Neither panobinostat or selinexor were evaluated in our initial GIST430 perturbational screen. The two OncoTarget predictions are incorporated into the Figure 2B heatmap, for simplicity, and are labeled as [# target protein]. However, the color intensity on the associated rows reports on the aberrant XPO1 and HDAC10 activity (-log10(FDR p -value)). Perhaps not surprisingly, several of the recurrently predicted drugs for cluster C 2 tumors (pharmacotypes III, IV, and V) are TKIs with partial selectivity for c-KIT, including masitinib, ponatinib, regorafenib, bosutinib and cabozantinib ( Figure 2B ). Intriguingly, imatinib itself is not predicted by OncoTreat for tumors in either cluster. It is possible that our analytical approach, which amplified the signal of MR proteins involved in implementing the imatinib-resistant state, cancels out the signal of MR programs that remain active in both clusters, and perhaps it is these programs that are most efficiently reversed by imatinib. In addition, our analysis is not effectively geared to identify drugs whose activity is predominantly mediated by targeting mutated proteins. Selection of GIST Patient-derived Xenograft Models To assess the efficacy of predicted drugs in vivo , we sought to identify GIST patient-derived xenograft (PDX) mouse models that significantly recapitulate the MR protein activity signature of imatinib-resistant patient tumors. We contracted with Crown Bioscience, Inc, a company that specializes in PDX drug testing, who provided us with RNASeq profiles of 18 existing GIST models in their tumor bank. VIPER and OncoMatch analysis ( see Methods ) identified only two PDX models, GS5106 and GS5108 , both representing very high-fidelity models, at the MR level, for all 10 imatinib-resistant patient tumors ( Figure 2C-D ), with each tumor being matched by at least one of the two models at an extremely statistically significant level ( p < 10 -30 ). Indeed, GS5106 and GS5108 provide highly significant matches to all 17 tumors comprising pharmacotypes I and II (columns representing tumors in Figures 2B and 2D are in the same ordering), which include all of the imatinib resistant tumors. Notably, while OncoMatch analysis was completely agnostic to any known clinical or genotypic PDX characteristics, pre-existing therapeutic data—made available by Crown Bioscience and later reproduced in our own studies—confirmed that both models were indeed imatinib resistant. GS5108 harbors a canonical KIT exon 11 mutation and a co-occurring (presumed secondary) KIT exon 17 mutation. GS5106 harbors a KIT exon 13 mutation and a co-occurring (presumed secondary) KIT exon 17 mutation. To ensure that the drugs predicted from patient profiles were also predicted for these PDX models, we repeated the OncoTarget and OncoTreat analyses based on their RNASeq profiles ( GS5106 and GS5108 ), as well as on the profiles of two additional PDX models ( GS11351 and GS11354 ) which were found to represent optimal matches for pharmacotypes III-V, which included no imatinib-resistant tumors. Of the seven candidate drugs predicted from human tumor profiles, linifanib and selumetinib (as well as aberrant MAP2K2 activity) were predicted as significant in both imatinib-resistant PDX models but not in models predicted as imatinib sensitive ( Figure 3A ). In contrast, tamoxifen and AZD1775 were only predicted as significant in the GS5106 PDX , selinexor only in the GS5108 (based on aberrant XPO1 activity), and crizotinib was not predicted for either model ( Figure 3A ). Download figure Open in new tab Figure 3. Selection of drug predictions that are conserved in the selected PDX models and assignment to the therapeutic study. (A) Heatmap displays the analogous OncoTreat predictions for four GIST PDX models, GS5108 and GS5106 that are strong matches to C 1 (imatinib-resistant) patient tumors and GS11354 and GS11351 which are strong matches to C 2 patient tumors. Rows are organized in the same order as in Figure 2B . A few recurrent OncoTarget predictions are incorporated into the heatmap for simplicity and are denoted as [# target protein], with corresponding OncoTarget -log10 (FDR p -value). (B) PDX therapeutic study design for GS5106 and GS5108 . Only OncoTreat/OncoTarget predictions that are conserved for the PDX tumor were incorporated in the study. Initial maximum tolerated dose (MTD) studies in non-tumor bearing NOD/SCID mice were performed at three doses levels for each drug, using reported doses from the literature as an initial starting point. Dose and schedule for all but the final part of the study was based on the MTD. Notably, assessment of crizotinib as a prediction for the GS5106 model did not quite meet our statistical significance threshold but was close ( p ∼ 10 -4 ), while it was predicted to have no meaningful activity for GS5108 . This discordance is consistent with the fact that crizotinib was found to invert the activity of a subset of TCM-specific MRs detectable in only 3 of 10 imatinib-resistant tumors, and that the activity of this subset of MRs was not strongly conserved in the two PDX models. Thus, crizotinib was eliminated, leaving six candidate drugs for in vivo efficacy evaluation. The resulting PDX therapeutic study design is summarized in Figure 3B . OncoTreat and OncoTarget Predict Treatment Response in PDX Models In vivo studies took place in four phases: 1) maximum tolerated dose (MTD) finding studies in non-tumor bearing NOD/SCID mice using three dose levels for each drug; 2) therapeutic studies with standard of care drugs for GIST (imatinib and regorafenib) to assess for phenotypic resistance in the passage corresponding to the therapeutic cohort; 3) therapeutic studies with OncoTreat or OncoTarget -predicted drugs in P3 passages of the models; 4) testing of active predicted drugs at titrated doses lower than the MTD. Early on-treatment biopsy specimens were collected in each of the therapeutic arms (in phases 2-4), for pharmacodynamic testing. These mice were excluded from therapeutic response assessment. The following MTDs were identified for further study: imatinib 50 mg/kg twice daily; regorafenib 10 mg/kg daily (5 days on, 2 days off); linifanib 100 mg/kg daily; selumetinib 100 mg/kg daily; tamoxifen 10 mg/kg daily; AZD1775 50 mg/kg daily; panobinostat 50 mg/kg daily; and selinexor 15 mg/kg three times weekly. Predicted drugs were tested only in the PDX model(s) in which they were also predicted, as summarized in Figure 3B . Mice were treated for 21 days, with tumor measurements continuing for 28 days. Treatment doses were held as needed based on total body weight loss parameters, per protocol, and other observed toxicities. Therapeutic studies in both PDX models confirmed complete phenotypic resistance to imatinib with no significant difference in tumor growth between imatinib and vehicle control-treated mice ( Figure 4A &C ). This is reassuring as these models were selected agnostic to any known clinical, genotypic, or phenotypic characteristics of the PDX, but rather based on their ability to recapitulate the MR signatures we saw in imatinib-resistant patient tumors. Importantly, the multi-kinase inhibitor regorafenib demonstrated significant tumor growth inhibition (TGI) relative to vehicle control in both models at the end of treatment (EoT) timepoint, achieving 55% TGI in GS5106 ( p = 0.022 based on a linear mixed model for repeated measurements as implemented in 59 ) and 120% TGI in GS5108 (i.e., 20% regression from baseline, p = 0.002) ( Figure 4A &C ). While regorafenib is clinically approved for imatinib-resistant GIST, it is often poorly tolerated for long term use in patients due to multiple off-target side effects 60 . Download figure Open in new tab Figure 4. Treatment response in GIST PDX models. Therapeutic PDX studies were conducted in 5 parts ( A-E ) in collaboration with Crown Bioscience, each with a vehicle control arm. (A) Standard of care drugs imatinib and regorafenib were tested in GS5106 . Tumor growth curves (left panel) and relative tumor growth inhibition (TGI) compared to vehicle control at the end of treatment (EoT: right panel) are shown. As expected, imatinib did not demonstrate TGI, while regorafenib induced 55% TGI ( p = 0.022). (B) Drugs predicted by OncoTreat (AZD1775, linifanib, selumetinib, and tamoxifen) were tested in GS5106 . Linifanib demonstrates marked tumor regression (TGI of 137%, p = 0.015), tamoxifen demonstrated minimal TGI (22%, p = 0.47), and AZD1775 and selumetinib were ineffective. (C) Standard of care drugs imatinib and regorafenib were tested in GS5108 . As expected, imatinib did not demonstrate TGI, while regorafenib resulted in 120% TGI. (D) Drugs predicted by OncoTreat (linifanib, selumetinib) and OncoTarget (selinexor based on XPO1 activation and panobinostat based on HDAC10/7 activation) were tested at the maximum tolerated dose in GS5108 . Linifanib again demonstrated marked tumor regression (TGI 169%, p = 0.033), while selinexor resulted in 45% TGI and selumetinib was ineffective. Panobinostat resulted in accelerated tumor growth compared to vehicle control. (E) Because part 4 of the study ( panel D ) suffered from unexpected toxicity resulting in weight loss and early death in mice across multiple arms of unclear etiology, which prompted significant dose interruptions and dropout of mice, an additional study was performed evaluating linifanib, selinexor, and selumetinib at multiple dose levels. Linifanib continued to demonstrate significant TGI across dose levels (80%, p = 0.003 and 75%, p = 0.005). Likewise, both selinexor arms and the selumetinib arm all demonstrated statistically significant TGI in the 50-60% range, at the lower dose levels, but with more consistent dosing. Linifanib, which was predicted by OncoTreat for 8 of 10 imatinib-resistant LRG tumors, resulted in marked tumor regression in both PDX models, with 137% TGI ( p = 0.015) and 169% TGI ( p = 0.033) in GS5106 and GS5108 , respectively, when evaluated at the MTD ( Figure 4B &D ). Selinexor, which was predicted by OncoTarget for 7 of 10 imatinib-resistant tumors based on XPO1 activation, was evaluated in GS5108 and resulted in 45% TGI at EoT ( Figure 4D ). This is close to the commonly cited TGI ≥ 50% benchmark in PDX therapeutic studies 61 – 63 that predicts clinical benefit. Selumetinib, which was predicted for 5 of 10 imatinib-resistant patient tumors and in both PDX models by OncoTreat , unfortunately did not result in significant TGI in either GS5106 or GS5108 ( Figure 4B &D ). Likewise, tamoxifen and AZD1775, which were predicted for a smaller subset of imatinib-resistant tumors and only tested in GS5106 , demonstrated no significant TGI ( Figure 4B ). Finally, panobinostat was evaluated in GS5108 and demonstrated no significant TGI and in fact apparently resulted in acceleration of tumor growth with -57% TGI ( Figure 4D ). There were concerns with the health of the cohort of mice used for these therapeutic studies, including premature deaths across multiple arms, despite all drug doses being well tolerated in the prior MTD studies in non-tumor bearing mice. As these studies were being conducted in the early days of the COVID-19 pandemic, it is possible that sterility conditions in the testing facility were suboptimal. The dropout of mice in the therapeutic study further limited statistical power to confirm meaningful TGI—e.g., the 45% TGI with selinexor in GS5108 did not meet statistical significance due to fewer mice making it to the EoT timepoint. We thus proceeded to perform a final phase of therapeutic testing, evaluating the three strongest predictions in GS5108 , linifanib, selumetinib, and selinexor, at multiple dose levels. The efficacy of linifanib at inducing significant TGI was demonstrated at doses significantly lower than the MTD, 50 mg/kg and 25 mg/kg daily, with TGI of 80% ( p = 0.003) and 75% ( p = 0.005), respectively, although there was some clear dose dependence on the extent of response ( Figure 4E ). There was also moderate TGI with selinexor, which after a one week run in at the MTD, was administered at the lower tested doses, 10 mg/kg twice weekly and 5 mg/kg three times weekly, with TGI of 50% ( p = 0.045) and 53% ( p = 0.017), respectively. Selumetinib also resulted in moderate TGI of 54% ( p = 0.049) at the alternative dosing schedule, 50 mg/kg twice daily, in GS5108 ( Figure 4E ). It is likely that the greater benefit seen with selumetinib at the lower dose was due to the ability to more consistently administer the dose, as the mice were generally healthier in this final phase of testing. A comprehensive summary of TGI assessment for all therapeutic arms and statistical significance testing is presented in Table S4 . Pharmacodynamic Assessment of TCM-inversion by Predicted Drugs In Vivo Pharmacodynamic (PD) studies are critical to drug development to elucidate mechanism and to characterize primary and acquired resistance. Specifically, we were interested in learning if OncoTreat -predicted drugs that demonstrate efficacy in the PDX, recapitulate in vivo the reversal of TCM MR activity that occurs in the cognate cell line model and forms the basis of prediction. Additionally, since a number of our predictions were ineffective, our PD analysis attempted to determine if these drugs were ineffective due to an inability of achieve TCM MR activity reversal in vivo , perhaps due to pharmacokinetic factors, or ineffective despite achieving reversal, for instance due to later cell adaptation. Finally, as regorafenib, one of the standard of care control arms, demonstrated effective TGI in the PDXs despite not being predicted for imatinib-resistant tumors by OncoTreat , we wished to learn if regorafenib induced TCM MR activity reversal in vivo in these two models. Biopsy specimens were collected from two randomly selected mice in each therapeutic arm, three hours following the third dose, and RNASeq was performed. We assessed for the reversal of TCM MRs, as defined from the baseline PDX tumor profile ( i.e. , 25↑+25↓), in the VIPER-derived differential protein activity signatures comparing drug vs. vehicle control-treated PDX samples. In GS5106 , ineffective drugs including imatinib and the predicted drugs AZD1775, tamoxifen, and selumetinib, demonstrated no statistically significant evidence of in vivo TCM MR activity reversal ( Figure 5A ). On the other hand, both effective drugs, including the standard of care control regorafenib ( p = 9 x 10 - 17 ) and linifanib ( p = 2 x 10 -18 , by OncoTreat ) demonstrate significant TCM-inversion. Download figure Open in new tab Figure 5. Pharmacodynamic assessment of TCM-inversion in vivo , in early on-treatment biopsy samples and activity changes in select OncoTarget proteins. In the two GIST PDX models, two mice from each drug arm were sacrificed after the 3 rd dose. VIPER was used to generate a differential protein activity signature for each drug-treated versus corresponding vehicle control-treated arm. Reversal of activated and inactivated TCM MRs, as defined from the baseline PDX profiles, was assessed in the treatment signature by enrichment analysis (aREA) ( A-C ) . Statistically significant TCM-inversion in vivo (FDR p < 10 -5 , highlighted by surrounding purple boxes), was confirmed for linifanib in both PDX models, including all three dose level arms in GS5108 . This is consistent with the initial OncoTreat prediction, based on in vitro reversal of the TCM in the cognate cell line (GIST430) model. Regorafenib, an approved clinical option for GIST after progression on imatinib, also demonstrated significant TCM-inversion in both models, consistent with its significant TGI in vivo . Selumetinib, which resulted in TGI in GS5108 at the lower 50 mg/kg dose, demonstrated significant TCM-inversion only at this dose level. Drug predictions by OncoTreat that did not induce significant TGI, generally did not induce significant TCM-inversion in vivo , including the AZD1775, selumetinib, and tamoxifen arms in GS5106 . (D) Changes in activity of select druggable proteins are shown in the heatmap. The activity of c-KIT and PDGFRA were not consistently reversed by any of the tested drugs in these imatinib-resistant PDXs, with the exception of modest reduction in c-KIT activity with selumetinib. Selinexor, which demonstrated some TGI across dose levels in GS5108 , consistently reversed the activity of XPO1, which was the basis for its prediction by OncoTarget . Likewise, selumetinib which demonstrated some TGI at the lower dose level in GS5108 , and was predicted based on both OncoTreat and OncoTarget analysis (MAP2K2 [MEK2] activation), consistently demonstrated reversal of MEK2 activity in vivo , although interestingly linifanib also resulted in marked deactivation of MEK2. Finally, panobinostat which was predicted based on HDAC10/HDAC7 activation in C 1 patient tumors, did not reverse HDAC10/HDAC7 activity in vivo . Likewise, in GS5108 , imatinib demonstrated no significant in vivo TCM MR activity reversal ( Figure 5B ). On the other hand, standard of care control regorafenib demonstrated borderline TCM-inversion ( p = 3 x 10 -5 ) and linifanib demonstrated marked TCM-inversion across multiple dose levels ( p = 9 x 10 -11 at 100 mg/kg daily, p = 1 x 10 -15 at 50 mg/kg daily, and p = 4 x 10 -12 at 25 mg/kg, all by OncoTreat ) ( Figure 5B-C ). Interestingly, selumetinib, which demonstrated moderate TGI at the lower and better tolerated dose level, demonstrated TCM-inversion at this lower 50 mg/kg dose ( p = 5 x 10 -6 , by OncoTreat ) but not at the 100 mg/kg dose ( Figure 5B-C ). Selinexor, which also demonstrated significant TGI compared to control, did not achieve in vivo TCM-inversion ( Figure 5B-C ). This is perhaps not surprising, as the basis of the selinexor prediction was the singular activation of XPO1, and not based on TCM MR protein activity reversal in our in vitro drug screens. Indeed, selinexor does markedly decrease XPO1 activity in vivo , whereas panobinostat, which was ineffective, did not significantly reduce HDAC10/7 activity ( Figure 5D ). In fact, panobinostat is noted to significantly enhance TCM MR activity in GS5108 ( Figure 5B ), corresponding to the accelerated tumor growth we observed. Interestingly, selumetinib was the only drug that demonstrated a trend toward reduction in c-KIT activity in these PDX models. Discussion The development of high-affinity inhibitors of the c-KIT tyrosine kinase, most notably imatinib 64 , has greatly benefited patients with GIST, a disease characterized by aberrant c-KIT activity 3 . C-KIT targeting therapy has reduced the rate of relapse after surgical resection of high-risk GIST 65 and significantly prolonged survival in the metastatic setting 12 , 25 . Unfortunately, the eventual development of imatinib-resistance is common, at which point cancer progression is often inexorable, resulting in patients succumbing to disease 66 . While most tumors that develop acquired resistance to imatinib harbor secondary mutations in KIT 16 , 17 , the development of next generation c-KIT inhibitors like ripretinib, that are predicted to overcome these mutations, has only achieved modest efficacy in the clinic, with objective response rates as low as 9% and mPFS of 6.3 months in the imatinib-resistant setting 26 . In this study, we demonstrate the application of two RNA-based, NY and California CLIA certified assays, OncoTarget and OncoTreat , to identify candidate drugs for the effective treatment of imatinib-resistant GIST. These approaches complement traditional oncogene-centered precision cancer medicine, by shifting the therapeutic focus from mutated oncoproteins to highly conserved proteins dubbed Master Regulators (MRs), responsible for canalizing the aberrant signals arising from the entire mutational spectrum of a tumor. Oncogene mutations represent only one of many potential mechanisms for inducing the aberrant activity of a protein 37 . Indeed, the remarkable clinical benefit derived from androgen receptor blockade in prostate cancer and Bruton’s tyrosine kinase inhibitors in B-cell malignancies are examples of cancer dependencies implemented by genes that are almost never mutated at initial diagnosis. Conversely, presence of a mutation is no guarantee of drug sensitivity, as mutations downstream from the primary oncogene can rapidly induce selection of drug resistant cells 38 . Compared to mutation-based approaches, OncoTarget significantly expands the search field for pharmacologically actionable, aberrantly activated proteins, by identifying individual MRs for which a high-affinity inhibitor is already available. Its use can lead to systematic discovery of effective treatment—from the current repertoire of clinically relevant inhibitors in oncology—for tumors that have failed to respond to multiple lines of therapy, as shown both preclinically 28 , 44 and clinically 31 . OncoTreat implements a more holistic approach to identify drugs capable of inverting the activity of the 25 most aberrantly active and 25 most aberrantly inactive MRs of a tumor—a protein module dubbed the Tumor Checkpoint (TCM) 37 and shown to be responsible for canalizing the effect of >80% of functional mutations in individual tumors 54 . OncoTreat leverages large-scale drug perturbation screens with RNASeq profiling of the molecular response of cognate tumor cell models to identify clinically relevant drugs that invert TCM activity. As shown here, OncoTreat provides a scalable tool to identify effective therapies for drug resistant tumors by matching their mechanistic drivers to proteome-wide drug mechanism of action, as dissected by VIPER. While OncoTarget and OncoTreat predictions have been systematically tested in PDX models 28 , a critical novelty here is the use of these methods to identify the unique dependencies of drug resistance in tumors that fail to respond or relapse following standard of care therapy. Indeed, the development of imatinib-resistance in advanced GIST is the key determinant of poor outcome. Our analysis shows that MR protein activity is not only virtually orthogonal in imatinib-resistant vs. imatinib-sensitive tumors ( Figure 1 ), but that aberrant activation and inactivation of these proteins is remarkably conserved across the entire subset of imatinib-resistant GISTs. This conservation provides significant mechanistic insight, suggesting that imatinib-resistance is not fully accounted for through the clonal selection of secondary mutations that reduce imatinib binding. Rather, imatinib-resistance is associated with a critical, proteome-wide reprogramming of cell state, which appears independent of whether imatinib-resistance was primary or acquired, since the cluster of imatinib-resistant tumors shows conservation across both forms. Critically, seven tumors that had not yet developed overt imatinib-resistance present almost perfect MR activity overlap with those presenting frank imatinib-resistance, five of which would also be predicted to be imatinib-resistant based on genotype (three harboring PDGFRA D842V and two being KIT / PDGFRA -wildtype). Additional follow up would be required to determine whether tumors presenting with the imatinib-resistance MR signature will ultimately develop clinical imatinib-resistance—although, some may have been cured by timely surgical removal alone. Equally important, we show that the VIPER-based OncoMatch algorithm—originally developed to identify high-fidelity cognate models of human tumors that recapitulate their MR activity 29 — could identify only 2 of 18 established GIST PDX models as optimally conserving imatinib-resistance at the MR level. This is critical for two reasons. First, it allows effective validation of predicted drugs in models that most closely capture the clinical drug-resistant state. Second, and perhaps more importantly, it prevents testing drugs in models which, by failing to replicate the activity of the ultra-conserved MRs of imatinib-resistant GIST, would potentially lead to idiosyncratic findings. As expected, not all predictions were effective. However, the PD studies included in this manuscript show that many of the ineffective OncoTreat- predicted drugs were not able to recapitulate TCM MR activity reversal in vivo ( Figure 5 ). This suggests that pharmacokinetic factors may have played a significant role, since in a previous study, 15 of 18 OncoTreat predicted drugs induced TCM activity reversal in vivo and resulted in disease control. Subsequent to the original validation studies, we have completed a much larger perturbational screen in GIST430, incorporating 333 clinically relevant oncology drugs. These new profiles have helped identify several additional candidate drugs by OncoTreat that are predicted to induce TCM activity inversion in a sizable subset of imatinib-resistant tumors ( Figure 6 : pharmacotypes III & IV). Among others, these include axitinib, MLN-2480 (pan-RAF inhibitor), AZD-5069 (CXCR2 antagonist), savolitinib (MET inhibitor), prinomastat (matrix metalloproteinase inhibitor), ipatasertib (pan-AKT inhibitor), abexinostat (pan-HDAC inhibitor), enasidenib (IDH2 inhibitor), conventional alkylating chemotherapeutics [carmustine and mechlorethamine], multiple hormonal signaling modulators [enzalutamide, fulvestrant], and differentiating agents [ATRA (vitamin A metabolite) and arsenic trioxide]. Additionally, since most predicted drugs did not significantly reduce c-KIT activity, it is also possible that greater tumor response may be achieved by combining drugs predicted to reverse the resistant state with imatinib or other TKIs that directly target c-KIT. It is reasonable to expect that imatinib-sensitive and resistant cell states may co-exist in most tumors 67 , albeit at different ratios. Download figure Open in new tab Figure 6. OncoTreat predictions for LRG patient tumors after completion of an expanded high throughput drug screen in GIST430, using 333 drugs, including 122 FDA approved antineoplastics and 192 late-stage experimental drugs in phase II or III trials. ( A ) Heatmap of OncoTreat predictions using data from the expanded drug screen in GIST430. Color intensity in the heatmap corresponds to the -log10 (FDR p -value), and the predictions are clustered into pharmacotypes ( columns ) and by drug ( rows ), with cluster reliability indices for cluster assignment shown as barplots on the top and righthand side of the heatmap. A few recurrent OncoTarget predictions are incorporated into the heatmap for simplicity and are denoted as [# target protein], with corresponding OncoTarget -log10 (FDR p -value). Clinical annotation is provided at the bottom of the heatmap. Several additional candidate drugs to target imatinib-resistant tumors emerge, including axitinib, MLN-2480, AZD-5069, savolitinib, ipatasertib, prinomastat, enasidenib, carmustine, multiple hormonal signaling modulators [enzalutamide, fulvestrant], and differentiating agents [ATRA (Vitamin A) and arsenic trioxide]. Interestingly, while panobinostat is not predicted based on its measured effect in GIST430, an alternative pan-HDAC inhibitor abexinostat is strongly predicted for several imatinib-resistant tumors. Selinexor, on the other hand, is in fact predicted by OncoTreat for the vast majority of tumors with XPO1 activation. It should be noted that the framework discussed in this study is not specific to GIST and could be applied broadly across virtually any drug-resistant tumor, to identify dependencies associated with common molecular mechanisms. While not tested in this study, the approach could also be applied iteratively, to identify additional therapeutic options as treatment resistance evolves over time, an approach whose feasibility will increase as blood-based tumor RNA profiling technologies become available 68 . Importantly, we have demonstrated the value of VIPER-based protein activity inference to single cell (scRNASeq) profiles 69 , 70 , and OncoTreat and OncoTarget can directly be applied to identify optimal therapeutic targeting of dominant subpopulations within tumors. Targeting of multiple subpopulations, either sequentially or in combination, may ultimately be necessary to durably overcome resistance. Methods Generation of Gene Regulatory Networks To support context-specific regulatory protein activity inference by VIPER and metaVIPER, we have generated comprehensive molecular interaction networks (interactomes) using the Algorithm for the Reconstruction of Accurate Cellular Networks (ARACNe) 71 , 72 . Most networks were reverse engineered through the application of ARACNe to cohorts from The Cancer Genome Atlas (TCGA), whenever ≥ 100 RNASeq profiles are available. While there is a soft tissue sarcoma cohort in TCGA (“sarc”), GISTs were specifically excluded. The TCGA RNASeq level 3 data were downloaded from NCI Genomics Data Commons 73 . Raw counts were normalized and variance stabilized by fitting the dispersion to a negative-binomial distribution as implemented in the DESeq2 R-package 74 (RRID: SCR_000154). ARACNe was run with 100 bootstrap iterations using an input set of candidate regulators including: (a) 1,877 transcription factors annotated in the Gene Ontology (GO) 75 “Molecular Function database” as DNA-binding transcription factor activity (GO:0003700), DNA binding (GO:0003677), Transcription regulator activity (GO:0030528), or Regulation of transcription, DNA-templated (GO:0003677 and GO:0045449); (b) 677 transcriptional co-factors manually curated from genes annotated as Transcription Coregulator Activity (GO:0003712), Plays a Role in Regulating Transcription (GO:0030528), or Regulation of Transcription (GO:0045449); and (c) 3,895 genes encoding for signal transduction proteins, dually annotated in the GO “Biological Process database” as Signal Transduction (GO:0007165) and in the GO “Cellular Component database” as either Intracellular (GO:0005622) or Plasma membrane (GO:0005886). The Data Processing Inequality (DPI) parameter of ARACNe was set to 0, the Mutual Information (MI) threshold set to p = 10 −8 , and the mode of regulation computed based on the correlation between regulator and target gene expression as previously described 32 . The version of n=43 generated networks used in our work are provided (see Figshare: https://figshare.com/articles/dataset/Available_Interactomes_Networks_/30246550 ). Collaboration with Life Raft Group We have worked closely with the patient advocacy organization, Life Raft Group (LRG), who agreed to share with us deidentified tumor bank specimens and linked clinical annotation. Available clinical variables included primary tumor location, stage at diagnosis, risk classification at diagnosis based on tumor size and mitotic rate, clinical genotype testing (specific KIT and PDGFRA variants), subsequent disease status (relapse/metastasis), survival from diagnosis, and prior treatments and durations. Nucleic acid extraction, sequencing, and analysis All formalin-fixed paraffin-embedded (FFPE) samples received from LRG were sectioned into fresh slides and stained by H&E, evaluated by a pathologist, and micro-dissected to ensure that a minimum of 70-80% viable tumor was present for subsequent extraction and analyses. RNA was extracted using the QIAGEN RNeasy (FFPE) Kit. Strand specific sequencing libraries were generated from rRNA depleted total RNAs using the NEBNext Ultra Directional RNA Library Prep Kit for Illumina. Sequencing was performed on the Illumina NextSeq 500 sequencer with 100bp x 2 paired-end reads to a depth of 20M. We initially performed RNASeq on about 50 samples from the LRG tumor bank, including several imatinib-resistant cases. Some samples did not meet quality control standards following library generation or sequencing, and thus the final analysis dataset included 34 unique tumors, including 10 that had clinical progression on imatinib. Alignment was performed with STARv2.5.3 76 to the Genome Reference Consortium GRCh38.v90 build with gene-level raw counts summarization with STAR quant-mode. metaVIPER Analysis We have previously extensively validated the Virtual Proteomics by Enriched Regulon (VIPER) algorithm as a highly robust and specific tool for the accurate inference of regulatory protein activity in a tissue context-dependent manner 32 , 37 , 42 . VIPER leverages accurate tissue-specific gene regulatory networks to measure differential protein activity from bulk or single-cell gene expression signatures (GES). Specifically, akin to a multiplexed gene-reporter assay, VIPER measures a protein’s differential transcriptional activity through a probabilistic enrichment framework that assesses the normalized enrichment score (NES) of its activated and repressed transcriptional targets ( regulon ) in genes over and under expressed in a sample of interest compared to a set of control samples ( reference model ). Given significant batch effects introduced through various mRNA enrichment strategies for library generation from FFPE samples, the only suitable reference model for our analysis of LRG GIST tumors was the centroid of all gene expression profiles comprising the dataset. Thus, for each tumor sample, we generate a differential gene expression signature (DGES)— computed as the gene-wise relative expression to the distribution of the expression of that gene across all 34 LRG samples—and expressed as its Z-score relative to the reference model . Next, VIPER, using each regulatory protein’s set of regulated genes, as assigned in the network, performs an enrichment test to determine if the protein’s transcriptional footprint is over-, under-, or normally represented in the tumor’s DGES, in the process assigning each protein a normalized enrichment score (NES) as a measure of its regulatory activity. Specifically, VIPER implements a rapid analytic-rank based enrichment analysis (aREA) test 32 , which efficiently provides test statistics similar to permutation-based gene set enrichment analysis 51 . When a suitable context-specific gene regulatory network is not available, as was the case for GIST, we have successfully employed metaVIPER, which essentially first runs VIPER using all available networks independently, and then for each protein performs Stouffer’s Z-score integration for each of the independently derived NES’s, which can optionally be weighted 45 . The most differentially active proteins identified by VIPER comprise candidate Master Regulator (MR) proteins that mechanistically control a tumor’s transcriptional identity, as shown by multiple studies, see 37 for a comprehensive perspective. VIPER reproducibility is extremely high, such that Spearman correlation of activity profiles generated from RNASeq at 30M to as low as 50K read depth 32 is p > 0.8, even though correlation of the underlying gene expression profiles is low p < 0.3. While VIPER is uniquely suited for assessing regulatory proteins that directly control gene expression, including transcription factors, co-factors, and chromatin remodeling enzymes, we have shown that the algorithm is equally effective in monitoring activity of signaling proteins 32 and cell surface markers 45 . Optimal Clustering Analysis Consensus and optimal clustering were performed on the metaVIPER derived differential protein activity profiles of the 34 LRG GIST tumors, as implemented in the ConsensusClusterPlus 46 and OptCluster 47 packages for R. Specifically, a partition around medoids (PAM) approach, using Spearman-rank correlation as the distance metric between samples, considering k=2 to 8 clusters (8 representing 0.25 * the total number of samples), and performing n=10,000 iterations was employed for the analysis presented in Figure 1 . All relevant statistical analyses were performed using R software (v3.5.0). Following optimal clustering of samples into two roughly even partitions, metaVIPER was run iteratively, this time using the centroid of samples from the opposing cluster (instead of the entire dataset) to generate the DGES, in order to further amplify the protein activity signal derived from the imatinib-resistant (cluster C 1 ) state. The obtained protein activity signatures were used for further downstream analysis, including OncoTarget and OncoTreat predictions. Pathway Enrichment Analysis Pathway analysis on the integrated differential protein activity signature of cluster C 1 imatinib-resistant tumors, as presented in supplementary Figure S1, was performed with ‘Gene Ontology’, ‘Cancer Hallmarks’, ‘Immunological Signatures’, and ‘Oncogenic Signatures’ gene sets provided in the Broad MSigDB collections 50 – 53 . Pathway enrichment analysis was performed by using the single-tail aREA method described above as a rapid approximation to the Kolmogorov-Smirnov (KS)-test used in GSEA 51 , inputting the pathway genes and the sorted differential protein activity signature. Unlike other statistical tests used for pathway enrichment analysis, such as the Fisher’s exact test on a thresholded list of differentially expressed genes/proteins or the repeated application of the KS test in classical GSEA, aREA does not require the arbitrary binarization into significant/non-significant hits for target genes in the signature but takes into account their relative position in the signature. OncoTarget Analysis Through the use of (a) DrugBank 77 , (b) the SelleckChem database (RRID:SCR_003823) 78 , (c) published literature, and (d) publicly available information on pharmaceutical company drug development pipelines, we have curated a refined list of 180 actionable proteins representing validated targets of high-affinity pharmacological inhibitors, either FDA approved or in clinical trials ( Table S1 ). This manually curated target-drug(s) database is dominated by signaling proteins and established oncoproteins, including c-KIT. Pharmacological agents with narrow therapeutic indices—such as those targeting neurotransmitters, ion channels, and vasoactive drugs—were purposefully removed from the database as they are less likely to be successfully repurposed in oncology. OncoTarget simply analyzes the VIPER outputted protein activity measurements for these 180 actionable proteins, and provides a multiple-testing corrected significance value for the corresponding NES. We used a conservative threshold (FDR p < 10 -5 ) to identify candidate proteins eliciting essentiality when targeted by a pharmacological inhibitor, for in vivo validation. OncoTreat Analysis We have previously described OncoTreat 28 , 33 as a methodology to systematically elucidate compounds capable of significantly inverting the activity of the 25↑+25↓ MRs comprising the tumor checkpoint module (TCM) that regulates the tumor state in general, or more specifically that which regulates metastatic progression or therapeutic resistance. For the current study, we adapted OncoTreat to identify TCM-inverter compounds to target the imatinib-resistant state. The requisite steps include: Identifying cell lines (typically 1 or 2) or organoid lines jointly representing high-fidelity (i.e., cognate) in vitro models for the samples of interest (e.g., imatinib-resistant LRG GISTs), based on TCM-conservation (FDR-corrected p < 10 -5 ), as assessed by enrichment analysis, see next section on OncoMatch. Generating RNASeq profiles within each cognate cell line, from Step 1, at 24h following perturbation with a library of clinically relevant oncology drugs. Drugs were titrated at their maximum sub-lethal concentration (i.e., 48h EC 20 ), as determined by 10-point dose response profiles. Profiled drugs included FDA-approved and late-stage experimental oncology drugs (in phase II and III clinical trials) (see Table S2 for the drugs tested in the initial GIST430 and GIST-T1 perturbation screens and Table S3 for the drugs tested in the subsequent expanded GIST430 screen). For expanded screens, we have attempted to be all inclusive, excluding therapeutic antibodies and antibody drug conjugates due to a lack of availability and immunotherapy agents due to lack of appropriateness to screen in vitro . Additional miscellaneous compounds with EC 50 ≤ 2μM in the selected cell line were also included. Most compounds were purchased from SelleckChem or Tocris. DMSO was selected as a universal in vitro solvent (vehicle). Multiplexed, low depth (1M to 2M reads) RNASeq profiles were generated using 96-well plates via the PLATESeq technology, using fully automated microfluidics for increased throughput and reproducibility 58 . Eight DMSO-treated controls were included in each plate, to avoid plate-dependent batch effects and to mitigate the inherent variability of DMSO treatment. Generating a subproteome-wide context-specific Drug Mechanism of Action (MoA) for each drug, as represented by the differential activity of each protein in drug- vs. vehicle control (DMSO)-treated cells. Differential protein activity was assessed by VIPER analysis, see above. Identifying tumor-specific candidate MRs and the TCMs they comprise, by metaVIPER analysis of the sample’s DGES, using the opposing cluster of LRG samples as the reference model . Finally, prioritizing pharmacological agents based on the statistical significance of the enrichment of the tumor sample’s TCM-activity signature ( i.e. , 25↑+25↓ MRs) in proteins inactivated and activated in drug- vs. DMSO-treated cells, respectively, with negative NES indicating TCM-inversion (FDR-corrected p < 10 -5 , 1-tailed aREA). The number of candidate MR proteins (n=50) used to assess TCM-inversion, which for this step is restricted to only transcription factors and co-factors, was selected because we have shown that, on average, across all of TCGA, the vast majority of functionally-relevant genomic events can be found upstream of the top 50 VIPER-inferred candidate MR proteins 54 . We also demonstrated in 28 that a wide range of number of candidate MRs to include in the TCM has minimal effect on the ordering of OncoTreat predictions for drug sensitivity. OncoMatch, Cell Line and Patient-Derived Xenograft (PDX) Model Fidelity Analysis Model fidelity was assessed based on the statistical significance of TCM-activity conservation between cluster C 1 (imatinib-resistant) LRG GIST tumors and a model-derived sample. For the purposes of OncoMatch analysis to cell line models only, protein activity was computed by generating a DGES comparing each cell line against a large repository of cancer cell lines ( reference model ), which includes both the Cancer Cell Line Encyclopedia (CCLE) 79 and the Genentech Cell Line Screening Initiative (gCSI) 80 . Meanwhile, protein activity for LRG tumors was computed against an analogous reference model — RNA profiles of all tumor samples in TCGA. Lineage matched normal or adjacent tissue is not used as a reference model for OncoMatch, as differential cell proliferation signaling of tumor v. normal overwhelms the resulting signature. Next, we computed the enrichment of the TCM-activity signature (25 most active and 25 most inactive patient tumor-specific MRs) in differentially active and inactive proteins in the model (OncoMatch). The aREA 32 test was again used, but any suitable enrichment analysis algorithm could be substituted. All cell lines used in these analyses were re-sequenced on-site, to capture any potential drift effects. A threshold of FDR p < 10 -5 was used to identify at least moderate-fidelity models. The analysis was used to evaluate the suitability of the GIST430 and GIST-T1 models for the generation of perturbational profiles that optimally capture the effects of tested drugs on the activity of TCM proteins. We worked with Crown Bioscience to characterize 18 of their existing GIST patient-derived xenograft (PDX) mouse models and identified a small subset that is remarkably representative of imatinib-resistant GIST. For the purposes of OncoMatch analysis between patient tumors and PDX models, protein activity was computed by first generating a DGES relative to the centroid of the LRG GIST and Crown Bioscience PDX datasets, respectively, followed by metaVIPER analysis and application of aREA to evaluate TCM-activity conservation. We thus evaluated the fidelity of PDXs prior to testing drug predictions for imatinib-resistant patient tumors, and further confirmed that those drug predictions were also conserved for the PDX tumor before proceeding. Therapeutic drug testing All in vivo studies were performed by Crown Bioscience. Initial maximum tolerated dose (MTD) studies were performed at 3 dose levels for imatinib, regorafenib, and the six candidate drugs in non-tumor bearing NOD/SCID mice. Reported doses in the literature for PDX studies were used as a starting point for the dose levels evaluated. Dosing continued for 14 days, with an additional 7 days of post-treatment monitoring for toxicity. Clinical observations were performed twice weekly and body weight measurements daily or every other day at minimum. For the efficacy studies in GS5106 and GS5108 , tumor fragments from stock mice were harvested and used for inoculation into mice. Mice are ear tagged. Each mouse is inoculated subcutaneously in the right front flank with a specific PDX tumor fragment (3x3x3 mm). Implanted mice are monitored for palpable tumors, or any changes in appearance or behavior. Daily monitoring takes place for mice showing any signs of morbidity. Once tumors are palpable, tumors are measured twice weekly using calipers. Tumor volume is estimated using the following equation: (longest diameter * shortest diameter 2 )/2. Once average tumor volume reaches 150-250mm 3 , mice are randomly assigned to the respective treatment groups and dosed on day 1. Dosing continued for up to 21 days or until death/sacrifice, with an additional 7 days of observation post-treatment. Tumor volume is measured twice weekly for the duration of the study. One individual is responsible for tumor measurements for the duration of the study. Body weight is also measured at least twice weekly for the duration of the study, and frequency of monitoring was increased to every other day when toxicity was noted. If body weight loss of >10% is observed, Dietgel is given ad libitum . If body weight loss of >15% is observed, the animal is given a dosing holiday until weight loss is 20% is observed, the animal is monitored daily for signs of recovery for up to 72 hours. If there are no signs of recovery, the animal is sacrificed as per Crown Bioscience’s IACUC protocol regulations. The DRAP package for R 59 was used to plot tumor volume curves and compute mean tumor growth inhibition (TGI) at the EoT (21-days) timepoint for each drug arm versus vehicle control. TGI was calculated using the formula 1 – F(Vt)/F(Vc), where F(Vt) is the difference in mean tumor volume from baseline to EoT in the treatment arm, and F(Vc) is the difference in mean tumor volume from baseline to EoT in the vehicle control arm. TGI ≥ 1.0 (i.e., ≥ 100%) indicates that mean tumor volume has regressed from baseline in the treatment arm, while a negative value, TGI < 0 (i.e., a negative percentage), indicates that the tumor volume increase in the treatment arm exceeded that seen in the control arm. The linear mixed model for repeated measures 59 , 81 , as implemented in DRAP, was used for statistical significance testing to determine if a drug significantly impacted tumor growth from time 0 to the EoT timepoint. For both mean TGI and statistical significance testing, drugs were compared to the concurrent vehicle control arm only for the five separate parts of the PDX therapeutic study as presented in Figure 4 . Pharmacodynamic (PD) Assessments of TCM-inversion Samples for PD assessment were procured from two mice per treatment arm in GS5106 and GS5108 . Mice were randomly selected for early sacrifice, independent of tumor size, three hours following the third dose, and were excluded from response assessment. We performed RNASeq and subsequent VIPER on paired drug vs. vehicle control-treated PDX tumor samples. TCM-inversion was assessed based on the statistical significance of the enrichment of the TCM-activity signature (i.e., 25↑+25↓ MRs of the baseline PDX tumor profile) in proteins inactivated and activated in drug vs. vehicle control-treated mice, respectively, again using aREA, although alternative enrichment tests may be used. Negative NES indicates TCM-inversion (FDR p < 10-5). This is in effect the same analysis as OncoTreat . Data Availability RNASeq raw data of 34 LRG patient tumors and RNASeq profiles of on-treatment (pharmacodynamic) PDX tumor samples generated as part of this work have been deposited to gene expression omnibus (GEO). RNASeq data of 18 established GIST PDX models was obtained via a data use agreement with Crown Bioscience. Relevant high throughput drug perturbation data used for de novo assessment of drug MoA is available on Figshare (FS: https://figshare.com/articles/dataset/Drug_Perturbation_Profiles_in_GIST_Cell_Lines/30247381 ). The version of TCGA and TARGET processed data (RNASeq counts) used to generate networks are available on FS ( https://figshare.com/articles/dataset/TCGA_and_TARGET_RNASeq_Counts_Data_EntrezID_/30247285 ). The ARACNe generated networks used to run VIPER are available on FS ( https://figshare.com/articles/dataset/Available_Interactomes_Networks_/30246550 ). Additional requests should be directed to the corresponding author. Conflict of Interest Statement Dr. Califano is founder, equity holder, and consultant of DarwinHealth, Inc., a company that has licensed some of the algorithms used in this manuscript from Columbia University. Columbia University is also an equity holder in DarwinHealth Inc. Dr. Alvarez is an employee and equity hold of DarwinHealth, Inc. Dr. Ingham is a stockholder in Regeneron Pharmaceuticals, Inc., with no direct relevance to the work presented herein. Dr. Schwartz has been involved in clinical trials of selinexor sponsored by Karyopharm, with no direct relevance to the work presented herein. Supplementary Figures and Tables Figure S1. Pathway enrichment analysis on the integrating protein activity signature of all cluster C 1 (imatinib-resistant) tumors. The MSigDB gene set collections for Gene Ontology, Cancer Hallmarks, Immunological Signatures, and Oncogenic Signatures are used for this analysis. The analysis suggests imatinib-resistance is associated with a complex cellular reprogramming affecting transcriptional machinery, DNA repair, oxidative phosphorylation, immunological signaling, and MYC proliferative signaling, amongst others. Figure S2. Heatmap for the 24-hour expanded drug perturbation screen in GIST430. The heatmap shows the differential protein activity profile of 333 drugs in GIST430 cells compared to vehicle control, annotated by their canonical mechanism and the sublethal EC 20 concentration at which it was screened. VIPER monitored proteins are shown in the columns and we use unsupervised hierarchical clustering to highlight drugs that induce similar transcriptional response, i.e., context-specific observed mechanism of action. Table S1. Curated list of proteins with high-affinity inhibitor drugs for OncoTarget analysis. Table S2. List of drugs in the initial GIST430 perturbation screen. The EC 20 and lower concentration at which each drug was tested is reported, along with FDA approval status. Table S3. List of drugs in a subsequent expanded GIST430 perturbation screen. The EC 20 at which each drug was tested is reported, along with FDA approval status. Table S4. Detailed summary of therapeutic response at end of treatment time point, organized by PDX model and separate parts of the study performed with independent control arms. The presented t-statistic and p -value are from a linear mixed model accounting for repeat measures over time of tumor volume, as implemented in the DRAP package. Acknowledgements This work was supported by the NCI Office of Cancer Target Discovery and Development (CTD2) award U01CA272610, the NCI Outstanding Investigator award R35CA197745, and the NIH Shared Instrumentation Grant S10OD032433, all to A. Califano. Also, this publication was supported in part through the National Cancer Institute Cancer Center Support Grant P30CA013696 to CUIMC and by the National Center for Advancing Translational Sciences, National Institutes of Health, through Grant Number UL1TR001873. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. Funder Information Declared National Cancer Institute, https://ror.org/040gcmg81 , U01CA272610 , R35CA197745 , P30CA013696 National Institutes of Health, https://ror.org/01cwqze88 , S10OD032433 National Center for Advancing Translational Sciences, https://ror.org/04pw6fb54 , UL1TR001873 References 1. ↵ Weinstein , I . Cancer: addiction to oncogenes . Sciences 297 , 63 – 64 ( 2002 ). OpenUrl FREE Full Text 2. ↵ Miettinen , M. & Lasota , J . Gastrointestinal stromal tumors: pathology and prognosis at different sites . Semin Diagn Pathol 23 , 70 – 83 ( 2006 ). OpenUrl CrossRef PubMed Web of Science 3. ↵ Hirota , S. et al. Gain-of-function mutations of c-kit in human gastrointestinal stromal tumors . Science 279 , 577 – 80 ( 1998 ). OpenUrl Abstract / FREE Full Text 4. Kindblom , L.G. , Remotti , H.E. , Aldenborg , F. & Meis-Kindblom , J.M . Gastrointestinal pacemaker cell tumor (GIPACT): gastrointestinal stromal tumors show phenotypic characteristics of the interstitial cells of Cajal . Am J Pathol 152 , 1259 – 69 ( 1998 ). OpenUrl PubMed Web of Science 5. Sarlomo-Rikala , M. , Kovatich , A.J. , Barusevicius , A. & Miettinen , M . CD117: a sensitive marker for gastrointestinal stromal tumors that is more specific than CD34 . Mod Pathol 11 , 728 – 34 ( 1998 ). OpenUrl PubMed Web of Science 6. ↵ Sircar , K. et al. Interstitial cells of Cajal as precursors of gastrointestinal stromal tumors . Am J Surg Pathol 23 , 377 – 89 ( 1999 ). OpenUrl CrossRef PubMed Web of Science 7. ↵ Edmonson , J. , Marks , R. , Buckner , J. & Mahoney , M . Contrast of response to dacarbazine, mitomycin, doxorubacin and cisplatin plus GM-CSF between patients with advanced gastrointestinal stromal tumors and patients with other advanced leiomyosarcomas . Cancer Invest 20 , 605 – 612 ( 2002 ). OpenUrl CrossRef PubMed Web of Science 8. ↵ Heinrich , M.C. et al. PDGFRA activating mutations in gastrointestinal stromal tumors . Science 299 , 708 – 10 ( 2003 ). OpenUrl Abstract / FREE Full Text 9. ↵ Boikos , S.A. et al. Molecular Subtypes of KIT/PDGFRA Wild-Type Gastrointestinal Stromal Tumors: A Report From the National Institutes of Health Gastrointestinal Stromal Tumor Clinic . JAMA Oncol 2 , 922 – 8 ( 2016 ). OpenUrl PubMed 10. Hostein , I. et al. BRAF mutation status in gastrointestinal stromal tumors . Am J Clin Pathol 133 , 141 – 8 ( 2010 ). OpenUrl CrossRef PubMed 11. ↵ Mandelker , D. et al. Expanded genetic testing of GIST patients identifies high proportion of non-syndromic patients with germline alterations . NPJ Precis Oncol 7 , 1 ( 2023 ). 12. ↵ Demetri , G.D. et al. Efficacy and safety of imatinib mesylate in advanced gastrointestinal stromal tumors . N Engl J Med 347 , 472 – 80 ( 2002 ). OpenUrl CrossRef PubMed Web of Science 13. ↵ Blanke , C. , Rankin , C. , Demetri , G. , Ryan , C. & von Mehren , M . Phase III randomized intergroup trial assessing imatinib mesylate at two dose levels in patients with unresectable or metastatic gastrointestinal stromal tumors expressing the KIT receptor tyrosine kinase . J Clin Oncol 26 , 626 – 632 ( 2008 ). OpenUrl Abstract / FREE Full Text 14. ↵ Heinrich , M. , Corless , C. , Demetri , G. , Blanke , C. & von Mehren , M . Kinase mutations and imatinib response in patients with metastatic gastrointestinal stromal tumor . J Clin Oncol 21 , 4342 – 4349 ( 2003 ). OpenUrl Abstract / FREE Full Text 15. ↵ Patel , S . Long-term efficacy of imatinib for treatment of metastatic GIST . Cancer Chemother Pharmacol 72 , 277 – 86 ( 2013 ). OpenUrl CrossRef PubMed 16. ↵ Antonescu , C. , Besmer , P. , Guo , T. , Arkun , K. & Hom , G . Acquired resistance to imatinib in gastrointestinal stromal tumors occurs through secondary gene mutations . Clin Cancer Res 11 , 4182 – 4190 ( 2005 ). OpenUrl Abstract / FREE Full Text 17. ↵ Antonescu , C.R. & DeMatteo , R.P . CCR 20th Anniversary Commentary: A Genetic Mechanism of Imatinib Resistance in Gastrointestinal Stromal Tumor-Where Are We a Decade Later? Clin Cancer Res 21 , 3363 – 5 ( 2015 ). OpenUrl Abstract / FREE Full Text 18. ↵ Gramza , A.W. , Corless , C.L. & Heinrich , M.C . Resistance to Tyrosine Kinase Inhibitors in Gastrointestinal Stromal Tumors . Clin Cancer Res 15 , 7510 – 7518 ( 2009 ). OpenUrl Abstract / FREE Full Text 19. ↵ Heinrich , M.C. et al. Molecular correlates of imatinib resistance in gastrointestinal stromal tumors . J Clin Oncol 24 , 4764 – 74 ( 2006 ). OpenUrl Abstract / FREE Full Text 20. ↵ Tarn , C. et al. Insulin-like growth factor 1 receptor is a potential therapeutic target for gastrointestinal stromal tumors . Proc Natl Acad Sci U S A 105 , 8387 – 92 ( 2008 ). OpenUrl Abstract / FREE Full Text 21. ↵ Burger , H. et al. Chronic imatinib mesylate exposure leads to reduced intracellular drug accumulation by induction of the ABCG2 (BCRP) and ABCB1 (MDR1) drug transport pumps . Cancer Biol Ther 4 , 747 – 52 ( 2005 ). OpenUrl CrossRef PubMed Web of Science 22. Delbaldo , C. et al. Pharmacokinetic-pharmacodynamic relationships of imatinib and its main metabolite in patients with advanced gastrointestinal stromal tumors . Clin Cancer Res 12 , 6073 – 8 ( 2006 ). OpenUrl Abstract / FREE Full Text 23. ↵ Prenen , H. et al. Everolimus alters imatinib blood partition in favour of the erythrocyte . J Pharm Pharmacol 58 , 1063 – 6 ( 2006 ). OpenUrl PubMed 24. ↵ Demetri , G.D. et al. Efficacy and safety of regorafenib for advanced gastrointestinal stromal tumours after failure of imatinib and sunitinib (GRID): an international, multicentre, randomised, placebo-controlled, phase 3 trial . Lancet 381 , 295 – 302 ( 2013 ). OpenUrl CrossRef PubMed Web of Science 25. ↵ Demetri , G.D. et al. Efficacy and safety of sunitinib in patients with advanced gastrointestinal stromal tumour after failure of imatinib: a randomised controlled trial . Lancet 368 , 1329 – 38 ( 2006 ). OpenUrl CrossRef PubMed Web of Science 26. ↵ Blay , J.Y. et al. Ripretinib in patients with advanced gastrointestinal stromal tumours (INVICTUS): a double-blind, randomised, placebo-controlled, phase 3 trial . Lancet Oncol 21 , 923 – 934 ( 2020 ). OpenUrl CrossRef PubMed 27. ↵ Heinrich , M.C. et al. Avapritinib in advanced PDGFRA D842V-mutant gastrointestinal stromal tumour (NAVIGATOR): a multicentre, open-label, phase 1 trial . Lancet Oncol 21 , 935 – 946 ( 2020 ). OpenUrl CrossRef PubMed 28. ↵ Mundi , P.S. et al. A Transcriptome-Based Precision Oncology Platform for Patient-Therapy Alignment in a Diverse Set of Treatment-Resistant Malignancies . Cancer Discov 13 , 1386 – 1407 ( 2023 ). OpenUrl CrossRef PubMed 29. ↵ Vasciaveo , A. et al. OncoLoop: A Network-Based Precision Cancer Medicine Framework . Cancer Discov 13 , 386 – 409 ( 2023 ). OpenUrl CrossRef PubMed 30. ↵ Jamison , J.K. et al. Entinostat in patients with relapsed or refractory abdominal neuroendocrine tumors . Oncologist 29 , 817 – e1213 ( 2024 ). OpenUrl PubMed 31. ↵ Zeleke , T.Z. et al. Network-based assessment of HDAC6 activity predicts preclinical and clinical responses to the HDAC6 inhibitor ricolinostat in breast cancer . Nat Cancer 4 , 257 – 275 ( 2023 ). OpenUrl PubMed 32. ↵ Alvarez , M.J. et al. Functional characterization of somatic mutations in cancer using network-based inference of protein activity . Nature genetics 48 , 838 – 47 ( 2016 ). OpenUrl CrossRef PubMed 33. ↵ Alvarez , M.J. et al. A precision oncology approach to the pharmacological targeting of mechanistic dependencies in neuroendocrine tumors . Nat Genet 50 , 979 – 989 ( 2018 ). OpenUrl CrossRef PubMed 34. ↵ Rajbhandari , P. et al. Cross-Cohort Analysis Identifies a TEAD4-MYCN Positive Feedback Loop as the Core Regulatory Element of High-Risk Neuroblastoma . Cancer Discov 8 , 582 – 599 ( 2018 ). OpenUrl Abstract / FREE Full Text 35. ↵ Aytes , A. et al. Cross-species regulatory network analysis identifies a synergistic interaction between FOXM1 and CENPF that drives prostate cancer malignancy . Cancer Cell 25 , 638 – 51 ( 2014 ). OpenUrl CrossRef PubMed 36. ↵ Carro , M.S. et al. The transcriptional network for mesenchymal transformation of brain tumours . Nature 463 , 318 – 25 ( 2010 ). OpenUrl CrossRef PubMed Web of Science 37. ↵ Califano , A. & Alvarez , M.J . The recurrent architecture of tumour initiation, progression and drug sensitivity . Nat Rev Cancer 17 , 116 – 130 ( 2017 ). OpenUrl CrossRef PubMed 38. ↵ Chen , J.C. et al. Identification of Causal Genetic Drivers of Human Disease through Systems-Level Analysis of Regulatory Networks . Cell 159 , 402 – 14 ( 2014 ). OpenUrl CrossRef PubMed Web of Science 39. ↵ Piovan , E. et al. Direct reversal of glucocorticoid resistance by AKT inhibition in acute lymphoblastic leukemia . Cancer Cell 24 , 766 – 76 ( 2013 ). OpenUrl CrossRef PubMed Web of Science 40. ↵ Scotto , L. et al. Generation of pralatrexate resistant T-cell lymphoma lines reveals two patterns of acquired drug resistance that is overcome with epigenetic modifiers . Genes Chromosomes Cancer 59 , 639 – 651 ( 2020 ). OpenUrl PubMed 41. ↵ Rodriguez-Barrueco , R. et al. Inhibition of the autocrine IL-6-JAK2-STAT3-calprotectin axis as targeted therapy for HR-/HER2+ breast cancers . Genes & development 29 , 1631 – 48 ( 2015 ). OpenUrl Abstract / FREE Full Text 42. ↵ Bisikirska , B. et al. Elucidation and Pharmacological Targeting of Novel Molecular Drivers of Follicular Lymphoma Progression . Cancer Res 76 , 664 – 74 ( 2016 ). OpenUrl Abstract / FREE Full Text 43. ↵ Mitrofanova , A. , Aytes , A. , Shen , C. , Abate-Shen , C. & Califano , A . A systems biology approach to predict drug response for human prostate cancer based on in vivo preclinical analyses of mouse models . Cell Reports 12 , 1 – 12 ( 2015 ). OpenUrl PubMed 44. ↵ Coutinho , D.F. et al. Validation of a non-oncogene encoded vulnerability to exportin 1 inhibition in pediatric renal tumors . Med 3 , 774 – 791 e7 ( 2022 ). OpenUrl PubMed 45. ↵ Ding , H. et al. Quantitative assessment of protein activity in orphan tissues and single cells using the metaVIPER algorithm . Nat Commun 9 , 1471 ( 2018 ). OpenUrl CrossRef PubMed 46. ↵ Wilkerson , M.D. & Hayes , D.N . ConsensusClusterPlus: a class discovery tool with confidence assessments and item tracking . Bioinformatics 26 , 1572 – 3 ( 2010 ). OpenUrl CrossRef PubMed Web of Science 47. ↵ Sekula , M. , Datta , S. & Datta , S . optCluster: An R Package for Determining the Optimal Clustering Algorithm . Bioinformation 13 , 101 – 103 ( 2017 ). OpenUrl CrossRef PubMed 48. ↵ Yoo , C. et al. Efficacy of Imatinib in Patients with Platelet-Derived Growth Factor Receptor Alpha-Mutated Gastrointestinal Stromal Tumors . Cancer Res Treat 48 , 546 – 52 ( 2016 ). OpenUrl PubMed 49. ↵ Sun , S. , Zhong , B. , Zeng , X. , Li , J. & Chen , Q . Transcription factor E4F1 as a regulator of cell life and disease progression . Sci Adv 9 , eadh1991 ( 2023 ). OpenUrl PubMed 50. ↵ Liberzon , A. et al. Molecular signatures database (MSigDB) 3.0 . Bioinformatics 27 , 1739 – 40 ( 2011 ). OpenUrl CrossRef PubMed Web of Science 51. ↵ Subramanian , A. , et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles . Proc Natl Acad Sci U S A 102 , 15545 – 50 ( 2005 ). OpenUrl Abstract / FREE Full Text 52. ↵ Liberzon , A. et al. The Molecular Signatures Database (MSigDB) hallmark gene set collection . Cell Syst 1 , 417 – 425 ( 2015 ). OpenUrl PubMed 53. ↵ Godec , J. et al. Compendium of Immune Signatures Identifies Conserved and Species-Specific Biology in Response to Inflammation . Immunity 44 , 194 – 206 ( 2016 ). OpenUrl CrossRef PubMed 54. ↵ Paull , E.O. et al. A modular master regulator landscape controls cancer transcriptional identity . Cell 184 , 334 – 351 e20 ( 2021 ). OpenUrl CrossRef PubMed 55. ↵ Taguchi , T. et al. Conventional and molecular cytogenetic characterization of a new human cell line, GIST-T1, established from gastrointestinal stromal tumor . Lab Invest 82 , 663 – 5 ( 2002 ). OpenUrl CrossRef PubMed Web of Science 56. ↵ Bansal , M. et al. A community computational challenge to predict the activity of pairs of compounds . Nat Biotechnol 32 , 1213 – 22 ( 2014 ). OpenUrl CrossRef PubMed 57. ↵ Woo , J.H. et al. Elucidating Compound Mechanism of Action by Network Perturbation Analysis . Cell 162 , 441 – 451 ( 2015 ). OpenUrl CrossRef PubMed 58. ↵ Bush , E.C. et al. PLATE-Seq for genome-wide regulatory network analysis of high-throughput screens . Nat Commun 8 , 105 ( 2017 ). OpenUrl CrossRef PubMed 59. ↵ Li , Q. , Dai , W. , Liu , J. , Li , Y.X. & Li , Y.Y . DRAP: a toolbox for drug response analysis and visualization tailored for preclinical drug testing on patient-derived xenograft models . J Transl Med 17 , 39 ( 2019 ). OpenUrl CrossRef PubMed 60. ↵ Elseud , Y.A. , Shaaban , A. , Mohanty , A. & Albarrak , J . Safety and tolerability of regorafenib: a real-life experience . J Gastrointest Cancer ( 2021 ). 61. ↵ Dickinson , J. , de Matas , M. , Dickinson , P.A. & Mistry , H.B . Exploring a model-based analysis of patient derived xenograft studies in oncology drug development . PeerJ 9 , e10681 ( 2021 ). OpenUrl PubMed 62. Gould , S.E. , Junttila , M.R. & de Sauvage , F.J . Translational value of mouse models in oncology drug development . Nat Med 21 , 431 – 9 ( 2015 ). OpenUrl CrossRef PubMed 63. ↵ Nemati , F. et al. Patient Derived Xenografts (PDX) Models as an Avatar to Assess Personalized Therapy Options in Uveal Melanoma: A Feasibility Study . Curr Oncol 30 , 9090 – 9103 ( 2023 ). OpenUrl PubMed 64. ↵ Buchdunger , E. et al. Abl protein-tyrosine kinase inhibitor STI571 inhibits in vitro signal transduction mediated by c-kit and platelet-derived growth factor receptors . J Pharmacol Exp Ther 295 , 139 – 45 ( 2000 ). OpenUrl Abstract / FREE Full Text 65. ↵ Blay , J.Y. et al. A randomized study of 6 versus 3 years of adjuvant imatinib in patients with localized GIST at high risk of relapse . Ann Oncol 35 , 1157 – 1168 ( 2024 ). OpenUrl PubMed 66. ↵ Casali , P.G. et al. Ten-Year Progression-Free and Overall Survival in Patients With Unresectable or Metastatic GI Stromal Tumors: Long-Term Analysis of the European Organisation for Research and Treatment of Cancer, Italian Sarcoma Group, and Australasian Gastrointestinal Trials Group Intergroup Phase III Randomized Trial on Imatinib at Two Dose Levels . J Clin Oncol 35 , 1713 – 1720 ( 2017 ). OpenUrl CrossRef PubMed 67. ↵ Aissa , A.F. et al. Single-cell transcriptional changes associated with drug tolerance and response to combination therapies in cancer . Nat Commun 12 , 1628 ( 2021 ). OpenUrl CrossRef PubMed 68. ↵ Negishi , R. et al. Transcriptomic profiling of single circulating tumor cells provides insight into human metastatic gastric cancer . Commun Biol 5 , 20 ( 2022 ). OpenUrl PubMed 69. ↵ Obradovic , A. et al. Systematic elucidation and pharmacological targeting of tumor-infiltrating regulatory T cell master regulators . Cancer Cell 41 , 933 – 949 e11 ( 2023 ). OpenUrl CrossRef PubMed 70. ↵ Obradovic , A. et al. Single-cell protein activity analysis identifies recurrence-associated renal tumor macrophages . Cell 184 , 2988 – 3005 e16 ( 2021 ). OpenUrl CrossRef PubMed 71. ↵ Margolin , A.A. et al. ARACNE: an algorithm for the reconstruction of gene regulatory networks in a mammalian cellular context . BMC Bioinformatics 7 Suppl 1 , S7 ( 2006 ). OpenUrl CrossRef PubMed 72. ↵ Margolin , A.A. et al. Reverse engineering cellular networks . Nat Protoc 1 , 662 – 71 ( 2006 ). OpenUrl CrossRef PubMed 73. ↵ Zhang , Z. et al. Uniform genomic data analysis in the NCI Genomic Data Commons . Nat Commun 12 , 1226 ( 2021 ). OpenUrl CrossRef PubMed 74. ↵ Love , M.I. , Huber , W. & Anders , S . Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2 . Genome Biol 15 , 550 ( 2014 ). OpenUrl CrossRef PubMed 75. ↵ Ashburner , M. et al. Gene ontology: tool for the unification of biology. The Gene Ontology Consortium . Nat Genet 25 , 25 – 9 ( 2000 ). OpenUrl CrossRef PubMed Web of Science 76. ↵ Dobin , A. et al. STAR: ultrafast universal RNA-seq aligner . Bioinformatics 29 , 15 – 21 ( 2013 ). OpenUrl CrossRef PubMed Web of Science 77. ↵ Wishart , D.S. et al. DrugBank: a knowledgebase for drugs, drug actions and drug targets . Nucleic Acids Res 36 , D901 – 6 ( 2008 ). OpenUrl CrossRef PubMed Web of Science 78. ↵ FDA-approved & Passed Phase I Drug Library Contents . ( SelleckChem ). 79. ↵ Barretina , J. et al. The Cancer Cell Line Encyclopedia enables predictive modelling of anticancer drug sensitivity . Nature 483 , 603 – 7 ( 2012 ). OpenUrl CrossRef PubMed Web of Science 80. ↵ Haverty , P.M. et al. Reproducible pharmacogenomic profiling of cancer cell line panels . Nature 533 , 333 – 7 ( 2016 ). OpenUrl CrossRef PubMed 81. ↵ Oberg , A.L. et al. Statistical analysis of comparative tumor growth repeated measures experiments in the ovarian cancer patient derived xenograft (PDX) setting . Sci Rep 11 , 8076 ( 2021 ). OpenUrl PubMed View the discussion thread. Back to top Previous Next Posted October 14, 2025. Download PDF Supplementary Material Email Thank you for your interest in spreading the word about bioRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Systematic elucidation and pharmacologic targeting on non-oncogene dependencies in imatinib-resistant gastrointestinal stromal tumor Message Subject (Your Name) has forwarded a page to you from bioRxiv Message Body (Your Name) thought you would like to see this page from the bioRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Systematic elucidation and pharmacologic targeting on non-oncogene dependencies in imatinib-resistant gastrointestinal stromal tumor Prabhjot S. Mundi , Adina Grunn , Arsenije Kojadinovic , Charles Karan , Ronald Realubit , Cristina I. Caescu , Hanina Hibshoosh , Mahalaxmi Aburi , Mariano J. Alvarez , Matthew Ingham , Denisse Evans , Sara Rothschild , Gary K. Schwartz , Andrea Califano bioRxiv 2025.10.12.681609; doi: https://doi.org/10.1101/2025.10.12.681609 Share This Article: Copy Citation Tools Systematic elucidation and pharmacologic targeting on non-oncogene dependencies in imatinib-resistant gastrointestinal stromal tumor Prabhjot S. Mundi , Adina Grunn , Arsenije Kojadinovic , Charles Karan , Ronald Realubit , Cristina I. Caescu , Hanina Hibshoosh , Mahalaxmi Aburi , Mariano J. Alvarez , Matthew Ingham , Denisse Evans , Sara Rothschild , Gary K. Schwartz , Andrea Califano bioRxiv 2025.10.12.681609; doi: https://doi.org/10.1101/2025.10.12.681609 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Cancer Biology Subject Areas All Articles Animal Behavior and Cognition (7622) Biochemistry (17648) Bioengineering (13871) Bioinformatics (41880) Biophysics (21423) Cancer Biology (18558) Cell Biology (25460) Clinical Trials (138) Developmental Biology (13364) Ecology (19866) Epidemiology (2067) Evolutionary Biology (24290) Genetics (15589) Genomics (22475) Immunology (17711) Microbiology (40327) Molecular Biology (17145) Neuroscience (88473) Paleontology (666) Pathology (2827) Pharmacology and Toxicology (4816) Physiology (7635) Plant Biology (15114) Scientific Communication and Education (2044) Synthetic Biology (4286) Systems Biology (9815) Zoology (2268)