Skip to main content
NCBI home page
eLife logoLink to eLife
. 2018 Sep 4;7:e37294. doi: 10.7554/eLife.37294

Tumor copy number alteration burden is a pan-cancer prognostic factor associated with recurrence and death

Haley Hieronymus 1, Rajmohan Murali 2, Amy Tin 3, Kamlesh Yadav 4, Wassim Abida 1,5, Henrik Moller 6, Daniel Berney 7, Howard Scher 5,8, Brett Carver 9, Peter Scardino 9, Nikolaus Schultz 10, Barry Taylor 1,3,10, Andrew Vickers 3, Jack Cuzick 11, Charles L Sawyers 1,12,✉
Editors: Michael R Green13, Jeffrey Settleman14
PMCID: PMC6145837  PMID: 30178746

Abstract

The level of copy number alteration (CNA), termed CNA burden, in the tumor genome is associated with recurrence of primary prostate cancer. Whether CNA burden is associated with prostate cancer survival or outcomes in other cancers is unknown. We analyzed the CNA landscape of conservatively treated prostate cancer in a biopsy and transurethral resection cohort, reflecting an increasingly common treatment approach. We find that CNA burden is prognostic for cancer-specific death, independent of standard clinical prognosticators. More broadly, we find CNA burden is significantly associated with disease-free and overall survival in primary breast, endometrial, renal clear cell, thyroid, and colorectal cancer in TCGA cohorts. To assess clinical applicability, we validated these findings in an independent pan-cancer cohort of patients whose tumors were sequenced using a clinically-certified next generation sequencing assay (MSK-IMPACT), where prognostic value varied based on cancer type. This prognostic association was affected by incorporating tumor purity in some cohorts. Overall, CNA burden of primary and metastatic tumors is a prognostic factor, potentially modulated by sample purity and measurable by current clinical sequencing.

Research organism: Human

eLife digest

Cancer cells carry different types of mutations that are associated with the cell starting to multiply uncontrollably. Certain changes only affect one or a few letters of the genetic code. Others, known as copy number alterations, or CNA, involve larger portions of the genome that can either be lost (deletions) or duplicated (amplifications). Tumors in different patients carry variable amounts of these deletions or amplifications, which together are known as the CNA burden.

New technologies allow scientists to scan the genomes of tumors and examine the type of mutations present in each patient. The results can help to decide on the best course of action. For example, in prostate cancer, patients whose tumors have a high CNA burden are at greater risk of relapse after treatment. However, it has been unclear whether these people also have lower survival rates, and if CNA burden can predict outcome of other types of cancers.

Hieronymus et al. conducted genetic analyses on over a hundred samples from prostate cancer patients who were not treated with surgery or radiation. The results showed that a higher CNA burden in the tumors is correlated with more deaths due to the disease. The findings in prostate cancer were also true across different types of cancers. These conclusions also emerged when Hieronymus et al. then looked at genomic data obtained from patients with various cancers using a different DNA sequencing test, which is certified for clinical use. This demonstrates that CNA burden could be a useful marker in clinical settings to help assess risk in cancer patients.

Introduction

Somatic copy number alterations (CNAs) are nearly ubiquitous in cancer (Zack et al., 2013; Heitzer et al., 2016) and alter a greater portion of the cancer genome than any other type of somatic genetic alteration (Heitzer et al., 2016). Different cancer types vary in their balance of copy number alterations to somatic point mutations, with prostate cancer having relatively high rates of CNA compared to point mutation. Given the prevalence of CNAs in cancer, significant effort has been directed towards identifying specific CNAs associated with cancer clinical characteristics and prognosis as well as the potential driver genes they contain (Liang et al., 2016; Wang et al., 2016; Nibourel et al., 2017). There are well demonstrated associations between specific CNAs and CNA signatures to cancer state and characteristics (Visakorpi et al., 1995; Williams et al., 2014; Taylor et al., 2010). CNV patterns or clusters have been associated with high Gleason prostate cancer (Gleason 8 + compared to Gleason 6–7 [Williams et al., 2014]) and recurrent disease (compared to primary [Visakorpi et al., 1995; Cancer Genome Atlas Research Network, 2015; Viswanathan et al., 2018]). Nonetheless, most CNAs are large, (Zack et al., 2013; Beroukhim et al., 2010) and their associations with cancer outcome may not be well identified by gene-specific approaches. Increasing evidence indicates that large CNAs harbor multiple drivers (Tschaharganeh et al., 2016; Liu et al., 2016), emphasizing the need to study their biological and clinical significance beyond individual gene-focused standpoints.

The CNA burden of a tumor is the degree to which a tumor's genome is altered as a percentage of genome length and represents a fundamental measure of genome copy number alteration level. As such, tumor CNA burden, rather than individual CNAs, may be associated with cancer outcomes. While tumor mutational burden (TMB) predicts response to immunotherapy across multiple cancer types (Bergerot et al., 2018; Goodman et al., 2017), tumor CNA burden may be prognostic for outcomes such as recurrence and survival. Indeed, we and others have previously found CNA burden and genome-wide CNA patterns to be associated with biochemical recurrence and metastasis in primary prostate cancer, the most common cancer in men, across multiple cohorts (Taylor et al., 2010; Hieronymus et al., 2014; Camacho et al., 2017). This prognostic significance of tumor CNA burden extends to low and intermediate risk prostate cancer (Gleason scores of 7 and less) (Hieronymus et al., 2014) and has the potential to better stratify risk in patients who are considering conservative treatment approaches such as active surveillance to reduce overtreatment (Chen et al., 2016; Tosoian et al., 2016).

In addition to questions about the prognostic potential and overall landscape of CNA in conservatively treated prostate cancer, it is unknown whether CNA burden is prognostic for prostate cancer survival, rather than only recurrence and metastasis. Nor is it known whether the prognostic significance of tumor CNA burden extends to other cancer types. Here we set out to address these questions, as well as whether tumor CNA burden can be prognostic in a clinical practice setting, including (i) in cancers treated conservatively rather than through immediate surgery or radiation, (ii) in biopsy or resection samples, and (iii) using a clinical targeted sequencing that allows rapid and cost-effective measurement of tumor CNA burden.

To address these questions, we first examine the genomic CNA landscape of conservatively treated prostate cancer in more than a hundred diagnostic biopsy and resection specimens from a conservatively treated cohort; this cohort consisted of patients with localized prostate who were not treated with surgery or radiation within six months of diagnosis. We demonstrate that tumor CNA burden is associated with cancer-specific death, independent of standard clinical predictors. To explore the prognostic significance of tumor CNA burden more broadly in other cancer types, we find that tumor CNA burden is also associated with disease-free and overall survival in TCGA cohorts of primary breast, endometrial, renal clear cell, thyroid, and colorectal cancer in addition to prostate cancer, with the degree of association varying in some cancer types. We then establish the clinical feasibility of measuring tumor CNA burden using the FDA-cleared MSK-IMPACT clinical next generation sequencing (NGS) assay and confirm that tumor CNA burden is associated with overall and disease-specific survival in both primary and metastatic tumors across cancer types. In all, we demonstrate that tumor CNA burden is a prognostic factor associated with cancer recurrence and death in multiple cancer types, including in conservatively treated prostate cancer which would benefit from increased risk stratification.

Results

The genomic copy number landscape of conservatively treated prostate cancer

To explore the genomic copy number landscape of conservatively treated prostate cancer, we set out to analyze copy number alteration (CNA) in cancer obtained non-surgically through biopsy and transurethral prostate resection (TURP) using a widely studied, conservatively treated primary prostate cancer cohort (Cuzick et al., 2006). This retrospective Transatlantic Prostate Group 1 (TAPG1) cohort (n = 1675) consists of men below age 76 with clinically localized prostate cancer and prostate-specific antigen (PSA) below 100 ng/ml who did not receive surgery or radiation within 6 months of diagnosis (Cuzick et al., 2006). This population-based cohort was drawn from six cancer registries in Great Britain, and the majority of the cohort was followed without treatment, while a subset received hormonal therapy. The original diagnostic samples, either biopsy or TURP, were obtained and centrally reviewed to obtain consistent pathological evaluation to the current standards. Drawing from this cohort, we carried out genome-wide CNA analysis by array-based comparative genomic hybridization (aCGH) of 107 biopsies or TURP samples from the TAPG1 cohort, as tissue availability is limited for much of the full cohort. The subset of cases used for CNA analysis, which make up our conservative treatment CNA cohort, have similar clinical characteristics to the full TAPG1 cohort, including median diagnosis age, baseline PSA, hormonal treatment, and clinical stage, with the exception of higher Gleason score distribution, likely due to selection for cases with sufficient DNA for analysis (Supplementary file 2). As expected for a cohort not subject to PSA screening, the patients are older and have higher grade at diagnosis than is typical for contemporary US cohorts. Among the cohort, 47 patients developed metastasis and 43 died of prostate cancer. The median follow-up time for survivors was 10.3 years from diagnosis.

The copy number alteration landscape of the conservative treatment cohort revealed canonical copy number alterations of prostate cancer, including gain of chromosome 8q and losses on chromosomes 6 p, 8 p, 13q and 16 p, though with lower frequency than seen in prostate cancer cohorts analyzed by our group (MSKCC cohort) (Taylor et al., 2010) and TCGA (Cancer Genome Atlas Research Network, 2015) (Figure 1a). The percentage of the cancer genome showing copy number changes, termed tumor CNA burden (TCB), is similar between the conservative treatment CNA cohort and other cohorts (Figure 1b), with a mean tumor CNA burden of 5.7% (median 1.5%, IQR 0.05–8.5%) compared to 5.2% (median 3.0%, IQR 0.04–6.9%) for the 2010 MSKCC primary prostate cancer cohort (Taylor et al., 2010) and 4.0% (median 0.7%, IQR 0.08–5.1%) for the 2014 MSKCC primary prostate cancer cohort (Hieronymus et al., 2014). The tumor CNA burden of the conservative treatment CNA cohort is, however, somewhat lower than the 8.7% average tumor CNA burden of the TCGA prostate cohort (Cancer Genome Atlas Research Network, 2015) (mean 8.7%, median 6.2%, IQR 1.7–11.9%).

Figure 1. Tumor copy number landscape of conservatively treated primary prostate cancer, compared to other primary prostate cancer cohorts.

Figure 1.

Open in a new tab

(a) Heat map of copy number alterations in conservative treatment CNA cohort, as well as TCGA, MSKCC, and IMPACT primary prostate cancer cohorts. (b) Frequency distribution of CNA burden, as log of percentage of genome copy number altered, for the conservative treatment prostate cancer cohort and three other primary prostate cancer cohorts.

Tumor CNA burden is prognostic for prostate cancer-specific death

Since tumor CNA burden is associated with prostate cancer recurrence and metastasis in prostatectomy cohorts (Taylor et al., 2010; Hieronymus et al., 2014), we sought to determine whether tumor CNA burden was prognostic for cancer-specific death in biopsies of conservatively treated prostate cancer. In our conservative treatment CNA cohort, we find that tumor CNA burden as a continuous variable is significantly associated with prostate cancer-specific death (per 5% tumor CNA burden, HR 1.49; 95% CI 1.30, 1.70; p<0.0001; Table 1). Greater tumor CNA burden correlates with an increase in death from disease compared to a lower tumor CNA burden (Figure 2a). The risk of death due to prostate cancer within 5 years of diagnosis increases with tumor CNA burden over the majority of the tumor CNA burden distribution (Figure 2b). For example, the 5 year risk of death due to prostate cancer would be 13% for patients with a 2% tumor CNA burden and 28% for patients with a 10% tumor CNA burden (Figure 2b). Tumor CNA burden may therefore serve as a prognostic factor for cancer-specific death in patients who undergo increasingly common conservative treatment approaches.

Table 1. Tumor CNA burden is associated with prostate cancer-specific death in conservative treatment cohort independent of Gleason sum score and CAPRA score.

Cox Regression model assessing the association between CNA burden (per 5%) and cancer specific survival. N = 107*

Model HR 95% CI P-value
Univariate, tumor CNA burden 1.49 1.30, 1.70 <0.0001
Multivariable – adjusting for Gleason sum (≤6, 7, ≥8) 1.44 1.24, 1.67 <0.0001
Multivariable – adjusting for UCSF-CAPRA score utilizing multiple imputation 1.44 1.24, 1.68 <0.0001
Multivariable – adjusting for UCSF-CAPRA score without utilizing multiple imputation
* N = 60 (excludes 47 patients with unknown stage)		 1.57 1.29, 1.92 <0.0001
Open in a new tab

Figure 2. Tumor copy number alteration burden is associated with death from prostate cancer in conservatively treated patients.

(a) Cumulative Incidence of death from disease (dashed lines) and death from other causes (solid lines) based in cases with high CNA burden (red lines, CNA Burden greater than or equal to the median CNA burden of this cohort, 1.48) or non-high CNA burden (black lines, CNA Burden < median). (b) Risk for death from prostate cancer within 5 years of diagnosis. Univariate risk for 5 year prostate cancer-specific death, calculated by locally weighted Kaplan–Meier estimates (solid black line) with 95% confidence interval (dashed black lines) overlaid on the distribution of CNA burden (gray). (c) Association of tumor CNA burden with available cancer outcomes in the conservative treatment primary prostate cancer TAPG1 cohort, TCGA primary cancer cohorts, and the MSK-IMPACT clinical sequencing prostate and pan-cancer cohorts of primary and metastatic tumors. Forest plot of hazard ratio (per 5% CNA burden) with 95% confidence interval shown for cancer-specific mortality (dark blue), overall mortality (light blue), and cancer recurrence (green). Supplementary Tables and Figures.

Figure 2.

Figure 2—figure supplement 1. Kaplan-Meier plot of biochemical recurrence in TCGA primary prostate cohort.

Figure 2—figure supplement 1.

The highest quartile tumor CNA burden (above 75 percentile CNA burden, green) is compared to lower three quartiles (blue) with risk table showing the number of patients present at each time point.
Figure 2—figure supplement 2. Tumor CNA burden in multiple cancers is associated with disease free survival and overall survival.

Figure 2—figure supplement 2.

Kaplan-Meier plot of disease free survival (left) and overall survival (right) of TCGA cohorts of (a) endometrial cancer and (b) colorectal cancer. The highest quartile CNA (above 75 percentile CNA burden, green) is compared to lower three quartiles (blue).
Figure 2—figure supplement 3. Correlation between CNA burden from IMPACT targeted sequencing assay and whole exome sequencing (WES) of same samples, pan-cancer.

Figure 2—figure supplement 3.

The relationship between CNA burden determined by IMPACT targeted sequencing and WES in a subset of pan-cancer IMPACT cohort samples analyzed by both approaches (n = 1005) is shown (rho = 0.88, p-value=0).
Figure 2—figure supplement 4. Tumor CNA burden in primary prostate cancer is prognostic for overall survival when assayed by clinically approved sequencing panel.

Figure 2—figure supplement 4.

Kaplan-Meier plot of overall survival of IMPACT primary prostate cancer cohort by CNA burden quartile in (a) primary and (b) metastatic tumors. The highest quartile CNA (above 75 percentile CNA burden, green) is compared to lower three quartiles (blue).
Figure 2—figure supplement 5. Forest Plot of Hazard Ratios (individual and pooled) for meta-analysis assessing the association between tumor CNA burden and overall survival in (a) primary cancer and (b) patients with metastatic cancer in the pan-cancer IMPACT cohort.

Figure 2—figure supplement 5.

Open in a new tab

We next asked whether tumor CNA burden was associated with outcome after adjusting for established prognostic variables, including Gleason sum score and the UCSF Cancer of the Prostate Risk Assessment (CAPRA) score (Cooperberg et al., 2005; Brajtbord et al., 2017) which combines PSA, Gleason score, percentage positive biopsy cores, clinical stage, and age (Figure 2c). Tumor CNA burden is significantly associated with cancer-specific death even after adjusting for biopsy Gleason score (per 5% tumor CNA burden, HR 1.44; 95% CI 1.24, 1.67; p<0.0001) or CAPRA score (per 5% tumor CNA burden, HR 1.44; 95% CI 1.24, 1.68; p<0.0001) (Table 1, Figure 2c). The addition of tumor CNA burden into the model with the CAPRA score increased Harrell’s concordance index from 0.756 to 0.805 for cancer-specific survival in our cohort of men with conservatively treated prostate cancer.

Tumor CNA burden is prognostic for cancer-free and overall survival in multiple cancer types

Large, clinically annotated cancer genomic efforts such as TCGA now provide an opportunity to examine whether CNA burden is prognostic for primary cancer outcomes across many cancer types. In the TCGA primary prostate cancer cohort (Cancer Genome Atlas Research Network, 2015), tumor CNA burden is significantly associated with biochemical recurrence individually (p<0.0001; per 5% tumor CNA burden, HR = 1.27; 95% CI, 1.13, 1.42) and after adjustment for Gleason score and mutation burden (p=0.015; per 5% tumor CNA burden, HR = 1.18; 95% CI, 1.03, 1.35), validating our findings from other prostate cancer cohorts (Figure 2c, Figure 2—figure supplement 1, Table 2). There were insufficient deaths in this cohort to analyze survival. CNA burden was still significantly associated with biochemical recurrence after adjusting for tumor sample purity determined by ABSOLUTE (p<0.003; per 5% CNA burden, HR = 1.22; 95% CI, 1.07, 1.40; Table 2). Since tumor CNA burden could potentially reflect simply the prognostic significance of aneuploidy as determined by cytometric DNA index in various cancers (Walther et al., 2008; Danielsen et al., 2016), we examined the tumor CNA burden in a multivariable model together with ploidy. Ploidy, generated by CLONET and previously published for this cohort, estimates the average DNA index of the tumor cells (Carter et al., 2012; Prandi et al., 2014). Tumor CNA burden was associated with recurrence independent of tumor ploidy (p=0.002; per 5% tumor CNA burden, HR = 1.32; 95% CI 1.11, 1.56; Table 2). Moreover, for a multivariable model that includes tumor CNA burden, Gleason grade, and mutation burden, the Harrell’s C-index is 0.691. In contrast, the C-index for a model including ploidy instead of tumor CNA burden is only 0.606, indicating that a model with clinical factors and ploidy does not perform as well as a model with the same clinical factors and tumor CNA burden.

Table 2. Tumor CNA burden is associated with recurrence and overall survival independent of disease stage in multiple cancer types.

Cohort Model Disease free Time Overall Survival
Cases HR 95% CI P Cases HR 95% CI P
Prostate cancer TCGA Tumor CNA burden, per 5% tumor CNA burden, univariate 280 1.27 1.13, 1.42 <0.0001 Insufficient events


Tumor CNA burden, per 5% tumor CNA burden, adjusted for Gleason grade and mutation burden 279 1.18 1.03, 1.35 0.015
Tumor CNA burden, per 5% tumor CNA burden, adjusted for purity (ABSOLUTE)* 243 1.22 1.07, 1.40 0.003
Tumor CNA burden, per 5% tumor CNA burden, adjusted for ploidy 243 1.32 1.11, 1.56 0.002
Breast cancer TCGA Tumor CNA burden, per 5% tumor CNA burden, univariate 709 1.07 1.01, 1.14 0.028 794 1.08 1.03, 1.13 0.0005
Tumor CNA burden, per 5% tumor CNA burden, Multivariable, adjusted for disease stage 695 1.07 1.00, 1.14 0.049 777 1.08 1.03, 1.13 0.002
Endometrial Cancer TCGA Tumor CNA burden, per 5% tumor CNA burden, univariate 496 1.10 1.06, 1.14 <0.0001 536 1.13 1.08, 1.17 <0.0001
Tumor CNA burden, per 5% tumor CNA burden, multivariable, adjusted for disease stage 496 1.08 1.04, 1.13 <0.0001 536 1.10 1.05, 1.15 <0.0001
Renal clear cell cancer TCGA Tumor CNA burden, per 5% tumor CNA burden, univariate 425 1.05 1.01, 1.09 0.028 525 1.02 0.98, 1.06 NS (0.4)
Tumor CNA burden, per 5% tumor CNA burden, multivariable, adjusted for disease stage 423 1.05 1.00, 1.11 0.035 522 1.01 0.97, 1.06 NS (0.7)
Thyroid cancer TCGA Tumor CNA burden, per 5% tumor CNA burden, univariate 483 1.17 1.01, 1.35 0.033 497 1.30 1.04, 1.63 0.021
Tumor CNA burden, per 5% tumor CNA burden, multivariable, adjusted for disease stage 481 1.18 1.00, 1.39 0.048 Insufficient events
Colorectal cancer TCGA Tumor CNA burden, per 5% tumor CNA burden, univariate 512 1.05 1.00, 1.11 0.037 587 1.06 1.01, 1.12 0.012
Tumor CNA burden, per 5% tumor CNA burden, multivariable, adjusted for disease stage 496 1.03 0.98, 1.09 NS (0.3) 567 1.03 0.97, 1.09 NS (0.3)
Tumor CNA burden, per 5% tumor CNA burden, multivariable, adjusted for tumor stage 511 1.06 1.01, 1.12 0.028 585 1.07 1.02, 1.13 0.009
Open in a new tab

*Result differed with similar adjustment in IMPACT prostate cancer cohort utilizing FACETS, see Supplementary file 3.

The prognostic significance of tumor CNA burden in prostate cancer led us to ask whether tumor CNA burden is prognostic in other cancer types. Towards this end, we examined published TCGA cohorts for multiple cancer types with available disease-free survival and overall survival data, including breast (Ciriello et al., 2015), endometrial (Cancer Genome Atlas Research Network et al., 2013), renal clear cell (Cancer Genome Atlas Research Network, 2013), thyroid (Cancer Genome Atlas Research Network, 2014), and colorectal (Cancer Genome Atlas Network, 2012) cancers. We found that tumor CNA burden is associated with recurrence (disease-free survival) in these cancer types (Figure 2c, Figure 2—figure supplement 2, Table 2). This association between tumor CNA burden and lower disease-free survival was independent of disease stage in all cancer types except colorectal cancer, where the association was independent of tumor stage but not disease stage (Table 2). In addition to lower disease-free survival, higher tumor CNA burden was also significantly associated with lower overall survival in breast, endometrial, thyroid, and colorectal cancer (Table 2). This association with overall survival was independent of disease stage in breast and endometrial cancer and independent of tumor stage in colorectal cancer (Table 2). There were insufficient cases of thyroid cancer with stage data for this analysis. In summary, tumor CNA burden is prognostic for recurrence and/or overall survival in multiple cancer types beyond prostate cancer, including breast, endometrial, colorectal, renal clear cell, and thyroid cancer.

Tumor CNA burden determined by clinical targeted sequencing of primary and metastatic tumors is prognostic for survival

We next wanted to determine whether CNA burden’s prognostic associations could be observed using panel-based targeted sequencing assays that are increasingly entering clinical use, in contrast to CGH array-based determination of tumor CNA burden. The Memorial Sloan Kettering-Integrated Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT) assay is a clinical laboratory improvement amendments (CLIA)-certified sequencing-based assay (Cheng et al., 2015) of several hundred cancer genes and 1042 common single nucleotide polymorphisms (SNPs) that has been used to profile 504 prostate tumors (Abida et al., 2017) and more than ten thousand tumors across other cancer types (Zehir et al., 2017). The IMPACT assay identifies both somatic point mutations and copy number alterations in the genes included in the panel. Overall copy number burden is calculated across the whole genome (Figure 1a) using segmentation derived from a combination of the profiled SNPs to provide low resolution copy number data and the genes sequenced in the panel (Cheng et al., 2015; Abida et al., 2017; Zehir et al., 2017). To address the possibility that CNA burden from the IMPACT panel might differ from that derived from more comprehensive sequencing, we directly compared CNA burden calculations from 1005 tumors that were profiled using both IMPACT and whole exome sequencing. CNA burden determined by the two methods were highly correlated (p-value<0.0001, rho = 0.88, n = 1005), indicating that CNA burden is not significantly affected by the reduced resolution in moving from whole exome to targeted panel sequencing (Figure 2—figure supplement 3).

We find that tumor CNA burden assayed by targeted clinical sequencing is significantly associated with overall survival in primary prostate tumors (per 5% tumor CNA burden, HR = 1.17; 95% CI, 1.04, 1.3; p=0.007; Table 3, Figure 2c, Figure 2—figure supplement 4) in the IMPACT prostate cohort (Abida et al., 2017). As clinical sequencing assays such as MSK-IMPACT are principally used in the metastatic patient population, the IMPACT cohorts also provide an opportunity to investigate the prognostic significance of tumor CNA burden in late stage disease. We find that tumor CNA burden of metastatic prostate tumors assayed by clinical sequencing is also significantly associated with survival (per 5% tumor CNA burden, HR = 1.07; 95% CI, 1.01, 1.14; p=0.020; Table 3, Figure 2c, Figure 2—figure supplement 4).

Table 3. Tumor CNA burden determined by clinically approved sequencing panel is associated with overall survival in primary and metastatic tumors.

Model Overall Survival
Primary tumors Metastatic tumors
HR 95% P HR 95% P
Prostate Cancer*,†
Univariate, tumor CNA burden, per 5% 1.17 1.04, 1.31 0.007 1.07 1.01, 1.14 0.020
Multivariable
Tumor CNA burden, per 5%
Mutation burden (per mutation)		 1.11
1.22		 0.98, 1.26
1.12, 1.33		 0.10
<0.0001		 1.08
1.05		 1.02, 1.15
1.02, 1.08		 0.011
0.001
Multivariable
Tumor CNA burden, per 5%
TP53 CN loss or mutation		 1.17
4.12		 1.04, 1.31
2.02, 8.41		 0.007
<0.0001		 1.06
1.24		 1.00, 1.13
0.76, 2.02		 NS (0.069)
NS (0.4)
Multivariable
Tumor CNA burden, per 5%
RB1 CN loss or mutation		 1.15
3.24		 1.02, 1.30
0.70, 14.98		 0.026
NS (0.13)		 1.06
1.68		 0.99, 1.13
0.94, 2.99		 NS (0.091)
NS (0.080)
Multivariable
Tumor CNA burden, per 5%
PTEN CN loss or mutation		 1.17
2.38		 1.04, 1.32
1.03, 5.51		 0.008
0.042		 1.07
1.15		 1.01, 1.14
0.70, 1.89		 0.023
NS (0.6)
Pan- Cancer
Univariate, tumor CNA burden, per 5%ठ1.04 1.02, 1.05 <0.0001 1.02 1.01, 1.03 0.005
Univariate, mutation burden (per five units)ठ0.98 0.97, 1.00 NS (0.072) 0.99 0.97, 1.01 NS (0.4)
Meta-analysis, tumor CNA burden (per 5%)# 1.04 1.02, 1.05 <0.0001** 1.02 1.01, 1.04 0.005††
Meta-analysis, tumor CNA burden (per 5%), excluding outlier cancer types‡‡ 1.05 1.03, 1.07 <0.0001§§ 1.03 1.01, 1.04 0.002##
Open in a new tab

*Prostate primary tumors: patient n = 261 for all models except multivariable model with mutation burden, where n = 227; event n = 33; median follow-up time for survivors 40 (IQR 25,81) months.

†Prostate metastatic tumors: patient n = 216 for all models except multivariable model with mutation burden, where n = 205; event n = 80; median follow-up time for survivors 59.5 (IQR 32, 129) months.

‡Pan-cancer primary tumors, univariate models: patient n = 6610, event n = 1535, median follow-up time for survivors 24 (IQR 11, 61) months.

§Pan-cancer metastatic tumors, univariate models: patient n = 4864, event n = 1467, median follow-up time for survivors 51 (IQR 23, 109) months.

#Pan-cancer meta-analysis, among ten most prevalent cancer types: primary tumor patient n = 4863, metastatic tumor patient n = 3676. Estimates are based on overall fixed effects.

**p-value corresponds with test of effects size. Chi-square test for heterogeneity p-value=0.003.

††p-value corresponds with test of effects size. Chi-square test for heterogeneity p-value=0.024.

‡‡Exclusion of cancer types to reduce heterogeneity: primary tumor patient n = 3887, metastatic tumor patient n = 3098. Estimates are based on overall fixed effects.

§§Excluding pancreatic and colorectal cancer, test of effects size p-value. Chi-square test for heterogeneity p-value=0.3.

##Excluding pancreatic and prostate cancer, test of effects size p-value. Chi-square test for heterogeneity p-value=0.8.

Since clinical sequencing assays also provide point mutation information for several hundred cancer genes, we asked if tumor CNA burden is prognostic after adjusting for known prostate cancer driver alterations. In separate multivariable regression models adjusting for TP53, RB1, or PTEN loss and/or mutation, tumor CNA burden is still associated with overall survival independent of these alterations in primary prostate tumors (Table 3). In metastatic tumors, these specific gene mutations do not reach prognostic significance when combined with tumor CNA burden (Table 3). Notably, tumor CNA burden remains significant in metastatic tumors after adjusting for overall tumor mutation burden (per 5% tumor CNA burden, HR = 1.08; 95% CI = 1.02, 1.15; p=0.011; Table 3).

As targeted clinical sequencing is applied to a wide range of cancer types, we expanded our survival analysis to a pan-cancer cohort, consisting of 6610 primary tumors and 4864 metastatic tumors across 53 cancer types assayed by MSK-IMPACT sequencing panel (Materials and methods and Supplementary file 2). We find that tumor CNA burden is prognostic for overall survival pan-cancer in primary tumors (p<0.0001; per 5% tumor CNA burden, HR = 1.04; 95% CI, 1.02, 1.05) and in metastatic tumors (p=0.005; per 5% tumor CNA burden, HR = 1.02; 95% CI, 1.01, 1.03) in a univariate analysis of these pan-cancer cohorts (Table 3, Figure 2c). Tumor CNA burden is also prognostic for cancer-specific death in the metastatic tumor cohort (p=0.026; per 5% tumor CNA burden, HR = 1.05; 95% CI, 1.01, 1.10). Adjustment for sample tumor purity determined by FACETS (Shen and Seshan, 2016) found that CNA burden was still significantly associated with overall survival in primary tumors in the pan-cancer analysis and approached significance for metastatic tumors (p=0.06; Supplementary file 3), though purity-adjusted CNA burden was no longer significantly associated with overall survival in the prostate tumor subsets (Supplementary file 3). Adjustment for sample tumor purity determined by FACETS (Shen and Seshan, 2016) found that CNA burden was still significantly associated with overall survival in primary tumors in the pan-cancer analysis approached significance for metastatic tumors (p=0.06; Supplementary file 3), though purity-adjusted CNA burden was no longer significantly associated with overall survival in the prostate tumor subsets (Supplementary file 3). Tumor mutation burden (TMB), in contrast to tumor CNA burden, was not associated with overall survival or cancer-specific survival (p=0.4 and p>0.9, respectively; Table 3).

Since the pan-cancer prognostic significance of tumor CNA burden is likely to be influenced by the distribution of cancer types within the IMPACT cohorts, we stratified the primary and metastatic pan-cancer IMPACT cohorts by their ten most prevalent cancer types, which make up nearly three-quarters of the cohort (Supplementary file 2). A multivariable Cox model was used for each cancer type to adjust for mutation burden and extract the effect size, which was then entered into a meta-analysis. After stratifying by cancer type, the CNA burden of primary tumors measured by the MSK-IMPACT assay is still significantly associated with death (overall fixed effects HR = 1.04; 95% CI 1.02, 1.05; test of effects size p<0.0001; Table 3; Figure 2c). Similarly, metastatic tumor CNA burden was associated with death (overall fixed effects HR = 1.02; 95% CI 1.01, 1.04; test of effects size p=0.005; Table 3; Figure 2c).

A closer look at the pan-cancer analysis reveals statistically significant heterogeneity in the relationship between tumor CNA burden and survival across tumor types (p=0.003 and p=0.024 in primary and metastatic tumor cohorts respectively, Figure 2—figure supplement 4). In primary tumors, heterogeneity appears to be driven by colorectal and pancreatic cancers, where an inverse association between tumor CNA burden and death is seen (Figure 2—figure supplement 5a). After excluding colorectal and pancreatic cancers, heterogeneity is no longer statistically significant (overall fixed effects HR = 1.05; 95% CI 1.03, 1.07; test of effects size p<0.0001; test for heterogeneity p=0.3; Figure 2—figure supplement 5a). In metastatic tumors, two outlying cancer types drive this heterogeneity: pancreatic cancer, which shows the same inverse association of tumor CNA burden with death as in primary pancreatic tumors, and prostate, which shows the opposite effect (Figure 2—figure supplement 5b). Exclusion of either cancer type eliminates the significant heterogeneity in effects size, such that higher tumor CNA burden is associated with increased death in the remaining homogenous set of cancer types (overall fixed effects HR = 1.03; 95% CI 1.01, 1.04; test of effects size p=0.002; test for heterogeneity p=0.8, Figure 2—figure supplement 5b). These results indicate that tumor CNA burden can have differing levels of prognostic effect depending on the cancer type, while a core set of cancer types show a statistically similar association between overall survival and tumor CNA burden assayed by targeted sequencing. More generally, we find that tumor CNA burden determined by a clinically-certified sequencing panel is associated with overall and disease-specific mortality in a large multi-cancer population, including in patients with metastatic cancer where clinical sequencing is increasingly applied.

Discussion

Many specific genes altered by CNA have been associated with cancer outcomes (Liang et al., 2016; Wang et al., 2016; Nibourel et al., 2017), however the relationship between outcome and the overall level of CNA harbored by a tumor is less well studied. Here we expanded on our previous work showing that tumor CNA burden is associated with recurrence in surgically treated primary prostate cancer (Taylor et al., 2010; Hieronymus et al., 2014) by showing a significant association with death from prostate cancer, including in conservatively treated patients where the tumor CNA burden measurement was made from biopsies. Importantly, this association remains significant even after adjusting for Gleason score or for CAPRA score, demonstrating that CNA burden is independent of previously identified associations with these measures of cancer pathology or disease state. Thus, tumor CNA burden assessment from prostate biopsies could have a role in deciding between surgery and surveillance for men at the low end of intermediate risk. Conversely, it may also have role in men at high risk where multimodal treatment may be needed.

An unanticipated outcome of our analysis beyond prostate cancer is the prognostic role of tumor CNA burden across a range of tumor types. The pan-cancer tumor CNA burden association is significant but also heterogeneous depending on cancer type. Recent work has similarly found that the presence of any CNA, regardless of gene identity, is associated with overall and event-free survival in pediatric AML (Vujkovic et al., 2017) and that the percentage of subclonal CNAs but not subclonal somatic point mutations is associated with overall survival in non-small cell lung cancer (Jamal-Hanjani et al., 2017). Moreover, survival time was associated with a CNA signature derived from supervised analyses in prostate cancer and extended to breast and lung cancer (Pearlman et al., 2018). Prognostic individual CNAs or sets of CNAs, as opposed to the broader measure of genome-wide CNA level examined here may be specific to individual cancer types, whereas we have demonstrated the prognostic potential of a generalized measure of overall copy number dysregulation. Further work will be needed to address the trade-offs between generalizability of CNA burden and discriminatory power. In addition, it will be important to investigate whether the prognostic associations of CNA burden from the pan-cancer analysis are independent of known cancer- or subtype-specific prognostic factors, such as ER receptor status in breast cancer, ultra- and hypermutated (POLE and MSI+) status in endometrial cancer and MSI-positive/CIN-negative status in colorectal cancer (Walther et al., 2008).

We find it notable that tumor CNA burden assessment using a targeted sequencing can serve as a surrogate for tumor CNA burden calculated using more comprehensive genomic assays such as array CGH. With the proliferation of different clinical sequencing panels for mutation detection, it will be of interest to see how much resolution, depth, and coverage can be reduced and still retain the prognostic association of CNA burden; future work in this area will also need to incorporate the predictive clinical utility of the point mutation data to address the multimodal uses of clinical sequencing assays. Another important variable is tumor purity. The prognostic significance of CNA burden can be affected by sample tumor purity, with purity being independently associated with outcome. The effect of purity on the association between CNA burden and outcome appears complex and may be influenced by the analysis platform, cancer type, and outcome type. For example, pan-cancer CNA burden from clinical sequencing panel remained prognostic for survival after purity adjustment in primary tumors and was just below significance for metastatic tumors, though the CNA burden of the prostate tumor subset assayed by IMPACT sequencing panel did not. However, the CNA burden of prostate tumors assayed by SNP array showed continued association with recurrence after adjustment for purity. Tumor purity alone was also independently associated with survival, revealing a complex interaction between these tumor features that will need further exploration. As targeted sequencing moves from tumor samples to liquid biopsy in the form of cell-free DNA (cfDNA) (Heitzer et al., 2016; Xia et al., 2015; Hyman et al., 2017), it will be important to determine whether tumor CNA burden determined by analysis of cfDNA has similar prognostic utility as that determined by direct analysis of tumor DNA. There is already some evidence this may be possible, as the summed CNA level of the most highly copy number altered genes assayed from whole genome sequencing of cfDNA in twenty metastatic prostate cancer patients correlated with overall survival (Xia et al., 2015). As sequencing costs continue to drop and computational power improves, it would be interesting to investigate low pass whole genome sequencing as an alternative approach for determining tumor CNA burden that provides complete genome coverage.

Another interesting feature of the association of tumor CNA burden with outcome demonstrated here is that it has prognostic significance independent of tumor mutation burden (TMB). This is consistent with recent work in glioblastoma, breast, lung, and ovarian cancer showing that CNA-derived signatures have more prognostic power than somatic point mutation-based signatures, as measured by concordance index (Gómez-Rueda et al., 2015). Thus, tumor CNA burden could complement clinical analyses of actionable driver mutations using a single panel-based sequencing assay.

The prognostic significance of tumor CNA burden raises intriguing questions regarding the underlying biology. Tumor CNA burden may be a simple measure that correlates with the extent of oncogenic driver alterations. Yet, we show that tumor CNA burden retains its prognostic significance after adjustment for a number of known oncogenic alterations in primary prostate cancer, including PTEN loss associated with increased tumor CNA burden (Castro et al., 2015; Williams et al., 2014). In metastatic tumors, combining tumor CNA burden with TP53 or RB1 loss in multivariable analyses renders both slightly below conventional significance thresholds, raising the possibility of biological interplay between these genes (particularly TP53) and subsequent copy number alteration that develops during tumor evolution. Further, the prognostic associations of tumor CNA burden are independent of tumor ploidy, which suggests that tumor CNA burden may not simply reflect aneuploidy, defined as abnormal DNA content (Danielsen et al., 2016). It is also possible that tumor CNA burden captures prognostic information about currently unidentified driver alterations and/or the rate of ongoing CNA within a tumor that may generate additional driver alterations, including those reflecting intratumoral heterogeneity, thereby affecting outcome. Ongoing work by others has begun to develop genomic methods for identifying mechanisms of somatic CNA (Wala et al., 2017); and identify prognostic CNA signatures and the mechanisms underlying the component CNA (Macintyre et al., 2018). Ultimately, the biology underlying the significant association of tumor CNA burden with multiple cancer outcomes will be a fruitful area for future investigation.

Materials and methods

aCGH copy number analysis of conservative-treatment TAPG cohort.

Of the TAPG1 cohort (Cuzick et al., 2006), FFPE prostate tumor tissue from 180 patients was macrodissected from slides. DNA was isolated (Agilent FFPE DNA isolation for aCGH protocol) and quantified by picogreen-based quantification. 107 cases yielded greater than 500 ug DNA and were analyzed by Agilent 180K human CGH arrays (Agilent, 4 × 180K G4449A arrays, per manufacturer's instructions). Copy number data from patients in the TAPG copy number cohort were quantified, normalized, segmented, and analyzed with RAE, as previously described (Taylor et al., 2010; Hieronymus et al., 2014). The conservative treatment TAPG copy number cohort array data was deposited in NCBI GEO under accession number GSE103665 (Gene Expression Omnibus, RRID:SCR_007303).

Tumor CNA burden (tumor CNA burden) was analyzed as percent CNA burden, defined as the length of the genome altered by copy number alteration multiplied by 100. For regression analyses, tumor CNA burden was scaled as per five percent so that the estimates of our hazard ratios were more interpretable. All statistical analyses were performed using Stata 13 (RRID:SCR_012763, StataCorp, College Station, TX).

TAPG copy number cohort statistical analyses

For Cox regression analyses, the primary aim was to determine whether tumor CNA burden is associated with cancer specific survival (CSS). First, we assessed whether there was an association between tumor CNA burden and CSS by utilizing a univariate Cox model, censoring patients who did not die at the date of their last follow-up and patients who died of other causes at their death date. Secondly, in order to assess whether there is information from tumor CNA burden over and above biopsy Gleason score, we utilized a multivariable Cox model, adjusting for biopsy Gleason sum categorized as ≤6, 7, and ≥8. Finally, to assess whether there is an association between tumor CNA burden and CSS after accounting for the preoperative predictors of CSS, we utilized a multivariable Cox model, adjusting for the UCSF-CAPRA score, a preoperative risk score calculated by incorporating the patient’s age at diagnosis, PSA at diagnosis, primary and secondary Gleason score at biopsy and clinical tumor stage. As percent of positive biopsy cores was not available for the cohort, a modified CAPRA score was utilized not incorporating this information. Among our cohort of 107 patients, 47 patients were missing clinical tumor stage; multiple imputation was used to impute the missing values. Statistical analyses were performed utilizing the measured and imputed values combined across 10 imputations using Rubin’s method. Furthermore, to evaluate the discriminative accuracy of the model including tumor CNA burden, we calculated bootstrap optimism-corrected Harrell’s C-index. It should be noted that the discrimination of the CAPRA score is lower in the TAPG1 conservative treatment CNA cohort than seen in some other prostate cancer cohorts, and this may impact the degree to which tumor CNA burden increases the concordance index. All data used for these analyses are available in Supplementary file 4.

For illustrative purposes, we utilized competing risk methods to estimate the probability of death from prostate cancer in the setting of death from other causes. Cumulative incidence was shown for patients who died from prostate cancer, or died from other causes, stratified on tumor CNA burden in relation to the median tumor CNA burden among the cohort, using the stcompet command in Stata.

Statistical analyses of IMPACT cohorts

For analysis of the prostate cancer MSK-IMPACT cohort (Abida et al., 2017), the published cases were analyzed by Cox regression for association between overall survival and tumor CNA burden (Supplementary file 5 and 6). The IMPACT cases were separated into groups consisting of primary tumors or metastatic tumors, including loco-regional, non-resistant to treatment, and treatment resistant, though primary tumor samples include cases sampled after metastatic spread. Among our primary and metastatic IMPACT prostate cancer cohorts, we excluded men with unknown overall survival status and unknown time until overall survival status, leaving us with a final cohort of 261 and 216 men, respectively. Among these two groups of patients, we assessed the association between tumor CNA burden and overall survival using a univariate Cox model. Multivariable Cox models were then used to determine whether the association between tumor CNA burden and overall survival remained after accounting for purity determined by FACETS (Shen and Seshan, 2016), the overall point mutation burden, or specific somatic gene alterations (shallow or deep copy number loss or mutation) occurring in prostate cancer (BRCA1, BRCA2, ATM, TP53, RB1, and PTEN), using separate models for each alteration. As the overall point mutation burden was not available for all patients, 34 patients with primary prostate cancer and 11 patients with metastatic prostate cancer were excluded from this portion of the analysis in their respective cohorts.

For analysis of our pan-cancer IMPACT cohort (MSK-IMPACT cohort (Zehir et al., 2017) and additionally accrued IMPACT samples), outcome data at time of analysis, mutation burden, and fraction genome altered data used were derived and available in updated form the cBio Portal (RRID:SCR_002877, http://www.cbioportal.org/study?id=msk_impact_2017, samples and annotation used at time of analysis available as Supplementary file 7 and 8). A cohort of 7305 primary tumor cases across 53 different cancer types and a cohort 5907 metastatic tumor cases, across 47 different cancer types, were identified. Within the primary and metastatic disease cohorts, we excluded patients with unknown tumor CNA burden, overall survival status, unreported follow-up time, death or censoring immediately after treatment, unknown cancer type, and unknown mutation burden. The final cohort used here therefore included 6610 and 4864 patients, respectively. Within both of these cohorts, univariate Cox models were used to determine whether CNA or mutation burden is associated with overall survival. Reported follow-up time was used. As it is probable that the association between tumor CNA burden and survival likely varies based on the particular cancer type, we focused on patients with the ten most prevalent cancer types in both of the respective cohorts (Supplementary file 2, 5198 and 3886 patients with primary and metastatic disease respectively) and proceeded with a meta-analysis in order to stratify by cancer type. In particular, we utilized a multivariable Cox model, adjusting for mutation burden for each cancer type and extracted the effect size. The effect size for each cancer type was then entered into a meta-analysis using the metan command in Stata. Both fixed effects and random effects estimates were calculated. Fixed effects weights were calculated using inverse-variance weighting, metan weights were calculated using the DerSimonian and Laird method.

Statistical analyses of TCGA cohorts

For analyses of TCGA cohorts, the following published cohorts were filtered for only primary, non-neoadjuvantly treated cases and analyzed: TCGA prostate adenocarcinoma (2015) (Cancer Genome Atlas Research Network, 2015), breast carcinoma (Ciriello et al., 2015), uterine endometriod cancer (Cancer Genome Atlas Research Network et al., 2013), renal clear cell carcinoma (Cancer Genome Atlas Research Network, 2013), papillary thyroid carcinoma (Cancer Genome Atlas Research Network, 2014), and colorectal adenocarcinoma (Cancer Genome Atlas Network, 2012). The number of cases and exclusions based on unavailable data are detailed in Supplementary file 9. Cox regression was used to test the association of tumor CNA burden as a continuous variable with (i) cancer free status and (ii) overall survival in univariate models and in multivariable models with disease stage. For the TCGA colorectal cancer cohort, tumor stage was also used. For the TCGA prostate adenocarcinoma cohort, multivariable Cox regression models that included Gleason score, mutation count, ploidy, and/or ABSOLUTE purity (Carter et al., 2012) originally reported with this cohort were also used. Analyses including purity exclude 37 patients without absolute tumor purity measured, resulting in analysis with 243 men, 29 of whom had BCR, and a median followup time for survivors of 20.1 (7.0, 37.9) months.

Data access. The conservative treatment TAPG copy number cohort array data was deposited in NCBI GEO (Gene Expression Omnibus, under accession number GSE103665 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE103665).

Acknowledgements

We thank the members of the Prostate Cancer Oncogenome Group for critical contributions. This work was supported by HHMI (CLS), CA193837, CA092629, CA155169, the Prostate Cancer Foundation Young Investigator Award (to KKY), Orchid (DMB). We thank the MSKCC Integrated Genomics Operation Core for technical work. The MSKCC Integrated Genomics Operation Core is funded by P30 CA08748, Cycle for Survival and the Marie-Josée and Henry R Kravis Center for Molecular Oncology.

Funding Statement

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Contributor Information

Charles L Sawyers, Email: sawyersc@mskcc.org.

Michael R Green, Howard Hughes Medical Institute, University of Massachusetts Medical School, United States.

Jeffrey Settleman, Calico Life Sciences, United States.

Funding Information

This paper was supported by the following grants:

  • Prostate Cancer Foundation to Kamlesh Yadav.

  • American Cancer Society RSG-15-067-01-TBG to Barry Taylor.

  • Prostate Cancer Foundation to Barry Taylor.

  • National Cancer Institute R01 CA204749 to Barry Taylor.

  • Howard Hughes Medical Institute to Charles L Sawyers.

  • National Institutes of Health CA193837 to Charles L Sawyers.

  • National Institutes of Health CA092629 to Charles L Sawyers.

  • National Institutes of Health CA155169 to Charles L Sawyers.

  • National Institutes of Health CA008748 to Charles L Sawyers.

Additional information

Competing interests

Senior Editor, eLife; Board of Directors of Novartis; co-founder of ORIC Pharm; co-inventor of enzalutamide and apalutamide; Science advisor to Agios, Beigene, Blueprint, Column Group, Foghorn, Housey Pharma, Nextech, KSQ, Petra and PMV; co-founder of Seragon, purchased by Genentech/Roche in 2014.

No competing interests declared.

Author contributions

Conceptualization, Data curation, Formal analysis, Supervision, Investigation, Visualization, Methodology, Writing—original draft, Project administration, Writing—review and editing.

Data curation, Methodology, Project administration.

Data curation, Formal analysis, Visualization, Methodology, Writing—review and editing.

Investigation, Project administration.

Data curation.

Resources, Data curation, Methodology.

Resources, Data curation.

Conceptualization, Resources, Supervision.

Conceptualization, Writing—review and editing.

Conceptualization, Funding acquisition.

Data curation, Writing—review and editing.

Conceptualization, Investigation, Visualization, Writing—review and editing.

Conceptualization, Formal analysis, Writing—review and editing.

Data curation, Validation, Investigation, Project administration.

Conceptualization, Resources, Supervision, Funding acquisition, Writing—review and editing.

Additional files

Supplementary file 1. Cohort characteristics.
elife-37294-supp1.docx (14.7KB, docx)
DOI: 10.7554/eLife.37294.013
Supplementary file 2. Distribution of cancer types in IMPACT cohorts.
elife-37294-supp2.docx (18KB, docx)
DOI: 10.7554/eLife.37294.014
Supplementary file 3. Association between overall survival and CNA burden after adjustment for purity in IMPACT prostate and pan-cancer cohorts.

Purity was determined by FACETS (Shen and Seshan, 2016).

elife-37294-supp3.docx (17.7KB, docx)
DOI: 10.7554/eLife.37294.015
Supplementary file 4. TAPG1 conservative treatment primary prostate CNA cohort.
elife-37294-supp4.csv (16.5KB, csv)
DOI: 10.7554/eLife.37294.016
Supplementary file 5. MSK-IMPACT primary prostate tumor cohort annotation.
elife-37294-supp5.csv (18.5KB, csv)
DOI: 10.7554/eLife.37294.017
Supplementary file 6. MSK-IMPACT metastatic prostate tumor cohort annotation.
elife-37294-supp6.csv (15.1KB, csv)
DOI: 10.7554/eLife.37294.018
Supplementary file 7. MSK-IMPACT primary pan-cancer cohort annotation.
elife-37294-supp7.csv (381.2KB, csv)
DOI: 10.7554/eLife.37294.019
Supplementary file 8. MSK-IMPACT metastatic pan-cancer cohort annotation.
elife-37294-supp8.csv (493KB, csv)
DOI: 10.7554/eLife.37294.020
Supplementary file 9. TCGA Cohort statistics: patient exclusion, events, and follow-up.
elife-37294-supp9.docx (14.4KB, docx)
DOI: 10.7554/eLife.37294.021
Transparent reporting form
elife-37294-transrepform.pdf (319.4KB, pdf)
DOI: 10.7554/eLife.37294.022

Data availability

All data generated or analysed during this study are included in the manuscript and supporting files and reference materials. The conservative treatment TAPG copy number cohort array data was deposited in NCBI GEO under accession number GSE103665 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE103665).

The following dataset was generated:

Hieronymus H, author; Taylor BS, author; Sawyers CL, author. Copy number alteration burden is a pan-cancer prognostic factor associated with metastasis and death in conservatively treated prostate cancer: TAPG1 CNA cohort aCGH data. 2018 https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE103665 Publicly available at the NCBI Gene Expression Omnibus (accession no: GSE103665).

References

  1. Abida W, Armenia J, Gopalan A, Brennan R, Walsh M, Barron D, Danila D, Rathkopf D, Morris M, Slovin S, McLaughlin B, Curtis K, Hyman DM, Durack JC, Solomon SB, Arcila ME, Zehir A, Syed A, Gao J, Chakravarty D, Vargas HA, Robson ME, Vijai J, Offit K, Donoghue MTA, Abeshouse AA, Kundra R, Heins ZJ, Penson AV, Harris C, Taylor BS, Ladanyi M, Mandelker D, Zhang L, Reuter VE, Kantoff PW, Solit DB, Berger MF, Sawyers CL, Schultz N, Scher HI. Prospective genomic profiling of prostate Cancer across disease states reveals germline and somatic alterations that may affect clinical decision making. JCO Precision Oncology. 2017;2017:1–16. doi: 10.1200/PO.17.00029. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Bergerot PG, Hahn AW, Bergerot CD, Jones J, Pal SK. The role of circulating tumor DNA in renal cell carcinoma. Current Treatment Options in Oncology. 2018;19:10. doi: 10.1007/s11864-018-0530-4. [DOI] [PubMed] [Google Scholar]
  3. Beroukhim R, Mermel CH, Porter D, Wei G, Raychaudhuri S, Donovan J, Barretina J, Boehm JS, Dobson J, Urashima M, Mc Henry KT, Pinchback RM, Ligon AH, Cho YJ, Haery L, Greulich H, Reich M, Winckler W, Lawrence MS, Weir BA, Tanaka KE, Chiang DY, Bass AJ, Loo A, Hoffman C, Prensner J, Liefeld T, Gao Q, Yecies D, Signoretti S, Maher E, Kaye FJ, Sasaki H, Tepper JE, Fletcher JA, Tabernero J, Baselga J, Tsao MS, Demichelis F, Rubin MA, Janne PA, Daly MJ, Nucera C, Levine RL, Ebert BL, Gabriel S, Rustgi AK, Antonescu CR, Ladanyi M, Letai A, Garraway LA, Loda M, Beer DG, True LD, Okamoto A, Pomeroy SL, Singer S, Golub TR, Lander ES, Getz G, Sellers WR, Meyerson M. The landscape of somatic copy-number alteration across human cancers. Nature. 2010;463:899–905. doi: 10.1038/nature08822. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Brajtbord JS, Leapman MS, Cooperberg MR. The CAPRA Score at 10 years: contemporary perspectives and analysis of supporting studies. European Urology. 2017;71:705–709. doi: 10.1016/j.eururo.2016.08.065. [DOI] [PubMed] [Google Scholar]
  5. Camacho N, Van Loo P, Edwards S, Kay JD, Matthews L, Haase K, Clark J, Dennis N, Thomas S, Kremeyer B, Zamora J, Butler AP, Gundem G, Merson S, Luxton H, Hawkins S, Ghori M, Marsden L, Lambert A, Karaszi K, Pelvender G, Massie CE, Kote-Jarai Z, Raine K, Jones D, Howat WJ, Hazell S, Livni N, Fisher C, Ogden C, Kumar P, Thompson A, Nicol D, Mayer E, Dudderidge T, Yu Y, Zhang H, Shah NC, Gnanapragasam VJ, Isaacs W, Visakorpi T, Hamdy F, Berney D, Verrill C, Warren AY, Wedge DC, Lynch AG, Foster CS, Lu YJ, Bova GS, Whitaker HC, McDermott U, Neal DE, Eeles R, Cooper CS, Brewer DS, CRUK-ICGC Prostate Group Appraising the relevance of DNA copy number loss and gain in prostate cancer using whole genome DNA sequence data. PLOS Genetics. 2017;13:e1007001. doi: 10.1371/journal.pgen.1007001. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Cancer Genome Atlas Network Comprehensive molecular characterization of human colon and rectal cancer. Nature. 2012;487:330–337. doi: 10.1038/nature11252. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Cancer Genome Atlas Research Network Comprehensive molecular characterization of clear cell renal cell carcinoma. Nature. 2013;499:43–49. doi: 10.1038/nature12222. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Cancer Genome Atlas Research Network. Kandoth C, Schultz N, Cherniack AD, Akbani R, Liu Y, Shen H, Robertson AG, Pashtan I, Shen R, Benz CC, Yau C, Laird PW, Ding L, Zhang W, Mills GB, Kucherlapati R, Mardis ER, Levine DA. Integrated genomic characterization of endometrial carcinoma. Nature. 2013;497:242–273. doi: 10.1038/nature12113. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Cancer Genome Atlas Research Network Integrated genomic characterization of papillary thyroid carcinoma. Cell. 2014;159:676–690. doi: 10.1016/j.cell.2014.09.050. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Cancer Genome Atlas Research Network The molecular taxonomy of primary prostate Cancer. Cell. 2015;163:1011–1025. doi: 10.1016/j.cell.2015.10.025. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Carter SL, Cibulskis K, Helman E, McKenna A, Shen H, Zack T, Laird PW, Onofrio RC, Winckler W, Weir BA, Beroukhim R, Pellman D, Levine DA, Lander ES, Meyerson M, Getz G. Absolute quantification of somatic DNA alterations in human cancer. Nature Biotechnology. 2012;30:413–421. doi: 10.1038/nbt.2203. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Castro E, Jugurnauth-Little S, Karlsson Q, Al-Shahrour F, Piñeiro-Yañez E, Van de Poll F, Leongamornlert D, Dadaev T, Govindasami K, Guy M, Eeles R, Kote-Jarai Z, UKGPCS, EMBRACE and IMPACT studies High burden of copy number alterations and c-MYC amplification in prostate cancer from BRCA2 germline mutation carriers. Annals of Oncology. 2015;26:2293–2300. doi: 10.1093/annonc/mdv356. [DOI] [PubMed] [Google Scholar]
  13. Chen RC, Rumble RB, Loblaw DA, Finelli A, Ehdaie B, Cooperberg MR, Morgan SC, Tyldesley S, Haluschak JJ, Tan W, Justman S, Jain S. Active surveillance for the management of localized prostate cancer (cancer care ontario guideline): american society of clinical oncology clinical practice guideline endorsement. Journal of Clinical Oncology. 2016;34:2182–2190. doi: 10.1200/JCO.2015.65.7759. [DOI] [PubMed] [Google Scholar]
  14. Cheng DT, Mitchell TN, Zehir A, Shah RH, Benayed R, Syed A, Chandramohan R, Liu ZY, Won HH, Scott SN, Brannon AR, O'Reilly C, Sadowska J, Casanova J, Yannes A, Hechtman JF, Yao J, Song W, Ross DS, Oultache A, Dogan S, Borsu L, Hameed M, Nafa K, Arcila ME, Ladanyi M, Berger MF. Memorial sloan kettering-integrated mutation profiling of actionable cancer targets (MSK-IMPACT): A hybridization Capture-Based Next-Generation sequencing clinical assay for solid tumor molecular oncology. The Journal of Molecular Diagnostics : JMD. 2015;17:251–264. doi: 10.1016/j.jmoldx.2014.12.006. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Ciriello G, Gatza ML, Beck AH, Wilkerson MD, Rhie SK, Pastore A, Zhang H, McLellan M, Yau C, Kandoth C, Bowlby R, Shen H, Hayat S, Fieldhouse R, Lester SC, Tse GM, Factor RE, Collins LC, Allison KH, Chen YY, Jensen K, Johnson NB, Oesterreich S, Mills GB, Cherniack AD, Robertson G, Benz C, Sander C, Laird PW, Hoadley KA, King TA, Perou CM, TCGA Research Network Comprehensive molecular portraits of invasive lobular breast cancer. Cell. 2015;163:506–519. doi: 10.1016/j.cell.2015.09.033. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Cooperberg MR, Pasta DJ, Elkin EP, Litwin MS, Latini DM, Du Chane J, Carroll PR. The university of california, san francisco cancer of the prostate risk assessment score: a straightforward and reliable preoperative predictor of disease recurrence after radical prostatectomy. The Journal of Urology. 2005;173:1938–1942. doi: 10.1097/01.ju.0000158155.33890.e7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Cuzick J, Fisher G, Kattan MW, Berney D, Oliver T, Foster CS, Møller H, Reuter V, Fearn P, Eastham J, Scardino P, Transatlantic Prostate Group Long-term outcome among men with conservatively treated localised prostate cancer. British Journal of Cancer. 2006;95:1186–1194. doi: 10.1038/sj.bjc.6603411. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Danielsen HE, Pradhan M, Novelli M. Revisiting tumour aneuploidy - the place of ploidy assessment in the molecular era. Nature Reviews Clinical Oncology. 2016;13:291–304. doi: 10.1038/nrclinonc.2015.208. [DOI] [PubMed] [Google Scholar]
  19. Gómez-Rueda H, Martínez-Ledesma E, Martínez-Torteya A, Palacios-Corona R, Trevino V. Integration and comparison of different genomic data for outcome prediction in cancer. BioData Mining. 2015;8:32. doi: 10.1186/s13040-015-0065-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Goodman AM, Kato S, Bazhenova L, Patel SP, Frampton GM, Miller V, Stephens PJ, Daniels GA, Kurzrock R. Tumor mutational burden as an independent predictor of response to immunotherapy in diverse cancers. Molecular Cancer Therapeutics. 2017;16:2598–2608. doi: 10.1158/1535-7163.MCT-17-0386. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Heitzer E, Ulz P, Geigl JB, Speicher MR. Non-invasive detection of genome-wide somatic copy number alterations by liquid biopsies. Molecular Oncology. 2016;10:494–502. doi: 10.1016/j.molonc.2015.12.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Hieronymus H, Schultz N, Gopalan A, Carver BS, Chang MT, Xiao Y, Heguy A, Huberman K, Bernstein M, Assel M, Murali R, Vickers A, Scardino PT, Sander C, Reuter V, Taylor BS, Sawyers CL. Copy number alteration burden predicts prostate cancer relapse. PNAS. 2014;111:11139–11144. doi: 10.1073/pnas.1411446111. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Hyman DM, Taylor BS, Baselga J. Implementing genome-driven oncology. Cell. 2017;168:584–599. doi: 10.1016/j.cell.2016.12.015. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Jamal-Hanjani M, Wilson GA, McGranahan N, Birkbak NJ, Watkins TBK, Veeriah S, Shafi S, Johnson DH, Mitter R, Rosenthal R, Salm M, Horswell S, Escudero M, Matthews N, Rowan A, Chambers T, Moore DA, Turajlic S, Xu H, Lee SM, Forster MD, Ahmad T, Hiley CT, Abbosh C, Falzon M, Borg E, Marafioti T, Lawrence D, Hayward M, Kolvekar S, Panagiotopoulos N, Janes SM, Thakrar R, Ahmed A, Blackhall F, Summers Y, Shah R, Joseph L, Quinn AM, Crosbie PA, Naidu B, Middleton G, Langman G, Trotter S, Nicolson M, Remmen H, Kerr K, Chetty M, Gomersall L, Fennell DA, Nakas A, Rathinam S, Anand G, Khan S, Russell P, Ezhil V, Ismail B, Irvin-Sellers M, Prakash V, Lester JF, Kornaszewska M, Attanoos R, Adams H, Davies H, Dentro S, Taniere P, O'Sullivan B, Lowe HL, Hartley JA, Iles N, Bell H, Ngai Y, Shaw JA, Herrero J, Szallasi Z, Schwarz RF, Stewart A, Quezada SA, Le Quesne J, Van Loo P, Dive C, Hackshaw A, Swanton C, TRACERx Consortium Tracking the evolution of non-small-cell lung cancer. New England Journal of Medicine. 2017;376:2109–2121. doi: 10.1056/NEJMoa1616288. [DOI] [PubMed] [Google Scholar]
  25. Liang L, Fang JY, Xu J. Gastric cancer and gene copy number variation: emerging cancer drivers for targeted therapy. Oncogene. 2016;35:1475–1482. doi: 10.1038/onc.2015.209. [DOI] [PubMed] [Google Scholar]
  26. Liu Y, Chen C, Xu Z, Scuoppo C, Rillahan CD, Gao J, Spitzer B, Bosbach B, Kastenhuber ER, Baslan T, Ackermann S, Cheng L, Wang Q, Niu T, Schultz N, Levine RL, Mills AA, Lowe SW. Deletions linked to TP53 loss drive cancer through p53-independent mechanisms. Nature. 2016;531:471–475. doi: 10.1038/nature17157. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Macintyre G, Goranova TE, De Silva D, Ennis D, Piskorz AM, Eldridge M, Sie D, Lewsley LA, Hanif A, Wilson C, Dowson S, Glasspool RM, Lockley M, Brockbank E, Montes A, Walther A, Sundar S, Edmondson R, Hall GD, Clamp A, Gourley C, Hall M, Fotopoulou C, Gabra H, Paul J, Supernat A, Millan D, Hoyle A, Bryson G, Nourse C, Mincarelli L, Sanchez LN, Ylstra B, Jimenez-Linan M, Moore L, Hofmann O, Markowetz F, McNeish IA, Brenton JD. Copy number signatures and mutational processes in ovarian carcinoma. Nature Genetics. 2018 doi: 10.1038/s41588-018-0179-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Nibourel O, Guihard S, Roumier C, Pottier N, Terre C, Paquet A, Peyrouze P, Geffroy S, Quentin S, Alberdi A, Abdelali RB, Renneville A, Demay C, Celli-Lebras K, Barbry P, Quesnel B, Castaigne S, Dombret H, Soulier J, Preudhomme C, Cheok MH. Copy-number analysis identified new prognostic marker in acute myeloid leukemia. Leukemia. 2017;31:555–564. doi: 10.1038/leu.2016.265. [DOI] [PubMed] [Google Scholar]
  29. Pearlman A, Upadhyay K, Cole K, Loke J, Sun K, Fineberg S, Freedland SJ, Shao Y, Ostrer H. Robust genomic copy number predictor of pan cancer metastasis. Genes & Cancer. 2018;9:66–77. doi: 10.18632/genesandcancer.165. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Prandi D, Baca SC, Romanel A, Barbieri CE, Mosquera JM, Fontugne J, Beltran H, Sboner A, Garraway LA, Rubin MA, Demichelis F. Unraveling the clonal hierarchy of somatic genomic aberrations. Genome Biology. 2014;15:439. doi: 10.1186/s13059-014-0439-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Shen R, Seshan VE. FACETS: allele-specific copy number and clonal heterogeneity analysis tool for high-throughput DNA sequencing. Nucleic Acids Research. 2016;44:e131. doi: 10.1093/nar/gkw520. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Taylor BS, Schultz N, Hieronymus H, Gopalan A, Xiao Y, Carver BS, Arora VK, Kaushik P, Cerami E, Reva B, Antipin Y, Mitsiades N, Landers T, Dolgalev I, Major JE, Wilson M, Socci ND, Lash AE, Heguy A, Eastham JA, Scher HI, Reuter VE, Scardino PT, Sander C, Sawyers CL, Gerald WL. Integrative genomic profiling of human prostate cancer. Cancer Cell. 2010;18:11–22. doi: 10.1016/j.ccr.2010.05.026. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Tosoian JJ, Carter HB, Lepor A, Loeb S. Active surveillance for prostate cancer: current evidence and contemporary state of practice. Nature Reviews Urology. 2016;13:205–215. doi: 10.1038/nrurol.2016.45. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Tschaharganeh DF, Bosbach B, Lowe SW. Coordinated tumor suppression by chromosome 8p. Cancer Cell. 2016;29:617–619. doi: 10.1016/j.ccell.2016.04.011. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Visakorpi T, Kallioniemi AH, Syvänen AC, Hyytinen ER, Karhu R, Tammela T, Isola JJ, Kallioniemi OP. Genetic changes in primary and recurrent prostate Cancer by comparative genomic hybridization. Cancer Research. 1995;55:342–347. [PubMed] [Google Scholar]
  36. Viswanathan SR, Ha G, Hoff AM, Wala JA, Carrot-Zhang J, Whelan CW, Haradhvala NJ, Freeman SS, Reed SC, Rhoades J, Polak P, Cipicchio M, Wankowicz SA, Wong A, Kamath T, Zhang Z, Gydush GJ, Rotem D, Love JC, Getz G, Gabriel S, Zhang CZ, Dehm SM, Nelson PS, Van Allen EM, Choudhury AD, Adalsteinsson VA, Beroukhim R, Taplin ME, Meyerson M, PCF/SU2C International Prostate Cancer Dream Team Structural alterations driving castration-resistant prostate cancer revealed by linked-read genome sequencing. Cell. 2018;174:433–447. doi: 10.1016/j.cell.2018.05.036. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Vujkovic M, Attiyeh EF, Ries RE, Goodman EK, Ding Y, Kavcic M, Alonzo TA, Wang YC, Gerbing RB, Sung L, Hirsch B, Raimondi S, Gamis AS, Meshinchi S, Aplenc R. Genomic architecture and treatment outcome in pediatric acute myeloid leukemia: a Children's Oncology Group report. Blood. 2017;129:3051–3058. doi: 10.1182/blood-2017-03-772384. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Wala JAL, Craft Yilong;, Schumacher David;, Steven E, Imielinski M, Haber JE, Roberts N, Yao X, Stewart C, Zhang C-Z, Tubio J, Ju YS, Campbell P, Weischenfeldt J, Beroukhim R. Selective and mechanistic sources of recurrent rearrangements across the Cancer genome. BioRxiv. 2017 doi: 10.1101/187609. [DOI]
  39. Walther A, Houlston R, Tomlinson I. Association between chromosomal instability and prognosis in colorectal cancer: a meta-analysis. Gut. 2008;57:941–950. doi: 10.1136/gut.2007.135004. [DOI] [PubMed] [Google Scholar]
  40. Wang H, Liang L, Fang JY, Xu J. Somatic gene copy number alterations in colorectal cancer: new quest for cancer drivers and biomarkers. Oncogene. 2016;35:2011–2019. doi: 10.1038/onc.2015.304. [DOI] [PubMed] [Google Scholar]
  41. Williams JL, Greer PA, Squire JA. Recurrent copy number alterations in prostate cancer: an in silico meta-analysis of publicly available genomic data. Cancer Genetics. 2014;207:474–488. doi: 10.1016/j.cancergen.2014.09.003. [DOI] [PubMed] [Google Scholar]
  42. Xia S, Kohli M, Du M, Dittmar RL, Lee A, Nandy D, Yuan T, Guo Y, Wang Y, Tschannen MR, Worthey E, Jacob H, See W, Kilari D, Wang X, Hovey RL, Huang CC, Wang L. Plasma genetic and genomic abnormalities predict treatment response and clinical outcome in advanced prostate cancer. Oncotarget. 2015;6:16411–16421. doi: 10.18632/oncotarget.3845. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Zack TI, Schumacher SE, Carter SL, Cherniack AD, Saksena G, Tabak B, Lawrence MS, Zhsng CZ, Wala J, Mermel CH, Sougnez C, Gabriel SB, Hernandez B, Shen H, Laird PW, Getz G, Meyerson M, Beroukhim R. Pan-cancer patterns of somatic copy number alteration. Nature Genetics. 2013;45:1134–1140. doi: 10.1038/ng.2760. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Zehir A, Benayed R, Shah RH, Syed A, Middha S, Kim HR, Srinivasan P, Gao J, Chakravarty D, Devlin SM, Hellmann MD, Barron DA, Schram AM, Hameed M, Dogan S, Ross DS, Hechtman JF, DeLair DF, Yao J, Mandelker DL, Cheng DT, Chandramohan R, Mohanty AS, Ptashkin RN, Jayakumaran G, Prasad M, Syed MH, Rema AB, Liu ZY, Nafa K, Borsu L, Sadowska J, Casanova J, Bacares R, Kiecka IJ, Razumova A, Son JB, Stewart L, Baldi T, Mullaney KA, Al-Ahmadie H, Vakiani E, Abeshouse AA, Penson AV, Jonsson P, Camacho N, Chang MT, Won HH, Gross BE, Kundra R, Heins ZJ, Chen HW, Phillips S, Zhang H, Wang J, Ochoa A, Wills J, Eubank M, Thomas SB, Gardos SM, Reales DN, Galle J, Durany R, Cambria R, Abida W, Cercek A, Feldman DR, Gounder MM, Hakimi AA, Harding JJ, Iyer G, Janjigian YY, Jordan EJ, Kelly CM, Lowery MA, Morris LGT, Omuro AM, Raj N, Razavi P, Shoushtari AN, Shukla N, Soumerai TE, Varghese AM, Yaeger R, Coleman J, Bochner B, Riely GJ, Saltz LB, Scher HI, Sabbatini PJ, Robson ME, Klimstra DS, Taylor BS, Baselga J, Schultz N, Hyman DM, Arcila ME, Solit DB, Ladanyi M, Berger MF. Mutational landscape of metastatic cancer revealed from prospective clinical sequencing of 10,000 patients. Nature Medicine. 2017;23:703–713. doi: 10.1038/nm.4333. [DOI] [PMC free article] [PubMed] [Google Scholar]

Decision letter

Editor: Michael R Green1
Reviewed by: Cory Abate-Shen2, Mark A Rubin3

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Tumor copy number alteration burden is a pan-cancer prognostic factor associated with recurrence and death" for consideration by eLife. Your article has been reviewed by Jeffrey Settleman as the Senior Editor, Michael Green as the Reviewing Editor, and three reviewers. The following individuals involved in review of your submission have agreed to reveal their identity: Cory Abate-Shen (Reviewer #1); Mark A Rubin (Reviewer #3).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

The manuscript analyzed the CNA landscape of conservatively treated prostate cancer in a historic biopsy and transurethral resection cohort and found that increased abundance of CNA in tumors is correlated with recurrence and mortality. This finding was further extended across multiple cancer types by analyzing data from TCGA and other cohorts. Given the prevalence of CNA throughout cancer genomes, this study highlights the potential for incorporating overall CNA burden assessment as a molecular prognostic factor in treatment decision making.

The reviewers agreed that the approach is rigorous and addresses important aspects of biomarker evaluation including taking into account clinical and pathology parameters. The reviewers also felt that the manuscript is clear and well-written. However, the reviewers also expressed a concern about the novelty of the results in the light of previously published work, as elaborated below. The reviewers would like to see your response to these concerns in a point-by-point response letter and in a revised manuscript before committing to a decision regarding appropriateness for publication in eLife.

Essential revisions:

1) The authors state that they are surprised by the findings that CNVs have such a strong association with clinical progression for prostate and other cancers. Yet, this is known in the prostate cancer research community and has been previously reported on in different eras based on available technology. Therefore, it would be reasonable for the authors to consider their finding in the context of the prior knowledge. For example, comparative genomic hybridization (CGH) technology led to some of the first genome-wide observations regarding copy number alterations and disease state (Cher et al., 1994; Visakorpi et al., 1995). CGH identified somatic copy number alterations in a high percentage (~75%) of localized PCA. Losses were found to be 5 times more common than gains and most often involved 8p (32%), 13q (32%), 6q (22%), 16q (19%), 18q (19%), and 9p (16%). These early genome-wide studies also suggested that the pattern of copy number alterations changes with disease progression. For example, gains of 7, 8q, and X were more often observed in the CRPC state (Visakorpi et al., 1995). SNP arrays also very nicely accessed CNV (Dumur et al., 2003; Lin et al., 2004). These platforms were used by the TCGA studies to perform a comprehensive copy number analysis for localized PCA (Cancer Genome Atlas Research Network. Electronic address and Cancer Genome Atlas Research, 2015). As well as a prior meta-analysis that showed recurrent CNVs were associated with disease progression (Williams et al., 2014). (Note: This paper is referenced in the text but for a more focused point). These are just a selection. There are others.

2) Related to the above comment, if the major point is that these new datasets add substantially more information, it would be nice to show a down-sizing experiment in silico wherein decreasing the coverage means the signal that supports these associations is lost. What is the minimal coverage needed to make these associations? This is important due to prior work that used older technologies to make similar claims.

3) There is recent published work that appears to be very similar. Pearlman et al., (2018) report, "pan-cancer metastasis potential score (panMPS) based on observed CNAs". panMPS predicts metastasis and metastasis-free survival in cohorts of patients with prostate cancer, triple negative breast cancer and lung adenocarcinoma, and overall survival in the Metabric breast cancer cohort and three cohorts from The Cancer Genome Atlas (TCGA), including prostate, breast and lung adenocarcinoma. And Ross-Adams et al., (2015) explored risk in association with CNV and integrating transcriptomic data, albeit on smaller prostate cancer data sets. These studies and work reported in bioRχiv from the ICGC groups on structural variation might also be valuable to include in a discussion of recent relevant prior work in this field.

4) Is the association between CNA burden and survival in prostate cancer explainable by differences in purity? Impure samples will have low apparent CNA burden.

5) The manuscript points out that "aneuploidy" but not "CNA burden" has previously been associated with outcome. In many cases, "aneuploidy" referred not to overall changes in tumor ploidy, e.g. due to genome doubling, but indeed to something similar to CNA burden. "Chromosomal instability" has also been used to describe the same phenomenon and has been associated with prognosis in multiple cancers. For example: Murayama-Hosokawa et al., 2010 (endometrial cancer); Walther et al., 2008; Mouradov et al., 2013; and Berg et al., 2015 (colorectal cancer, and making the important point that the relationship with CIN or CNA burden may reflect absence of MSI); Karlsson et al., 2007 (breast cancer); and Carter et al., 2006 (multiple cancer types).

6) The overall association between CNA burden and survival may be obscuring negative associations between specific CNAs and survival. Have the authors evaluated the association between specific CNAs and survival, after controlling for overall CNA burden?

eLife. 2018 Sep 4;7:e37294. doi: 10.7554/eLife.37294.027

Author response

Essential revisions:

1) The authors state that they are surprised by the findings that CNVs have such a strong association with clinical progression for prostate and other cancers. Yet, this is known in the prostate cancer research community and has been previously reported on in different eras based on available technology. Therefore, it would be reasonable for the authors to consider their finding in the context of the prior knowledge. For example, comparative genomic hybridization (CGH) technology led to some of the first genome-wide observations regarding copy number alterations and disease state (Cher et al., 1994; Visakorpi et al., 1995). CGH identified somatic copy number alterations in a high percentage (~75%) of localized PCA. Losses were found to be 5 times more common than gains and most often involved 8p (32%), 13q (32%), 6q (22%), 16q (19%), 18q (19%), and 9p (16%). These early genome-wide studies also suggested that the pattern of copy number alterations changes with disease progression. For example, gains of 7, 8q, and X were more often observed in the CRPC state (Visakorpi et al., 1995). SNP arrays also very nicely accessed CNV (Dumur et al., 2003; Lin et al., 2004). These platforms were used by the TCGA studies to perform a comprehensive copy number analysis for localized PCA (Cancer Genome Atlas Research Network. Electronic address and Cancer Genome Atlas Research, 2015). As well as a prior meta-analysis that showed recurrent CNVs were associated with disease progression (Williams et al., 2014). (Note: This paper is referenced in the text but for a more focused point). These are just a selection. There are others.

We thank the reviewers for highlighting this and appreciate the well demonstrated connection of specific CNVs and CNV signatures (derived from supervised analyses) to disease state in prostate cancer. We have expanded our consideration of this in the Introduction and Discussion section. As the reviewers note, CNV patterns or clusters have been associated with Gleason 8+ disease (compared to Gleason 6-7, Williams et al., 2014) and recurrent disease (compared to primary, Visakorpi et al., 1995, 2015). Indeed, our group initially found CNA patterns associated with shorter time to biochemical recurrence in prostate cancer (Taylor et al., 2010). Nonetheless, previous studies have primarily looked at the association of individual or sets of CNVs with pathology and disease state, rather than outcomes themselves and time to outcome.

The novelty of the work presented in our study lies in the association of (1) high quality clinical endpoints across cancer types and in a pan-cancer cohort with (2) the overall level or burden of genomic CNA of a tumor, agnostic to the identity of the component CNAs. Moreover, we adjust for prognostic pathological features which have previously been linked to CNVs, showing that association of CNA burden with outcome is independentof the previously published association with pathology features or disease state. We have expanded on this in the discussion to address the reviewers’ comments.

The key novel insight is the following: specific prognostic CNVs are likely to be cancer specific, whereas here we show the prognostic potential of a generalized measure of overall copy number dysregulation, as well as the feasibility of measuring this dysregulation using a current clinical NGS test. We acknowledge that the concept of CNA burden as a prognostic factor could seem rather obvious, at first glance, if viewed as a summation that includes all specific prognostic CNVs. But CNA burden could include both positive and negative prognostic CNVs that counterbalance this. Our work demonstrates that the prognostic utility of overall CNA burden is not lost in the noise of CNAs that are unassociated or negatively associated with poor outcome.

2) Related to the above comment, if the major point is that these new datasets add substantially more information, it would be nice to show a down-sizing experiment in silico wherein decreasing the coverage means the signal that supports these associations is lost. What is the minimal coverage needed to make these associations? This is important due to prior work that used older technologies to make similar claims.

This is an interesting question that raises a number of issues about coverage, resolution, and depth of genome sampling in assaying tumor copy number, both in newer sequencing approaches and older approaches such as array-based methods. The effective coverage of the genome is similar between targeted sequencing by IMPACT (due to the inclusion of SNP probes to specifically address the coverage question) versus older aCGH approaches (used for our conservative treatment cohort) as well as whole exome sequencing. Specifically, the method of calculating burden from the IMPACT targeted sequencing data incorporates SNP probes throughout the genome. We have updated the description of how CNA burden is calculated from IMPACT to clarify this. To illustrate the similarity in breadth of coverage between these technologies, we have added the prostate cohort analyzed by targeted IMPACT assay to Figure 1A. As shown with this addition, the coverage is similar with the newer clinically approved technology.

Rather than running a downsizing experiment, we took advantage of a pan-cancer IMPACT cohort for which whole exome sequencing was also performed (n=1005) and ran an upsizing experiment. The CNA burden determined by targeted IMPACT assay and by whole exome sequencing is highly correlated (rho=0.88, p-value<0.0001, Figure 2—figure supplement 3). We have added this analysis to the Results section and Discussion section. While the suggestion for a further downsizing experiment is interesting, we feel the IMPACT versus WES comparison addresses the main point raised by the referee. Efforts to further downsize from IMPACT to determine the minimal coverage required for prognosis determination is beyond the scope of this initial report and would best benefit from incorporation of other prognostic clinical markers.

3) There is recent published work that appears to be very similar. Pearlman et al., (2018) report, "pan-cancer metastasis potential score (panMPS) based on observed CNAs". panMPS predicts metastasis and metastasis-free survival in cohorts of patients with prostate cancer, triple negative breast cancer and lung adenocarcinoma, and overall survival in the Metabric breast cancer cohort and three cohorts from The Cancer Genome Atlas (TCGA), including prostate, breast and lung adenocarcinoma. And Ross-Adams et al., (2015) explored risk in association with CNV and integrating transcriptomic data, albeit on smaller prostate cancer data sets. These studies and work reported in bioRχiv from the ICGC groups on structural variation might also be valuable to include in a discussion of recent relevant prior work in this field.

We thank the reviewers for bringing our attention to the recent work by Pearlman et al., (2018). We have added a discussion of this study and how it derives a 295-gene frequency-weighted CNA signature derived by supervised analysis of prostate metastases that is prognostic for metastasis and overall survival in prostate and breast cancer. It is interesting to note that CNA burden outperforms their signature in one prostate cohort, while their signature outperforms CNA burden in the other prostate cohort they analyze (one in which PSA and stage are not prognostic). As a set of CNAs selected from prostate cancer, it may have less generalizable prognostic significance beyond the two cancers examined. We expand on this with some of the recent studies showing that CNA frequency, independent of identity, is associated with survival in pediatric AML (Vujkovic et al., 2017) and NSC lung cancer (Jamal-Hanjani et al., 2017). We have also added some discussion of prepublication results from ICGC and others (available on BioRχiv) on somatic CNA in cancer with regards to CNA, prognosis, and mechanism.

4) Is the association between CNA burden and survival in prostate cancer explainable by differences in purity? Impure samples will have low apparent CNA burden.

This is an important question which we have addressed with several additional analyses. First, we looked at the association between CNA burden and outcome in the TCGA prostate cohort after adjusting for purity using the ABSOLUTE purity metric, which was reported with the first publication of this cohort and represents a standard computationally-determined measure of purity. We find that CNA burden is still prognostic for recurrence after adjusting for sample tumor purity in the TCGA cohort (Table 2 and Results section). It is also still prognostic after adjusting for tumor purity along with Gleason and/or mutation burden, but these additional multivariate models are overfitted and we therefore elected not to include them in the revised manuscript. Next, we analyzed the IMPACT cohorts after adjusting for purity. Importantly, purity in these cohorts is more variable since these samples are not subject to the quality control requirements mandated by TCGA (typically >70% tumor content). We found that CNA burden in prostate tumors falls below the significance of association with overall survival when adjusting for purity determined by FACETS. However, purity-adjusted CNA burden is still significant in the pan-cancer primary tumor cohort and approaches significance in the metastatic tumor cohort (Supplementary file 3 and Results section).

This variability may in part be influenced by the smaller hazard ratio of CNA burden determined by IMPACT (sequencing panel) assay versus arrays (TCGA and TAPG/conservative treatment cohort) in these respective cohorts. It is also worth noting that a subset of samples in the IMPACT prostate cohort could not be included in the purity-adjusted analysis due to computational limitations in the purity calculation. Due to this limitation, we saw a decrease in the significance of association between mortality and CNA burden even before adjusting for purity, given with this smaller number of cases (P = 0.007 in the full cohort versus P = 0.02 in the subset for which a purity calculation was available).

In summary, tumor purity is a potentially significant variable that modulates CNA prognostic significance, and it will be important to continue to explore its effect especially in the context of additional molecular prognostic factors. It is interesting that within the IMPACT cohort, purity itself is associated with outcome and may warrant further investigation in such work. We have added a discussion of these considerations to the Discussion section.

5) The manuscript points out that "aneuploidy" but not "CNA burden" has previously been associated with outcome. In many cases, "aneuploidy" referred not to overall changes in tumor ploidy, e.g. due to genome doubling, but indeed to something similar to CNA burden. "Chromosomal instability" has also been used to describe the same phenomenon and has been associated with prognosis in multiple cancers. For example: Murayama-Hosokawa et al., 2010 (endometrial cancer); Walther et al., 2008; Mouradov et al., 2013; and Berg et al., 2015 (colorectal cancer, and making the important point that the relationship with CIN or CNA burden may reflect absence of MSI); Karlsson et al., 2007 (breast cancer); and Carter et al., 2006 (multiple cancer types).

We thank the referees for calling attention to thee related concepts of CNA burden, aneuploidy, and chromosomal instability. Aneuploidy has predominantly been defined as having a non-diploid DNA index, generally assayed by cytometry (e.g. Walther et al., 2008, which we thank the reviewers for directing our attention to, provides a metastudy which uses this metric for aneuploidy; Danielsen et al., 2016). The ploidy metric (determined by CLONET) used in our manuscript to adjust for aneuploidy in our multivariable regression model reflects the same information as DNA index conventionally used to measure aneuploidy. Indeed, ploidy estimates from SNP and sequencing data, such as the one we use, have been successfully benchmarked against FACS and SKY-determined DNA index in human tumors and cell lines (Carter et al., 2012; Prandi et al., 2014). Therefore, we believe our conclusion that the prognostic significance of CNA burden is independent of aneuploidy reflects the traditional measure of aneuploidy in prognostic significance.

However, we understand that aneuploidy has various, often more mechanistic, definitions (e.g., numerical aneuploidy/CIN and structural aneuploidy/GIN), some of which are not captured by ploidy as numerical DNA index. To address the reviewers’ points, we have clarified how the ploidy used in our analysis was determined and its relationship to ploidy and DNA index used in prognostic studies of aneuploidy or chromosomal instability (Results). We have also qualified our conclusion from this finding in the discussion to take into consideration the way aneuploidy was measured. We have also expanded the discussion of how future work will be needed to determine if CNA burden is prognostic independent of cancer type-specific prognostic classifications, including MSI-positive status in colorectal cancer which, as the reviewers note, is largely exclusive with CIN-positive status and may be involved in the prognostic interplay between CNA burden and CIN status (including Berg et al., 2015).

6) The overall association between CNA burden and survival may be obscuring negative associations between specific CNAs and survival. Have the authors evaluated the association between specific CNAs and survival, after controlling for overall CNA burden?

We looked for the association between survival and selected specific CNAs in multivariable regression models with CNA burden in prostate cancer (Table 3 and Materials and methods section). This analysis was focused on the question of whether the specific CNAs we examined accounted for the prognostic significance of CNA burden, and we found that they did not. These specific CNAs did have negative associations with survival (positive association with mortality) for the most part (or no significant association) and their negative associations with survival were not strengthened by the adjustment for CNA burden through its inclusion in the model.

The question about whether there are specific CNAs that have a positive association with survival that might be obscured by CNA burden is interesting and worthy of future exploration. We believe that many individual prognostic CNAs are likely to be cancer type-specific, and current cohorts unfortunately do not have sufficient size to run individual regressions for all CNAs to uncover these in a rigorous fashion.

Associated Data

    This section collects any data citations, data availability statements, or supplementary materials included in this article.

    Supplementary Materials

    Supplementary file 1. Cohort characteristics.
    elife-37294-supp1.docx (14.7KB, docx)
    DOI: 10.7554/eLife.37294.013
    Supplementary file 2. Distribution of cancer types in IMPACT cohorts.
    elife-37294-supp2.docx (18KB, docx)
    DOI: 10.7554/eLife.37294.014
    Supplementary file 3. Association between overall survival and CNA burden after adjustment for purity in IMPACT prostate and pan-cancer cohorts.

    Purity was determined by FACETS (Shen and Seshan, 2016).

    elife-37294-supp3.docx (17.7KB, docx)
    DOI: 10.7554/eLife.37294.015
    Supplementary file 4. TAPG1 conservative treatment primary prostate CNA cohort.
    elife-37294-supp4.csv (16.5KB, csv)
    DOI: 10.7554/eLife.37294.016
    Supplementary file 5. MSK-IMPACT primary prostate tumor cohort annotation.
    elife-37294-supp5.csv (18.5KB, csv)
    DOI: 10.7554/eLife.37294.017
    Supplementary file 6. MSK-IMPACT metastatic prostate tumor cohort annotation.
    elife-37294-supp6.csv (15.1KB, csv)
    DOI: 10.7554/eLife.37294.018
    Supplementary file 7. MSK-IMPACT primary pan-cancer cohort annotation.
    elife-37294-supp7.csv (381.2KB, csv)
    DOI: 10.7554/eLife.37294.019
    Supplementary file 8. MSK-IMPACT metastatic pan-cancer cohort annotation.
    elife-37294-supp8.csv (493KB, csv)
    DOI: 10.7554/eLife.37294.020
    Supplementary file 9. TCGA Cohort statistics: patient exclusion, events, and follow-up.
    elife-37294-supp9.docx (14.4KB, docx)
    DOI: 10.7554/eLife.37294.021
    Transparent reporting form
    elife-37294-transrepform.pdf (319.4KB, pdf)
    DOI: 10.7554/eLife.37294.022

    Data Availability Statement

    All data generated or analysed during this study are included in the manuscript and supporting files and reference materials. The conservative treatment TAPG copy number cohort array data was deposited in NCBI GEO under accession number GSE103665 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE103665).

    The following dataset was generated:

    Hieronymus H, author; Taylor BS, author; Sawyers CL, author. Copy number alteration burden is a pan-cancer prognostic factor associated with metastasis and death in conservatively treated prostate cancer: TAPG1 CNA cohort aCGH data. 2018 https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE103665 Publicly available at the NCBI Gene Expression Omnibus (accession no: GSE103665).

    Articles from eLife are provided here courtesy of eLife Sciences Publications, Ltd

    RESOURCES