Cancer Biol Med 2021. doi: 10.20892/j.issn.2095-3941.2020.0560

ORIGINAL ARTICLE

Systematic screening reveals synergistic interactions that overcome MAPK inhibitor resistance in cancer cells

Yu Yu1*, Minzhen Tao2*, Libin Xu3, Lei Cao4, Baoyu Le5, Na An5, Jilin Dong5, Yajie Xu5, Baoxing Yang5, Wei Li5, Bing Liu5, Qiong Wu6, Yinying Lu7, Zhen Xie2, Xiaohua Lian1 1Department of Cell Biology, Basic Medical College, Army Medical University (Third Military Medical University), Chongqing 400038, China; 2MOE Key Laboratory of Bioinformatics and Bioinformatics Division, Center for Synthetic and System Biology, Department of Automation, Beijing National Research Center for Information Science and Technology, Tsinghua University, Beijing 100084, China; 3National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing 100021, China; 4Department of Thoracic Surgery, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences, Beijing 100730, China; 5Beijing Syngentech Co., Ltd, Beijing 102206, China; 6School of Life Sciences, Tsinghua University, Beijing 100084, China; 7The Comprehensive Liver Cancer Center, The 5th Medical Center of PLA General Hospital, Beijing 100039, China

ABSTRACT Objective: Effective adjuvant therapeutic strategies are urgently needed to overcome MAPK inhibitor (MAPKi) resistance, which is one of the most common forms of resistance that has emerged in many types of cancers. Here, we aimed to systematically identify the genetic interactions underlying MAPKi resistance, and to further investigate the mechanisms that produce the genetic interactions that generate synergistic MAPKi resistance. Methods: We conducted a comprehensive pair-wise sgRNA-based high-throughput screening assay to identify synergistic interactions that sensitized cancer cells to MAPKi, and validated 3 genetic combinations through competitive growth, cell viability, and spheroid formation assays. We next conducted Kaplan-Meier survival analysis based on The Cancer Genome Atlas database and conducted immunohistochemistry to determine the clinical relevance of these synergistic combinations. We also investigated the MAPKi resistance mechanisms of these validated synergistic combinations by using co-immunoprecipitation, Western blot, qRT- PCR, and immunofluorescence assays. Results: We constructed a systematic interaction network of MAPKi resistance and identified 3 novel synergistic combinations that effectively targeted MAPKi resistance (ITGB3 + IGF1R, ITGB3 + JNK, and HDGF + LGR5). We next analyzed their clinical relevance and the mechanisms by which they sensitized cancer cells to MAPKi exposure. Specifically, we discovered a novel protein complex, HDGF-LGR5, that adaptively responded to MAPKi to enhance cancer cell stemness, which was up- or downregulated by the inhibitors of ITGB3 + JNK or ITGB3 + IGF1R. Conclusions: Pair-wise sgRNA library screening provided systematic insights into elucidating MAPKi resistance in cancer cells. ITGB3- + IGF1R-targeting drugs (cilengitide + linsitinib) could be used as an effective therapy for suppressing the adaptive formation of the HDGF-LGR5 protein complex, which enhanced cancer stemness during MAPKi stress. KEYWORDS Pair-wise sgRNA library; genetic interactions; MAPKi resistance; combinatorial therapy; cancer stemness

Introduction *These authors contributed equally to this work. Correspondence to: Xiaohua Lian and Zhen Xie E-mail: [email protected] and [email protected] Aberrant activation of the mitogen-activated protein kinase ORCID ID: https://orcid.org/0000-0002-3682-2087 (MAPK)/extracellular signal-regulated kinase (ERK) path- and https://orcid.org/0000-0001-8798-9592 way is common in many cancer species. In melanomas, the Received September 16, 2020; accepted January 13, 2021. mutation of valine to glutamic acid at amino acid 600 in Available at www.cancerbiomed.org V600E ©2021 Cancer Biology & Medicine. Creative Commons BRAF (BRAF ) occurs in more than 50% of patients, Attribution-NonCommercial 4.0 International License and can boost kinase activity and constitutively promote 2 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance the activation of MEK/ERK signaling, which endows cancer mainly on MAPK signaling itself and its best known bypassing cells with increased proliferation, survival, and metastasis1-4. pathways, such as PI3K41. They therefore lack comprehensive Although MAPK inhibitors (MAPKi) such as vemurafenib insight into the cooperative mechanisms responsible for the (PLX4032), dabrafenib, and trametinib have been approved as occurrences of effective gene pairs. clinical drugs for late-stage tumors, most patients ultimately Elevated signals related to receptor tyrosine kinases develop lethal resistance to MAPKi treatment5. Understanding (RTKs)17, the EMT, and cancer stem cells (CSCs)42 have the molecular mechanism of MAPKi resistance is therefore emerged as 3 major causes of drug resistance in diverse essential to improving the efficacy of MAPKi treatment and species of cancer cells receiving various kinds of therapies. preventing disease relapse. However, most studies have focused on only 1 of these regu- To systematically analyze potential MAPKi resistance latory processes43,44. How these kinds of signals interact with mechanisms, we selected A375 melanoma cells with typical one another to generate adaptive resistance in refractory can- MAPK hyperactivation characteristics as our high-throughput cer cells remains poorly understood. In this study, we aimed screening model. We included regulators that influenced sev- to identify genetic interactions among 84 genes reported eral critical biological and cellular processes that have been to be involved in the RTK-, EMT-, and CSC-induced drug shown to be involved in bypassing the BRAFV600E inhibition in resistances of cancer cells, to identify potent combinatorial melanoma cells; for example, modulators of MAPK, GTPases, regimens that are universally applicable in sensitizing a wide and epigenetic modifications; regulators of cell proliferation variety of cancer cells. and cell death; facilitators of the epithelial-mesenchymal To identify the genetic interactions among RTK-, EMT-, transition (EMT); and factors that sustain cancer stem cell and CSC-relevant genes, we developed an assembly method maintenance6-27. Additionally, the STAGA and MEDIATOR to efficiently construct a library of more than 40,000 pair-wise complexes, which regulate transcription from RNA polymerase sgRNA combinations. We then performed pooled CRISPR/ II promoters, are essential for vemurafenib resistance, possibly Cas9 screening in cultured human A375 melanoma cells that due to their contributions to excessive metabolism or repres- contained the BRAFV600E mutation. Next, we evaluated the sion of cell apoptosis28-30. However, combinations of multiple genetic interaction of each gene pair by calculating the phe- resistance mechanisms may occur in melanoma cells, and fur- notypic value of vemurafenib resistance, validated 3 identified ther studies are needed to elucidate the functional relationships novel gene pairs with strong GI scores by using a cell growth between individual genes involved in MAPKi resistance31,32. competition assay and small molecule inhibition assay, and The functional relationships between 2 genes, also termed experimentally identified the underlying synergetic mech- genetic interaction (GI), can be evaluated by comparing the anisms of multidrug regimens. These results provided com- phenotypes of cells with simultaneous mutations in 2 genes prehensive information about potential critical connections to the phenotypes of cells with a single mutation33. Systematic between essential genes that may influence vemurafenib resist- identification of genetic interactions in mammalian cells is ance in melanoma cells. The methods developed in this study first facilitated by using pooled screening or large-scale auto- enabled the efficient assembly of pair-wise sgRNA libraries mated transfection, in which 2 small hairpin RNAs (shRNAs) and facilitated systematic analyses of genetic interaction net- are introduced into individual cells to simultaneously per- works of complex biological processes. turb 2 genes34,35. More recently, a multiplexed CRISPR/Cas9 To deliver an adjuvant sensitizing effect to MAPKi resist- screening approach has been developed by using a library of ance in an expanded panel of cancer cell lines with hyper- barcoded pair-wise single-guide RNA (sgRNA) combinations activated MAPK signaling, we also tested our adjuvant to search for potential growth-inhibiting gene pairs in cancer sensitizing pairs across cell lines within multiple cancer cells, which can facilitate the development of combinatorial species, including melanoma (MeWo), hepatocarcinoma cancer therapy36-39. Furthermore, concurrent suppression of (MHCC97H), lung adenocarcinoma (H1299), and breast BRAF, MEK, and ERK has been used to prevent the propaga- cancer (MCF7) cells. tion of cells with high level amplification of BRAF mutations,­ By analyzing clinical samples from melanoma patients with highlighting a practical strategy for overcoming vemurafenib different grades of stemness, we noted that HDGF and LGR5 resistance40. However, until now, these studies have been correlated with the expression of a commonly used CSC restricted to limited gene combinations, and have been focused marker, CD133, which was co-localized in the microvascular Cancer Biol Med Vol xx, No x Month 2021 3 regions of melanoma tissues. Next, using cell line mod- NCBI Biosample (https://www.ncbi.nlm.nih.gov/biosample). els, we found that the simultaneous suppression of HDGF The cell lines tested negative for mycoplasma contamination and LGR5 synergistically disrupted the formation of cancer and were regularly treated with mycoplasma removing agent. spheroids and downregulated the level of cancer stem cell TransStble3 and Trans5 a competent cells were purchased markers. Consistently, we found that they paired as a phys- from TransGen Biotech (Beijing, China). ical complex that allowed cancer cells to adapt to the vemu- rafenib/dabrafenib/trametinib environment. To the best of Reagents and enzymes our knowledge, this is the first study to identify and analyze a MAPKi stress-responsive protein complex that adaptively Q5 High-Fidelity DNA polymerase, T4 DNA ligase, and induced resistance. We also provided a detailed study of the restriction endonucleases of BsaI and Esp3I were pur- coordination of HDGF-LGR5 in achieving adaptive drug chased from New England Biolabs (Ipswich, MA, USA). resistance by sustaining EMT signals and CSC maintenance. Polyethylenimine (PEI) was purchased from Polysciences These observations suggested that HDGF and LGR5 acted in (Warrington, PA, USA). Polybrene and puromycin were pur- concert to affect the stemness and survival of BRAFi (BRAF chased from Sigma-Aldrich (St. Louis, MO, USA). Hoechst/ inhibitor)-treated cancer cells, which is worth developing as a propidium iodide (PI) double staining system, kanamycin, future therapy. and spectinomycin were obtained from Solarbio (Beijing, China). Mycoplasma removing agent was purchased from Materials and methods Solarbio. Vemurafenib, cilengitide, linsitinib, and JNK-IN-8 were acquired from Selleck (Houston, TX, USA). TRIzol Patients and SYBR Green PCR Master Mix were purchased from Beyotime Biotechnology and Thermal Fisher Scientific Melanoma patient specimens (No. 882775, No. 848557, No. (Waltham, MA, USA). Primary antibody to BIRC5 was pur- 842601, and No. 844583) used for immunohistochemistry chased from Solarbio. Primary antibodies to phospho-ERK- (IHC) and immunofluorescence (IF) staining were collected 1Thr202/Tyr204/ERK2Thr185/Tyr187, cleaved caspase-3, β-catenin, at the National Cancer Center of Chinese Academy of Medical phospho-FAK, BMI1, OCT4, nestin, p75 NGFR, CD133, Sciences and Peking Union Medical College, after informed β-actin, and secondary antibodies conjugated to horseradish written consent was obtained from each patient or each peroxidase were purchased from Beyotime Biotechnology. patient’s guardian according to the institutional guidelines Primary antibody to HDGF was purchased from Proteintech and the principles of the Declaration of Helsinki. Tumor (Rosemont, IL, USA) and primary antibody to LGR5 was tissue sections were stained with anti-ITGB3, anti-IGF1R, obtained from Sino Biological (Beijing, China). or anti-phospho-JNK antibodies for IHC analyses, and anti-CD133 + anti-BMI1, anti-HDGF + anti-LGR5 antibod- Efficiency test of 2 promoters in the pair-wise ies for IF analyses. sgRNA vector Cell culture The coding region of the fluorescence protein of mKate or EYFP was split into 2 fragments, and a designed linker that was The HEK293 (293-FT) cell line was purchased from Life recognized by the CRISPR/Cas9 system was added between Technologies (Carlsbad, CA, USA). The A375 cell line was the fragments (Supplementary Figure S1A and S1C). The obtained from the National Infrastructure of Cell Line promoters of U6 and 7SK were then assembled into the vec- Resource (Beijing, China), MeWo cells were purchased from tors, which are listed (the detailed sequence information of Fenghbio (Beijing, China), H1299 and MCF7 cells were sgRNA is shown in Supplementary Table S1). acquired from Procell Life Science & Technology (Waltham, MA, USA), and MHCC97H cells were purchased from U6-mKate_linker_sgRNA-7SK-negative_sgRNA-EF1α- Beyotime Biotechnology (Jiangsu, China). No cell line used in spCas9; this study was found in the database of commonly misiden- U6-negative_sgRNA-7SK-mKate_linker_sgRNA-EF1α- tified cell lines maintained by ICLAC (https://iclac.org/) and spCas9; 4 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance

U6-mKate_linker_sgRNA-7SK-EYFP_linker_sgRNA-EF1α- oligos were annealed and cloned into 7SK-pHS-BVC-LW002 spCas9; or U6-pHS-BVC-LW003 vectors by using BsaI. To generate the U6-EYFP_linker_sgRNA-7SK-mKate_linker_sgRNA-EF1α- pooled sgRNA vector sublibraries, 84 annealed sgRNA spacer spCas9; oligos in each pool were mixed with 12 annealed negative con- U6-negative_sgRNA-7SK-negative_sgRNA-EF1α-spCas9; trol sgRNA spacer oligos and the annealed S1:S2 spacer oli- gos (S1 sequence: 5′-ACCGATCCTCACGGTGAACGTCT-3′; Next, the PEI induced transfection assays were conducted S2 sequence: 5′-AAACAGACGTTCACCGTGAGGAT-3′) by combining the plasmid carrying expression fragments of with equal molar ratios and cloned them into the 7SK-pHS- mKate with either of the first 2 vectors as well as the negative BVC-LW002 or U6-pHS-BVC-LW003 vectors in a Golden Gate control vector V packaged by using PEI at the ratio of 1:3 for reaction by using BsaI, generating single-wise sgRNA libraries the single activation test. Additionally, the plasmids carrying (7SK-pHS-BVC-LW002-library and U6-pHS-BVC-LW003- expression fragments of mKate and EYFP were combined with library). The destination pair-wise sgRNA lentiviral vector vectors III, IV, or V above, and the same method was separately (pHS-BGL-YR001) and single-wise sgRNA libraries (7SK-pHS- used for double activation testing. These transfection mixtures BVC-LW002-library and U6-pHS-BVC-LW003-library) were were added into 24-well plates of 293FT cells using 3 experi- mixed at equal molar ratios in the second round of the Golden- mental replicates. Gate reaction by using Esp3I, resulting in a pair-wise sgRNA After 2 days, the cells were harvested and analyzed by fluo- library. The pair-wise sgRNA library plasmid DNAs were pre- rescence-activated cell sorting. Subsequently, the percentages pared by using TransStble3 competent cells, and then purified. of mKate in the experimental groups of vectors I and II or the percentages of mKate and EYFP in the experimental groups Lentivirus production and library screening of vectors III and IV were divided by the corresponding per- centage of leaky fluorescence in the negative control groups of Pooled lentiviral library plasmid DNA was a mixture with vector V, which was then defined as the relative fluorescence equal molar ratios of 6 pair-wise sgRNA libraries, 3 single-wise expression level. Finally, we used the 2-tailed unpaired t-test to sgRNA libraries, and a negative control library. Pooled lenti- evaluate the statistical significance of the relative fluorescence viruses were produced and packaged as described with minor activated by different vectors with adversely located promoters modifications. Briefly, 20 µg of pooled lentiviral library plas- of U6 and 7SK. mid DNA, 20 µg of pCMV-dR8.2-dvpr, and 15 µg of pCMV- VSV-G were transfected into HEK293FT cells in a 15 cm petri Construction of plasmid DNA and sgRNA dish by using PEI transfection reagents. Approximately 1 × 108 libraries Cas9-A375 cells were infected with pooled sgRNA libraries at 0.3 MOI to maintain library coverage at approximately 1,000- The vectors used in this study were constructed using standard fold. One day after infection, the cells were selected with 1 mg/mL molecular cloning techniques, including restriction enzyme puromycin prior to further experiments. Selected A375 cells digestion, ligation, PCR, and Golden Gate Assembly. Custom were then treated with vemurafenib or dimethyl sulfoxide oligonucleotides were purchased from Genewiz (Beijing, (DMSO) for 10 days. The A375 genomic DNA was extracted China). The vector constructs were transformed into E. coli from harvested A375 cells. Approximately 569 bp DNA frag- strain, Trans5a, and 50 µg/mL of kanamycin/spectinomycin ments containing the pair-wise sgRNA region were amplified was used to isolate colonies harboring the constructs. DNA by PCR with 25 thermal cycles from A375 genomic DNA was then extracted and purified. Sequences of the vector con- using primers (5′-ATTTGTCTCGAGGTCGAGAATTC-3′ structs were verified by DNA sequencing. and 5′-TCATATGCTTACCGTAACTTG-3′) using Q5 high- We synthesized a control oligo pool for constructing 12 fidelity DNA polymerase. The DNA fragments were used to negative control sgRNAs, and 3 oligo pools for constructing construct DNA libraries (BGI, Yantien, China) by combining sgRNAs. Each oligo pool included 84 sgRNA spacers targeting the fragmented DNA and end repair mix (BGI), then purify- 84 genes. To construct the sgRNA vector containing the U6 pro- ing them before conducting the A-tailing reaction to obtain moter (U6p) or 7SK promotor (7SKp)-driven expression of a the product of the adenylate 3′-end DNA. PCR free index sgRNA that targeted a specific gene, the 20 bp sgRNA spacer adapters, (BGI) and the T4 DNA ligase mix, were added to Cancer Biol Med Vol xx, No x Month 2021 5 the A-tailing product for ligation of adapters. Finally, adapter- targeting the same gene or gene pair and excluded the minimal ligated DNA samples were purified for deep sequencing using and maximal scores to obtain the quantitative growth pheno- HiSeq 2500 (Illumina, San Diego, CA, USA) with the 250 type for the corresponding single- or pair-wise perturbation. nucleotide paired-end mode. The value of P for a given gene pair was calculated by compar- ing the distribution of r scores targeting the pair of interest to Sequencing data processing the negative control using the Mann-Whitney U test. To calculate the genetic interaction strength from the phe- The quality of deep sequencing data was first evaluated by notypic r score of high-throughput assays, we defined GI using FastQC [Babraham Bioinformatics (https://www.bio- scores as the deviation of the observed phenotype (robserved) informatics.babraham.ac.uk/projects/fastqc/)]. The paired- from the expected phenotype (rexpected) calculated by linear end sequencing reads were mapped to designed sequences of regression analysis using the R package (The R Project for sgRNA expressing library constructs, where an exact match was Statistical Computing, Vienna, Austria). The GI and r scores required in the seed region of sgRNA, while two mismatches below the 10th percentile and above the 90th percentile were were allowed elsewhere. The paired-end reads that contained selected to construct the edge-weighted hierarchical network. duplicates of 2 identical promoters and harbored dislocated Network hubs were identified by using Eigenvector. sgRNAs were eliminated in the subsequent analysis. The num- To define the synergistic or antagonistic effect of gene com- bers of paired-end reads for each unique pair-wise sgRNA binations, the genetic interaction strength was calculated using combination in a given sample were normalized as follows: Loewe’s additivity principle (GIadditivity = |pair-wise pheno- typic r score|-|single-wise phenotypic r score 1 + single-wise Normalized reads perc ombination phenotypic r score 2|)45. Readsp er combination = Total read sforal lcombinationsi nt he sample ××1.61007 + .05 Kaplan-Meier survival curve plots where 1.6 × 107 represented the maximal total reads among all The BRAF mutant cohort from The Cancer Genome Atlas samples, and 0.05 was added to facilitate calculating pheno- (TCGA) database was selected, and the RNA expression lev- typic values in a logarithm scale. els of NF2, ITGB3, and IGF1R from the patients with mutated To quantify the growth phenotypes of Cas9-A375 cells with BRAF were discretized into 5 classes as “higher”, “high”, different sgRNA perturbations, we defined the r score, which “medium”, “low”, or “lower” using the BurStMisc::ntile() func- was adapted from a previous study34 as follows: tion. The lower expressions of “ITGB3” and “IGF1R” in the PLXPLX 1 NNAW/ T figure were defined as patients with ITGB3 and IGF1R expres- ρA =×log2 DMSO DMSO kt NNAW/ T sions below the “low” level. The high expression of NF2 in the figure was defined as patients with NF2 expressions above the  PLXPLX  1 NNBW/ T “medium” level. ρB =×log2  DMSO DMSO  kt  NNBW/ T  Cell growth competition assays PLXPLX 1 NNAB / WT ρAB =×log,2 DMSO DMSO kt NNAB / WT Cas9-A375 cell lines with sgRNA perturbations of either the single or double gene of 3 gene pairs (ITGB3 + IGF1R, where k was the cell doubling difference of wild-type cells and NF2 + CCNC) were made using the lentivirus-mediated­ cultured between vemurafenib and DMSO, and t denoted the integration system as previously mentioned. The sgRNA cultured time duration. NX was the normalized read count of sequences used are listed in Supplementary Table S2. For sgRNA combinations targeting single genes (A or B) and gene each gene pair, 2 × 104 cells of individual cell lines for a sin- pairs (AB), while NWT represented the median read counts of gle- or pair-wise gene perturbation were mixed and cultured the negative pair-wise sgRNA control in DMSO or the vemu- in DMEM media plus 10% fetal bovine serum (FBS). One day rafenib sample. After calculating the r score for each pair-wise after cell seeding, the cells were split into 2 halves and treated sgRNA combination, we averaged the sgRNA combinations with 2 µM vemurafenib or DMSO in equal volumes for up 6 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance to 12 days. FACS analysis was conducted using LSRII Fortessa Table 1 The primer sequences for qPCR amplification of (Becton Dickinson, Franklin Lakes, NJ, USA) as previously E-cadherin, vimentin and β-actin described46. At least 2 × 104 cell events were acquired per sam- Primer name Sequence ple. FACS data were analyzed using FlowJo software (https:// E-cadherin-F 5’-AAAGGCCCATTTCCTAAAAACCT-3’ www.flowjo.com/). E-cadherin-R 5’-TGCGTTCTCTATCCAGAGGCT-3’

Small molecule inhibition assays Vimentin-F 5’-GACGCCATCAACACCGAGTT-3’ Vimentin-R 5’-CTTTGTCGTTGGTTAGCTGGT-3’

Small molecule inhibitors (cilengitide for ITGB3, linsitinib for β-actin-F 5’-CATGTACGTTGCTATCCAGGC-3’

IGF1R, and JNK-IN-8) were used in the assays. Approximately β-actin-R 5’-CTCCTTAATGTCACGCACGAT-3’ 1 × 104 A375 cells were seeded in a 96-well plate 1 day before the inhibition assay with 2 drug combinations of cilengitide Western blot assays and linsitinib, or cilengitide and JNK-IN-8. For each combi- natorial drug treatment, A375 cells were treated with either Cell lysates were extracted by using RIPA lysis buffer (Beyotime 2 μM vemurafenib or DMSO in equal volumes, plus individ- Biotechnology), and the protein concentrations of different ual drugs or drug combinations for 2 days. To label live and samples were determined using the BCA Protein Assay Kit dead cells after drug treatments, 100 µL of Hoechst 33342/PI (Beyotime Biotechnology). A total of 25 μg of each protein sam- mixed solution was added into each well of the 96-well plate ple was separated by SDS-PAGE and transferred to 0.45 μm pol- and incubated at 4 °C for 20 min. After washing with 100 µL yvinylidene difluoride membranes (Beyotime Biotechnology). phosphate-buffered saline (PBS), A375 cells were measured using After blocking for 1 h and 30 min, the membranes were treated the high content cellular screening system using UV/488 nm with primary antibodies to cleaved caspase-3, phospho- dual excitation and emission wavelengths at 460 nm for ERK1Thr202/Tyr204/ERK2Thr185/Tyr187, BIRC5, β-catenin, BMI1, Hoechst 33342 and 575 nm for PI. The number of live cells CD133, OCT4, nestin, or NGFR overnight at 4 °C. The mem- (referred to as N) was calculated as follows: N = total cell branes were then washed with TBST (Solarbio) and incubated counts - high blue fluorescence counts - high red fluorescence with horseradish peroxidase-conjugated­ secondary antibodies counts. The r scores were calculated using the formula as pre- (Beyotime Biotechnology) for 1 h at 25 °C. The protein bands viously described. were detected using a chemiluminescence detection kit (Beyotime Biotechnology), and the optical densities of protein signals were Quantitative RT-PCR calculated using ImageJ software (National Institutes of Health, Bethesda, MD, USA), with β-actin as the loading control. Total RNAs were extracted from A375 cells using TRIzol (Invitrogen, Carlsbad, CA, USA), and the amounts of Proliferation assays RNA from different samples were measured using a Nanodrop2000 (Thermo Fisher Scientific). Then, 1.5 μg Cell proliferation inhibition was assayed using the Cell RNA of the samples was diluted into a whole RNA amplifi- Counting Kit-8 (Dojindo Molecular Technologies, Rockville, cation mixture containing oligo(dT) and dNTPs. A reverse MD, USA) on an iMark™ Microplate Absorbance Reader transcription mixture containing reverse transcriptase, (Bio-Rad, Hercules, CA, USA). Cells (1 × 105 cells/mL) were dithiothreitol, and first strand buffer were added to the plated and treated for 72 h in triplicate for each experimental amplified RNA mixture to synthesize cDNA. Next, SYBR condition. Each well was incubated with 10 µL of CCK8 rea- Green was added with primers for the E-cadherin, ­vimentin, gent in 100 µL of cell culture medium, then the 96-well plate and β-actin loading control into the cDNA template (at a was incubated for 1 h at 37 °C in a humidified atmosphere final dilution of 1:10). Finally, the expression-fold change containing 5% CO2. The absorbances were then measured at was calculated using 2−ΔΔCT, where ΔΔCT = ΔCT (gene of 450 nm, which were corrected relative to blank wells contain- interest) - ΔCT(loading control) (Table 1). ing CCK8 reagent and medium only. Cancer Biol Med Vol xx, No x Month 2021 7

Co-IP assays cells/mL in F12K/DMEM basic culture medium with 20 µg/mL EGF, 20 µg/mL bFGF, 5 µg/mL LIF, 10 µg/mL insulin, 1% The cells were cultured to 80%–90% confluence, and treated N2, and 1% B27. After 24 h, the siRNAs of HDGF and LGR5 with drug regimens consisting of vemurafenib, dabrafenib, were transfected into A375, H1299, and MHCC97H cells vemurafenib + trametinib, dabrafenib + trametinib, or DMSO. according to the treatment groups, and after another 24 h, 2 After a 72 h incubation with the small molecular weight drugs, µM vemurafenib and 2 µM vemurafenib + 10 nM trametinib the cell lysates were diluted with lysis buffer (NP-40, supple- were added to the A375 cells, and 5 µM dabrafenib and 5 mented with protease inhibitors and phosphatase inhibitors), µM dabrafenib + 100 nM trametinib were added to the and incubated with 10 µg polyclonal rabbit anti-LGR5 (Sino H1299 and MHCC97H cells Then, spheroid images were Biological, Beijing, China) or 10 µg monoclonal mouse anti- captured by using a 4× microscope objective, every 2 days HDGF (Proteintech), for 10–14 h at 4 °C with end-over-end until the 7th day. The spheroid formation efficiency was eval- mixing. Protein A/G-agarose beads (Beyotime Biotechnology) uated by AnaSP and ReViSP, which are MATLAB programs were then added and the reaction mixture was further mixed that can calculate multiple morphological parameters of for 2 h at 4 °C. The immunoprecipitates were separated from ­spheroids47, with the reconstruction of three-dimensional the supernatants by centrifugation and washed with PBS con- (3D) images, which reflected the stereoscopic morphology of taining 1% NP-40. The proteins were extracted from the aga- the spheroids48,49. rose beads by boiling in 1× SDS gel loading buffer and resolved using 10% SDS-PAGE. Immunofluorescence

A total of 4 × 104 cells were initially plated in cover glasses Co-inhibition efficiency tests of siHDGF and that had been set in the bottom of a 24-well plate, and after siLGR5 24 h, the siRNAs of HDGF and LGR5 were transfected into cells according to the treatment groups. After another 48 h, A total of 800 µL cells were initially plated in 12-well plates 2 µM vemurafenib or DMSO was added to the cells. After (Corning, Corning, NY, USA) at a density of 2 × 105 cells/mL 72 h of drug treatment, the cells were fixed in 4% formal- for each treatment with the RPMI 1640 + 10% FBS culture dehyde (Beyotime Biotechnology) in 1× TBS for 10 min at medium. Then, 100 pmol siRNA of HDGF and LGR5, or a room temperature, permeabilized in 0.3% TritonX-100- scrambled negative control siRNA, were co-transfected with TBS or Saponin-TBS for 10 min, washed in 0.1% Tween20- Lipofectamine 3000™ into A375, H1299, or MHCC97H cells TBS (TBST) and incubated in blocking solution (Beyotime when the cell confluence reached 70%–80%. The total RNA of Biotechnology) for 1 h. The cells were then treated with pri- samples was extracted on the 3rd day after the siRNA trans- mary antibodies overnight in TBST with 10% FBS at 4 °C, fection. The total RNA was reverse-transcribed into cDNA washed in TBST, and treated with secondary antibodies at using the PrimeScript™ RT Master Mix (Takara, Shiga, Japan). 37 °C for 1 h in TBST. After washing, the cells were exam- Then, approximately 1 µg cDNA of each sample was added to ined using an A1 confocal microscope (Nikon, Tokyo, Japan). the 2× Taq PCR Master Mix (Syngentech, Beijing, China) with Primary antibodies were anti-mouse-SOX2 (Sino Biological), corresponding primers to amplify HDGF, LGR5, and β-actin anti-rabbit-KLF4 (Beyotime Biotechnology), anti-rabbit- sequences. HDGF, LGR5, and β-actin primer sequences were Phospho-FAK (Beyotime Biotechnology), and anti-nestin obtained from PrimerBank (https://pga.mgh.harvard.edu/ (Beyotime Biotechnology). primerbank), where the PrimerBank IDs of HDGF, LGR5, and -actin were 186928818c2, 24475886c1, and 4501885a1, β Ethical approval respectively.

Written informed consent was obtained from all patients, and Spheroid formation assays the study protocol was approved by the Ethics Committee of the National Cancer Center/Center Hospital, Chinese A total of 100 µL of cells were initially plated in an ultra-low Academy of Medical Sciences and Peking Union Medical attachment 96-well plate (Corning) at a density of 2 × 105 College (Approval No. NCC20190-008). 8 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance

Results cell populations were quantified by paired-end deep sequenc- ing of 2 variable sgRNA spacer sequences. The abundances Construction of the pair-wise sgRNA library were used as a proxy for cell fitness under different treatments and the strategy for screening experiments (Figure 1A). In addition, over 92% of sgRNA combinations were recovered in samples of the plasmid library, and cells from We selected 84 genes based on previous CRISPR/Cas9 ­single replicates 1 or 2 in the baseline day (Supplementary Figure gene knockout and transcriptional activation screening S2C). We showed that the abundances of individual pair-wise ­studies13,21, and other mechanistic reports concerning the sin- sgRNA combinations in plasmid samples were highly corre- gle key genes involved in MAPKi resistance (Supplementary lated with those in either T0 sample, with Pearson’s correla- Table S3)7,11,12,14-16,18,19,24,27. These genes function in dif- tion coefficient > 0.75 (Supplementary Figure S2D and S2E), ferent biological processes, including basic transcription and the plasmid frequencies of the 3 samples were also evenly regulation, mitotic progression, cell death regulation, reg- distributed (Supplementary Figure S1). ulation of GTPase activity, chromatin covalent modifica- tion, and the transmembrane receptor protein tyrosine Phenotypic evaluation of the pair-wise sgRNA kinase signaling pathway (Supplementary Figure S2A). We library then used CRISPR-ERA to design sgRNAs that targeted the genes of interest, as well as a set of negative control sgRNAs To quantify the growth phenotype of Cas9-A375 cells with (Supplementary Table S4)50. different pair-wise sgRNA perturbations in response to To build the pair-wise sgRNA library, the synthesized and vemurafenib treatment, we defined the r score, which was annealed sgRNA oligos were grouped into a negative con- adapted from previous studies (see details in the Materials and trol sgRNA pool and 3 positive sgRNA pools. The 3 positive Methods)34. When mutant Cas9-A375 cells showed resistance sgRNA pools contained 3 sets of sgRNAs targeting 84 single to vemurafenib treatment compared to wild-type cells, the r genes (Supplementary Figure S2B). Each sgRNA pool was score was greater than 0. In contrast, a r score below 0 indi- separately cloned into intermediate vectors with either the cated that cells were more sensitive to vemurafenib treatment human U6 or 7SK promoter using the Golden Gate clon- than wild-type cells. From the testing experiments above, we ing method, to generate single sgRNA libraries with ~100× found that sgRNA driven by either U6 or 7SK could equally transformation coverage (Supplementary Figure S2B). To cause mutations in a given target gene (Supplementary link the single-wise sgRNA libraries into pair-wise sgRNA Figure S3). We used the averaged r scores of pair-wise sgRNA libraries, 2 single sgRNA libraries were simultaneously cloned combinations targeting the same gene or gene pair after into the lentiviral vector by using a second Golden Gate reac- excluding the minimal and maximal scores as the r score of tion, to generate sublibraries with 10× transformation cov- the corresponding single gene or gene pair perturbation for erage (Supplementary Figure S2B). All sublibraries were further analyses. To identify strong growth phenotypes, we mixed in equimolar amounts to generate the total pair-wise analyzed the distribution of obtained r scores and arbitrar- sgRNA library that contained more than 40,000 pair-wise ily set the 10th and 90th percentiles as thresholds, selecting 68 sgRNA combinations targeting zero (negative control), gene pairs with strong sensitizing phenotypes and 159 gene 1 (single-wise), or 2 genes (pair-wise) among the 84 selected pairs with strong protective phenotypes, respectively, for fur- genes (Supplementary Figure S2B). Cas9-expressing A375 ther analysis (Supplementary Figure S4A and Supplementary cells (Cas9-A375) were infected with a low titer of the pair- Table S5). In addition, we found that r scores below the 10th wise sgRNA library to increase the likelihood of single copy percentile threshold and above the 90th percentile threshold lentiviral integration into each genome (Figure 1A). After were well correlated in 2 screening replicates, with a Pearson’s puromycin selection to ensure sufficient lentiviral integra- correlation coefficient of 0.949 (Supplementary Figure S4B). tion and Cas9 knockout, infected A375 cells were treated with We then constructed a network of identified gene pairs with vemurafenib or DMSO for 10 days (Figure 1A). To assess the strong r scores. We found 8 genes as the top 10% strong hubs, reproducibility of our high-throughput method, we inde- including 6 genes involved in the protective phenotype and 2 pendently performed the screening experiment twice. Then, genes involved in the sensitizing phenotype (Supplementary the abundances of pair-wise sgRNA combinations in different Figure S4C). Cancer Biol Med Vol xx, No x Month 2021 9

A Pairwise sgRNA pladmid library Pairwise sgRNA lentivirus library

Lentiviral delivery of pairwise sgRNA Virus packaging Lentivirus infection lentivirus library

Engineered Cas9-A375 cell line Vemurafenib

Vemurafenib

Puromycin Readcount selection 10 days Deep sequencing Pairwise SgRNA species Comparative analysis DMSO

DMSO Readcount

Pairwise SgRNA species B C

Data ≥90th percentile Replicate 1 Data between 10th & 90th percentile Data ≤10th percentile CCNC+NF2 Strongly interacted Weakly interacted Strongly interacted region region region HDGF+LGR5 0.5 ITGB3+Jun 2.0 IGF1R+ITGB3 CCNC+NF2 Undefined data 1.5 HDGF+LGR5 ITGB3+Jun 0.0 1.0

Density (a.u.) 0.5 IGF1R+ITGB3 –0.5 0.0 10th 50th 90th

Gl score of replicate 2 (a.u.) PCC (data ≥90th & ≤10th percentile) = 0.924 Percentile of Gl scores (a.u.) –1.0 PCC (defined data) = 0.808 –0.5 0.0 0.5 GI score of replicate 1 (a.u.)

D Hub node Hit node Hub/Hit node Top node

GBP2 PRKRARAP1GAP2ITGB3

BUB1B YAP1 HDAC6

IGFIR TCEB2 MAP3K11 METTL21B HMMR TNFRSF1B Gl score (a.u.) PODN ZNF182 SPRED1 RBL1 TAF6L RHOG TNRC18 NF2 0.559 ARHGAP6 MECOM PDCD10 VAX1 UBE2M Jun CRB2 TFAP2C EED PRKCE RANBP9 PLK1 NF1 CHN2 0 MED12LAMA3 CCNC PRPF4B EGER MAPKAPK3 RAPGEF1 LPAR5 HOXC13 MED 15 PARP3 –0.678 CUL3 GPR98 EAPP NGFR GPR35 KRAS P2RY8 EZH2 C9orf50 BAP1 MRFAP1L1 APBB1 DCLK1 PYG01 ARHGEF5 ARHGEF2 ADI 1 EphA2 LGR5

EPAS1 HDGF

Figure 1 Screening combinatorial gene hits by using a pair-wise sgRNA library. (A) Functional screening using the pair-wise sgRNA library of the Cas9-A375 cell line. Cas9-A375 cells were infected with the pooled pair-wise sgRNA library and treated with vemurafenib or dimethyl sulfoxide. The abundances of pair-wise sgRNA combinations after different treatments were analyzed by paired-end deep sequencing. (B) The histogram distribution of averaged GI scores derived from replicate 1. GI scores below the 10th percentile or above the 90th percentile were recognized as strong interactions. (C) Correlation of GI scores of 2 independent replicates. Blue and pink dots represent the top and bottom 10% of reproducible GI scores; black dots indicate the middle 80% of reproducible GI scores; and gray dots denote undefined GI scores in 2 independent replicates. Four selected gene pairs are shown in red, green, blue, and orange. (D) The hierarchical network representation of the top listed GI score pairs. The thickness of edges indicates the GI strength. Blue and red designate the sensitizing and protective phenotypes, respectively. Purple nodes inside the yellow rectangle denote hub gene nodes. Blue nodes represent the top gene nodes. Yellow nodes were selected for experimental validation, as shown in subsequent figures. 10 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance

Calculation of GI scores from phenotypic scores LGR5) fell outside the linear fit with a 99% confidence inter- val (Figure 2A–2C, Supplementary Figure S6A and S6B, S6D To effectively analyze the genetic interaction strength, we per- and S6E, and S6G and S6H). Subsequently, we compared the formed a linear regression analysis by using the r scores of 84 normalized read counts of pair-wise sgRNA perturbations to single gene perturbations versus the r scores of gene pair per- the normalized read counts of single sgRNA perturbations turbations that shared a common bait gene (Supplementary and negative control sgRNAs (Figure 2C and Supplementary Figure S5A). This strategy for calculating GI scores was based Figure S6C, S6F, and S6I). We confirmed that pair-wise genetic on the assumption that strong GIs are usually rare in large perturbations of ITGB3 + IGF1R and ITGB3 + Jun resulted gene networks38. The linear fit represented the expected phe- in strong sensitizing phenotypes, and perturbation of NF2 + notype (expected) of gene pair perturbation, and GI scores of CCNC caused a strong protective phenotype (Figure 2C and gene pair perturbations were calculated as the deviation from Supplementary Figure S6C, S6F, and S6I). the expected phenotype to the observed phenotype (observed) For gene pairs of ITGB3 + IGF1R and NF2 + CCNC, we (Supplementary Figure S5A). We observed a large number of engineered 3 Cas9-A375 cell lines that expressed either the gene pair perturbations resulting in small GI scores outside EBFP or EYFP fluorescent protein along with a single sgRNA the linear fit with a 99% confidence interval, suggesting that targeting individual genes or expressed 2 sgRNAs and the further selection criteria may be needed to choose strong GI mKate fluorescent protein. The 3 engineered Cas9-A375 cell gene pairs (Supplementary Figure S5A). lines with either single or pair-wise gene perturbation(s) were To select strong and reliable GIs, we also analyzed the over- mixed with Cas9-A375 cells in 4 equal proportions and all distribution of obtained GI scores. Similarly, we defined cultured with vemurafenib or DMSO for up to 12 days the 10th and 90th percentiles of the GI score distributions as (Figure 2D). By using flow cytometry analysis, we confirmed arbitrary thresholds for selecting strong GIs (Supplementary that perturbations of ITGB3 + IGF1R gene pairs generated a Figure S5B). It has been reported that weak genetic interac- sensitizing growth phenotype (Figure 2E and Supplementary tions often produce a low correlation between GI scores derived Figure S7A) and that perturbation of the NF2 + CCNC gene from different screening replicates45. Thus, we analyzed the pair caused a protective growth phenotype (Figure 2F and association of GI scores between 2 independent screening rep- Supplementary Figure S7B). Next, we assayed the growth licates. We found that 59 gene pairs below the 10th percentile phenotypes of A375 cells after treatment with clinically threshold and 43 gene pairs above the 90th percentile threshold approved small molecular weight inhibitors, including cilen- were highly consistent between the 2 replicates, with an over- gitide for ITGB3, linsitinib for IGF1R, and JNK-IN-8 for the all Pearson’s correlation coefficient of 0.924 (Supplementary downstream effector of Jun-c-Jun N-terminal kinase 1 (JNK). Figure S5C and Supplementary Table S6). In contrast, weak Consistent with our previous findings, when A375 cells were GIs were less correlated between 2 replicates compared to the treated with a combinatorial treatment of varying amounts of strong GIs. To assay the relationships among these identified cilengitide and linsitinib, a significant sensitizing growth phe- genes and thus identify important biomarkers51, we constructed notype and a strong synergistic effect were observed (Figure a GI network by using the strong GI gene pairs (Figure 1D). 2G). Additionally, because Jun lacks an inhibitor at the pro- Seven genes, including ITGB3, GBP2, BUB1B, PRKRA, YAP1, tein level, the small molecule JNK-IN-8 was used to inhibit RAP1GAP2, and HDAC6, were identified as hubs with top JNK, which is the upstream regulator of Jun52. We confirmed ranked dense connectivities with other gene nodes (Figure 1D). that co-targeting JNK and ITGB3 also cooperatively sensitized Notably, only RAP1GAP2 was recognized as the hub node in A375 cells to vemurafenib (Figure 2H). We subsequently asked both networks constructed with strong GI scores and strong r whether these combinatorial hits were also potentially applica- scores (Figure 1D and Supplementary Figure S4C). ble in other cancer types and whether the use of vemurafenib + trametinib or dabrafenib + trametinib plus the combina- Phenotypic confirmation of candidate genetic torial adjuvant hits would achieve a superior effect compared interaction pairs with the sole use of vemurafenib or dabrafenib. We thus tested cell viabilities by using the combinatorial drug regimens of We confirmed that the GI scores of all 4 selected gene pairs cilengitide + linsitinib and cilengitide + JNK-IN-8 across mul- (ITGB3 + IGF1R, ITGB3 + Jun, NF2 + CCNC, and HDGF + tiple cancer cell lines, including melanoma (MeWo), head and Cancer Biol Med Vol xx, No x Month 2021 11

A BC P = 3.451 × 10–3 Negative control 0.7 Cl = 99% 0.7 Cl = 99% 15 IGF1R 0.5 0.5 Gl score Gl score ITGB3 10 0.3 0.60 0.3 IGF1R + ITGB3 0.50 0.1 0.30 0.1 5 –0.1 0.00 –0.1 0.00 –0.30

–0.3 with ITGB 3 –0.3 –0.50 with IGF1 R 0 (normalized readcount –0.5 2 –0.5 in vemurafenib group) Lo g

–0.7 IGF1R + ITGB3 Phenotype of double-gene –0.7 –5 Phenotype of double-gene ITGB3 + IGF1R –0.9 –0.9 –0.5 00.5 –0.50 0.5 –5 0510 15 Log (normalized readcount Phenotype of individual gene ( ρ) Phenotype of individual gene ( ρ) 2 in DMSO group)

D Vemurafenib Perturbation of gene I

Vemurafenib 12 days

Pairwise sgRNA constructs

Perturbation of gene II Abundancy U6 Neg 7SK Gene I hEF1α BFP Pairwise sgRNA species Lentivirus infection FACS analysis U6 Neg 7SK Gene II hEF1α EYFP 21 days Comparative Perturbation of gene I & II DMSO U6 Gene I 7SK Gene II analysis hEF1α mKate 1 day

U6 NegN7SK eg

Negative control DMSO 12 days Abundancy

Pairwise sgRNA species

E G ρ score Linsitinib (nM) –0.20 **

–0.25 05070 ** 0.2

–0.30 0.0 0 –0.2 ρ scor e –0.35 Protecting –0.4

0.70 ρ scor e 0.4 –0.6 –0.40 Cilengitide ( µ M) –0.8 0 –0.45 –0.0 0.5

IGF1R +– – –0.70 Linsitinib (70 nM)– ++ ITGB3 – + – Cilengitide (0.4 µM) +–+ Sensitizing IGF1R + ITGB3 – – +

F H

1.5 ρ score cilengitide (µM) * * ** 1.0 00.4 0.6

–0.5 0.5 0 scor e

ρ –1.0 Protecting 1.87 0.0 –1.5 60 ρ scor e –2.0

JNK-IN-8 (nM) 0 –0.5 –2.5 90 NF2 +– – –1.87 JNK-IN-8 (90 nM)– ++ CCNC – + – Sensitizing Cilengitide (0.4 µM) ++– NF2 + CCNC – – +

Figure 2 Combinatorial growth competition and small molecule intervention assays. (A, B) Scatter plots of r scores of pair-wise sgRNA per- turbations against a common bait gene and single sgRNA perturbations. GI scores were calculated as the deviation from the observed r score to a linearly fitted value (black line). The dotted line denotes the 99% confidence interval. The color bar represents GI scores from negative (blue) to positive (red) effects. (C) Scatter plots of read count abundances of single and pair-wise sgRNA species in vemurafenib- and dimethyl sulfoxide-treated samples. Red dots, pair-wise sgRNA species; green and blue, single sgRNA species; and gray dots, negative control sgRNA. Linear fit lines for pair-wise (red), single (green, blue), and negative control (gray) sgRNA species are shown. Data were tested for significance using the 2-tailed Mann-Whitney U test. The P-value indicates the results of Mann-Whitney U tests comparing the distributions of r scores of the gene pair perturbation to those of the negative control. (D) Schematic illustration of the competitive growth assay. (E and F) Results of the 12 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance

combinatorial growth competition assay. The r scores were calculated by using the relative abundances of individual Cas9-A375 derivatives (see details in the Materials and Methods). The data are the mean ± s.d. (N = 3) from 3 independent replicates. “*” designates significant difference (P < 0.01) between single and pair-wise sgRNA perturbations tested by unpaired 2-tailed Student’s t-tests. (G, H) Cilengitide, lins- itinib, and JNK-IN-8 are small molecular weight inhibitors of ITGB3, IGF1R, and JNK, respectively. A375 cells were treated with the indicated amounts of inhibitors. Vemurafenib resistance strength was calculated based on the number of live cells from different treatment groups. Data points are the mean ± s.d. (N = 4) from 3 independent replicates. “*” designates significant difference (P < 0.01) between single drug and combinatorial drug treatments tested by unpaired 2-tailed Student’s t-tests. neck squamous carcinoma (FaDu), and breast cancer (MCF7) Western blot, we found that linsitinib was essential for cells. The cell viability results showed that the melanoma repressing the expression of phospho-ERK1/2 (Figure 3A), MeWo cell line was sensitized by all 3 combinations, except suggesting that IGF1R rather than ITGB3 might be the main cilengitide + JNK-IN-8 paired with vemurafenib + trametinib relay for the reactivation of ERK. In addition, vemurafenib (Supplementary Figure S8). In contrast, FaDu was sensitized blocked the activity of BRAFV600E. Notably, vemurafenib and only by cilengitide + JNK-IN-8 paired with dabrafenib alone linsitinib cooperatively inhibited the promotion of β-catenin or dabrafenib + trametinib. However, in the breast cancer cell by cilengitide (Figure 3B). Furthermore, cilengitide and line MCF7, only cilengitide + JNK-IN-8 paired with dabrafenib vemurafenib synergistically enhanced the level of cleaved produced sensitization (Supplementary Figure S8). Notably, caspase-3 (Figure 3C). These results suggested that lins- we found that the combinatorial use of MAPKi had a minor itinib + cilengitide cooperated to decrease the downstream function in enhancing the cooperative adjuvant combinatorial oncogenic signal of BRAFV600E, cilengitide played a key role sensitizing effects. These results indicated that our pair-wise in inducing apoptosis with vemurafenib, and linsitinib lim- sgRNA screening strategy could be used to efficiently identify ited the oncogenic effect of cilengitide in upregulating the gene pairs with both sensitizing and protective phenotypes. expression of β-catenin. We also examined the protein levels of cleaved caspase-3 and The synergistic sensitizing mechanisms phospho-ERK1/2 in the 8 subcombinations of vemurafenib + of combinatorial drug treatments cilengitide + JNK-IN-8, showing that among the groupings of these 3 drugs, vemurafenib + JNK-IN-8 epistatically reduced To understand the mechanisms responsible for the synergis- the expression of phospho-ERK1/2 relative to treatment with tic sensitization produced by using the drug combinations vemurafenib alone (Figure 3D). Triple drug treatment (vemu- of vemurafenib + cilengitide + linsitinib and vemurafenib + rafenib + cilengitide + JNK-IN-8) had a considerable impact cilengitide + JNK-IN-8, A375 cells were treated with 8 sub- on the induction of cleaved caspase-3 (Figure 3E). These results combinations of each group of triple drugs. Using qRT-PCR, showed that the cooperative sensitizing phenotype of cilengitide we found that the relative mRNA expression of E-cadherin + JNK-IN-8 during vemurafenib pressure was achieved by the was dramatically elevated in cells treated with vemurafenib + joint stimulation of apoptotic facilitators through co-targeting cilengitide + linsitinib compared to cells treated with vemu- of ITGB3 + JNK and the inhibition of the oncogenic signal of rafenib alone or cilengitide + linsitinib (Supplementary phosphorylated ERK1/2 through targeting of JNK. Figure S9B). In contrast, the mRNA expression of vimentin in A375 cells treated with vemurafenib + cilengitide + linsitinib Clinical relevance of combinatorial targets sharply declined compared to that in A375 cells treated with cilengitide + linsitinib (Supplementary Figure S9C). These To bridge the gap between our in vitro results and the demands results indicated that a potential cellular transformation from of clinical applications, we aimed to examine the expression of the mesenchymal status to the epithelial status was induced genes and proteins from the candidate synergistic pairs in mel- in A375 cells with the administration of triple drugs (vemu- anoma patients, which might have a potential impact on cur- rafenib + cilengitide + linsitinib). rent clinical practices for melanoma treatment. Because NF2 As a hub node in the genetic interaction network of vemu- was reported as a vital tumor suppressor54, we constructed a rafenib resistance (Figure 1D), both ITGB3 and IGF1R are Kaplan-Meier plot of data from 8,277 patients in the ICGC55 responsible for the reactivation of ERK1/223,53. By using melanoma database to identify the relationship of NF2 gene Cancer Biol Med Vol xx, No x Month 2021 13

A BC Treatment I II III IV VVIVII VIII Treatment III III IV VVIVII VIII Treatment III III IV VVIVII VIII Vemurafenib + –+– + –+–Vemurafenib + –+– +–+– Vemurafenib + –+ – +–+– Cilengitide – – + +++–– Cilengitide – – + + + +–– Cilengitide – – + ++ +–– Linsitinib –– + + ––++ Linsitinib ––+ + ––++ Linsitinib ––+ + ––++ β-catenin Cleaved caspase-3 Phospho-ERK 1/2 β-actin β-actin β-actin 2.0 1.8 2.0 ** n 1.6 * 1.8 1.4 1.6 3.0 2 * * 1.2 1.4 2.5 * 1.0 * 1.2 2.0 0.8 1.0 0.6

0.8 ** 1.5 cleaved caspase-3 0.4 Relative expression of 0.6 ** ** 1.0 0.2 phospho-ERK 1/ 0.4 0.0 Relative expression of 0.5 ** ** 0.2 III III IV VVIVII VIII 0.0

0.0 Relative expression of β -cateni III III IV VVIVII VIII III III IV VVIVII VIII

DE Treatment III III IV VVIVII VIII Treatment III III IV VVIVII VIII Vemurafenib + –+ – +–+– Vemurafenib + –+– +–+– Cilengitide – – + ++ +– – Cilengitide – – + +++–– JNK-IN-8 – – + + ––++ JNK-IN-8 – – + + ––++ Phospho-ERK 1/2 Cleaved caspase-3 β-actin β-actin

1.8 * 1.6 1.6 **

2 1.4 1.4 1.2 1.2 0.1 0.1 0.8 * ** 0.8 0.6 ** **

phospho-ERK 1/ 0.6 cleaved caspase-3 Relative expression of 0.4

Relative expression of 0.4 0.2 0.0 0.2 III III IV VVIVII VIII 0.0 III III IV VVIVII VIII

Figure 3 Cooperative sensitizing mechanisms of cilengitide + linsitinib and cilengitide + JNK-IN-8 during vemurafenib treatment. (A-C) The relative expressions of β-catenin, cleaved caspase-3, and phospho-ERK1/2 during different combination treatments with cilengitide + linsitinib with or without vemurafenib and its underlying cooperative mechanism. (D, E) The relative expressions of cleaved caspase-3 and phospho- ERK1/2 during different combination treatments with cilengitide + JNK-IN-8 with or without vemurafenib and its underlying cooperative mech- anism. Bar plots are the mean ± s.d. (N = 4) from 3 independent replicates. “*” and “**” designate significant differences (P < 0.05 and 0.01, respectively) between the single drug vemurafenib treatment vs. combinatorial drug treatments tested by the unpaired 2-tailed Student’s t-test.

expressions in melanomas and patient survivals (Figure 4A). clinical survival information of melanoma patients with BRAF We found that a decreased Neurofibromin 2 (NF2) mRNA level mutations using TCGA database57, we performed survival was associated with a shortened survival in melanoma patients analyses that associated the discretized NF2, ITGB3, and IGF1R -9 (PNF2 < 5.204 × 10 ). It has been reported that integrin subu- expression status (higher, high, medium, low, and lower) of nit beta 3 (ITGB3) is positively related to a higher frequency every individual patient with overall survivals. Notably, we of malignant melanomas than benign lesions, and insulin-like found that BRAF mutation patients with lower expressions of growth factor 1 receptor was found to be involved in adap- ITGB3 and IGF1R exhibited a significantly higher survival than tive bypass of BRAFV600E inhibition44,56. After analyses of the patients in the other groups (Figure 4B and 4C). 14 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance

BRAF mutant + NF2 (higher expression) + (ITGB3 + IGF1R) (higher expression) n = 10 A BCBRAF mutant + (ITGB3 + IGF1R) (higher expression) n = 39 BRAF mutant + NF2 (higher expression) + (ITGB3 + IGF1R) (lower expression) n = 4 1.00 BRAF mutant + (ITGB3 + IGF1R) (lower expression) n = 11 NF2 BRAF mutant + NF2 (higher expression) + ITGB3 (lower expression) n = 10 P value = 5.204–9 BRAF mutant + ITGB3 (lower expression) n = 24 BRAF mutant + NF2 (higher expression) + IGF1R (lower expression) n = 11 < –16.06 (n = 4,114) BRAF mutant + IGF1R (lower expression) n = 28 BRAF mutant + NF2 (higher expression) n = 95 0.75 >= –16.06 (n = 4,163) 1.00 1.00 y y 0.75 0.75 0.50 0.50 0.50

0.25 Survival probabilit 0.25 0.25 Survival probabilit Survival probabilit y Log-rank P = 0.025 Log-rank P = 0.24 0.00 0.00 0.00 2,500 5,000 7,500 10,000 03,000 6,000 9,000 12,000 02,500 5,000 7,500 10,000 Survival time (day) Survival time (day) Survival time (day)

Patient 882775 Patient 848557 Patient 842601 Patient 844583 D

E

F

G

H

Figure 4 The clinical relevance of pair-wise sgRNA library-selected candidates. (A) Kaplan-Meier survival plot of melanoma patients accord- ing to the RNA expression level of NF2 in the ICGC database. (B) Kaplan-Meier survival plot of BRAF-mutated melanoma patients according to the combination of ITGB3 and IGF1R RNA expression levels from The Cancer Genome Atlas (TCGA) database. The statistical significance of differences in survival was evaluated by the log-rank test. (C) Kaplan-Meier survival plot of BRAF-mutated melanoma patients according to the combination of NF2, ITGB3, and IGF1R RNA expression levels from TCGA database. The statistical significance of differences in sur- vivals was evaluated by the log-rank test. (D) Representative immunofluorescence (IF) micrographs showing CD133 (red) and BMI1 (green) expressions in patient-derived CD133LOW/BMI1LOW or CD133HIGH/BMI1HIGH melanoma samples 882775, 848557, 842601, and 844583, respec- tively. Scale bars, 1,000 µm (white). (E) Representative IF micrographs showing LGR5 (red) and HDGF (green) expressions in patient-derived CD133LOW or CD133HIGH melanoma samples 882775, 848557, 842601, and 844583. Scale bars, 1,000 µm (white). (F) Representative immunohis- tochemistry (IHC) micrographs showing ITGB3 expression in patient-derived CD133LOW/BMI1LOW or CD133HIGH/BMI1HIGH melanoma samples 882775, 848557, 842601, and 844583. Scale bars, 250 µm (black). (G) Representative IHC micrographs showing IGF1R expressions in patient- derived CD133LOW/BMI1LOW or CD133HIGH/BMI1HIGH melanoma samples 882775, 848557, 842601, and 844583. Scale bars, 250 µm (black). (H) Representative IHC micrographs showing phospho-JNK1/2 expression in patient-derived CD133LOW/BMI1LOW or CD133HIGH/BMI1HIGH mel- anoma samples 882775, 848557, 842601, and 844583. Scale bars, 250 µm (black). Cancer Biol Med Vol xx, No x Month 2021 15

Using CD133 and BMI1 as biological markers of cancer during dabrafenib treatment (Supplementary Figure S10). malignancy58,59, we found that melanoma patient tissues These data suggested that HDGF may physically associate with with high grade expression of CD133 and BMI1 were asso- LGR5 as a response mechanism to the stress of MAPK signa- ciated with high levels of ITGB3, IGF1R, and JNK, as well ling inhibition. In contrast to these results, a normal melano- as co-expression of LGR5 + HDGF (Figure 4D–4H), where cyte cell line, PIG1, showed decreased HDGF-LGR5 complex these hits were selected from the pair-wise sgRNA screening formation in response to the addition of vemurafenib + tra- library and calculated as efficient synergistic pairs for sensi- metinib or vemurafenib alone (Figure 5C), and the cell viabil- tizing BRAFi resistance. We also found that G-protein cou- ity of PIG1 stalled after 72 h with the adjuvant perturbation pled receptor 5 (LGR5) and hepatoma-derived growth factor of HDGF and LGR5 (Supplementary Figure S11), suggesting (HDGF) were usually co-expressed with CD133. Because that the complex might be essential for cell viability during LGR5 is a biomarker for hair follicle stem cells, it plays a role MAPKi stress. However, we found that vemurafenib + cilen- in sustaining skin homeostasis, and the cells expressing this gitide + linsitinib decreased HDGF-LGR5 complex formation, protein are thought to be a subset of cells with high stemness whereas complex formation was dramatically increased when that contribute to tumor development60,61. We also found vemurafenib + cilengitide was added along with the MAPK that HDGF and LGR5 co-localized morphologically in the inhibitor, JNK-IN-8 (Figure 5D). microvasculature regions, suggesting that HDGF and LGR5 We also found that the protein level of CD133 was increased might both be involved in tumor angiogenesis and cancer when A375 cells were treated with vemurafenib + cilengitide stemness. + JNK-IN-8, but the effective co-inhibition of HDGF and Thus, we were particularly interested in the combinatorial LGR5 in A375 cells abrogated the CD133 expression stimu- targets of ITGB3 + IGF1R, ITGB3 + JNK, and HDGF + LGR5, lated by vemurafenib + cilengitide + JNK-IN-8 (Figure 5E, not only because of their high clinical significance, but also Supplementary Figure S12, and Supplementary Table S7). because ITGB3 + IGF1R and ITGB3 + JNK are FDA-approved These results suggested that the expression of CD133 may be drug targets, and HDGF + LGR5 potentially interact in a phys- associated with the formation of the LGR5-HDGF protein ical complex to contribute to tumor angiogenesis and cancer complex. stemness. In summary, a potential relationship was found between responsive HDGF-LGR5 complex formation and emerging HDGF and LGR5 specifically respond to MAPK MAPKi resistance. These results indicated that LGR5 and inhibitors as protein complexes, and complex HDGF might be critical adaptive mechanisms in the process formation is regulated by cilengitide + of MAPKi resistance. linsitinib and cilengitide + JNK-IN-8 Disruption of both HDGF and LGR5 HDGF has been implicated in cancer cell proliferation and synergistically represses the emergence angiogenesis62,63. However, it has been usually studied as a of BRAFi-responsive oncogenic signals nuclear targeting mitogen that exerts its effect through inter- nalization64. How HDGF regulates intracellular signals by Kruppel-like factor 4 (KLF4) and sex-determining region interacting with cell surface receptors remains unknown. We Y-box 2 (SOX2) have been reported as essential transcription found that HDGF and LGR5 usually co-localized in the cel- factors that maintain CSC stemness and invasion activity in lular polarized tip apex of the A375 cell membrane, especially melanomas, in addition to many other cancer cells65-71. We in morphologically mesenchymal-like cells (Figure 5A). We further observed that digenic knockdown of HDGF and LGR5 further showed that HDGF physically interacted with the cell inhibited the nuclear localization of KLF4 (Figure 6A). In surface protein, LGR5, in A375 cells with or without vemu- addition, SOX2 could also be essential for the preservation of rafenib + trametinib treatment or vemurafenib treatment cancer stemness (Rybak and Tang, 2013), and the heterodimer- alone. However, the abundance of the HDGF-LGR5 com- ization of KLF4 and SOX2 is an important indicator of the ini- plex increasingly responded to MAPKi stress in A375 cells tiation of CSC pluripotency67. We found that co-localization (Figure 5B). Moreover, the A549 and H1299 cell lines also of KLF4-SOX2 was predominantly abrogated (Figure 6A). It showed highly active complex formation of LGR5-HDGF was also equally important that phospho-FAK, whose nuclear 16 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance

A

20 ×

40 ×

B A375 C PIG1 Vemurafenib –++ Vemurafenib –++ Trametinib ––+ Trametinib ––+

IP HDGF 40 kDa IP HDGF 40 kDa HDGF 95 kDa HDGF 40 kDa Input Input LGR5 95 kDa LGR5 95 kDa IgG HDGF 40 kDa IgG HDGF 40 kDa

β-actin 43 kDa β-actin 40 kDa

DE Vemurafenib + – ++++– Vemurafenib –++ + Cilengitide – – + ++– + siLGR5 ––+ – Linsitinib – – ++––+ siHDGF ––+ – Cilengitide – – ++ JNK-IN-8 – –––– – + JNK-IN-8 – – ++

IP HDGF CD133 HDGF Input β-actin LGR5

IgG HDGF

β-actin

Figure 5 HDGF and LGR5 form a protein complex to respond to MAPKi stress, and the formation of the complex is regulated by cilengitide + linsitinib and cilengitide + JNK-IN-8. (A) Representative immunofluorescence (IF) micrographs showing HDGF (green) and LGR5 (red) expres- sions in A375 cells. Scale bars, 10 µm (white). (B) Immunoprecipitation (IP) assays with anti-LGR5 antibody extracts from A375 cells expressing endogenous HDGF exposed to dimethyl sulfoxide (DMSO), BRAFi, or BRAFi + MEKi for 72 h. (C) IP assays with anti-LGR5 antibody extracts from SK-MEL-28 cells expressing endogenous HDGF exposed to DMSO, BRAFi, or BRAFi + MEKi for 72 h. (D) IP assays with anti-LGR5 antibody extracts from A375 cells expressing endogenous HDGF exposed to different drug combinations for 72 h. (E) Western blot showing CD133 levels in DMSO-, vemurafenib-, vemurafenib + JNK-IN-8 + cilengitide + siLGR5 + siHDGF-, and vemurafenib + JNK-IN-8 + cilengitide-treated A375 cell samples. Cancer Biol Med Vol xx, No x Month 2021 17 localization has been shown to be an indicator of accelerated spheroid parameters and then produced 3D image reconstruc- cell proliferation and enhanced cell survival72, was excluded tions47,48. The spheroid parameters were analyzed mainly in from the cell nucleus in the tri-drug group (Figure 6A). We terms of volume, convexity, solidity, sphericity, and quantity. also observed that the expression of nestin, another CSC We found that spheroids originating from different cancers marker that indicates poor prognosis73,74, was decreased exhibited different morphological patterns. For example, A375 when HDGF and LGR5 were simultaneously knocked down and H1299 grew as a single spheroid, but MHCC97H was and vemurafenib was added (Figure 6B). Consistently, BMI1, composed of many individualized clones (Figure 7A–7F and CD133, and NGFR were downregulated by HDGF and LGR5 Supplementary Figure S14). For A375 cells, we observed that double knockdown with exposure to BRAFi in A375 cells the volumes of spheroids during treatment with vemurafenib (Figure 6B). However, the combination of vemurafenib + tra- + siHDGF + siLGR5, vemurafenib + siHDGF, vemurafenib + metinib was the only drug pair that downregulated the expres- siLGR5, and vemurafenib + trametinib were not significantly sion of OCT4, but not other CSC markers, implying its limited different. However, the volumes associated with these com- function in cancer stemness inhibition. Together, these results binations were decreased compared to those of other combi- suggested that HDGF and LGR5 might play a necessary role in natorial groups, including the single use vemurafenib group the responsive activation of stemness signals and induction of (Figure 7B and 7C). We next analyzed the spheroid parameters MAPKi resistance. of convexity, solidity, and sphericity to evaluate the morpho- Typically, TGFβ3 is a positive modulator of the PI3-kinase/ logical quality of spheroid formation48, and found that vemu- AKT signaling pathway and a bypass signaling molecule rafenib + siHDGF + siLGR5 disrupted the A375 spheroid promoting survival when MAPK signaling is inhibited75. morphology by influencing convexity, solidity, and sphericity, Moreover, phospho-SMAD2, a tumor suppressor in mela- when compared to other groups (Figure 7B and 7C). In H1299 nomas76, was found to be essential for nelfinavir to sensitize cells, we found that dabrafenib + siHDGF + siLGR5 and dab- MAPKi resistance by repressing PAX3-mediated upregulation rafenib + trametinib resulted in equal decreases in spheroid of MITF77. We found that phospho-SMAD2 was downreg- volumes (Figure 7D and 7E). Simultaneously, these two com- ulated in SK-MEL-28 cells by the pair-wise perturbation of binations achieved the highest efficiency in controlling the HDGF and LGR5 (Figure 6C), and we did not observe com- volume of spheroids among these combinations of perturba- plex formation between LGR5 and HDGF in SK-MEL-28 cells tions (Figure 7E). Notably, the use of 5 µM dabrafenib or 5 (Supplementary Figure S13). In contrast, phospho-SMAD2 µM dabrafenib + 100 nM trametinib had an equivalent minor was not affected by the pair-wise perturbation of HDGF and impact on the prohibition of MHCC97H spheroid counts and LGR5 in A375 cells (Figure 6C). Together, these results sug- areas, when compared to the negative control group. However, gested that HDGF-LGR5 complex formation might be a pro- the MHCC97H spheroids exhibited remarkable reductions in tective mechanism allowing specific kinds of tumor cells to quantity and area in the HDGF + LGR5 interference group sense MAPK repression, and thus strengthen tumor stemness when treated with dabrafenib (Supplementary Figure S14). for enhanced survival (Figure 6D). Together, these results showed that pair-wise genetic pertur- bation of HDGF and LGR5 with the addition of MAPKi effi- BRAFi with adjuvant genetic perturbation of ciently disrupted the formation of tumor spheroids in the 3D HDGF-LGR5 sensitizes spheroid formation in growth environment. cancer cells Discussion Because LGR5 has been considered a specific stem cell marker within the hair follicle and intestine60,78, and to validate the Using conventional strategies of dissecting vemurafenib resist- MAPKi sensitizing effect of co-inhibition of HDGF and LGR5, ance in BRAFV600E melanomas, various key players in multi- we determined the spheroid formation ability in A375 cells ple signaling pathways have been identified10,14,24,31. Because and expanded our investigated cancer types to hepatoma of the complexity of BRAFV600E melanomas in response to and NSCLC cancer cells, which could better illustrate the vemurafenib intervention79,80, it is essential to use a system- stemness maintenance property of LGR5. We therefore used atic approach to analyze genetic interactions between impor- the AnaSP and ReViSP evaluation methods to analyze the tant gene nodes. To accomplish this goal, we developed a 18 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance

A DMSO Vermurafenib Vermurafenib + siLGR5 Vermurafenib + siHDGF siLGR5 + siHDGF Vermurafenib + siLGR5 + siHDGF

B C Vermurafenib – + + –– – + ++

Trametinib – – + –– ––+ – Vermurafenib ++– – siLGR5 – ––+–– + + + siHDGF +–+ – siHDGF siLGR5 ––––+ + – + + +–+ –

NGFR A375 TGFβ3

BMI1 A375 phospho-SMAD2 (Ser250)

A375 β-actin CD133

SK-MEL-28 TGFβ3 Nestin SK-MEL-28 phospho-SMAD2 (Ser250) OCT4 SK-MEL-28 β-actin β-actin

D

CD133 CD133

Cell membrane Cell membrane

Nucleus Nucleus

Spheroids formation Spheroids disruption

Figure 6 The disruption of both HDGF and LGR5 synergistically represses the emergence of BRAFi-responsive oncogenic signals. (A) Representative immunofluorescence micrographs showing KLF4 (red) and SOX2 (green) expressions, β-catenin (red), and CD133 (green) expressions, and nestin (red) and phospho-FAK (green) expressions in cells treated with dimethyl sulfoxide (DMSO), vemurafenib, vemurafenib + siLGR5, vemurafenib + siHDGF, siLGR5 + siHDGF, or vemurafenib + siLGR5 + siHDGF. Scale bars, 10 µm (white). (B) Western blot showing OCT4, NGFR, nestin, BMI1, and CD133 levels in DMSO-, vemurafenib-, vemurafenib + siLGR5-, vemurafenib + siHDGF-, siLGR5 + siHDGF-, and vemurafenib + siLGR5 + siHDGF-treated A375 cell samples. (C) Western blot showing TGFβ3 and phospho-SMAD2 (Ser250) levels in vemu- rafenib + siHDGF + siLGR5-, siHDGF + siLGR5-, vemurafenib-, and DMSO-treated A375 or SK-MEL-28 cell samples. (D) A proposed model for adaptive HDGF-LGR5 drug resistance to MAPK inhibitors. Cancer Biol Med Vol xx, No x Month 2021 19

A Negative control siLGR5 + siHDGF siLGR5 siHDGF Trametinib

Vemurafenib

DMSO

B Negative control siLGR5 + siHDGF siLGR5 siHDGF Trametinib

100 200 0 0 100 –100 –200 0 400 400 –100 400 400 0 Vemurafenib 600 200 100 0 –200 500 400 350 400 50 50 50 500 100 100 50 100 50 0

200 0 400 200 –200 DMSO 0 0 200 800 –400 1,000 0 –200 500 600 –200 800 800 600 800 200 200 200 200 200 0 100 200

C A375 spheroid convexity A375 spheroid solidity 1 ** ** ** 1 ** ** 0.95 ** ** 0.95 ** ** ** 0.9 ** ** 0.9 ** ** 0.85 0.85 0.8 ** 0.8 0.75 0.75 0.7 0.7 0.65 0.65

0.6 Solidity (a.u.) 0.6

Convexity (a.u.) 0.55 0.55 l 5

siLGR5 siLGR5 siHDGF siLGR siLGR5 siHDGF + siLGR5 + siLGR5 + siLGR5 Vemurafenib Vemurafenib Negative controsiHDGF + siLGR5siHDGF Negative control siHDGF

Vemurafenib + Vemurafenib + siHDGF Vemurafenib + Vemurafenib + siHDGF Vemurafenib + trametinib Vemurafenib + trametinib

Vemurafenib + siHDGF Vemurafenib + A375 spheroid sphericity A375 spheroid volume 1 ** ** ** 40,00,00,000 ** 0.9 ** ** ** ** 35,00,00,000 0.8 30,00,00,000 ** 25,00,00,000 ** 0.7 20,00,00,000 ** 0.6 15,00,00,000 10,00,00,000 ** 0.5

Volume (a.u.) * Sphericity (a.u.) 5,00,00,000 0.4 0 b l b 5 5 ol R siLGR5 siLGR5 siLGR5 siHDGF tinib GR GR5 + e iLG iL HDGF + siLGR5 siLGR5 siL s si siHDGF + Vemurafenib am + ib + s b Negative controsiHDGF siHDGF Vemurafeni en egative contr af siHDG siHDGF + enib + tr N Vemurafenib + Vemurafenib + siHDGF + Vemurafenib + trametini Vemur Vemurafeni Vemuraf Vemurafenib + Vemurafenib

D Negative control siLGR5 + siHDGF siLGR5 siHDGF Trametinib

Dabrafenib

DMSO

E Negative control siLGR5 + siHDGF siLGR5 siHDGF Trametinib F H1299 spheroid volume

35,00,00,000 300 200 200 200 100 30,00,00,000 ** 0 0 0 0 0 Dabrafenib –100 25,00,00,000 –200 –200 –300 100 –150 20,00,00,000 ** 100 600 500 500 400 500 600 400 500 15,00,00,000 ** ** 200 500 100 100 100 100 100 ** 500 100 0 10,00,00,000 Volume (a.u.) ** ** 5,00,00,000

300 0 400 200 b l 200 0 0 0 0 DMSO –200 siLGR5 siHDGF siHDGF –200 + siLGR5 + siLGR5 –400 800 –300 Dabrafeni 700 1,000 1,200 800 800 800 600 Negative controsiHDGF + siLGRsiHDGF5 200 200 200 200 100 200 200 200 Dabrafenib Dabrafenib + Dabrafenib + trametinib

Dabrafenib +

Figure 7 Pair-wise inhibition of HDGF and LGR5 with MAPKi affected the efficiency of three-dimensional (3D) spheroid formation in cancer cells. (A) Images of A375 spheroids during drug perturbations in different combinations. Images were captured with a 4× microscope objec- tive. (B) The 3D reconstruction images of A375 spheroids. (C) Estimated convexity, solidity, sphericity, and volume of A375 spheroids with 20 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance

combinatorial genetic perturbations of HDGF and LGR5 with or without MAPKi. Bar plots show the mean ± s.d. (N = 3) from 3 independent replicates. “*” and “**” designate significant differences (P < 0.05 and 0.01, respectively) between the combinatorial drug regimen of vemu- rafenib + siHDGF + siLGR5 and other drug treatments using the unpaired 2-tailed Student’s t-test. (D) Images of H1299 spheroids during drug perturbations in different combinations. Images were captured with a 4× microscope objective. (E) The 3D reconstruction images of H1299 spheroids. (F) Estimated volume of H1299 spheroids with combinatorial genetic perturbations of HDGF and LGR5 with or without MAPKi. Bar plots show the mean ± s.d. (N = 3) from 3 independent replicates. “**” designates significant difference (P < 0.01) between the combinatorial drug regimens of vemurafenib + siHDGF + siLGR5 and other drug treatments using the unpaired 2-tailed Student’s t-test.

CRISPR/Cas9-based pair-wise sgRNA functional screening their downstream effects on HDGF and LGR5 complex for- and data mining platform, enabling us to systematically eval- mation. Identifying multiple drug combinations for different uate paired functional interactions among 84 genes involved genetic conditions is therefore necessary to provide effec- in vemurafenib resistance. Although inhibiting a single target tive cancer therapy via mutual neutralization of side effects gene, such as ROCK1 and NGFR, can sensitize vemurafenib-­ caused by individual drug treatments81 and cooperative inhi- resistant melanoma cells14,24, our results showed that pair-wise bition of multiple key targets to induce drug resistance. More perturbation of ROCK1 + NF2 or NGFR + NF2 conferred a importantly, because drug resistance has been frequently strong protective phenotype during vemurafenib treatment used in clinical practice during treatment with vemurafenib, (Supplementary Figure S5A and S5B). Furthermore, our dabrafenib, vemurafenib + trametinib, and dabrafenib + tra- identified drug combinations (cilengitide + linsitinib and metinib82,83, finding a therapeutic solution that overcomes cilengitide + JNK-IN-8) strongly sensitized A375 cells to the setbacks encountered with the current treatments at the vemurafenib single drug intervention. These combinations molecular level may prolong patient survival. In the present have not previously been reported as effective adjuvant regi- study, we showed that the combined use of vemurafenib + mens for overcoming drug resistance in tumors. In conclusion, trametinib blocked the expression of OCT4, which also plays these results highlighted the importance of studying genetic a role in the maintenance of normal somatic stem cells84; interactions among the genes involved in drug resistance. however, CSC markers such as NGFR, BMI1, CD133 and nes- While r scores quantified the growth phenotypes of sgRNA tin were not downregulated in the vemurafenib + trametinib perturbations, GI scores reflected the extent of enhancement group (Figure 6B), although these highly malignant factors or inhibition relationships between 2 genes in response to have been thought to be the cause of tumorigenicity and can- drug treatment. Both types of information could be useful cer recurrence14,74,81,85,86. Thus, comparing the mechanistic and are sometimes complementary to each other. Although molecular differences among different therapeutic combi- the GI score alone may offer insights into strongly interacting nations is advantageous for identifying tumor vulnerabilities genes, the r score, but not the GI score, can provide informa- when resistance arises. It is well-known that LGR5 is one of tion on whether these strongly interacting genes are essential the best prognostic biomarkers for drug resistance or tumor in the drug resistance phenotype. Consistent with previous relapse for many types of cancers87-89 and has been usu- findings38, we showed that weak genetic interactions displayed ally reported as an activator of the Wnt/β-catenin signaling a low correlation among different screening replicates. In our pathway90. study, the number of gene pairs with strong GI scores was However, the abundance of nuclear-localized HDGF is smaller than the number of pairs with strong r scores, suggest- considered a remarkable prognostic factor in various can- ing that genetic interactions may be more sensitive to exper- cers91,92, but its function in cancer has never been identified. imental settings than cell growth phenotypes. Combining In the present study, we proposed a novel molecular mech- genetic interactions and growth phenotypes may therefore anism involving its prognostic significance, suggesting that provide insightful information regarding cellular responses to LGR5 formed a protein complex with HDGF on the cell drug interventions. membrane, to function as a MAPKi response element, and This study found that cilengitide + linsitinib and cilen- promote cell survival by enhancing cancer stemness and the gitide + JNK-IN-8 sensitized A375 cells to vemurafenib. nuclear localization of FAK to protect tumor cells against Although both drug combinations shared a common target, MAPKi agents. This adaptive complex-forming mechanism is the drug cooperation mechanisms might differ, especially in a newly discovered function that differs from the previously Cancer Biol Med Vol xx, No x Month 2021 21 well characterized roles of LGR5 in the R-spondin-LGR5- Conflict of interest statement Wnt/β-catenin axis and the previously defined functional mitogenic role of HDGF in the nucleus91,92. To the best of our The authors declare competing financial interests as following: knowledge, this is the first report of an adaptive protein com- Z.X. is one of the founders of Syngentech; Z.X. has filed a patent plex formation mechanism that affects resistance to MAPKi application (Grant No. 201710262042.6) to the State Intellectual in cancer cells, so it is important to characterize the protein Property Office of China based on aspects of this study. interaction interfaces between HDGF and LGR5 in future studies to develop potential drugs to block the formation of the HDGF-LGR5 protein complex. References

Conclusions 1. Cantwell-Dorris ER, O’Leary JJ, Sheils OM. BRAFV600E: implications for carcinogenesis and molecular therapy. Mol Cancer Ther. 2011; 10: 385-94. In summary, we used a pooled pair-wise sgRNA screening and 2. Das Thakur M, Salangsang F, Landman AS, Sellers WR, Pryer NK, data analysis pipeline to comprehensively study the genetic Levesque MP, et al. Modelling vemurafenib resistance in melanoma interactions involved in vemurafenib resistance, which may reveals a strategy to forestall drug resistance. Nature. 2013; 494: facilitate an understanding of vemurafenib resistance mech- 251-5. anisms in many types of cancer cells. In addition, our strategy 3. Joseph EW, Pratilas CA, Poulikakosb PI, Tadib M, Wangd W, to identify sensitizing and protective gene pairs led to the dis- Taylord BS, et al. The RAF inhibitor PLX4032 inhibits ERK signaling and tumor cell proliferation in a V600E BRAF selective covery of novel combinatorial targets with clinical relevance, manner. Proc Natl Acad Sci. 2010; 107: 14903-8. especially for the drug regimen of cilengitide + linsitinib and 4. Lu H, Liu S, Zhang G, Kwong LN, Zhu Y, Miller JP, et al. Oncogenic genotype combination of HDGF- + LGR5-, which were found BRAF-mediated melanoma cell invasion. Cell Rep. 2016; 15: to be potent in overcoming MAPKi resistance in multiple 2012-24. cancer cells, which provided novel biological insights into a 5. Ravnan MC, Matalka MS. Vemurafenib in patients with BRAF responsive protein complex formation mechanism that can V600E mutation-positive advanced melanoma. Clin Ther. 2012; 34: 1474-86. adapt cancer cells to environmental stress. Combined with the 6. Alonso SR, Tracey L, Ortiz P, Perez-Gomez B, Palacios J, Pollan M, clinical sample data, we propose that these insights can pro- et al. A high-throughput study in melanoma identifies epithelial- vide therapeutic opportunities. mesenchymal transition as a major determinant of metastasis. Cancer Res. 2007; 67: 3450-60. Acknowledgments 7. Carbone M, Yang H, Pass HI, Krausz T, Testa JR, Gaudino G. BAP1 and cancer. Nat Rev Cancer. 2013; 13: 153-9. 8. Easwaran H, Tsai HC, Baylin SB. Cancer epigenetics: tumor We thank the members of the Xie laboratory for helpful dis- heterogeneity, plasticity of stem-like states, and drug resistance. cussions and support. We thank Syngentech for technical sup- Mol Cell. 2014; 54: 716-27. port and for providing the Cas9-A375 cell line as a gift. We also 9. Hann SR. MYC Cofactors: molecular switches controlling diverse thank Ms. Cao from the School of Pharmaceutics in Tsinghua biological outcomes. Cold Spring Harb Perspect Med. 2014; 34: University for her technical support. We are also grateful to 1-23. Mr. Sun Yuxiang and Ms. Zhao Mingxin from The National 10. Holohan C, Van Schaeybroeck S, Longley DB, Johnston PG. Cancer drug resistance: an evolving paradigm. Nat Rev Cancer. 2013; 13: Center for Protein Sciences (Beijing, China) for their technical 714-26. assistance. 11. Ji Z, Kumar R, Taylor M, Rajadurai A, Marzuka-Alcala A, Chen YE, et al. Vemurafenib synergizes with nutlin-3 to deplete survivin and Grant support suppresses melanoma viability and tumor growth. Clin Cancer Res. 2013; 19: 4383-91. This work was supported by grants from the National 12. Kim MH, Kim J, Hong H, Lee S-H, Lee J-K, Jung E, et al. Actin remodeling confers BRAF inhibitor resistance to melanoma cells Natural Science Foundation of China (Grant Nos. 31471255, through YAP/TAZ activation. EMBO J. 2016; 35: 462-78. 31771483, 81171515, 31670991, and 61721003), and the 13. Konermann S, Brigham MD, Trevino AE, Joung J, Abudayyeh OO, National Key Research and Development Program (Grant Barcena C, et al. Genome-scale transcriptional activation by an Nos. 2017YFC0908400 and 2017YFC0908401). engineered CRISPR-Cas9 complex. Nature. 2015; 517: 583-8. 22 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance

14. Lehraiki A, Abbe P, Cerezo M, Rouaud F, Allegra M, Kluza J, et al. 30. Yuneva M, Zamboni N, Oefner P, Sachidanandam R, Lazebnik Y. Increased CD271 expression by the NF-kB pathway promotes Deficiency in glutamine but not glucose induces MYC-dependent melanoma cell survival and drives acquired resistance to BRAF apoptosis in human cells. J Cell Biol. 2007; 178: 93-105. inhibitor vemurafenib. Cell Discov. 2015; 1: 1-13. 31. Alcalá AM, Flaherty KT. BRAF inhibitors for the treatment of 15. Miao B, Ji Z, Tan L, Taylor M, Zhang J, Choi HG, et al. EPHA2 is a metastatic melanoma: clinical trials and mechanisms of resistance. mediator of vemurafenib resistance and a novel therapeutic target Clin Cancer Res. 2012; 18: 33-9. in melanoma. Cancer Discov. 2014; 5: 274-87. 32. Rizos H, Menzies AM, Pupo GM, Carlino MS, Fung C, Hyman 16. Nakanishi Y, Seno H, Fukuoka A, Ueo T, Yamaga Y, Maruno T, et al. J, et al. BRAF inhibitor resistance mechanisms in metastatic Dclk1 distinguishes between tumor and normal stem cells in the melanoma: spectrum and clinical impact. Clin Cancer Res. 2014; intestine. Nat Genet. 2012; 45: 98-103. 20: 1965-77. 17. Nazarian R, Shi H, Wang Q, Kong X, Koya RC, Lee H, et al. 33. Dixon SJ, Costanzo M, Baryshnikova A, Andrews B, Boone C. Melanomas acquire resistance to B-RAF(V600E) inhibition by RTK Systematic mapping of genetic interaction networks. Annu Rev or N-RAS upregulation. Nature. 2010; 468: 973-7. Genet. 2009; 43: 601-25. 18. Peng U, Wang Z, Pei S, Ou Y, Hu P, Liu W, et al. ACY-1215 34. Bassik MC, Kampmann M, Lebbink RJ, Wang S, Hein MY, Poser accelerates vemurafenib induced cell death of BRAF-mutant I, et al. A systematic mammalian genetic interaction map reveals melanoma cells via induction of ER stress and inhibition of ERK pathways underlying ricin susceptibility. Cell. 2013; 152: 909-22. activation. Oncol Rep. 2017; 37: 1270-6. 35. Horn T, Sandmann T, Fischer B, Axelsson E, Huber W, Boutros 19. Roesch A, Vultur A, Bogeski I, Wang H, Zimmermann KM, M. Mapping of signaling networks through synthetic genetic Speicher D, et al. Overcoming intrinsic multidrug resistance in interaction analysis by RNAi. Nat Methods. 2011; 8: 341-6. melanoma by blocking the mitochondrial respiratory chain of 36. Du D, Roguev A, Gordon DE, Chen M, Chen S-H, Shales M, et al. slow-cycling JARID1B high cells. Cancer Cell. 2013; 23: 811-25. Genetic interaction mapping in mammalian cells using CRISPR 20. Rossman KL, Der CJ, Sondek J. GEF means go: turning on Rho interference. Nat Methods. 2017; 14: 577-80. GTPases with guanine nucleotide-exchange factors. Nat Rev Mol 37. Guo Y, Bao C, Ma D, Cao Y, Li Y, Xie Z, et al. Network-based Cell Biol. 2005; 6: 167-80. combinatorial CRISPR-Cas9 screens identify synergistic modules in 21. Shalem O, Sanjana NE, Hartenian E, Shi X, Scott DA, Mikkelsen TS, human cells. ACS Synth Biol. 2019; 8: 482-90. et al. Genome-scale CRISPR-Cas9 knockout screening in human 38. Han K, Jeng EE, Hess GT, Morgens DW, Li A, Bassik MC. cells. Science. 2014; 343: 561-5. Synergistic drug combinations for cancer identified in a CRISPR 22. Shi H, Moriceau G, Kong X, Lee M-K, Lee H, Koya RC, et al. screen for pairwise genetic interactions. Nat Biotechnol. 2017; 35: Melanoma whole-exome sequencing identifies V600E B-RAF 463-74. amplification-mediated acquired B-RAF inhibitor resistance. Nat 39. Wong ASL, Choi GCG, Cui CH, Pregernig G, Milani P, Adam M, Commun. 2012; 3: 724-8. et al. Multiplexed barcoded CRISPR-Cas9 screening enabled by 23. Smalley KSM. A pivotal role for ERK in the oncogenic behaviour of CombiGEM. Proc Natl Acad Sci. 2016; 113: 2544-9. malignant melanoma? Int J Cancer. 2003; 104: 527-32. 40. Xue Y, Martelotto L, Baslan T, Vides A, Solomon M, Mai TT, et al. 24. Smit MA, Maddalo G, Greig K, Raaijmakers LM, Possik PA, van An approach to suppress the evolution of resistance in BRAF Breukelen B, et al. ROCK1 is a potential combinatorial drug target V600E-mutant cancer. Nat Med. 2017; 23: 929-37. for BRAF mutant melanoma. Mol Syst Biol. 2014; 10: 772. 41. Welsh SJ, Rizos H, Scolyer RA, Long GV. Resistance to combination 25. Villicaña C, Cruz G, Zurita M. The basal transcription machinery BRAF and MEK inhibition in metastatic melanoma: where to next? as a target for cancer therapy. Cancer Cell Int. 2014; 14: 1-15. Eur J Cancer. 2016; 62: 76-85. 26. Voulgari A, Pintzas A. Epithelial-mesenchymal transition in 42. Shibue T, Weinberg RA. EMT, CSCs, and drug resistance: the cancer metastasis: mechanisms, markers and strategies to mechanistic link and clinical implications. Nat Rev Clin Oncol. overcome drug resistance in the clinic. Biochim Biophys Acta. 2017; 14: 611-29. 2009; 1796: 75-90. 43. Irvine M, Stewart A, Pedersen B, Boyd S, Kefford R, Rizos H. 27. Zingg D, Julien Debbache, Schaefer SM, Tuncer E, Frommel SC, Oncogenic PI3K/AKT promotes the step-wise evolution of Cheng P, et al. The epigenetic modifier EZH2 controls melanoma combination BRAF/MEK inhibitor resistance in melanoma. growth and metastasis through silencing of distinct tumour Oncogenesis. 2018; 7: 1-11. suppressors. Nat Commun. 2015; 6: 1-17. 44. Villanueva J, Vultur A, Lee JT, Somasundaram R, Fukunaga-Kalabis 28. Liu X, Vorontchikhina M, Wang Y-L, Faiola F, Martinez E. M, Cipolla AK, et al. Acquired resistance to BRAF inhibitors STAGA recruits mediator to the MYC oncoprotein to mediated by a RAF kinase switch in melanoma can be overcome by stimulate transcription and cell proliferation. Mol Cell Biol. 2008; cotargeting MEK and IGF-1R/PI3K. Cancer Cell. 2010; 18: 683-95. 28: 108-21. 45. Kampmann M, Bassik MC, Weissman JS. Integrated platform for 29. Shim H, Chun YS, Lewis BC, Dang CV. A unique glucose- genome-wide screening and construction of high-density genetic dependent apoptotic pathway induced by c-Myc. Proc Natl Acad interaction maps in mammalian cells. Proc Natl Acad Sci. 2013; 15: Sci. 1998; 95: 1511-6. 2317-26. Cancer Biol Med Vol xx, No x Month 2021 23

46. Ma D, Peng S, Xie Z. Integration and exchange of split dCas9 62. Eguchi R, Wakabayashi I. HDGF enhances VEGF-dependent domains for transcriptional controls in mammalian cells. Nat angiogenesis and FGF-2 is a VEGF-independent angiogenic factor Commun. 2016; 7: 1-7. in non-small cell lung cancer. Oncol Rep. 2020; 44: 14-28. 47. Piccinini F. AnaSP: a software suite for automatic image analysis of 63. Ke Y, Zhao W, Xiong J, Cao R. Downregulation of miR-16 promotes multicellular spheroids. Comput Methods Programs Biomed. 2015; growth and motility by targeting HDGF in non-small cell lung 119: 43-52. cancer cells. FEBS Lett. 2013; 587: 3153-7. 48. Zanoni M, Piccinini F, Arienti C, Zamagni A, Santi S, Polico R, 64. Bao C, Wang J, Ma W, Wang X, Cheng Y. HDGF: a novel jack-of-all- et al. 3D tumor spheroid models for in vitro therapeutic screening: trades in cancer. Futur Oncol. 2014; 10: 2675-85. a systematic approach to enhance the biological relevance of data 65. Basu-Roy U, Seo E, Ramanathapuram L, Rapp TB, Perry JA, Orkin obtained. Sci Rep. 2016; 6: 19103-14. SH, et al. Sox2 maintains self renewal of tumor-initiating cells in 49. Piccinini F, Tesei A, Arienti C, Bevilacqua A. Cancer multicellular osteosarcomas. Oncogene. 2012; 31: 2270-82. spheroids: volume assessment from a single 2D projection. Comput 66. Boumahdi S, Driessens G, Lapouge G, Rorive S, Nassar D, Le Methods Programs Biomed. 2015; 118: 95-106. Mercier M, et al. SOX2 controls tumour initiation and cancer 50. Liu H, Wei Z, Dominguez A, Li Y, Wang X, Qi LS. CRISPR-ERA: stem-cell functions in squamous-cell carcinoma. Nature. 2014; 511: a comprehensive designer tool for CRISPR genome editing, 246-50. (gene) repression, and activation. Bioinformatics. 2015; 31: 67. Lee SH, Wottrich S, Bonavida B. Crosstalks between Raf-kinase 3676-8. inhibitor protein and cancer stem cell transcription factors (Oct4, 51. Zhu G, Zhao XM, Wu J. A survey on biomarker identification based KLF4, Sox2, Nanog). Tumor Biol. 2017; 39: 1-15. on molecular networks. Quant Biol. 2016; 4: 310-9. 68. Pang L, Xu L, Yuan C, Li X, Zhang X, Wang W, et al. Activation of 52. May GHW, Allen KE, Clark W, Funk M, Gillespie DAF. Analysis of EGFR-KLF4 positive feedback loop results in acquired resistance the interaction between c-Jun and c-Jun N-terminal kinase in vivo. to sorafenib in hepatocellular carcinoma. Mol Carcinog. 2019; 58: J Biol Chem. 1998; 273: 33429-35. 2118-26. 53. Yeh AH, Bohula EA, Macaulay VM. Human melanoma cells 69. Riverso M, Montagnani V, Stecca B. KLF4 is regulated by RAS/RAF/ expressing V600E B-RAF are susceptible to IGF1R targeting by MEK/ERK signaling through E2F1 and promotes melanoma cell small interfering RNAs. Oncogene. 2006; 25: 6574-81. growth. Oncogene. 2017; 36: 3322-33. 54. Cooper J, Giancotti FG. Molecular insights into NF2/Merlin tumor 70. Utikal J, Maherali N, Kulalert W, Hochedlinger K. Sox2 is suppressor function. FEBS Lett. 2014; 588: 2743-52. dispensable for the reprogramming of melanocytes and melanoma 55. Watanabe K, Montserrat E, Gallinger S, van de Vijver M, López- cells into induced pluripotent stem cells. J Cell Sci. 2009; 122: Bigas N, Bader GD, et al. International network of cancer genome 3502-10. projects. Nature. 2010; 464: 993-8. 71. Yu F, Li J, Chen H, Fu J, Ray S, Huang S, et al. Kruppel-like 56. Natali PG, Hamby CV, Felding-Habermann B, Liang B, factor 4 (KLF4) is required for maintenance of breast cancer stem Nicotra MR, Fiippo FD, et al. Clinical significance of aV3 cells and for cell migration and invasion. Oncogene. 2011; 30: integrin and intercellular adhesion molecule-i expression in 2161-72. cutaneous malignant melanoma lesions. Cancer Res. 1997; 57: 72. Lim ST, Chen XL, Lim Y, Hanson DA, Vo TT, Howerton K, et al. 1554-60. Nuclear FAK promotes cell proliferation and survival through 57. Liu J, Lichtenberg T, Hoadley KA, Poisson LM, Lazar AJ, Cherniack FERM-enhanced p53 degradation. Mol Cell. 2008; 29: 9-22. AD, et al. An integrated TCGA pan-cancer clinical data resource 73. Krupkova Jr O, Loja T, Zambo I, Veselska R. Nestin expression in to drive high-quality survival outcome analytics. Cell. 2018; 173: human tumors and tumor cell lines. Neoplasma. 2010; 60: 607-16. 400-16. 74. Qin Q, Sun Y, Fei M, Zhang J, Jia Y, Gu M, et al. Expression of 58. Simbulan-Rosenthal CM, Dougherty R, Vakili S, Ferraro AM, putative stem marker nestin and CD133 in advanced serous Kuo LW, Alobaidi R, et al. CRISPR-Cas9 knockdown and induced ovarian cancer. Neoplasma. 2012; 60: 607-16. expression of CD133 reveal essential roles in melanoma invasion 75. Walker L, Millena AC, Strong N, Khan SA. Expression of TGFβ3 and metastasis. Cancers. 2019; 11: 1-23. and its effects on migratory and invasive behavior of prostate 59. Siddique HR, Saleem M. Role of BMI1, a stem cell factor, in cancer cancer cells: involvement of PI3-kinase/AKT signaling pathway. recurrence and chemoresistance: preclinical and clinical evidences. Clin Exp Metastasis. 2013; 30: 13-23. Stem Cells. 2012; 30: 372-8. 76. Mnich CD, Hoek KS, Oberholzer PA, Seifert B, Hafner J, Dummer 60. Jaks V, Barker N, Kasper M, Van Es JH, Snippert HJ, Clevers H, R, et al. Reduced pSmad2 immunodetection correlates with et al. Lgr5 marks cycling, yet long-lived, hair follicle stem cells. Nat increased primary melanoma thickness. Melanoma Res. 2007; 17: Genet. 2008; 40: 1291-9. 131-6. 61. Da Silva-Diz V, Solé-Sánchez S, Valdés-Gutiérrez A, Urpí M, 77. Smith MP, Brunton H, Rowling EJ, Ferguson J, Arozarena I, Riba-Artés D, Penin RM, et al. Progeny of Lgr5-expressing hair Miskolczi Z, et al. Inhibiting drivers of non-mutational drug follicle stem cell contributes to papillomavirus-induced tumor tolerance is a salvage strategy for targeted melanoma therapy. development in epidermis. Oncogene. 2012; 32: 3732-43. Cancer Cell. 2016; 29: 270-84. 24 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance

78. Snippert HJ, van der Flier LG, Sato T, van Es JH, van den Born M, 86. Siddique HR, Saleem M. Concise review: role of BMI1, a stem cell Kroon-Veenboer C, et al. Intestinal crypt homeostasis results from factor, in cancer recurrence and chemoresistance: preclinical and neutral competition between symmetrically dividing Lgr5 stem clinical evidences. Stem Cells. 2012; 30: 372-8. cells. Cell. 2010; 143: 134-44. 87. Ryuge S, Sato Y, Jiang SX, Wang G, Kobayashi M, Nagashio R, et al. 79. Allen EM Van, Wagle N, Sucker A, Treacy DJ, Johannessen CM, The clinicopathological significance of Lgr5 expression in lung Goetz EM, et al. The genetic landscape of clinical resistance to RAF adenocarcinoma. Lung Cancer. 2013; 82: 143-8. inhibition in metastatic melanoma. Cancer Discov. 2013; 4: 94-109. 88. Sánchez-Danés A, Larsimont JC, Liagre M, Muñoz-Couselo E, 80. Romano E, Pradervand S, Paillusson A, Weber J, Harshman K, Lapouge G, Brisebarre A, et al. A slow-cycling LGR5 tumour Muehlethaler K, et al. Identification of multiple mechanisms of population mediates basal cell carcinoma relapse after therapy. resistance to vemurafenib in a patient with BRAFV600E-mutated Nature. 2018; 562: 434-58. cutaneous melanoma successfully rechallenged after progression. 89. Yang L, Xie X, Tang H, Kong Y, Xie X, Chen J, et al. LGR5 promotes Clin Cancer Res. 2013; 19: 5749-57. breast cancer progression and maintains stem-like cells through 81. Li R, Chen T, Li S. Network-based method to infer the activation of wnt/β-catenin signaling. Stem Cells. 2015; 33: contributions of proteins to the etiology of drug side effects. Quant 2913-24. Biol. 2015; 3: 124-34. 90. de Lau W, Peng WC, Gros P, Clevers H. The R-spondin/Lgr5/Rnf43 82. Carlino MS, Gowrishankar K, Saunders CAB, Pupo GM, Snoyman module: regulator of Wnt signal strength. Genes Dev. 2014; 28: S, Zhang XD, et al. Antiproliferative effects of continued mitogen- 305-16. activated protein kinase pathway inhibition following acquired 91. Chen X, Yun J, Fei F, Yi J, Tian R, Li S, et al. Prognostic value of resistance to BRAF and/or MEK inhibition in melanoma. Mol nuclear hepatoma-derived growth factor (HDGF) localization in Cancer Ther. 2013; 12: 1332-42. patients with breast cancer. Pathol Res Pract. 2012; 208: 437-43. 83. Long GV, Fung C, Menzies AM, Pupo GM, Carlino MS, Hyman 92. Kishima Y, Yamamoto H, Izumoto Y, Yoshida K, Enomoto H, J, et al. Increased MAPK reactivation in early resistance to Yamamoto M, et al. Hepatoma-derived growth factor stimulates dabrafenib/trametinib combination therapy of BRAF-mutant cell growth after translocation to the nucleus by nuclear metastatic melanoma. Nat Commun. 2014; 5: 1-9. localization signals. J Biol Chem. 2002; 277: 10315-22. 84. Greco SJ, Liu K, Rameshwar P. Functional similarities among genes regulated by OCT4 in human mesenchymal and embryonic stem cells. Stem Cells. 2007; 25: 3143-54. Cite this article as: Yu Y, Tao M, Xu L, Cao L, Le B, An N, et al. Systematic 85. Angelastro JM, Lamé MW. Overexpression of CD133 promotes screening reveals synergistic interactions that overcome MAPK inhibitor drug resistance in C6 glioma cells. Mol Cancer Res. 2010; 8: resistance in cancer cells. Cancer Biol Med. 2021; x: xx-xx. doi: 10.20892/j. 1105-15. issn.2095-3941.2020.0560 Cancer Biol Med Vol xx, No x Month 2021 25

Supplementary materials

AB C 89.63% 84.61% 79.49%

10,000 Gini index 10,000 Gini index 10,000 Gini index = 0.16 = 0.19 = 0.22

8,000 8,000 8,000

6,000 6,000 6,000 gRNA combinations gRNA combinations 4,000 gRNA combinations 4,000 4,000 of of of

2,000 2,000 2,000 Numbers Numbers Numbers 0 0 0 0 2 46810120246 81012 024681012

Log2 (readcount) – plasmid Log2 (readcount) – T0 replicate 1 Log2 (readcount) – T0 replicate 2

Figure S1 Histogram of pair-wise sgRNA combinations in different libraries. (A-C) Distributions of pair-wise sgRNA combinations in library samples after plasmid preparation (A) and after lentiviral infection and puromycin selection in 2 replicates (B, C). The Gini coefficients were calculated to evaluate the distribution of sgRNA combinations. The shaded areas represent the maximal coverage of sgRNA combinations within a 64-fold range in plasmid samples, and 2 T0 replicates, respectively. 26 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance

A Regulation of trancription from RNA polymerase II promoter

Regulation of cell proliferation

Positive regulation of cell communication

Cellular response to stress

Regulation of cell death

Regulation of MAPK cascade

Regulation of cell differentiation

Epithelium development

Regulation of GTPase activity

Covalent chromatin modification

Transmembrane receptor protein tyrosine kinase signaling pathway

Cancer stem cell marker

0510 15 20 25 30 Gene numbers

B sgRNA replicate pool 1 sgRNA replicate pool 2 sgRNA replicate pool 3Negative sgRNA pool

Negative sgRNA pool Negative sgRNA pool sgRNA pool 1 sgRNA pool 1 sgRNA pool 2 sgRNA pool 2 sgRNA pool 3 sgRNA pool 3

Oligonucleotides annealing

Esp3I U6 Bsal ccdB Bsal Scaffold Esp3I Esp3I 7SK Bsal ccdB Bsal Scaffold Esp3I

Bsal Spacer1 Bsal Bsal Spacer2 Bsal

Kan Kan

Golden gate Golden gate assembly assembly

sgRNA1 sgRNA2 Pairwise sgRNA plasmid construction details of each sgRNA pool combination Esp3I U6 Spacer1 Scaffold Esp3I Esp3I ccdB Esp3I hEF1a 2A Esp3I 7SK Spacer2 Scaffold Esp3I

Kan Spe Kan

Golden gate assembly

Pairwise linking the four sgRNA pools into total library sgRNA1 sgRNA2 for the downstream screening

U6 Spacer1 Scaffold 7SK Spacer2 Scaffold hEF1a 2A Negative sgRNA pool Negative sgRNA pool sgRNA pool 1 sgRNA pool 1 sgRNA pool 2 sgRNA pool 2 Spe sgRNA pool 3 sgRNA pool 3

CD E PCC = 0.9184696 Plasmid library T0 sample replicate 1T0 sample replicate 2 PCC = 0.7666164 1 1 100 15

10 ) 80

10 60

5 40 5 Cumulative frequency (% 20 (normalized readcounts) - T0 replicate (normalized readcounts) - T0 replicate 2 2 Lo g Lo g 0 0 0 024 6810 12 0 369 0510

Log2(normalized readcounts) – T0 replicate 2 Log2(normalized pairwise sgRNA readcounts) Log2(normalized readcounts) – plasmid

Figure S2 Construction of the pair-wise sgRNA library. (A) Genes were classified into functional groups according to the gene ontology information from the Molecular Signature Database. (B) Workflow of pair-wise sgRNA library construction. Synthesized and annealed oligos Cancer Biol Med Vol xx, No x Month 2021 27

were mixed and cloned into the single-wise sgRNA constructs in the first Golden Gate reaction with BsaI. The single-wise sgRNA constructs were cloned into a lentiviral vector in the second Golden Gate reaction with Esp3I. Kan, kanamycin resistance gene; Spe, spectinomycin resist- ance gene; ccdB, ccdB toxin coding gene; PuroR, puromycin resistance gene. (C) Cumulative distribution of normalized pair-wise sgRNA read counts in the indicated sgRNA libraries. The cumulative coverages of the plasmid library (blue line), and two independent replicated samples before drug treatment (red and green lines) were 97.26%, 97.25%, 92.76%, respectively. (D, E) Pearson’s correlations of the indicated samples.

A B 16

I hEF1a mKate fragment1 gRNA cutting site mKate fragment2 14 l 12

gRNA1 gRNA2

e 10 II U6 Neg Scaffold 7SK Neg Scaffold hEF1a Cas9 escence leve

mKat 8 of

gRNA1 gRNA2 6 lative fluor III U6 Spacer Scaffold 7SK Neg Scaffold hEF1a Cas9 Re 4

2 gRNA1 gRNA2

IV U6 Neg Scaffold 7SK Spacer Scaffold hEF1a Cas9 0 I+ + III +– IV –+ CD

I hEF1a mKate fragment1 gRNA cutting site mKate fragment2

20 l

II hEF1a EYFP fragment1 gRNA cutting site EYFP ragment2 15

gRNA1 gRNA2 escence leve III U6 Neg Scaffold 7SK Neg Scaffold hEF1a Cas9 10 mKate + EYFP of

gRNA1 gRNA2 lative fluor Re IV U6 Spacer Scaffold 7SK Spacer Scaffold hEF1a Cas9 5

gRNA1 gRNA2 0 V U6 Spacer Scaffold 7SK Spacer Scaffold hEF1a Cas9 I+ + II + + IV + – V – +

Figure S3 Balancing test of pair-wise promoter configurations. (A) Plasmids used for cutting insulating linker to catalyze the recombination of mKate fragments. (B) The relative fluorescence expression level of mKate, which was calculated from the induced fluorescence level of vector III or IV divided by the leaky fluorescence level of the negative vector II. The bar plot is the mean ± s.d. (N = 3) from 3 independent replicates. There was no significant difference between U6 initiated relative fluorescence and 7SK initiated relative fluorescence according to the unpaired 2-tailed Student’s t-test. (C) Plasmids used for cutting insulating linker to catalyze recombination of mKate and EYFP fragments. (D) Relative fluorescence level of mKate + EYFP was calculated from the induced fluorescence level of vector IV or V divided by the leaky fluorescence level of the negative vector III. Bar plot was the mean ± s.d. (N = 3) from 3 independent replicates. There was no significant dif- ference between U6 initiated relative fluorescence and 7SK initiated relative fluorescence according to the unpaired 2-tailed Student’s t-test. 28 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance

Data ≥ 90th percentile A B Data between 10th & 90th percentile Replicate 1 Data ≤ 10th percentile HDGF + LGR5 IGF1R + ITGB3 1.0 Strongly sensitizing Weakly effecting Strongly protecting ITGB3 + Jun region region region CCNC + NF2 NF2 + NGFR 0.5 NF2 + ROCK1 Undefined data 1.0 0.0 IGF1R + ITGB3 plicate 2 (a.u.) re

e of –0.5 NF2 + ROCK1

Density (a.u.) HDGF + LGR5 ITGB + Jun 0.5 NF2 + NGFR

CCN + NF2 ρ scor –1.0 PCC (data ≥ 90th & ≤ 10th percentile) = 0.949 PCC (defined data) = 0.847 0.0 10th 50th 90th –1.0 –0.50.0 0.5 1.0 Percentile of ρ scores (a.u.) ρ score of replicate 1 (a.u.)

C Hub node Hit node Hub/hit node Top node

ρ score (a.u.) 1.238

0

–0.839

Figure S4 Identification of gene pairs with strong growth phenotypes during vemurafenib stress. (A) Histogram distribution of r scores in replicate 1. The r scores below the 10th percentile or above the 90th percentile were recognized as gene pairs with strong growth phenotype. (B) Correlation of r scores of 2 independent replicates. Blue and pink points represent the reproducible top and bottom 10% of r scores; black points indicate the middle 80% reproducible r scores; and gray points represent the other r scores that showed a low correlation in 2 independent replicates. (C) The hierarchical network representation of the top listed effective gene pairs (Supplementary Table S4). The thickness of edges indicates the strength of r scores. Blue and red colors designate the sensitizing and protective growth phenotypes, respec- tively. Purple nodes inside yellow rectangles denote hub gene nodes. Blue nodes represent top gene nodes. Yellow nodes were selected for the experimental validations shown in subsequent figures.

AB C Data between 10th & 90th percentile Data ≤ 10th percentile 0.8 Cl = 99% Replicate 1 0.5 A + B 0.6 0.4 Strongly interacted Weakly interacted Strongly interacted 0.2 region region region 0.0 0.0 2.0 ρ score) ρ –0.2 Gl score 1.5 –0.4 0.50

with A ( Gl score –0.5 0.25 1.0 –0.6 0.00 Observer ρ –0.25 Densit y A + B Phenotype of doulbe-gene –0.8 0.5 Gene A + B –0.50 Gl scores of replicate 2 (a.u.) PCC (data ≥ 90th & 10th percentile) = 0.924 –1.0 0.0 –1.0 PCC (defined data) = 0.808 –0.5 00.5 10th 50th 90th –0.5 0.00.5 Phenotype of individual gene ( ρ score) Percentile of Gl scores Gl scores of replicate 1

Figure S5 The strategy for selecting strong genetic interactions. (A) Scatter plot of r scores of pair-wise sgRNA perturbations against a com- mon bait gene (for example, Gene A) and a series of single-wise sgRNA perturbations. GI scores were obtained by calculating the deviations between observed and expected r scores. The dotted line denotes the 99% confidence interval). (B) The histogram distribution of averagedGI scores for targeting the same gene pair with both sgRNA orientations derived from replicate 1. GI scores below the 10th percentile or above the 90th percentile were recognized as strong interactions. (C) The correlation of GI scores of 2 independent replicates. Blue and pink points represent the reproducible top and bottom 10% of GI scores; black points indicate the middle 80% reproducible GI scores; and gray points represented the undefinedGI scores that showed a low correlation in 2 independent replicates. Cancer Biol Med Vol xx, No x Month 2021 29

P = 0.029 A Cl = 99% B C Cl = 99% Negatice control 0.8 0.8 15 ITGB3 0.6 0.6 Jun 0.4 0.4 10 ITGB3 + Jun 0.2 0.2 Gl score Gl score 0.0 0.0 0.40 5 0.50 0.20 –0.2 –0.2 0.00 with Jun

with ITGB 3 0.00 –0.4 –0.4 –0.20 0 (normalized readcount 2

–0.6 –0.50 –0.6 Jun + ITGB3 –0.40 in vemurafenib group) Lo g Phenotype of doulbe-gene –0.8 Phenotype of doulbe-gene –0.8 –5 –1.0 –1.0 –0.5 0 0.5 –0.5 0 0.5 –5 0510 15 Log (normalized readcount Phenotype of individual gene ( ρ) Phenotype of individual gene ( ρ) 2 in DMSO group)

Cl = 99% Cl = 99% D E F P = 6.335 × 10–6 NF2 + CCNC CCNC + NF2 1.0 1.0 20 Negatice control NF2 0.8 0.8 Gl score Gl score CCNC 0.6 0.6 0.40 CCNC + NF2 0.20 0.20 10 0.4 0.00 0.4 –0.20 0.00

with NF 2 0.2 0.2 –0.40 with CCN C –0.20 –0.60 0.0 0.0 0 (normalized readcount 2 in vemurafenib group) Phenotype of doulbe-gene Phenotype of doulbe-gene –0.2 –0.2 Lo g –0.4 –0.4 –0.5 0 0.5 –0.5 0 0.5 –5 0510 15

Phenotype of individual gene ( ρ) Phenotype of individual gene ( ρ) Log2 (normalized readcount in DMSO group)

G HI–3 Cl = 99% P = 3.369 × 10 0.7 Cl = 99% 0.7 15 Negatice control 0.5 0.5 LGR5 HDGF 0.3 0.3 10 HDGF + LGR5 Gl score Gl score 0.1 0.1 0.60 –0.1 0.30 –0.1 0.40 5 0.00 –0.3 –0.3 0.00 –0.30 with LGR5 with HDGF (normalized readcount –0.5 –0.40 2 –0.5 in vemurafenib group) 0 Lo g

–0.7 Phenotype of doulbe-gene –0.7 Phenotype of doulbe-gene HDGF + LGR5 LGR5 + HDGF –0.9 –0.9 –5 –0.5 0.0 0.5 –0.5 0.0 0.5 –5 0510 15 Log (normalized readcount Phenotype of individual gene ( ρ) Phenotype of individual gene ( ρ) 2 in DMSO group)

Figure S6 Validation of identified genetic interactions. (A, B), (D, E), (G, H) Scatter plots of r scores of pairwise sgRNA perturbations against a common bait gene and single-wise sgRNA perturbations. GI scores were calculated as the deviation from the observed r scores to linearly fitted values (black line). The dotted line denotes the 99% confidence interval. The color bar representsGI scores from negative (blue) to pos- itive (red) effects. (C, F and I) Scatter plots of the read count abundances of single-wise and pair-wise sgRNA species in the vemurafenib- and dimethyl sulfoxide-treated samples. Red dots, pair-wise sgRNA species; green and blue, single-wise sgRNA species; and gray dots, negative control sgRNA. Linear fit lines for pair-wise (red), single-wise (green, blue), and negative control (gray) sgRNA species. Data were tested for significance using the 2-tailed Mann-Whitney U test. (C, F and I) TheP values show the results of Mann-Whitney U tests when comparing the distribution of r scores perturbing the gene pairs against the negative controls. 30 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance

AB *** ** ** 5 –0.3

4 –0.4

–0.5 3 valu e ρ valu e ρ –0.6 2

–0.7 1

–0.8 0

IGF1R +–– NF2 +–– ITGB3 –+– CCNC –+– IGF1R + ITGB3 –++ NF2 + CCNC –++

Figure S7 The sgRNA growth competition assay for GI pairs. (A, B) Schematic gene perturbations were conducted with lentiviral sgRNA expression vectors, which were different from those in Figure 3 (E, G). Mixed Cas9-A375 cells with or without single or double gene pertur- bations were treated with vemurafenib or dimethyl sulfoxide for up to 10 days before fluorescence-activated cell sorting. The r values were calculated by using the relative abundances of individual Cas9-A375 derivatives (see details in the Materials and Methods). Data are the mean ± s.d. (N = 3) from 3 independent replicates. “*” and “**” designate significant differences (P < 0.05 and P < 0.01, respectively) between single and pair-wise sgRNA perturbations according to the unpaired 2-tailed Student’s t-test.

MeWo V + C + O MeWo V + T + C + O MeWo V + C + J MeWo V + T + C + J ** ** ** n.s

** 1.0

0.8 0.8 0.8 0.8 y y y y 0.6 0.6 0.6 0.6 Cell vitalit Cell vitalit

Cell vitalit 0.4 Cell vitalit

0.4 0.4 0.4 0.2

0.0

C J V + C V + OC + OV + C + OV C + CV + JC + JV + C + J + + + + O T T C + J T T C + O + V + V + V + V + T + C + J V V + T + C + O

FaDu D + C + J FaDu D + T + C + J MCF7 D + C + J MCF7 D + T + C + J ** * n.s ** ** 3 3.0 3 ** 2.5

2.5 y

2 y 2

2.0 y 2.0 y Cell vitalit

1.5 Cell vitalit Cell vitalit Cell vitalit 1.5 1 1 1.0

0.5 1.0 0

D + CD + JC + JD + C + JD C + CD + JC + JD + C + J + + J C T T C + J + + J + T T C + J + D + D + T + C + J D + D D + T + C + J D

Figure S8 Cell viability tests of the MAPKi sensitizing effect by using different drug combinations. MeWo, FaDu, and MCF7 cells were used as models to validate the efficiency of MAPKi sensitizing effects when using screening selected combinatorial hits. V, D, C, O, J were abbre- viated as vemurafenib, dabrafenib, cilengitide, OSI-096 (linsitinib), and JNK-IN-8, respectively. Bar plots denote the mean ± s.d. (N = 3) from 3 independent replicates, and the unpaired 2-tailed Student’s t-test was used to evaluate the statistical significance, “*” denotes P < 0.05, “**” denotes P < 0.01. Cancer Biol Med Vol xx, No x Month 2021 31

A Epithelial-mesenchymal transition

E-cadherin Epithelial status Mesenchymal status

Vimentin

B C 5.0 25 ** ** 4.5 20 4.0

vimenti n 3.5 E-cadherin of

of 15 3.0 2.5 ession

ession 10 2.0 * * 1.5 * 5 1.0

lative expr 0.5 lative expr Re

Re 0 0.0 Treatment III III IV Treatment III III IV Vemurafenib+ ––+ Vemurafenib + ––+ Cilengitide –– + + Cilengitide –– + + Linsitinib –– + + Linsitinib –– + +

Figure S9 Gene expressions of E-cadherin and vimentin are affected by different combinatorial drug regimens. (A) Schematic illustration of E-cadherin and vimentin gene expressions as significant biomarkers of the epithelial-mesenchymal transition. (B) The mRNA expressions of E-cadherin in different portfolio treatments of clilengitide + linsitinib. (C) The mRNA expressions of vimentin in the combinatorial treatments of clilengitide + linsitinib. “*” and “**” designate significant differences (P < 0.05 and P < 0.01, respectively) between single drug vemurafenib vs. combinatorial drug treatments tested using the unpaired 2-tailed Student’s t-test.

AB A549 H1299

Dabrafenib – ++ Dabrafenib – ++ Trametinib ––+ Trametinib ––+

IP HDGF 40 kDa IP HDGF 40 kDa

HDGF 40 kDa HDGF 40 kDa Input Input LGR5 95 kDa LGR5 95 kDa

IgG HDGF 40 kDa IgG HDGF 40 kDa

β-actin 43 kDa β-actin 43 kDa

Figure S10 HDGF and LGR5 form a protein complex to respond to MAPKi stress. (A) Immunoprecipitation assays (IP) with anti-LGR5 anti- body extracts from A549 cells expressing endogenous HDGF exposed to dimethyl sulfoxide (DMSO), BRAFi, or BRAFi + MEKi for 72 h. (B) IP assays with anti-LGR5 antibody extracts from H1299 cells expressing endogenous HDGF exposed to DMSO, BRAFi, or BRAFi + MEKi for 72 h. 32 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance

PIG1 SK-MEL-28 Vemurafenib + siHDGF + siLGR5 Vemurafenib–++

Vemurafenib + trametinib + siHDGF + siLGR5 Trametinib – –+ 0.6 IP HDGF 40 kDa

HDGF 40 kDa 0.4

y Input

LGR5 95 kDa Cell vitalit 0.2 IgG HDGF 40 kDa

β-actin 0.0 43 kDa 020 40 60 80 Time (h) Figure S13 HDGF and LGR5 cannot form a protein complex dur- ing MAPKi stress. Immunoprecipitation assays with anti-LGR5 anti- Figure S11 The CCK8 assay for detecting PIG1 cell growth. The body extracts from SK-MEL-28 cells expressing endogenous HDGF cell viability assay showing the growth curve of PIG1 cells during exposed to dimethyl sulfoxide, BRAFi, or BRAFi + MEKi for 72 h. treatments with vemurafenib + siHDGF + siLGR5 and vemurafenib + trametinib + siHDGF + siLGR5.

siLGR5 + – + – + – siHDGF + – + – + – LGR5 HDGF β-actin

A375 H1299 MHCC97H

Figure S12 The siHDGF and siLGR5 co-knockdown efficiency in A375, H1299, and MHCC97H cell lines. The HDGF and LGR5 co-knockdown efficiency in A375, H1299, and MHCC97H cells. A375, H1299, and MHCC97H cells were simultaneously co-transfected with siHDGF and siLGR5 in the experimental groups for 48 h. Cancer Biol Med Vol xx, No x Month 2021 33

Replicate 1 Negative control siLGR5 + siHDGF siLGR5 siHDGF Trametinib

Dabrafenib

DMSO

Replicate 2

Negative control siLGR5 + siHDGF siLGR5 siHDGF Trametinib

Dabrafenib

DMSO

Replicate 3 Negative control siLGR5 + siHDGF siLGR5 siHDGF Trametinib

Dabrafenib

DMSO

Figure S14 Images of MHCC97H spheroids during perturbations by different drug combinations. The images were captured using a 4× microscope objective, from 3 independent replicates. Scale bars, 1,000 µm (white). 34 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance

Table S1 Information of sgRNAs used in the efficiency tests of 2 promoters Plasmids used in Supplementary Figure S3A sgRNA sequences U6-mKate_linker_sgRNA-7SK-negative_sgRNA-EF1ɑ-spCas9 ACGGAGGCTAAGCGTCGCAA ACGTTCGAGTACGACCAGCT

U6-negative_sgRNA-7SK-mKate_linker_sgRNA-EF1ɑ-spCas9 ACGTTCGAGTACGACCAGCT ACGGAGGCTAAGCGTCGCAA

U6-negative_sgRNA-7SK-negative_sgRNA-EF1ɑ-spCas9 ACGTTCGAGTACGACCAGCT ACGTTCGAGTACGACCAGCT Plasmid used in Supplementary Figure S3C sgRNA sequences U6-mKate_linker_sgRNA-7SK-EYFP_linker_sgRNA-EF1ɑ-spCas9 ACGGAGGCTAAGCGTCGCAA TACTAACGCCGCTCCTACAG

U6-EYFP_linker_sgRNA-7SK-mKate_linker_sgRNA-EF1ɑ-spCas9 TACTAACGCCGCTCCTACAG ACGGAGGCTAAGCGTCGCAA

U6-negative_sgRNA-7SK-negative_sgRNA-EF1ɑ-spCas9 ACGTTCGAGTACGACCAGCT ACGTTCGAGTACGACCAGCT

Table S2 Information of sgRNAs used in the growth competition assay Plasmid DNA used in Figure 2E sgRNA sequence pLV-hU6-gRNA-neg-7sk-IGF1R-hEF1a-EYFP-2A-Puro GATCCTCACGGTGAACGTCT CTTCGAGATGACCAATCTCA pLV-hU6-gRNA-neg-7sk-ITGB3-hEF1a-EBFP-2A-Puro GATCCTCACGGTGAACGTCT CTGGCGGGCGTTGGCGTAGG pLV-hU6-IGF1R-7sk-ITGB3-hEF1a-mKate-2A-Puro CTTCGAGATGACCAA TCTCA CTGGCGGGCGTTGGCGTAGG Plasmid DNA used in Figure 2F sgRNA sequence pLV-hU6-gRNA-neg-7sk-CCNC-hEF1a-EYFP-2A-Puro GATCCTCACGGTGAACGTCT TAGGCAAAGATCCGTTCTGT pLV-hU6-gRNA-neg-7sk-NF2-hEF1a-EBFP2-2A-Puro GATCCTCACGGTGAACGTCT ATCCTCACGGTGAACGTCTT pLV-hU6-CCNC-7sk-NF2-hEF1a-mKate-2A-Puro TAGGCAAAGATCCGTTCTGT ATCCTCACGGTGAACGTCTT Plasmid DNA used in Supplementary Figure S7A sgRNA sequence pLV-hU6-gRNA-neg-7sk-IGF1R-hEF1a-EBFP-2A-Puro ATTTGCCGTTCGAATGCGTA GCTCTCGAGGCCAGCCACT pLV-hU6-gRNA-neg-7sk-ITGB3-hEF1a-EYFP-2A-Puro ATTTGCCGTTCGAATGCGTA AGTGAGGCCCGAGTACTAG pLV-hU6-IGF1R-7sk-ITGB3-hEF1a-mKate-2A-Puro GCTCTCGAGGCCAGCCACT AGTGAGGCCCGAGTACTAG Plasmid DNA used in Supplementary Figure S7B sgRNA sequence pLV-hU6-gRNA-neg-7sk-CCNC-hEF1a-EYFP-2A-Puro ATTTGCCGTTCGAATGCGTA GATGCCAAAAACACACATGT pLV-hU6-gRNA-neg-7sk-NF2-hEF1a-EBFP2-2A-Puro ATTTGCCGTTCGAATGCGTA GATCCTCACGGTGAACGTCT pLV-hU6-CCNC-7sk-NF2-hEF1a-mKate-2A-Puro GATGCCAAAAACACACATGT GATCCTCACGGTGAACGTCT Cancer Biol Med Vol xx, No x Month 2021 35

Table S3 List of previously reported genes involved in vemurafenib resistance Gene Supporting literature BAP1 Carbone M, Yang H, Pass HI, Krausz T, Testa JR, Gaudino G. BAP1 and cancer. Nat Rev Cancer. 2013; 13: 153–159.

BIRC5 Ji Z, Kumar R, Taylor M, Rajadurai A, Marzuka-Alcala A, Chen YE, et al. Vemurafenib Synergizes with Nutlin-3 to Deplete Survivin and Suppresses Melanoma Viability and Tumor Growth. Clin Cancer Res. 2013; 19: 4383–4391.

DCLK1 Nakanishi Y, Seno H, Fukuoka A, Ueo T, Yamaga Y, Maruno T, et al. Dclk1 distinguishes between tumor and normal stem cells in the intestine. Nat Genet. 2012; 45: 98–103.

EPHA2 Miao B, Ji Z, Tan L, Taylor M, Zhang J, Choi HG, et al. EPHA2 Is a Mediator of Vemurafenib Resistance and a Novel Therapeutic Target in Melanoma. Cancer Discov. 2014; 5: 274–287.

EZH2 Zingg D, Julien Debbache, Schaefer SM, Tuncer1 E, Frommel SC, Cheng P, et al. The epigenetic modifier EZH2 controls melanoma growth and metastasis through silencing of distinct tumour suppressors. Nat Commun. 2015; 6: 1–17.

HDAC6 PENG U, WANG Z, PEI S, OU Y, HU P, LIU W, et al. ACY-1215 accelerates vemurafenib induced cell death of BRAF-mutant melanoma cells via induction of ER stress and inhibition of ERK activation. Oncol Rep. 2017; 37: 1270–1276.

KDM5B Roesch A, Vultur A, Bogeski I, Wang H, Zimmermann KM, Speicher D, et al. Overcoming Intrinsic Multidrug Resistance in Melanoma by Blocking the Mitochondrial Respiratory Chain of Slow-Cycling JARID1Bhigh Cells. Cancer Cell. 2013; 23: 811–825.

NGFR Lehraiki A, Abbe P, Cerezo M, Rouaud F, Allegra M, Kluza J, et al. Increased CD271 expression by the NF-kB pathway promotes melanoma cell survival and drives acquired resistance to BRAF inhibitor vemurafenib. Cell Discov. 2015; 1: 1–13.

ROCK1 Smit MA, Maddalo G, Greig K, Raaijmakers LM, Possik PA, Breukelen B van, et al. ROCK1 is a potential combinatorial drug target for BRAF mutant melanoma. Mol Syst Biol. 2014; 10: 772–772.

YAP1 Kim MH, Kim J, Hong H, Lee S-H, Lee J-K, Jung E, et al. Actin remodeling confers BRAF inhibitor resistance to melanoma cells through YAP/TAZ activation. EMBO J. 2016; 35: 462–478.

Table S4 Information of 84 selected genes and designed sgRNAs Gene sgRNA Function category ADI1 TGGTAGTTTATCTTTGCATA /

AGACATGGTGACGCTCCCCG

TATGGTTATGATGTCCATCC

APBB1 AAGGTCAACGTCCCCTCCTC Regulation of transcription from RNA polymerase II promoter; regulation of cell death; regulation of cell differentiation; covalent chromatin modification AGGCTGGGTAGAGATGACCG

ATTACTGGCACATCCCAACA

ARHGAP6 GAATGGTGATCTGACAGCTC Regulation of GTPase activity

CTCGGAATATCCCCACTGTC

GAAACTGGATTCACTAGGAA

ARHGEF2 CAAGGAGGAGCGGCCACGGC Regulation of transcription from RNA polymerase II promoter; regulation of cell proliferation; regulation of cell death; regulation of cell differentiation; regulation of GCTATTAGAACGCCGACGCC GTPase activity TTTGGGCCCGAGGGTCCATG

ARHGEF5 CCTCCGAGTCGTACCTGCAG Regulation of transferase activity; regulation of MAPK cascade; regulation of GTPase activity CCGCTGCAGGTACGACTCGG

CACGGTGCTGTTGCGCACCA 36 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance

Table S4 Continued Gene sgRNA Function category BAP1 GCTCTTCGATCCATTTGAAC Regulation of cell proliferation; covalent chromatin modification CACGGACGTATCATCCACCA

CTTCCACGAGCAGGGTGAAG

BCAR3 TCGGCAGGTAGCTCTCTGAC Regulation of GTPase activity

AATGGCCATCGGCATTGCAG

GAAGGAATAGAGGTCCCCCA

BIRC5 GCCAGGCAGGGGGCAACGTC Regulation of cell death

GCTCGTTCTCAGTGGGGCAG

CACTGAGAACGAGCCAGACT

BUB1B ACAGTTACTTGCACAACCAA Regulation of transferase activity

ATGAGCTGCAAGCAGGCCCT

GCTCCAATCATCCGTGTAGG

C9orf50 AATACACCCCTTTGATGTTG /

CTGCCTGGGCGCCGATTCCC

CCGCCGGGTCGCGGTGAGCA

CCDC101 GCGGGAATCGGCAGACACGA Covalent chromatin modification

GCATCCGCTCATGGGTCTTC

GGGTTTGTTTGATCAGCTGA

CCNC GTACAACACAGGCTACATGT Regulation of transcription from RNA polymerase II promoter; regulation of transferase activity GATGCCAAAAACACACATGT

TAGGCAAAGATCCGTTCTGT

CHN2 CAAATGATTCTCTTGGGACG Regulation of GTPase activity

AGGACACCGACGAGCCGGAC

GTGGATCATTAGGTACCGGA

CRB2 TGTGGGCCCATGGAGCCCCG Regulation of transcription from RNA polymerase II promoter; regulation of cell differentiation; epithelium development ACATCGATGAGTGTGCATCC

TGCCAGGCTACCGAGAGTGG

CUL3 GAGATCAAGTTGTACGTTAT Regulation of transcription from RNA polymerase II promoter; regulation of cell proliferation; regulation of transferase activity TGGATATGATTGCAAGAGAG

CGAGATCAAGTTGTACGTTA

DCLK1 GGAGTAGAGAGCTGACTACC Cancer Stem cell marker

GAGATACAGCCGAGGGTCGC

GCACTTCGACGAGCGGGATA

DRAP1 TGTCGGGAACAGATGCCACC Regulation of transcription from RNA polymerase II promoter;

TCCCCGTCCCCCTGCATGTC

GGACGGGGAAGACAACCACA Cancer Biol Med Vol xx, No x Month 2021 37

Table S4 Continued

Gene sgRNA Function category EAPP GGATGACTACGACCCCTACG Regulation of transcription from RNA polymerase II promoter; regulation of cell proliferation CTAGGCTCTTCAACCGCGTA

ACTAGATGATTCACTTTCTC

EED GCAGGCATGTCTGTTCCCGC Regulation of transcription from RNA polymerase II promoter; covalent chromatin modification CGCAGTCGACACTTCCCTCT

TGTTTGGAGTTCAGTTTAAC

EGFR CCCCCGGGAGAGCGACTGCC Regulation of transcription from RNA polymerase II promoter; regulation of cell proliferation; regulation of cell death; regulation of transferase activity; regulation of AATTCGCTCCACTGTGTTGA MAPK cascade; epithelium development; regulation of GTPase activity TGCAAATAAAACCGGACTGA

EPAS1 GTCCATCTGCTGGTCAGCTT Regulation of transcription from RNA polymerase II promoter;

GAAAGATCATGTCGCCATCT

CTGCTGGTCAGCTTCGGCTT

EphA2 CCACACAGGCACCGATATCC Regulation of MAPK cascade; epithelium development; regulation of GTPase activity

GTGTGCAAGGCATCGACGCT

CTACTATGCCGAGTCGGACC

EZH2 GGAAGCAGGGACTGAAACGG Regulation of transcription from RNA polymerase II promoter; regulation of cell proliferation; regulation of transferase activity; regulation of MAPK cascade; regulation GGGAAGCAGGGACTGAAACG of cell differentiation; regulation of GTPase activity; covalent chromatin modification ACACGCTTCCGCCAACAAAC

FSD1 CAGCCGTACCTACGAGCTGC /

ACAGATGCTGCTGAACGTGG

CTGCAGCTCGTAGGTACGGC

GAB2 CTGTTGGCCATGGAACGAGC Regulation of cell proliferation

ATACTCCTGAGGTGCGCTGG

TGCGCTCTCGGAGAAGGTGC

GBP2 AACTTTCGGATGCACAACCG Cancer Stem cell marker

CTCAGTGTCGAGCAGAACTA

GGAAGTGCTTCGTCTTCGAT

GPR35 TGGTTATGTACAGGGCGCGA /

TCGCAGGGAGTGCAGCACGA

CGGCCGTGTGCGCGGTCCTC

GPR98 GGTATACAGGGGCTCCCCCA Regulation of cell differentiation

CTGAAATCTAGATTCAGGCG

GATGTCATCCAGAAACACTG

HDAC6 CTCTATCCCCAATCTAGCGG Regulation of cell death; regulation of cell differentiation; covalent chromatin modification TCCATCCACCGCTACGAGCA

ACCTAATCGTGGGACTGCAA 38 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance

Table S4 Continued Gene sgRNA Function category HDGF TAGCCGGAAGCCTTGACAGT Regulation of transcription from RNA polymerase II promoter;

GCAAGCCCAACAAGAGGAAA

AGCCGGAAGCCTTGACAGTA

HMMR TGAAAGGCTGGTCAAGCAAT /

TAAAGAGAAGAATCTGTTTG

TGCCCAGGACAGGCGGATCC

HOXC13 GTACGTCTATGAGGACAGCG Regulation of transcription from RNA polymerase II promoter; epithelium development

AGTGCTGGTAGCCTTCGACG

GACGCCCTCATCCCCGTCGA

IGF1R GGCTCTCGAGGCCAGCCACT Regulation of cell proliferation; regulation of cell death; regulation of transferase activity; regulation of MAPK cascade TCGTTGCGGATGTCGATGCC

CTTCGAGATGACCAATCTCA

ITGB3 TTCTCCTTCAGGTCACAGCG Regulation of transferase activity; regulation of cell differentiation

GAGTGAGGCCCGAGTACTAG

CTGGCGGGCGTTGGCGTAGG

Jun GCGGTAGCCTCGGTGGCAGG Regulation of transcription from RNA polymerase II promoter; regulation of cell proliferation; regulation of cell death; regulation of MAPK cascade; regulation of cell CTGCTGGGGTTGCGCGGGAA differentiation; epithelium development; regulation of GTPase activity CAGACAGTGCCCGAGATGCC

KDM5B TGCTGAGATGCGTATCCGTT Regulation of cell proliferation; epithelium development; covalent chromatin modification; cancer Stem cell marker CAGTCTGCTCGGCTATGGGC

CGGCTATGGGCCGGATCTTG

KRAS CGAATATGATCCAACAATAG Regulation of cell proliferation; regulation of cell death; regulation of transferase activity; regulation of MAPK cascade; regulation of cell differentiation; epithelium GTAGTTGGAGCTGGTGGCGT development GAATATAAACTTGTGGTAGT

LAMA3 GCGTTCACAGTGCTCTCCCG Epithelium development

GGGAATCCCCAGAAATTCGG

GCCCATTGTTCTCCCCCTGG

LCOR AGATCCATTCCTCGAACTTG Regulation of transcription from RNA polymerase II promoter;

TAATCCAGATGGCCCAAGGC

AGTACCGCCCAGACGGACTT

LGR5 GCTCTGACATACATTCCCAA Regulation of cell proliferation; epithelium development; cancer Stem cell marker

CACTGTCATTGCGAGCCCGA

CGGAGACTGGGCAGGGGATT

LPAR5 CCAGAGGGCTAGCGCGTTGA /

ACTCGTCGGGCCGAGTCTTC

ATCTTCCAGATGAACATGTA Cancer Biol Med Vol xx, No x Month 2021 39

Table S4 Continued Gene sgRNA Function category MAP3K11 AGAGCTGCTACCGCGCTCTG Regulation of cell death; regulation of transferase activity; regulation of MAPK cascade

CAGCGCCCGCTGGATCAGCT

AGCTGCTACCGCGCTCTGCG

MAPKAPK3 ACTTCTGTCCAGTGCGCCGA Regulation of transferase activity; regulation of MAPK cascade

CTTGGACAACTGGTAGTCGT

TGTATGACAGCCCCAAGGCC

MECOM TTGTTTGAGGCCCGACGAAG Regulation of cell death; regulation of MAPK cascade; covalent chromatin modification

GGCCTGTGGTACAAGCCGGA

GATGATGTCAGTACACCAAG

MED12 AGTCCGAGTGGACCGGCGAG Regulation of transcription from RNA polymerase II promoter; epithelium development

GATTGCTGCATAGTAGGCAC

CCGGCCCCGCGCCTATTACC

MED15 CCCAGAGCCCAGTGACGGCG Regulation of transcription from RNA polymerase II promoter;

GGCACACAGTAAATCCAGCA

TCGAAAATGGATAATGAGCC

METRL21B AGGCAAGAAGGTGATCGAAC /

GAAGGTGATCGAACTGGGTG

AGGTGATCGAACTGGGTGCG

MRFAP1L1 CGAAGCCGACGAGAGAGTGT /

CTCTTATACGCGAGCACGGG

TCGGGTGCTGCACGTGATTG

NF1 AACTTCGGAATTCTGCCTCT Regulation of cell proliferation; regulation of cell death; regulation of transferase activity; regulation of MAPK cascade; regulation of cell differentiation; epithelium ACGGCCTGGACCCATTCCAC development; regulation of GTPase activity GAGAGAAAATAAAACCCCAG

NF2 GTCCATGGTGACGATCCTCA Regulation of cell proliferation; regulation of transferase activity; regulation of MAPK cascade; regulation of cell differentiation; epithelium development GATCCTCACGGTGAACGTCT

ATCCTCACGGTGAACGTCTT

NGFR CGACGGCACGTATTCCGACG Regulation of MAPK cascade

CTCGTCGGAATACGTGCCGT

GCAGCGGCACACGGCGTCGT

P2RY8 CTGGCGCCGCCGTCGTTACG /

GCGCCGCCGTCGTTACGCGG

CGGCCACCGCGTAACGACGG

PAK7 CTACACGACCGAAAAGTACA Regulation of cell death

TTGAACACAGGGTTCATACT

ATGCATCACACCCATCCAGC 40 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance

Table S4 Continued Gene sgRNA Function category PARP3 ACCGTCATCCCGCACAACTT /

GTACACACTTATCGAAGTAC

TCGATAAGTGTGTACTTGCC

PCDH7 TCACGGTGACCATCCCCACA /

ACCGACTGGCCGAACATGGG

GTACCGGAGGAGCTGCTTGG

PDCD10 CACGGAGTCCCTTCTTCGTA Regulation of cell proliferation; regulation of cell death; regulation of transferase activity; regulation of MAPK cascade; epithelium development ATAGAGGGGCATAGAAACCA

CATCAGCTGCCATACGAAGA

PLK1 AATTTGCCGTAGGTAGTATC Regulation of transcription from RNA polymerase II promoter; regulation of cell death; regulation of transferase activity AGCCAAGCACAATTTGCCGT

TACCTACGGCAAATTGTGCT

PODN CAGGAGGGCGTCGTGGACTG Regulation of cell proliferation; regulation of transferase activity

GCCCCAGGATTTGGCCGAAG

GCACCTGTACAACAACGCGC

PRKCE TCTTAAGATCAAAATCTGCG Regulation of MAPK cascade

CGTCAGGCGCAGGGTCCATC

AACGCATGCGGCCGAGGAAG

PRKRA ATTCAGGTATTACACGAATA Regulation of cell proliferation; regulation of cell death

TATTCGTGTAATACCTGAAT

TTCACCTTCAGAGTAACCGT

PRPF4B GAAAGTAGAAGTCGCGATCG /

GGAGCAGATCACGCTTGCGA

CACGCGGTGGTCGTAGACGA

Pygo1 TATGGATTTGACGAAGGTAG Regulation of transcription from RNA polymerase II promoter;

AGATGGTCAGAGTTTGGATT

GACGAATGGCCATGACGGTT

RANBP9 GTGAACATGAATAGACTACC Regulation of MAPK cascade

CTAATTCTTCTAGAACGGTC

GGCAGGACGACTCCGGAGAC

RAP1GAP2 CTCGACCTGCACGACGATGT Regulation of cell differentiation; regulation of GTPase activity

ACAAAGCTGCCATTTACCGA

GATCCTGTCCGTCAAGTGCG

RAPGEF1 TGAAGCCGTATACCTCCATG Regulation of cell proliferation; regulation of transferase activity; regulation of MAPK cascade; regulation of cell differentiation; epithelium development; regulation of GAGATGTTGTCATACTGCGA GTPase activity GCTCCCCCTGACAGACCGCG Cancer Biol Med Vol xx, No x Month 2021 41

Table S4 Continued Gene sgRNA Function category RASSF5 AGTCCAACTGGATCAGGCTG Regulation of cell proliferation

GGATCAGGCTGCGGCATTCT

TGGATCAGGCTGCGGCATTC

RBL1 GTACTGTGTGAACTGCATGA Regulation of transcription from RNA polymerase II promoter; regulation of transferase activity AAATATCTCCGTCCGGTCAG

CCGCAAAAGCATTATTCCCA

RHOG TGACACCCTACGGCGCCTCA Regulation of cell proliferation; regulation of GTPase activity

GCGCCCATCACACCGCAGCA

GCGCGCAGAGCGCAGTTGAC

ROCK1 GTAGTGAAGGTGATTGGTAG Regulation of cell death; regulation of cell differentiation; epithelium development

GTGATTGGTAGAGGTGCATT

TATGAAGTAGTGAAGGTGAT

SPRED1 TATCCGTGGAGAGCGACTCA Regulation of transferase activity; regulation of MAPK cascade

TTATCCGTGGAGAGCGACTC

GGTGGATGGTTACCACTTGG

TACC1 ACAGACCCAGTGGCACGAGA /

TCATGTGGTCAGAAATCAGC

TCAGCTGGTGCCGAGGTGAA

TADA1 AATCTGACAACGCGTGAGAA Covalent chromatin modification

GACAACGCGTGAGAATGGCC

ACTGGGCTAACCTAAAGCTG

TADA2B GTAGCTAGGAAGGCGGCTTT Regulation of transcription from RNA polymerase II promoter; covalent chromatin modification CGAGCTGAAGCGCGCCCACG

AACCTAGCCGGCTCCAAACG

TAF5L CAAACCGTTGTTCTCACTGC Regulation of transcription from RNA polymerase II promoter; covalent chromatin modification CACGTCAAGATGAATATGTA

AAACAGTCCGAAGAGCACAG

TAF6L CCGCTCTTCTCGCTCTGACA Regulation of transcription from RNA polymerase II promoter; covalent chromatin modification TGACGGTTGAGGACTTCAAC

TGAGCCGGACAGACTCCCGA

TCEB2 AGCGGCCTCCTGACGAGCAG Regulation of transcription from RNA polymerase II promoter;

CGAACTGAAGCGCATCGTCG

GCTCGTCAGGAGGCCGCTTG

TFAP2C TTAAATGCCTCGTTACTGGG Regulation of transcription from RNA polymerase II promoter; regulation of cell proliferation CTGTCTGATCGTGCAGCAAC

GAGGCATTTAAGCATTCAGG 42 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance

Table S4 Continued Gene sgRNA Function category TNFRSF1B CACCCGGAGTATGGCCCCAG Regulation of cell proliferation; regulation of cell death; regulation of MAPK cascade

CGATAAGGCCCGGGGTACAC

ACTTGCCTGCCGATAAGGCC

TNRC18 TATAAGGCTCTGCATGGCCG /

CCCGGCGCCGCCGTTGCACA

GAGCGGAAAGACAGGTCAGA

TRIM67 ACAAGCTGAGCTTGTCCGCG Regulation of cell differentiation

GGCAGCACGCGAACCCCGTT

CAGGCAGTACTCCATCAGTC

UBE2M TGATATCACACGTCTTGGGC Regulation of cell death

TGAGGCAGACGTTGCCCTCG

GCGCAGCTGCGGATCCAGAA

VAX1 GCCTTCCTCAAGGAGCCGCA Regulation of transcription from RNA polymerase II promoter; regulation of cell proliferation; regulation of cell differentiation CTCGGACGCCGAGGCTGCCC

CAGCCGATAGAGCTGCTCCG

YAP1 GACGTTCATCTGGGACAGCA Regulation of transcription from RNA polymerase II promoter; regulation of cell proliferation; regulation of cell death; regulation of cell differentiation; epithelium GGGGGCTGTGACGTTCATCT development TGGGGGCTGTGACGTTCATC

ZNF182 ATTCTTCTACCTCCAACTTG /

CTACATGCAGTACATTCAAA

CCACACTCTTTATAGCCATA

Negative CCTCGTTCACCGCCGTCGCG / control ATTTGCCGTTCGAATGCGTA

AAAGATAGGCCCTTCGCGGA

TCGACCGTAAAACCGGATAT

AAACTCACATGCTTGCGATC

ATACGCAAGCGTACCGACAA

TCGAAGGGTAACGAATCTA

AGTCCAAGGCGTTCGCGAAG

CCGTCAACGACGGCAAGCTG

TCATTGACGAATACCTCGAG

ACCGCGATAATCCTACAATT

AACGTCGACAGCGGGTACAC Cancer Biol Med Vol xx, No x Month 2021 43

Table S5 List of the top 10% protective/sensitizing gene pairs Protective gene pair r score in P value in Protective gene pair r score in P value in replicate 1 replicate 1 replicate 1 replicate 1 NF2 SPRED1 1.2378 1.58E-07 BCAR3 NF1 0.5706 6.38E-04

NF1 NF2 1.2333 2.69E-06 CUL3 TADA1 0.5632 8.58E-04

EZH2 NF2 1.0410 8.67E-06 CUL3 EAPP 0.5616 1.06E-03

CCNC NF2 1.0027 6.34E-06 RANBP9 SPRED1 0.5597 5.41E-04

NF2 TADA1 1.0023 8.82E-07 PCDH7 TAF5L 0.5566 6.38E-04

NF2 PCDH7 0.9780 4.59E-07 BCAR3 SPRED1 0.5564 1.94E-04

BCAR3 NF2 0.9682 5.36E-07 ARHGAP6 NF2 0.5533 4.17E-03

ARHGEF2 NF2 0.9222 4.14E-06 PARP3 RANBP9 0.5525 1.27E-03

MED12 NF2 0.9218 8.09E-06 ARHGAP6 BCAR3 0.5509 1.18E-03

NF2 TADA2B 0.9129 2.16E-06 MAP3K11 NF2 0.5473 5.26E-03

CUL3 NF2 0.8864 1.50E-05 NF2 PLK1 0.5468 1.69E-03

LPAR5 NF2 0.8787 2.32E-06 LPAR5 PCDH7 0.5391 7.92E-04

NF2 TAF5L 0.8658 2.59E-06 PCDH7 BIRC5 0.5362 7.71E-04

METTL21B NF2 0.8642 2.79E-06 BCAR3 EED 0.5343 6.74E-04

KDM5B NF2 0.8597 1.03E-05 NF1 RASSF5 0.5314 1.82E-03

EPAS1 NF2 0.8586 3.11E-05 NF2 P2RY8 0.5298 1.53E-03

NF2 TRIM67 0.8536 7.54E-06 DRAP1 PCDH7 0.5297 9.55E-04

KRAS NF2 0.8533 3.00E-06 EGFR NF1 0.5255 4.58E-03

NF2 TACC1 0.8426 8.37E-06 CUL3 TAF6L 0.5233 8.58E-04

NF2 PDCD10 0.8165 5.19E-05 NF2 UBE2M 0.5207 2.78E-03

EED NF2 0.8109 2.64E-05 MECOM NF2 0.5196 1.87E-03

DCLK1 NF2 0.8064 1.96E-05 RANBP9 RAP1GAP2 0.5196 1.27E-03

NF2 Pygo1 0.8013 2.82E-05 EphA2 NF2 0.5176 1.01E-03

NF2 TFAP2C 0.8003 4.78E-06 NF1 TFAP2C 0.5157 9.81E-04

EGFR NF2 0.7979 2.64E-05 NF2 RAP1GAP2 0.5155 2.10E-02

NF1 TAF5L 0.7859 1.31E-05 NF2 RBL1 0.5148 2.28E-03

NF2 ROCK1 0.7850 2.24E-05 NF1 TADA2B 0.5096 2.23E-03

NF2 PAK7 0.7783 7.29E-06 NF1 UBE2M 0.5084 3.53E-03

MED15 NF2 0.7782 4.57E-05 EphA2 PCDH7 0.5077 7.71E-04

EAPP NF2 0.7743 1.50E-05 DRAP1 NF2 0.5071 9.39E-03

MRFAP1L1 NF2 0.7709 1.35E-04 FSD1 NF1 0.4962 1.92E-03

BAP1 NF2 0.7679 1.22E-05 KRAS NF1 0.4946 5.96E-02

HMMR NF2 0.7673 1.66E-05 C9orf50 NF2 0.4942 6.31E-03

GPR35 NF2 0.7671 1.22E-05 CUL3 NF1 0.4926 2.60E-02

GAB2 NF2 0.7626 1.66E-05 CCDC101 RANBP9 0.4885 2.46E-03 44 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance

Table S5 Continued

Protective gene pair r score in P value in Protective gene pair r score in P value in replicate 1 replicate 1 replicate 1 replicate 1 NF2 YAP1 0.7620 2.82E-05 CCDC101 PCDH7 0.4880 2.07E-03

CHN2 NF2 0.7589 2.64E-05 BCAR3 ZNF182 0.4866 3.53E-03

CUL3 EZH2 0.7586 1.40E-05 CCNC RANBP9 0.4837 5.38E-03

CCDC101 NF2 0.7494 2.91E-05 BCAR3 TADA1 0.4834 2.99E-03

NF2 RASSF5 0.7480 4.87E-05 CUL3 LGR5 0.4828 2.40E-03

NF2 ZNF182 0.7426 2.91E-05 NF1 RANBP9 0.4812 2.17E-03

NF2 BIRC5 0.7422 1.22E-05 BCAR3 RANBP9 0.4760 4.27E-03

NF2 PODN 0.7410 1.44E-04 CUL3 UBE2M 0.4745 2.40E-03

CRB2 NF2 0.7384 5.53E-05 NF2 TNFRSF1B 0.4701 5.76E-03

HDGF NF2 0.7367 5.90E-05 MRFAP1L1 PCDH7 0.4677 6.03E-03

NF2 VAX1 0.7363 2.82E-05 BCAR3 GAB2 0.4622 3.62E-03

LAMA3 NF2 0.7262 2.12E-04 MED12 NF1 0.4532 5.90E-03

LCOR NF2 0.7260 1.96E-05 NF1 TACC1 0.4487 4.47E-03

MAPKAPK3 NF2 0.7246 3.54E-05 NF1 RHOG 0.4452 9.60E-03

FSD1 NF2 0.7237 8.06E-05 HDAC6 NF2 0.4410 1.56E-02

NGFR NF2 0.7147 4.87E-05 CUL3 SPRED1 0.4387 2.65E-02

NF1 PCDH7 0.7111 3.77E-05 BCAR3 Pygo1 0.4369 8.80E-03

HOXC13 NF2 0.7058 6.74E-04 BCAR3 CCDC101 0.4369 5.51E-03

NF2 TNRC18 0.7053 3.46E-04 PCDH7 RANBP9 0.4354 8.99E-03

NF2 TAF6L 0.7039 1.53E-04 CUL3 PARP3 0.4352 1.38E-02

NF2 RAPGEF1 0.7032 3.21E-05 BCAR3 UBE2M 0.4344 7.22E-03

NF1 PDCD10 0.7027 3.01E-05 PCDH7 TACC1 0.4337 1.02E-02

IGF1R NF2 0.7003 2.24E-05 NF1 BIRC5 0.4325 2.92E-03

NF2 RANBP9 0.6955 2.12E-04 MECOM PCDH7 0.4302 2.81E-02

LGR5 NF2 0.6842 6.48E-05 PDCD10 RANBP9 0.4301 1.19E-02

ITGB3 NF2 0.6838 1.44E-04 SPRED1 TAF5L 0.4288 4.91E-03

NF2 RHOG 0.6808 4.72E-05 EPAS1 PCDH7 0.4287 9.81E-03

HMMR PCDH7 0.6772 8.06E-05 PCDH7 PODN 0.4270 1.11E-02

KDM5B RANBP9 0.6756 2.55E-05 APBB1 NF2 0.4257 3.71E-02

CUL3 PCDH7 0.6748 1.88E-04 ARHGEF5 NF2 0.4224 6.90E-03

BCAR3 PCDH7 0.6709 5.36E-05 C9orf50 PCDH7 0.4170 1.14E-02

NF2 PRPF4B 0.6622 1.77E-04 ARHGAP6 TADA1 0.4153 1.24E-02

NF2 PRKRA 0.6520 3.00E-04 BAP1 NF1 0.4139 4.47E-03

ADI1 NF2 0.6463 5.88E-04 CUL3 Pygo1 0.4106 4.52E-02

CUL3 ZNF182 0.6426 2.78E-03 P2RY8 RANBP9 0.4088 2.65E-02 Cancer Biol Med Vol xx, No x Month 2021 45

Table S5 Continued

Protective gene pair r score in P value in Protective gene pair r score in P value in replicate 1 replicate 1 replicate 1 replicate 1 BCAR3 CUL3 0.6377 2.24E-04 CUL3 KDM5B 0.4088 2.19E-02

Jun NF2 0.6338 1.28E-04 BCAR3 BIRC5 0.4059 8.42E-03

ADI1 ARHGAP6 0.6171 8.84E-05 BCAR3 TAF5L 0.3987 1.09E-02

ARHGAP6 PCDH7 0.6092 3.45E-04 HOXC13 RANBP9 0.3961 4.36E-02

PCDH7 TNRC18 0.6080 1.24E-04 BCAR3 PDCD10 0.3957 1.49E-02

ADI1 RANBP9 0.6041 6.04E-04 PCDH7 PRPF4B 0.3922 1.87E-02

PCDH7 ZNF182 0.5942 2.38E-04 PCDH7 UBE2M 0.3908 2.23E-02

GPR98 NF2 0.5734 4.45E-04 NF2 PARP3 0.3870 6.59E-02

EZH2 NF1 0.5718 1.24E-02 CCNC PCDH7 0.3844 1.65E-02

CUL3 RANBP9 0.5716 2.18E-04

Sensitizing gene pair r score in P value in Sensitizing gene pair r score in P value in replicate 1 replicate 1 replicate 1 replicate 1 PRPF4B RAP1GAP2 −0.8391 3.54E-05 HOXC13 MECOM −0.5028 2.17E-03

GBP2 LAMA3 −0.7970 5.53E-05 IGF1R PARP3 −0.5015 7.71E-03

CRB2 RBL1 −0.7964 1.34E-03 EAPP METTL21B −0.5009 1.38E-03

ARHGEF2 C9orf50 −0.7725 9.70E-05 ARHGEF5 TNFRSF1B −0.5008 4.69E-03

RAP1GAP2 RBL1 −0.7538 3.86E-06 EphA2 METTL21B −0.4974 1.40E-02

EGFR PLK1 −0.7528 1.28E-04 ITGB3 RAP1GAP2 −0.4944 1.07E-02

GPR98 TAF6L −0.7287 2.67E-04 ARHGEF2 MED15 −0.4907 7.06E-03

NGFR MED15 −0.6944 7.71E-04 C9orf50 NGFR −0.4840 3.71E-02

BUB1B ITGB3 −0.6828 3.62E-03 GBP2 MAP3K11 −0.4827 3.14E-02

PRPF4B RHOG −0.6477 4.76E-02 ARHGEF5 GBP2 −0.4821 1.82E-03

HDGF LGR5 −0.6397 3.37E-03 EAPP RHOG −0.4770 5.63E-03

NGFR PRPF4B −0.6255 2.85E-03 CHN2 PODN −0.4701 5.38E-03

PARP3 VAX1 −0.6236 1.34E-03 HDAC6 PODN −0.4696 1.38E-02

PARP3 PODN −0.6228 6.31E-03 PRPF4B VAX1 −0.4687 1.14E-02

C9orf50 MED15 −0.6219 1.57E-03 MAPKAPK3 VAX1 −0.4671 6.17E-03

ADI1 P2RY8 −0.5932 6.38E-04 GPR98 MRFAP1L1 −0.4658 1.50E-01

NGFR MED12 −0.5902 1.27E-02 HMMR RAP1GAP2 −0.4657 4.93E-02

PRPF4B TNRC18 −0.5850 2.12E-03 ITGB3 PRKRA −0.4642 6.31E-03

C9orf50 ITGB3 −0.5840 4.27E-03 Pygo1 RAP1GAP2 −0.4624 4.17E-03

EAPP TAF6L −0.5825 2.71E-03 ITGB3 TNRC18 −0.4619 2.23E-02

IGF1R ITGB3 −0.5796 3.45E-03 GPR98 IGF1R −0.4571 7.27E-02

LAMA3 RAP1GAP2 −0.5737 3.21E-03 GBP2 GPR35 −0.4570 2.14E-02 46 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance

Table S5 Continued

Sensitizing gene pair r score in P value in Sensitizing gene pair r score in P value in replicate 1 replicate 1 replicate 1 replicate 1 BUB1B C9orf50 −0.5692 1.18E-03 PLK1 PRKRA −0.4488 8.24E-03

MECOM RBL1 −0.5596 1.27E-02 MAP3K11 PRPF4B −0.4479 1.98E-02

IGF1R PRPF4B −0.5593 2.85E-03 APBB1 ITGB3 −0.4449 9.60E-03

ARHGEF2 KRAS −0.5537 3.85E-02 ARHGEF2 Jun −0.4441 2.02E-02

ITGB3 Jun −0.5513 2.86E-02 ARHGEF2 LCOR −0.4431 1.40E-02

ARHGEF5 RAP1GAP2 −0.5479 1.92E-03 GPR98 PLK1 −0.4390 2.85E-03

ADI1 PRKCE −0.5336 8.61E-03 C9orf50 RAP1GAP2 −0.4328 9.38E-02

METTL21B PODN −0.5328 1.49E-02 METTL21B RAP1GAP2 −0.4308 1.25E-01

MED15 RAPGEF1 −0.5282 1.07E-02 ADI1 GBP2 −0.4289 1.87E-02

ARHGEF5 KRAS −0.5193 3.29E-03 GPR98 MED15 −0.4246 9.19E-03

GBP2 MECOM −0.5170 1.19E-02 PRPF4B RBL1 −0.4229 1.79E-02

MED15 TNRC18 −0.5083 2.41E-02 MED12 VAX1 −0.4089 1.65E-02

Table S6 List of strongly interacted protective/sensitizing gene pairs (data shown below were calculated using sample Replicate 1) Strongly interacted Score of Score of Score of P value GI score GI score (Loewe’s Interaction protective gene pair (A-B) gene pair gene A gene B (Linear fitting) additivity) type ADI1 ARHGAP6 0.6171 −0.1868 −0.0117 8.84E-05 0.559 0.4186 Synergistic

BUB1B RAPGEF1 0.4 −0.2735 −0.1869 3.64E-02 0.4882 −0.0604 Antagonistic

BUB1B YAP1 0.346 −0.2735 −0.2184 3.67E-01 0.4711 −0.1459 Antagonistic

BUB1B EAPP 0.263 −0.2735 −0.4878 1.44E-01 0.4691 −0.4983 Antagonistic

RANBP9 RAP1GAP2 0.5196 0.1274 −0.5324 1.27E-03 0.4677 0.1146 Synergistic

ARHGEF2 CCNC 0.2584 −0.3823 −0.3631 1.77E-01 0.4634 −0.487 Antagonistic

PARP3 RANBP9 0.5525 −0.6766 0.1274 1.27E-03 0.4601 0.0033 Synergistic

ARHGAP6 CRB2 0.4747 −0.0117 −0.346 3.71E-03 0.4214 0.117 Synergistic t

NF2 SPRED1 1.2378 0.6859 0.2327 1.58E-07 0.4191 0.3192 Synergistic

ADI1 RANBP9 0.6041 −0.1868 0.1274 6.04E-04 0.4128 0.5447 Synergistic

NF1 NF2 1.2333 0.3664 0.6859 2.69E-06 0.4112 0.181 Synergistic

HDAC6 ITGB3 0.2818 −0.0889 −0.2108 1.06E-01 0.4086 −0.0179 Antagonistic

IGF1R KRAS 0.2178 −0.2905 −0.1942 2.27E-01 0.4083 −0.2669 Antagonistic

ARHGEF2 MAP3K11 0.1864 −0.3823 −0.2784 2.45E-01 0.4071 −0.4743 Antagonistic

LAMA3 PRKRA 0.4808 −0.1123 0.0484 3.13E-03 0.3989 0.4169 Synergistic

EED PRKRA 0.4796 −0.2929 0.0484 6.03E-03 0.3794 0.2351 Synergistic

HOXC13 RAP1GAP2 0.1022 −0.1916 −0.5324 3.78E-01 0.3702 −0.6218 Antagonistic

NF1 PDCD10 0.7027 0.3664 −0.2751 3.01E-05 0.3661 0.6114 Synergistic

HOXC13 RAPGEF1 0.3506 −0.1916 −0.1869 6.92E-02 0.3655 −0.0279 Antagonistic Cancer Biol Med Vol xx, No x Month 2021 47

Table S6 Continued

Strongly interacted Score of Score of Score of P value GI score GI score (Loewe’s Interaction protective gene pair (A-B) gene pair gene A gene B (Linear fitting) additivity) type CCNC NF2 1.0027 −0.3631 0.6859 6.34E-06 0.3607 0.6799 Synergistic

CUL3 EZH2 0.7586 0.237 0.1242 1.40E-05 0.3577 0.3974 Synergistic

ARHGAP6 PARP3 0.2903 −0.0117 −0.6766 1.19E-01 0.356 −0.398 Antagonistic

ITGB3 TNFRSF1B 0.0462 −0.2108 −0.4666 7.41E-01 0.3543 −0.6312 Antagonistic

EZH2 PARP3 0.3828 0.1242 −0.6766 2.65E-02 0.3521 −0.1696 Antagonistic

GPR98 Pygo1 0.0055 −0.4642 −0.245 7.58E-01 0.3507 −0.7037 Antagonistic

HDAC6 HOXC13 0.3195 −0.0889 −0.1916 5.76E-02 0.3427 0.039 Synergistic

PRKCE UBE2M 0.2688 −0.1565 −0.1777 1.63E-01 0.3346 −0.0654 Antagonistic

ARHGEF2 TCEB2 0.26 −0.3823 0.0444 2.33E-01 0.333 −0.0779 Antagonistic

GBP2 HMMR 0.2354 −0.1525 −0.1555 1.23E-01 0.3314 −0.0726 Antagonistic

ADI1 APBB1 0.2406 −0.1868 −0.2178 8.40E-02 0.3205 −0.164 Antagonistic

P2RY8 RAPGEF1 0.2076 −0.2296 −0.1869 5.55E-01 0.3182 −0.2089 Antagonistic

EPAS1 EphA2 0.3336 −0.1449 0.0731 2.36E-02 0.3111 0.2618 Synergistic

CCNC UBE2M 0.2417 −0.3631 −0.1777 3.40E-01 0.3094 −0.2991 Antagonistic

PRKCE RAP1GAP2 −0.0144 −0.1565 −0.5324 7.92E-01 0.3067 −0.6745 Antagonistic

PLK1 TAF6L 0.1553 −0.2568 −0.1632 3.00E-01 0.3066 −0.2647 Antagonistic

EGFR NF2 0.7979 −0.1807 0.6859 2.64E-05 0.3011 0.2927 Synergistic

NGFR RAP1GAP2 −0.0859 −0.2402 −0.5324 4.33E-01 0.2988 −0.6867 Antagonistic

GPR98 HOXC13 0.0585 −0.4642 −0.1916 9.32E-01 0.294 −0.5973 Antagonistic

MED12 NF2 0.9218 −0.0491 0.6859 8.09E-06 0.2897 0.285 Synergistic

APBB1 RBL1 0.1863 −0.2178 −0.1266 3.83E-01 0.2729 −0.1581 Antagonistic

TFAP2C UBE2M 0.2464 −0.053 −0.1777 1.01E-01 0.2729 0.0157 Synergistic

EAPP HOXC13 0.1284 −0.4878 −0.1916 1.55E-01 0.2637 −0.551 Antagonistic

CCNC RANBP9 0.4837 −0.3631 0.1274 5.38E-03 0.2589 0.248 Synergistic

Strongly interacted Score of Score of Score of P value GI score GI score (Loewe’s Interaction sensitizing gene pair (A-B) gene pair gene A gene (Linear fitting) additivity) type GBP2 LAMA3 −0.797 −0.1525 −0.1123 5.53E-05 −0.6777 0.5322 Synergistic

EGFR PLK1 −0.7528 −0.1807 −0.2568 1.28E-04 −0.6345 0.3153 Synergistic

GPR98 ZNF182 −0.7592 −0.4642 −0.0667 2.24E-04 −0.5901 0.2283 Synergistic

HDGF LGR5 −0.6397 −0.0575 −0.2154 3.37E-03 −0.5871 0.3668 Synergistic

NGFR MED15 −0.6944 −0.2402 −0.0826 7.71E-04 −0.5732 0.3716 Synergistic

LPAR5 PARP3 −1.0143 −0.306 −0.6766 8.37E-06 −0.5652 0.0317 Synergistic

NGFR MED12 −0.5902 −0.2402 −0.0491 1.27E-02 −0.5631 0.3009 Synergistic

CUL3 PODN −0.3459 0.237 −0.0846 5.47E-02 −0.5486 0.1935 Synergistic 48 Yu et al. Synergistic interactions that overcome MAPK inhibitor resistance

Table S6 Continued

Strongly interacted Score of Score of Score of P value GI score GI score (Loewe’s Interaction sensitizing gene pair (A-B) gene pair gene A gene (Linear fitting) additivity) type PRPF4B TFAP2C −0.7627 −0.3752 −0.053 2.23E-03 −0.5335 0.3345 Synergistic

PRKRA YAP1 −0.4496 0.0484 −0.2184 4.68E-02 −0.5188 0.2796 Synergistic

MED15 TNRC18 −0.5083 −0.0826 −0.0969 2.41E-02 −0.5025 0.3288 Synergistic

MED15 RAPGEF1 −0.5282 −0.0826 −0.1869 1.07E-02 −0.5022 0.2587 Synergistic

BUB1B ITGB3 −0.6828 −0.2735 −0.2108 3.62E-03 −0.4725 0.1985 Synergistic

GPR98 TAF6L −0.7287 −0.4642 −0.1632 2.67E-04 −0.4468 0.1013 Synergistic

METTL21B PODN −0.5328 −0.0836 −0.0846 1.49E-02 −0.4412 0.3646 Synergistic

PLK1 PRKRA −0.4488 −0.2568 0.0484 8.24E-03 −0.4334 0.2404 Synergistic

ITGB3 PRKRA −0.4642 −0.2108 0.0484 6.31E-03 −0.4304 0.3018 Synergistic

HDAC6 PODN −0.4696 −0.0889 −0.0846 1.38E-02 −0.422 0.2961 Synergistic

ADI1 P2RY8 −0.5932 −0.1868 −0.2296 6.38E-04 −0.4139 0.1768 Synergistic

ADI1 PRKCE −0.5336 −0.1868 −0.1565 8.61E-03 −0.4134 0.1903 Synergistic

C9orf50 MED15 −0.6219 −0.3004 −0.0826 1.57E-03 −0.4089 0.2389 Synergistic

BAP1 EGFR −0.4587 −0.2036 −0.1807 8.99E-03 −0.4016 0.0744 Synergistic

RAP1GAP2 RBL1 −0.7538 −0.5324 −0.1266 3.86E-06 −0.3989 0.0948 Synergistic

MED12 TFAP2C −0.3342 −0.0491 −0.053 1.79E-02 −0.393 0.2321 Synergistic

EAPP TAF6L −0.5825 −0.4878 −0.1632 2.71E-03 −0.3919 −0.0685 Antagonistic

KRAS P2RY8 −0.5754 −0.1942 −0.2296 2.02E-02 −0.3917 0.1516 Synergistic

ITGB3 Jun −0.5513 −0.2108 −0.166 2.86E-02 −0.391 0.1745 Synergistic

MECOM RBL1 −0.5596 −0.2327 −0.1266 1.27E-02 −0.3875 0.2003 Synergistic

MED12 VAX1 −0.4089 −0.0491 −0.3051 1.65E-02 −0.3772 0.0547 Synergistic

GBP2 PRKRA −0.3831 −0.1525 0.0484 1.29E-02 −0.3564 0.279 Synergistic

PRPF4B TNRC18 −0.585 −0.3752 −0.0969 2.12E-03 −0.354 0.1129 Synergistic

HDAC6 YAP1 −0.3852 −0.0889 −0.2184 1.92E-01 −0.3535 0.0779 Synergistic

IGF1R ITGB3 −0.5796 −0.2905 −0.2108 3.45E-03 −0.342 0.0783 Synergistic

ITGB3 TNRC18 −0.4619 −0.2108 −0.0969 2.23E-02 −0.3286 0.1542 Synergistic

GBP2 MECOM −0.517 −0.1525 −0.2327 1.19E-02 −0.3196 0.1318 Synergistic

ARHGEF2 MED15 −0.4907 −0.3823 −0.0826 7.06E-03 −0.3191 0.0258 Synergistic

APBB1 ITGB3 −0.4449 −0.2178 −0.2108 9.60E-03 −0.3147 0.0163 Synergistic

DCLK1 EZH2 −0.0324 0.0796 0.1242 5.45E-01 −0.3146 −0.1714 Antagonistic

CHN2 PODN −0.4701 −0.3555 −0.0846 5.38E-03 −0.3136 0.03 Synergistic

ARHGEF5 KRAS −0.5193 −0.2593 −0.1942 3.29E-03 −0.3128 0.0658 Synergistic

PARP3 PODN −0.6228 −0.6766 −0.0846 6.31E-03 −0.3117 −0.1384 Antagonistic

PRPF4B RAP1GAP2 −0.8391 −0.3752 −0.5324 3.54E-05 −0.3074 −0.0685 Antagonistic

ITGB3 TAF6L −0.4514 −0.2108 −0.1632 4.06E-02 −0.3057 0.0774 Synergistic Cancer Biol Med Vol xx, No x Month 2021 49

Table S6 Continued

Strongly interacted Score of Score of Score of P value GI score GI score (Loewe’s Interaction sensitizing gene pair (A-B) gene pair gene A gene (Linear fitting) additivity) type EphA2 MRFAP1L −0.2778 0.0731 −0.1608 4.13E-02 −0.3022 0.1901 Synergistic

GBP2 MAP3K11 −0.4827 −0.1525 −0.2784 3.14E-02 −0.3009 0.0518 Synergistic

NGFR PRPF4B −0.6255 −0.2402 −0.3752 2.85E-03 −0.3 0.0101 Synergistic

RAPGEF1 TFAP2C −0.3292 −0.1869 −0.053 4.93E-02 −0.2991 0.0893 Synergistic

LAMA3 MAPKAPK3 −0.3801 −0.1123 −0.1867 1.87E-02 −0.299 0.0811 Synergistic

LAMA3 PODN −0.3477 −0.1123 −0.0846 7.15E-02 −0.2938 0.1508 Synergistic

EZH2 METTL21B −0.1172 0.1242 −0.0836 9.14E-01 −0.2867 0.0766 Synergistic

APBB1 NGFR −0.3974 −0.2178 −0.2402 1.72E-02 −0.2836 −0.0606 Antagonistic

MAPKAPK3 VAX1 −0.4671 −0.1867 −0.3051 6.17E-03 −0.2814 −0.0247 Antagonistic

EAPP METTL21B −0.5009 −0.4878 −0.0836 1.38E-03 −0.2799 −0.0705 Antagonistic

PRPF4B RHOG −0.6477 −0.3752 −0.3318 4.76E-02 −0.276 −0.0593 Antagonistic

BAP1 CUL3 −0.0696 −0.2036 0.237 4.82E-01 −0.2738 0.0362 Synergistic

GBP2 YAP1 −0.3912 −0.1525 −0.2184 4.13E-02 −0.2681 0.0203 Synergistic

ADI1 GBP2 −0.4289 −0.1868 −0.1525 1.87E-02 −0.268 0.0896 Synergistic

BUB1B C9orf50 −0.5692 −0.2735 −0.3004 1.18E-03 −0.2676 −0.0047 Antagonistic

GBP2 GPR35 −0.457 −0.1525 −0.1666 2.14E-02 −0.2675 0.1379 Synergistic

Table S7 Information of siRNAs used in the three-dimensional spheroid formation assay and subsequent mechanism dissection Item Sequence siLGR5-forward CUCUGAUGAUGUCGAAAAATT siLGR5-reverse UUUUUCGACAUCAUCAGAGTT siHDGF-forward CUAUGGAGGUGGAAAAGAATT siHDGF-reverse UUCUUUUCCACCUCCAUAGTT