Identification of synthetic chemosensitivity genes paired with BRAF for BRAF/MAPK inhibitors

Molecular-targeted approaches are important for personalised cancer treatment, which requires knowledge regarding drug target specificity. Here, we used the synthetic lethality concept to identify candidate gene pairs with synergistic effects on drug responses. A synergistic chemo-sensitivity response was identified if a drug had a significantly lower half-maximal inhibitory concentration (IC50) in cell lines with a pair of mutated genes compared with those in other cell lines (wild-type or one mutated gene). Among significantly damaging mutations in the Genomics of Drug Sensitivity in Cancer database, we found 580 candidate synergistic chemo-sensitivity interaction sets for 456 genes and 54 commercial drugs. Clustering analyses according to drug/gene and drug/tissue interactions showed that BRAF/MAPK inhibitors clustered together; 11 partner genes for BRAF were identified. The combined effects of these partners on IC50 values were significant for both drug-specific and drug-combined comparisons. Survival analysis using The Cancer Genome Atlas data showed that patients who had mutated gene pairs in synergistic interaction sets had longer overall survival compared with that in patients with other mutation profiles. Overall, this analysis demonstrated that synergistic drug-responsive gene pairs could be successfully used as predictive markers of drug sensitivity and patient survival, offering new targets for personalised medicine.


Supplementary
| Swarm plot of IC50 of BRAF related MM group cell lines. A, B, and C represents IC50 of dabrafenib, refametinib, and trametinib, respectively. Cell lines are separated based on number of mutated αC genes which represent multi-drug combined partner genes. Color mapping and Euclidian distance was used for hierarchical clustering. Only ERK/MAPK signaling pathway targeting drug kept being grouped.
Supplementary Figure S4 | Checking burden effect of α genes and αC genes with BRAF in ERK/MAPK pathway targeting drug. Statistical test was processed for both α gene group(drug specific, A, C, E) and αC gene group(combined, B, D, F) base separated cell line groups; BRAF wild, M0, M1, M2. All statistical test were Student's t-test. Cell lines of BRAF wild group have wild type BRAF. Cell lines of M0 group have mutated BRAF and wild type α or αC genes. Cell lines of M1 group has mutated BRAF and one mutated α or αC gene. Cell lines of M2 group has mutated BRAF and two or more mutated α or αC genes. α gene group: Drug specific group of α genes paired with BRAF; αC gene group: Combined α genes for all three ERK/MAPK pathway targeting drugs. Figure S5 | Melanoma-specific analysis for the nine partner genes in BRAF-related SCI sets. Student's t-test was processed for both drug specific gene group (drug specific, A, C, E) and multi-drug combined gene group (combined, B, D, F) base separated cell line groups. The numbers represent p values.

Supplementary Figure S6 | Gender-specific survival analysis of pan-cancer patients in the TCGA
In this analysis, we only included melanoma of 85 males (left) and 56 females (right). Patients were separated according to the mutation profiles of BRAF and GPR112 (ADGRG4). Figure S7 | Analysis pipeline to identify candidate SCI sets. To identify gene pairs showing synergistic chemotherapeutic interactions with a specific drug, we compared the IC50 values of the MM group and the other three groups using the Student t-test. Gene pairs with significantly lower IC50 values in MM were further filtered according to the following two criteria: 1) mean IC50 value of MM groups was the smallest among the four groups, and 2) the maximum IC50 value of the MM group was smaller than the median IC50 value of the WW group. Figure S8 | Detailed flowchart of data process for damaging mutation data. GDSC's preprocessed whole exome sequencing (WES) data and copy number variant (CNV) data were used to filter and extract damaging mutation data. Damaging missense mutation, loss of function (LoF) mutation, copy number deletion were assembled for damaging mutation data. SIFT and PolyPhen-2 score mapping was processed at WES data for filtering damaging missense mutation. Then variants were filtered for damaging (PolyPhen-2) and deleterious (SIFT) mutations. Finally, missense mutations were only remaining as damaging missense data. Stop loss and nonsense mutation were defined as LoF mutations. Cell linegene mutation table were made for all three mutation data groups and merged into single table. After merging, gene filtering based on number of mutated cell lines were processed and final damaging mutation data table was made.

Supplementary
Supplementary Table S1 | Gene-cell line damaging mutation profile binary table (row: cell lines, n = 990; column: genes, n = 19100). We made the table which contains the mutation profile of 19,100 genes for 990 cell lines. The numbers represent the combined count of mutations including loss of function mutation, damaging missense and copy number deletions.
Supplementary Table S2 | Detailed information of final SCI sets. Total 580 SCI sets' detailed information is shown. In this table, name of drug and genes, number of cell lines of separated 4 groups, mean / SD values of IC50 for 4 groups, p-value and FDR value is included for each SCI sets. Mean, standard deviation, minimum and maximum value of gene pairs, genes and cell lines per drug of final SCS interaction sets are shown.

Mean / SD Min Max
Gene (3.1%) † In MM group, four patients used several drugs. In WM group, two patients used several drugs. ‡ Patients belong to MM and MW group were all skin cutaneous melanoma. In WM group, there were eight types of cancers (hepatocellular carcinoma, 13 cases; kidney cancer including papillary, clear cell and kidney chromophobe, 5 cases; and one case for each adrenocortical carcinoma, lung squamous cell carcinoma, sarcoma, skin cutaneous melanoma). In WW group, there were nine types of cancers (liver hepatocellular carcinoma, 15 cases; kidney cancer including papillary, clear cell and kidney chromophobe, 10 cases; and one case for each brain lower grade glioma, colon adenocarcinoma, sarcoma, glioblastoma multiforme, uterine corpus endometrial carcinoma). MM represent both BRAF and multi-drug combined partner genes mutated patient group. MW represent BRAF mutated and wild type partner genes. WM represent wild type BRAF and mutated type partner genes. WW represent both BRAF and partner genes wild type. * Described as a BRAF inhibitor or GSK BRAF inhibitor.  We compared the mutational profile of well-known resistance genes including BRAF amplification between the BRAF-related SCS groups that we identified from the previous analysis in the GDSC database. In this comparison, the partner genes were eleven genes derived from the clustering analysis including all three BRAF-inhibitors. * The front capital character denotes the BRAF gene and the latter of the two capital characters denotes the mutational status of the multi-drug combined partner genes. † Only cell lines tested with any of dabrafenib, refametinib, and trametinib are included and 936 cell lines were able to check CNV data.

MM (n = 11)
1 (9.1%) -1 (9.1%) -1 (9.1%) -1 (9.1%) -1 (9.1%) -1 (9.1%) 6 (54.5%) We compared the mutational profile of well-known resistance genes including BRAF amplification between the BRAF-related SCS groups' patients in the TCGA database. In this comparison, the partner genes were eleven genes derived from the clustering analysis including all three BRAF-inhibitors from the previously derived from the analysis of GDSC. Only 62 patients were able to check CNV data. * The front capital character denotes the BRAF gene and the latter of the two capital characters denotes the mutational status of the multi-drug combined partner genes.

SCS: synergistic chemo-sensitivity
Supplementary We compared the mutational profile of well-known resistance genes including BRAF amplification between the BRAF-related SCS groups' patients in the TCGA database. In this comparison, the partner genes were ten genes except TTN derived from the clustering analysis including all three BRAF-inhibitors from the previously derived from the analysis of GDSC. Only 62 patients were able to check CNV data.