Genomic and prognostic heterogeneity among RAS/BRAF V600E/TP53 co‐mutated resectable colorectal liver metastases

Colorectal cancer commonly metastasizes into multiple liver foci, but intermetastatic heterogeneity remains poorly described. Here, we demonstrate that mutations of RAS/BRAF/TP53 are homogeneous within patients, while DNA copy number aberrations can vary greatly. RAS/BRAF/TP53 co‐mutations conferred a dismal prognosis, and co‐mutations combined with a high burden and heterogeneity of copy number aberrations identified patients with the poorest outcome.


Abstract Background
Hepatic resection is potentially curative for patients with colorectal liver metastases, but the treatment bene t varies. KRAS/NRAS (RAS)/TP53 co-mutations are associated with a poor prognosis after resection, but there is large variation in patient outcome within the mutation groups, and genetic testing is currently not used to determine the likely bene t from surgery. We have investigated the potential for improved prognostic strati cation by combined biomarker analysis with DNA copy number aberrations (CNAs) in a tumor heterogeneity setting.

Methods
We determined the mutation status of KRAS, NRAS, BRAF, and TP53 in 441 liver lesions from 171 patients treated by partial hepatectomy for metastatic colorectal cancer. CNAs were pro led in 232 tumors from 67 of the patients.

Results
Driver mutations in KRAS, NRAS, BRAF V600E , and TP53, and high-level ampli cations affecting cancercritical genes such as ERBB2 and EGFR, were predominantly homogeneous within patients.
Conclusions DNA copy number pro ling identi ed genomic and prognostic heterogeneity among patients with resectable colorectal liver metastases with co-mutated RAS/BRAF V600E /TP53.

Background
Approximately 30% of all colorectal cancer (CRC) patients develop metastases to the liver during their disease course, of whom 20% undergo hepatic resection as a potentially curable treatment (1,2). In a Norwegian study, the ve-year overall and disease free survival was 46% and 24%, respectively, after partial hepatectomy (3), compared to a 5-year relative survival rate of 15-22% for patients with distant metastases from CRC overall (4). Around one third of the patients experience early recurrence following resection (3,5,6), and there are currently no strong markers for prediction of long-term bene t from surgery.
Mutations in RAS have consistently been associated with a poor prognosis among patients with resectable colorectal liver metastases (CRLM) (7)(8)(9), and it has been suggested that surgical treatment is less bene cial in patients with RAS mutations (10). However, the prognostic effect size is modest (11) and it was recently proposed that the effect is limited to tumors with co-occurring TP53 mutations (12,13), or co-occurring TP53 and SMAD4 mutations (14). BRAF V600E mutations have a stronger prognostic effect size, but the prognostic value is limited by the low prevalence of this marker among patients with resectable CRLM (15). CRLM commonly presents with multiple distinct liver lesions. Cancer-critical genes with a high mutation prevalence in CRC generally have a homogenous mutation pattern across metastatic lesions from the same patient (16,17), although treatment pressure may cause subclonal expansion, as illustrated by the emergence of resistant subclones with pre-existing or acquired KRAS mutations during anti-EGFR therapy (18,19). More extensive mutation heterogeneity has been demonstrated in other protein-coding genes, both in intra-tumor and inter-tumor comparisons (20)(21)(22). We have previously shown that there is considerable intra-patient inter-metastatic heterogeneity also on the DNA copy number level (23). The clinical impact of such inter-metastatic molecular heterogeneity remains poorly de ned (20,21,24), although our study suggested that a high degree of heterogeneity of DNA copy number aberrations (CNAs) are associated with a poor prognosis. We have previously also reported differential radiological responses to standard neoadjuvant treatment among metastatic lesions in a subgroup of approximately 10% of patients with resectable CRLM (25). How this phenotypic heterogeneity relates to molecular heterogeneity is currently not clear, but the poor survival rate of this patient subgroup after surgery highlights the potential clinical importance of inter-metastatic heterogeneity.
Here, we performed combined biomarker analyses in a tumor heterogeneity setting and in relation to outcome among patients with resectable CRLM. We investigated mutations in KRAS, NRAS, BRAF V600E , and TP53, combined with the overall burden and inter-metastatic heterogeneity of CNAs.

Patient samples
The study included 460 liver metastases from 176 patients who underwent resection for CRLM at Oslo University Hospital, Oslo, Norway, between October 2013 and February 2018. Five patients were excluded due to mucinous tumor tissue, poor DNA quality, or suspicion of low tumor cell content. In total, 441 liver metastases from 171 patients were included for mutation analyses (Fig. 1), of which 102 patients had multiple lesions analyzed (median of 3 metastatic lesions per patient, range 2-9).

Mutation and microsatellite instability analyses
A total of 355 metastatic tumor samples from 103 patients have previously been analyzed for hotspot mutations in BRAF exon 15 and KRAS and NRAS exons 2-4 by Sanger sequencing (17). The remaining 86 tumor samples and 68 patients were analyzed in the present study.
All 441 tumor samples were also sequenced for all coding regions of TP53 (exons 2-11). In summary, three singleplex PCR reactions were used to analyze TP53 exons 2-4, 5-6 and 7-9, respectively, by amplifying 50 ng of DNA in a reaction mix containing 10x HotStar-buffer, dNTP, HotStar Taq polymerase (Qiagen), and the primers described in Supplementary Table 1. TP53 exons 10 and 11 were analyzed in a separate multiplex PCR reaction by ampli cation of 50 ng of DNA using the 2x Multiplex PCR-kit (Qiagen). PCR products were puri ed using Illustra ExoProStar 1-step (GE Healthcare, Chicago, Illinois, U.S.A.), and the Applied Biosystems BigDye Terminator v1.1 Cycle Sequencing Kit and Applied Biosystems 3730 DNA Analyzer were used for sequencing (both Thermo Fisher Scienti c). DNA from the blood of two healthy donors was used as controls. The results were analyzed using Applied Biosystems Sequencing Analysis software v5.3.1 and SeqScape software v2.5 (Thermo Fisher Scienti c) and scored independently by two investigators. All mutations were validated in independent PCR reactions, and synonymous mutations were not called as mutations.
Five cases with intra-patient mutation heterogeneity of TP53 among metastatic lesions were re-analyzed. Heterogeneous cases were rst veri ed by a second Sanger sequencing, and two of the patients were subsequently re-analyzed also with ultra-deep targeted sequencing with the Illumina TruSight Tumor 15 gene panel as described in (17). High-sensitivity analyses discon rmed TP53 mutation heterogeneity in one patient. The other patient had inter-metastatic heterogeneity in the speci c loci affected by TP53 mutation, and displayed p.Asp184fs mutations in two lesions and p.Arg273His mutations in three lesions (all ve lesions had the same KRAS mutation). The three patients for whom high-sensitivity data were not available were classi ed as mutated.
All tumors were analyzed for microsatellite instability (MSI) status using PCR-based marker analyses, either as previously described using BAT25/BAT26 (26), or using the ve markers incorporated in the MSI Analysis System version 1.2 (Promega, Fitchburg, WI, USA). Uncertain cases after analyses of BAT25/BAT26 were re-analyzed with the MSI Analysis System. All patients had microsatellite stable (MSS) tumors except for one. DNA copy number analyses A total of 232 lesions from the rst 67 patients with multiple metastases sampled were analyzed by genome-wide DNA copy number pro ling using the Applied Biosystems CytoScanHD array (Thermo Fisher Scienti c). The procedure was conducted according to the manufacturer's instructions, following the CytoScan Assay Manual Protocol. Resulting raw-intensity CEL-les were pre-processed with the R package rawcopy (v1.1) (27), and subsequently segmented by ASCAT (v2.5) (28), with penalty parameter set to 25 and chromosomes X and Y excluded. A primary interest was to estimate the level of CNA heterogeneity among samples from the same patient. This estimate is highly sensitive to poor data quality, and strict quality control was therefore performed on the segmented data by careful visual inspection of copy number pro les and Sunrise plots produced by ASCAT. Samples with non-aberrant pro les (no/few CNAs) or poor Sunrise plots were excluded, retaining 192 lesions from 64 patients for further analyses. Copy number gain and loss was called for segments with ≥ 1 or ≤-1 copies relative to the median genome-wide copy number estimated by ASCAT, respectively. For comparison with the data processing approach used in our previous study (23), the pre-processed data from rawcopy was additionally segmented by the PCF algorithm implemented in the R package copynumber (29) with the penalty parameter (gamma) set to 100.
To enable analyses across samples, the sample-wise segmented data were further split into their smallest genomic regions of overlap by computationally introducing breakpoints at every unique breakpoint occurring in any sample in the total dataset.
The COSMIC Cancer Gene Census version 86 (30) was used to de ne cancer-critical genes, (both Tier 1 and Tier 2 genes considered). Of the 719 genes, 672 were covered in the CNA data.
A sample-wise estimate of the overall CNA burden was calculated as the fraction of the genome (per cent of base pairs) with aberrant copy number. For patients with multiple lesions, the mean CNA burden was used for patient-wise analyses. Estimates of ploidy were derived from ASCAT.
Unpublished DNA copy number data was available for three matching primary tumors for comparison of ampli cation status in the metastases.

Estimation of intra-patient inter-metastatic copy number heterogeneity
High-quality DNA copy number data were available for at least two metastatic lesions from 48 patients ( Fig. 1), including a total of 176 tumors and a median of 4 tumors per patient (range 2-8). These were analyzed for intra-patient inter-metastatic DNA copy number heterogeneity by three different approaches. First, the genome-wide matrix of estimated copy numbers was used to perform pairwise comparisons among metastatic lesions from each patient based on Euclidean distances, using the dist function implemented in the R stats package. To obtain one heterogeneity measure per patient, the mean Euclidean distance of all pairwise comparisons was calculated, in accordance with the approach used in our previous study (23). Second, the pairwise distance was calculated as in the rst approach but using Pearson correlation-based distance. Third, CNA heterogeneity was assessed by a gene-wise estimation (protein-coding genes from UCSC known genes) of the fraction of CNAs within a patient that were not common across the lesions, i.e. genes with aberrant copy number in one or more lesions but not in all. The heterogeneity-calling was more conservative with this approach, as only events exceeding the copy number gain/loss thresholds were considered heterogeneous, while genes consistently affected by gain (or loss) but with varying amplitudes were regarded as homogeneous CNA events.
The patient-wise CNA heterogeneity measure was categorized as high or low relative to the median across the patients.
For robustness, data segmented with the PCF algorithm was used to estimate copy number heterogeneity (distance-based) in the same manner as in our previous study (23), by calculating the average pair-wise Euclidean distance between DNA segments with a variance of > 0.03 among samples from each patient.

Statistical analyses
Pairwise comparisons of variables between groups were done by non-parametric Wilcoxon rank-sum tests for continuous variables and with Fisher's exact test for categorical data, both implemented in the R stats package.
Survival analyses were performed with 5-year cancer-speci c survival (5y-CSS) as the end point. Time to death from CRC was measured from start of treatment (either neoadjuvant systemic treatment or surgery) and deaths from other causes were censored (31). Only patients with MSS cancers and R0 or R1 status in the liver after resection were included in survival analyses (n = 165 of 171 patients in the full cohort, n = 62 of 64 patients in CNA burden analyses, n = 46 of 48 in CNA heterogeneity analyses). Kaplan-Meier estimates and log rank tests were used for comparisons of variables with only two groups, using the survdiff function in the R survival package. For comparisons of more than two groups, log rank tests for trend were performed using the comp function in the R survMisc package. All Kaplan-Meier plots were made with the ggsurvplot function in the R survminer package. Univariable and multivariable Cox regression analyses were performed with the coxph function in the R survival package.

Results
Concordant driver gene mutation pro les among multiple resected CRLM Among resected CRLM from 171 patients, the patient-wise mutation prevalence was 42 and BRAF V600E mutations were mutually exclusive, while RAS/BRAF V600E co-occurred with TP53 mutations in 31% of the patients (Fig. 2a). The mutation status of all four genes was homogeneous in all metastatic deposits analyzed from each patient when ultra-deep targeted sequencing was applied (three patients had uncon rmed heterogeneity; Fig. 2b). DNA copy number pro ling indicated larger intra-patient inter-metastatic variation in these genes, and heterogenous DNA copy number status was found in 16/48 patients (33%) for BRAF, 24/48 patients (50%) for KRAS, 19/48 patients (40%) for NRAS, and 9/48 (19%) for TP53 (Supplementary Fig. 1b). However, this was associated with a larger genome-wide level of CNA heterogeneity in the same patients ( Supplementary Fig. 1c), indicating that these four genes were not speci cally targeted.
None of the four genes had any high-level ampli cations events (≤ 6 additional copies), but a genomewide search identi ed high-level ampli cations (≥ 15 additional copies) in CRLM from 22 (34%) of the 64 patients analyzed. Among cancer-critical genes (de ned in the COSMIC Cancer Gene Census), recurrent high-level ampli cations were found only of ERBB2 in two patients, while EGFR and the cell cycle genes CDK6, CCND2, and CCND3 were ampli ed in one patient each (Table 2). Additionally, the nominated target TOX3 (32) was ampli ed in one patient. The ampli cation events were commonly concordant in intrapatient inter-metastatic comparisons, albeit with variation in the amplitude (Fig. 2b). Among patients with multiple metastases analyzed, 31% of high-level ampli cations were homogenous, and an additional 38% of the ampli cations had lower-amplitude gains (≥ 5 additional copies) in all other metastases from the same patient. Corresponding numbers for ampli cation events affecting cancer-critical genes were 50% and 25% (Supplementary Fig. 2). Notably, intra-patient concordance was also found for the clinically relevant target genes ERBB2 and EGFR, including in comparison with the primary tumor of one patient with ERBB2 ampli cation (9 additional copies in the primary and 17 and 18 additional copies in the two metastases). For the patient with two CRLM with CCND2 ampli cations (29 and 47 additional copies), the primary tumor had 32 additional copies of this gene. The patient with CCND3 ampli cation in the range of 14-16 additional copies in all 7 metastases did not have a detectable CCND3 ampli cation in the primary tumor. Frequent inter-metastatic DNA copy number heterogeneity on the genome-wide scale The genome-wide CNA frequencies, summarized patient-wise, were in accordance with the well-known aberration pro les of CRC (among 192 lesions from 64 patients; Supplementary Fig. 3). Frequent copy number gains were found on chromosome arms 7p and q, 8q, 13q and 20q, and copy number losses on 1p, 4p and q, 8p, 17p and 18p and q. For the 48 patients with multiple metastases analyzed, intermetastatic CNA heterogeneity was estimated by three different approaches (Methods) and with three different sets of input data of varying width of genomic coverage (across the whole genome, from protein coding genes, or from only the subset of 672 cancer-critical genes). The different estimates were strongly correlated, indicating robustness to both the approach (Spearman's rho ≥ 0.63, p < 0.001) and to the width of genomic coverage (Spearman's rho ≥ 0.93, p < 0.001; Supplementary Fig. 4). Further analyses were performed using the genome-wide Euclidean distance-derived heterogeneity measure, consistent with our previous study (23). There was a large variation among patients in the degree of inter-metastatic CNA heterogeneity (Fig. 3a). This CNA heterogeneity was independent of the number of lesions analyzed per patient, the patient-wise median aberrant cell fraction and the RAS/BRAF V600E mutation status, but was correlated with the patient-wise median ploidy state and ploidy range, and the TP53 mutation status (Fig. 3b and c; Supplementary Table 2). The CNA heterogeneity score was also weakly correlated with the mean patient-wise CNA burden of the metastases (analyzed as the fraction of the genome with aberrant copy numbers; Spearman's rho 0.33, p = 0.02; Fig. 3c).
Co-mutated RAS/BRAF V600E and TP53 is associated with poor patient outcome The 165 patients with MSS cancers that were treated with R0 or R1 hepatic resection had a median cancer-speci c survival of 48 months and a 5y-CSS rate of 40%. The 139 patients with R0 or R1 status overall had a median cancer-speci c survival of 50 months and a 5y-CSS rate of 44%. Several clinicopathological factors (Table 1) were associated with poor patient outcome in univariable Cox regression analysis, and gender, size of the largest metastasis, R-status in the liver and presence of extrahepatic disease remained signi cant in multivariable analyses (Supplementary Table 3). for wild-type, p = 0.1; Fig. 4). Co-mutations of RAS/BRAF V600E and TP53 had a strong prognostic impact, with a 5y-CSS of 25%, compared to 46% for patients with RAS mutation only, 42% for TP53 mutation only, and 71% in patients with wild-type status for all four genes (p = 0.001, test for trend, Fig. 4). Co-mutated RAS/BRAF V600E and TP53 was not signi cant compared to patients with RAS mutations only (p = 0.2).
The prognostic role of co-mutations was not driven by patients with BRAF V600E mutations, as the analyses remained signi cant upon exclusion of 3 patients with BRAF V600E mutations ( Supplementary   Fig. 5a). Co-mutations of RAS/BRAF V600E and TP53 were enriched in patients with a right-sided primary tumor location and with extrahepatic metastases and depleted among patients with positive nodal status and those receiving neoadjuvant anti-EGFR or VEGF treatment (Supplementary Table 4). However, comutation was signi cant in multivariable analyses including clinicopathological factors (Supplementary Table 3).

Genome-wide CNA pro les have poor prognostic associations
Two measures of the CNA pro les of the CRLM were analyzed for prognostic associations among patients with MSS cancers and R0/R1 resection: the genome-wide CNA burden (the mean across lesions for patients with multiple CRLM analyzed) and the intra-patient inter-metastatic CNA heterogeneity estimate. Both these patient-wise CNA measures were categorized into a high and low group relative to the respective median in the patient series. High CNA heterogeneity or CNA burden was not overrepresented according to any of the clinical variables listed in Table 1. A high overall CNA burden was signi cantly associated with a poor 5y-CSS rate, with survival rates of 15% and 44% in the high and low group, respectively (p = 0.02; Fig. 5). CNA burden was also signi cantly associated with a poor patient outcome when analyzed as a continuous variable (HR 1.03, 95% CI 1.01-1.05, p = 0.009). Furthermore, patients with high intra-patient inter-metastatic CNA heterogeneity also had a poorer survival rate than patients with a low heterogeneity, although not statistically signi cant in this smaller patient subgroup (5y-CSS of 23% and 37%, respectively, p = 0.2; Fig. 5). The combination of a high CNA burden and a high CNA heterogeneity was associated with a particularly poor patient outcome, and patients in this subgroup had a 5 year-CSS rate of 9%, compared to a 30% among patients with only one of the variables high and 50% among patients low for both CNA measures (p = 0.02, test for trend; Fig. 5). The median survival rates in the three groups were 25, 36 and 50 months, respectively.
Combined biomarker analyses suggest potential for strati cation of the RAS/BRAF V600E /TP53 mutated subgroup by CNA pro les Both CNA heterogeneity and CNA burden were signi cantly higher in patients with TP53 mutated compared to wild-type tumors, but the CNA estimates were not associated with RAS/BRAF V600E mutation status. Furthermore, there was a substantial variation in the CNA estimates within the mutational subgroups (Fig. 3b), motivating us to analyze the different prognostic biomarkers individually and combined. Within the RAS/BRAF V600E mutated subgroup, the 5y CSS was 15% in patients with a high level of inter-metastatic CNA heterogeneity versus 42% in patients with low CNA heterogeneity (p = 0.08).
Similarly, the 5y CSS was 0% in the RAS/BRAF V600E mutated patients with a high CNA burden versus 44% in RAS/BRAF V600E mutated patients with a low CNA burden (p = 0.02; Fig. 6a). Prognostic strati cation of the TP53 mutated subgroup by either of the CNA estimates was not statistically signi cant (p ≥ 0.2; Supplementary Fig. 5b). The triple combination of co-mutation in RAS/BRAF V600E and TP53 and high inter-metastatic CNA heterogeneity was associated with a worse 5y-CSS compared with comutations/low heterogeneity and the remaining patients (p = 0.02 for analysis of trend among the three groups; Fig. 6b). Similar strati cation of patients with co-mutations by the CNA burden showed a prognostic association also for patients with a triple combination of co-mutations and high CNA burden (p = 0.01, test for trend; Fig. 6b). Both associations were also supported by univariable Cox regression analyses (Supplementary Table 5). Similar results were found when excluding patients with extrahepatic metastases from the analyses (Supplementary Fig. 5c).

Discussion
Intra-patient molecular heterogeneity is anticipated to have clinical implications (33), and current evidence in metastatic CRC suggests that heterogeneity on the DNA copy number level is more widespread than heterogeneity of single nucleotide variants (SNVs) and small insertions/deletions (indels), at least in cancer critical genes (21,23,24). We have shown that mutations in KRAS, NRAS, BRAF V600E (17), and TP53 are predominantly homogeneously present among multiple resected CRLM from each patient. The DNA copy number states of the four genes were more heterogeneous among metastases and correlated with the genome-wide inter-metastatic CNA heterogeneity, consistent with a lower selection pressure for these genes on the DNA copy number level than on the point mutation level. Furthermore, high-level ampli cations targeting cancer-critical genes, including the therapeutic targets ERBB2 and EGFR, were also typically homogeneously present within patients, both among multiple metastatic lesions and in the primary tumor. The timing of cancer-critical ampli cations is poorly studied in CRC, and our results suggest that driver amplicons commonly arise before metastatic dissemination. In contrast, the level of genome-wide inter-metastatic DNA copy number heterogeneity beyond ampli cation events varied substantially among patients. There was no enrichment or depletion of cancer-related genes among genomic regions with heterogeneous DNA copy number, suggesting that CNA heterogeneity is a genome-wide and target-ignorant characteristic.
There is an urgent clinical need for markers that can identify CRLM patients who likely have limited bene t from surgery in combination with presently recommended perioperative oncological treatment.
BRAF V600E and RAS mutations are the molecular markers with best documented prognostic value, but their use in selection of patients for hepatectomy is currently not supported. BRAF V600E has been shown to have the strongest prognostic effect size, but a low prevalence of only 3-5% among patients with resectable CRLM (15), and < 2% in this study. RAS mutations identify a larger patient subgroup, but have weaker prognostic value, which suggests molecular heterogeneity among patients with RAS-mutated cancers. In primary CRC, the prognostic value of RAS has been suggested to be limited to MSS cancers and to depend on the consensus molecular subtypes (34). In patients with resectable CRLM, the prognostic value may depend on co-occurring TP53 mutations (12,13) or TP53/SMAD4 mutations (14).
Here, we con rm the potential for improved prognostic strati cation of CRLM after hepatic resection by RAS/BRAF V600E and TP53 co-mutations. Furthermore, we nd that high inter-metastatic genomic heterogeneity confers poor outcome within the RAS mutated subgroup and show a potential for further prognostic strati cation of the RAS/BRAF V600E and TP53 co-mutated subgroup by combined analyses with genome-wide CNA pro les. Although CNA burden and the level of CNA heterogeneity were independent of RAS mutation status, patients with TP53-mutated tumors had more extensive intermetastatic CNA heterogeneity and a higher CNA burden than patients with wild-type tumors, suggesting a confounding prognostic effect. Loss of normal TP53-expression has previously been associated with tolerability to aneuploidy (35)(36)(37)(38)(39)(40)(41), and it is conceivable that TP53 mutations are needed for a submissive state that allows extensive copy number heterogeneity to evolve. The CNA heterogeneity alone tended to be associated with poor outcome, while patients with higher than median CNA burden had signi cantly worse cancer-speci c survival than patients with low CNA burden. The latter is in line with a recent pancancer study describing a signi cant association between a high CNA burden and poor disease-speci c survival in patients with metastatic cancer (42). Our study was not powered to conclude on the independent prognostic value of CNA heterogeneity and TP53 mutations in patients with RAS-mutated CRLM, although there was a signi cant trend for poorer patient survival in the RAS/BRAF V600E /TP53 comutated/high CNA heterogeneity group versus co-mutated/low heterogeneity versus remaining patients.
In accordance with the recent report from the Vauthey group (12), multivariable analysis with clinicopathological variables con rmed that co-mutation of RAS/BRAF V600E and TP53 was independently associated with poor patient survival.
It has been debated whether the association between residual disease and outcome may re ect underlying cancer biology, as mutated RAS is associated with both a positive resection margin and early development of lung metastases (8,9,43). However, excluding the patients with extra-hepatic metastases did not impact on the prognostic associations found in this study.

Conclusions
We have described genomic heterogeneity on the DNA copy number level in patients with resectable CRLM, also within patient subgroups de ned by RAS/BRAF V600E and TP53 mutations. By combined biomarker analyses we con rm the superior prognostic value of RAS/BRAF V600E and TP53 co-mutations compared with either mutation alone. Furthermore, a high level of intra-patient inter-metastatic CNA

Consent for publication
Not applicable.

Availability of data and materials
The datasets supporting the conclusions of this article can be obtained from the authors upon reasonable request.

Competing interests
The authors declare no competing interests. Overview of the included patients and samples in the study.  average CNA burden (proportion of the genome with aberrant copy number), the patient-wise median and range of ploidy states among the metastases, and RAS/BRAFV600E and TP53 mutation status. b) CNA heterogeneity was signi cantly associated with TP53, but not RAS/BRAFV600E mutation status (n=42/n=6 TP53 mutated/wild-type; n=29/n=19 RAS/BRAFV600E mutated/wild-type). TP53 mutation was also associated with higher CNA burden, while RAS/BRAFV600E mutations were associated with Page 24/26 lower CNA burden (n=51/n=13 TP53 mutated/wild-type; n=42/n=22 RAS/BRAFV600E mutated/wildtype). The CNA estimates still varied within the mutational subgroups, with inter-quartile range between 27 and 65 for CNA heterogeneity (Euclidean distance) and 10-33 for CNA burden (%). c) CNA heterogeneity assessed as the mean Euclidean distance was not correlated with the number of lesions analyzed, and only weakly to the overall CNA burden.

Figure 4
Five-year CSS according to mutation status. *Log rank test for trend. For pair-wise comparisons, RAS/BRAFV600E/TP35 co-mutation was associated with signi cantly worse survival than double wildtype (p=0.006) and TP53 mutation only (p=0.01), but not compared to RAS/BRAFV600E mutations only (p=0.2). Wt = wild-type.

Figure 5
Five-year CSS according to CNA burden (left), CNA heterogeneity (middle) and both measures combined (right).