Clinical and genomic data of patients with colorectal cancer were extracted from the colorectal cancer cohort of the Cancer Genome Atlas (TCGA) [12]. The cohort includes a total of 594 patients, of whom 534 patients were profiled for mutations. TCGA used whole exome sequencing for the genomic analyses. In addition to mutations, TCGA provided analyses of copy number alterations and structural variants. Single nucleotide mutation calling was conducted with input from various pipelines [16]. Copy number alterations were analyzed in TCGA studies with the GISTIC (Genomic Identification of Significant Targets in Cancer) algorithm, which assigns a score of 2 or above in genes with putative amplification [17]. For quantification of chromosomal instability (CIN), TCGA used a score (aneuploidy score, AS) which was calculated in each sample as the sum of the number of chromosome arms in the sample that had copy number gains or losses. For the calculation of AS, chromosome arms were defined as copy number altered if more than 80% of their length contained somatic copy number alterations. In contrast, chromosome arms with somatic copy number alterations extending from 20% to 80% of their length were considered indeterminate and chromosome arms with somatic copy number alterations in less than 20% of their length were considered not altered. For the calculation of the AS from Affymetrix 6.0 SNP arrays an algorithm called ABSOLUTE was used in TCGA [18].

The Cancer Cell Line Encyclopedia (CCLE) is an extensive collection of cancer cell lines procured from various agencies and analyzed using whole exome sequencing [19]. Similar to TCGA, the determination of copy number alterations in the CCLE collection was performed with the GISTIC algorithm [17]. Data on drug sensitivity of cell lines from colorectal cancer with APC and other WNT/β-catenin pathway mutations were obtained from the Genomics of Drug Sensitivity in Cancer (GDSC) dataset (http://www.cancerrxgene.org) [20]. The GDSC project contains two experimental datasets, GDSC1 and GDSC2, that differ in the experimental conditions used. GDSC1 experiments were performed between 2009 and 2015. These experiments used media alone in the negative control cell lines not exposed to drugs. The most recent dataset GDSC2 performed after 2015 and used in the current study employed media with vehicle (DMSO, dimethylsulfoxide) in the negative controls. Dependencies on knock down or knock out of specific genes in colorectal cancer cell lines with APC and other WNT/β-catenin pathway mutations were obtained from the DepMap portal that contains data from RNA interference (RNAi) arrays and CRISPR (Clustered Regularly Interspaced Short Palindromic Repeats) arrays of cell lines from the CCLE collection [21, 22]. CRISPR and RNAi arrays identify essential genes that are important for the survival of assayed cell lines and, as a result, the absence of these essential genes has a significant survival and proliferation effect in vitro [23–25]. The genes and dependencies discovered with the two array methodologies do not completely overlap, as the depth of suppression differs. RNAi experiments in DepMap were from project Achilles which used the DEMETER algorithm for analysis [24]. CRISPR arrays in DepMap were from project SCORE containing 323 cancer cell lines from various cancers and a library of 18,009 targeted genes [26]. Computational modeling of experiments in SCORE was performed with the CERES and the CHRONOS algorithms [27, 28].

Interrogation of both TCGA cohort and colorectal cancer cell lines at the individual case and cell line level was performed online at the cBioPortal for Cancer Genomics platform (cBioportal, http://www.cbioportal.org). cBioportal is a site containing genomic and associated clinical data from publicly available studies, maintained by Memorial Sloan Kettering Cancer Center (MSKCC) and other academic institutions [29, 30]. Two groups of alterations in the WNT/β-catenin pathway, the first consisting of APC alterations and the second consisting of alterations in three other WNT/β-catenin pathway genes (RNF43, CTNNB1, and TCF7L2) were considered in the TCGA cohort and groups of patients were constructed in a 2 by 2 manner resulting in a total of four groups.

Statistical comparisons of categorical and continuous data were carried out with the Fisher exact test or the χ²-test and the t-test or ANOVA. All statistical comparisons were considered significant if P < 0.05.

The frequently altered colorectal cancer tumor suppressor, APC showed alterations (most commonly mutations and a small number of deletions) in 395 of the 534 cases (74%) in the colorectal cancer cohort of TCGA with mutation data. Of the APC altered cases, 95 cases (17.8% of the entire cohort) possessed also concomitant alterations (mostly mutations and a small number of amplifications or deletions) in one or more of three other WNT/β-catenin pathway genes, RNF43, CTNNB1, and TCF7L2 (termed quadruple altered cohort) and 300 cases (56.2% of the entire cohort) had no alterations in these three genes (termed APC only altered cohort). Colorectal cancers without APC alterations constituted 26% of the cases in the colorectal cancer TCGA cohort and among them, 44 cases (8.2% of the entire cohort) had alterations in RNF43, CTNNB1, or TCF7L2 (termed triple altered cohort) and 95 cases (17.8% of the entire cohort) had no alterations in RNF43, CTNNB1, and TCF7L2 (quadruple wild type cohort). The four groups did not differ significantly in their average age, percentage of older (above 65 years old) patients, the distribution of sexes, or the stage of colorectal cancer at diagnosis (Table 1). The majority of cancers in the triple altered cohort (93.2%) were located in the colon and only 6.8% of those cancers were rectal, while in the three other groups, 25% to 30% of the cases were rectal (χ²-test P = 0.01, Table 1). Significant differences were also observed between the groups in their genomic characteristics (Table 2). The triple altered group had a majority (60%) of microsatellite instability (MSI) cases, which were less frequent in the other cohorts (22% in the quadruple altered and quadruple wild type cohorts and only 3.2% in the APC only altered cohort). Chromosome instability (CIN), on the other hand, was more prevalent in the APC only altered cohort (83% of cases), followed by the quadruple wild type cohort (66% of cases) and the quadruple altered cohort (53.5% of cases), while the triple altered group presented the lowest CIN prevalence (35% of cases). Consistent with the genomic sub-type prevalence, the triple altered group had the highest frequency of high tumor mutation burden (TMB) (above 10 mutations/Mb) and the highest frequency of low AS and FGA scores compared with the three other groups (Table 2).

Mutations in the gene encoding for tumor suppressor p53, TP53, were more frequent in the two groups with APC alterations (54.7% of cases in the quadruple altered group and 71% of cases in the APC only altered group) compared with the two groups without APC alterations (43.2% of cases in the triple altered group and 31.6% of cases in the quadruple wild type group, χ²-test P < 0.00001, Figure 1). Mutations in oncogene KRAS, and in the less prevalent NRAS, were also more frequent in the two groups with APC alterations (χ²-test P < 0.00001 for KRAS, but not reaching significance, P = 0.14, for NRAS). BRAF mutations were significantly more prevalent in the triple altered group, with 47.7% of cases showing mutations, compared with 4% to 16.8% of cases in the other groups (χ²-test P < 0.00001, Figure 1). PIK3CA mutations were more prevalent in the two groups with RNF43, CTNNB1, or TCF7L2 alterations (χ²-test P = 0.0003).

The genes encoding for receptor tyrosine kinases, including members of the EGFR family, members of the FGFR family, RET, ALK, and NTRK3 were significantly more frequently mutated in the two groups with RNF43, CTNNB1, or TCF7L2 alterations (Table 3 and Figure 2). Moreover, several proteins of the KRAS/BRAF/MEK and the PI3K/AKT/mTOR pathways had a higher prevalence in the same groups (Table 3 and Figure 3).

Consistent with the higher prevalence of MSI and high TMB, the two groups with RNF43, CTNNB1, or TCF7L2 alterations possessed statistically significant higher mutation rates of genes encoding for MMR-associated proteins and proofreading polymerase epsilon (Table 3 and Figure 4). The quadruple wild type group showed a low rate of mutations in these genes, despite containing 22% of MSI high cases, suggesting that epigenetic modifications such as MLH1 promoter methylation may be the responsible alterations involved in producing MMR defects in this group.

DNA damage response genes, including BRCA1, BRCA2, ATM, and CDK12 were more frequently mutated in the two groups with RNF43, CTNNB1, or TCF7L2 alterations, both with (quadruple altered group) and without (triple altered group) APC alterations compared with the two groups without RNF43, CTNNB1, or TCF7L2 alterations (Table 3 and Figure 5). ATM mutations had the highest prevalence of mutations in the two RNF43, CTNNB1, or TCF7L2 altered groups with about one-fourth of the cases being mutated (Figure 5).

In addition to RNF43, CTNNB1, and TCF7L2 alterations, the quadruple altered and triple altered groups displayed more frequent mutations in other components of the WNT/β-catenin/APC pathway, such as LRP1B, AXIN2, and the atypical cadherins FAT1 and FAT4 (Table 3 and Figure 6). LRP1B mutations were present in one-fourth of the cases, while FAT1 and FAT4 had an even higher prevalence of mutations in these groups.

Similar to colorectal cancer patient samples, APC mutations were the most common genomic alterations of the WNT/β-catenin pathway in the colorectal cancer cell line cohort of the CCLE collection consisting of 84 colorectal cancer cell lines. The three other genes of the pathway that are frequently mutated in colorectal cancer samples, RNF43, CTNNB1, and TCF7L2, were also frequently mutated in colorectal cancer cell lines, although less frequently than APC. Among APC mutated cell lines, 11 cell lines possessed pathogenic mutations in both APC and one or more of the three other genes of the pathway (the quadruple mutant group, Table 4). The cell lines with pathogenic APC mutations in the absence of mutations in RNF43, CTNNB1, or TCF7L2 formed the APC only mutated group consisting of 13 cell lines. Nine colorectal cancer cell lines had mutations in one or more of the three other genes, in the absence of APC mutations (triple mutated cell lines, Table 4). A fourth group consisting of four cell lines were wild type for all four frequently mutated WNT/β-catenin pathway genes (APC, RNF43, CTNNB1, and TCF7L2: quadruple wild type cell lines) (Table 4).

Although data from some cell lines in the quadruple mutated group were not available, all cell lines of the group with data available were MSI high and had a high TMB (Table 4). All but one cell line was diploid and most had a low CIN as measured by the FGA. In contrast, and consistent with patient samples, all but one cell line in the group with APC only mutations were microsatellite stability (MSS), and had low TMB and high CIN. The group of cell lines with non-APC WNT/β-catenin pathway gene mutations (triple mutated group) was comprised of both MSS and MSI cell lines and a corresponding mixture of high and low TMB (Table 4). Lastly, the quadruple wild type group was comprised of cell lines with low TMB and MSS status. Pathogenic TP53 mutations were present in 8 of the 11 (82.7%) cell lines of the quadruple mutated group, in 8 of the 13 (61.5%) cell lines of the APC only mutated group, in 6 of the 9 (66.7%) cell lines of the triple mutated group and in 1 of the 4 (25%) cell lines of the quadruple wild type group (Table 5). Classic codon 12 and 13 pathogenic KRAS mutations were present in 4 of the 11 (36.4%) quadruple mutated cell lines (one other cell line of the group had a mutation of unknown significance at codon 140). Pathogenic KRAS mutations were present in 7 of 13 (53.8%) cell lines of the APC only mutated group, in 5 of 9 (55.6%) cell lines of the triple mutated group, and in 1 of the 4 (25%) cell lines of the quadruple wild type group (Table 5). No cell lines in the four groups possessed mutations in the homologous NRAS gene, but 2 cell lines from the quadruple mutated group had homo-deletions of the NRAS locus at chromosome 1p13.2, an alteration that is rarely observed in colorectal cancer patient samples, with a prevalence of 0.7% in the colorectal cohort from TCGA.

V600E or other pathogenic BRAF mutations were observed in 3 of 11 (27.3%) quadruple mutated cell lines, in 3 of 13 (23.1%) cell lines of the APC only mutated group, in 2 of 9 (22.2%) cell lines of the triple mutated group, and no cell lines of the quadruple wild type group (Table 5). Overall, the colorectal cancer cell lines in the four groups capture to a significant degree, albeit with some variability, the corresponding prevalence of the pathogenic mutations in TP53, KRAS, and BRAF in patient samples (Figure 7). For example, pathogenic BRAF mutations were more frequent in cell lines of the quadruple altered and APC only altered groups than the corresponding patient samples, while in the triple altered group, where pathogenic BRAF mutations were most prevalent in patient samples, corresponding cell lines had a lower frequency of BRAF mutations (Figure 7).

Comparison of the sensitivities of the four groups of colorectal cancer cell lines to representative porcupine inhibitors, tankyrase inhibitors, and EGFR inhibitors was performed using data from the GDSC project. Although the small number of in vitro assayed cell lines prevented a formal statistical comparison from reaching significance, the two quadruple cell lines with data available seemed to be more sensitive to two of the three porcupine inhibitors tested, all three tankyrase inhibitors with data and to be more sensitive to the EGFR inhibitor erlotinib than cell lines from the other groups (Table 6).

RNAi and CRISPR arrays showed that knock out or knock down of at least one of the four WNT/β-catenin pathway genes of interest (APC, RNF43, CTNNB1, and TCF7L2) in all but one of the cell lines of the quadruple mutated group produced significant growth delay gene effect (Z-score < –2.5). Most cell lines of the group were sensitive to the knock out or knock down of CTNNB1 (Tables 7 and 8). In the group of APC only altered colorectal cancer cell lines, knock out or knock down of APC had in general no significant effect. In contrast, growth delay was observed with RNAi knock down and even more with CRISPR knock out of CTNNB1 and TCF7L2, which are downstream of the pathway defect in these cells (Tables 7 and 8). Some cell lines of the triple altered group were also sensitive to knock out or knock down of CTNNB1 and TCF7L2, but others were resistant to all four genes knock out or knock down. Data for quadruple wild type cell lines were limited and were available for one cell line for RNAi and three cell lines for CRISPR (Tables 7 and 8). Cell line LS513 displayed sensitivity to knock out and knock down of TCF7L2 (Z-score < –2.5), while cell lines C10 and C99 appeared to be resistant to knock down of all four genes.