Frequent clonal relations between metastases and non-index prostate cancer lesions

Prostate cancer (PCa) has the highest incidence of all cancers in men worldwide (1). The majority of patients present with a tumor confined to the prostate and can be treated with curative intent. Prevention of metastatic outgrowth by local treatment of the prostate is a key objective of PCa treatment, as systemic metastatic spread is the main cause of mortality. For primary local treatment, a choice is made between radiotherapy (RT) and radical prostatectomy (RP), which are considered equally effective (2). However, both types of treatment are associated with significant adverse side effects, including urinary incontinence, impotence, and bladder and rectum dysfunction (3).

A more focal treatment of identified tumors in the prostate could spare men the toxicity of whole-gland treatments. However, PCa is often multifocal (4). In approximately 60% of patients, multiple subclones with distinct clinicopathological features are identified in the prostate at the time of diagnosis (5). Conventionally, the largest primary tumor with the highest grade is defined as the index lesion (6, 7). There is limited evidence that progression of the disease into disseminated PCa is linked to the index lesion (8, 9). Several approaches to focal therapy (FT) have been developed in recent years, aimed at ablating the index lesion and maintaining the oncological benefit of active treatments while preserving genitourinary and other organ functions (10).

However, a number of small reports indicated that metastatic disease does not uniformly originate from the index lesion but can also derive from small, obscure non-index primary lesions. Reports linking non-index lesions with metastases used fluorescence in situ hybridization analyses in 23 cases (11), targeted next-generation sequencing in a single patient case (12), and deep whole-genome sequencing in a cohort study of 10 cases (6). However, because most of these studies are statistically underpowered or used technologies that prevented a comprehensive interpretation of genomic data, it remains to be determined which percentage of lymph node (LN) metastases is clonally derived from a non-index lesion. Structural genomic rearrangements (SGRs) are common in PCa (13), and all SGRs cumulatively constitute a copy number alteration (CNA) profile. Profiling tumors for clonal SGRs reveals tumor evolutionary trajectories, and therefore CNA profiling can be applied to reveal clonality between primary lesions and metastases (12, 14, 15).

In this study, we used low-coverage whole-genome sequencing (LC-WGS) to uncover CNA profiles of matched primary lesions and local metastases. In total, we have interrogated CNA profiles of 70 primary lesions and matched them with 30 local metastases from 30 patients. By comparing CNA profiles in patients, we aim to shed light on the primary clonal origins of local metastases of PCa. Interestingly, in 23.3% of cases, the index lesion was not the clonal origin of the local metastasis. This suggests that focal ablation of only the index lesion might be an undertreatment of a significant proportion of PCa patients.

To investigate clonality between primary lesions and LN metastases, we obtained clinical data from a cohort of 30 patients who underwent RP and a pelvic LN dissection, with at least 1 primary tumor and 1 LN metastasis identified (Table 1). We hypothesized that CNA profiles would enable us to assess clonality between metastases and primary lesions within the same patient. Multiple primary tumors are observed in the prostate gland (Figure 1A), of which the largest tumor with the highest grade is considered the index lesion (P₁). Other primary lesions (P_2–n) can vary in size and grade, and pelvic LNs are a frequent site of regional PCa metastases.

Figure 1

Case study of patient 11. (A) Schematic representation of a prostate with multiple primary lesions (P) and a pelvic LN metastasis (N). (B) Example CNA profiles showing different SGR features in the index lesion P₁, other primary lesion P₂, and LN metastasis N. Breakpoint features are shared between N and P₁. (C) H&E staining of the FFPE whole-mount prostatectomy and an LN metastasis of patient 11. The index lesion P₁ and 2 other primary lesions (P₂ and P₃) as well as the LN metastasis are indicated. Marks in P₁ where 3 cores were taken are indicated. (D) Selected CNA profiles of patient 11 acquired by LC-WGS and processed with QDNAseq (see Methods). Black dots represent log2 ratio values in bins of 30 kb; red lines represent segmented log2 ratio values. (E) Pearson correlation heatmap of segmented CGHcall (15) values of all isolated tumors of patient 11. (F) Principal component analysis plot of segmented CGHcall values of P₁, P_2–n, and N.

Table 1

Cohort characteristics

Because genetically different primary lesions have unique CNA profiles, we here directly compared the CNA profiles of LN metastases with different primary lesions in the same patient to identify the disseminating primary lesion (Figure 1B). Our expert pathologist defined P₁ based on tumor size and Gleason score and delineated the areas of the other primary lesions (P_2–6) in the prostate and site N in LNs (Figure 1C). N was selected based on the largest lesion to maximize the chance of successful DNA isolation. A total of 63 primary tumors were identified in 30 patients.

As an example, we discuss patient 11 (Supplemental Table 1; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.124756DS1), from whom 3 primary tumors and a large LN metastasis are shown (Figure 1C) and whose primary tumors were also detected on MRI (Supplemental Figure 1). DNA was isolated from all visible primary tumors P_1–6 and N. Additionally, as a control, we isolated DNA from 2 extra sites of index lesion P₁ (denoted P₁* and P₁** in figures) and performed LC-WGS for CNA profiles of all DNA samples. CNA profiles of P₁ and N shared common CNA profile features, while P₂ and N did not, suggesting that N disseminated from P₁ (Figure 1D). Pearson correlation (Figure 1E) and a principal component analysis (Figure 1F) both showed a correlation between P₁ and N versus P_2–6.

To estimate the percentage of patients with regional metastases that did not disseminate from the primary PCa index lesion, we selected 30 FFPE prostatectomy specimens and concurrent regional metastases (Table 1). Among all 30 patients, previously reported frequent CNAs (16, 17) were identified, including deletions at TMPRSS2-ERG, PTEN, and chromosome 8p (Figure 2A). Globally, no major differences were observed in aggregate CNAs between primary lesions and regional metastases (Figure 2, B and C). We did not find clear copy number enrichment of specific regions when comparing primary tumors that metastasized versus those that did not. However, given the relatively limited resolution of the data, focal gains and losses are likely overlooked, and thus, conclusive statements on focal aberrations cannot be made at this stage.

Figure 2

CNA profiles of PCa lesions and LN metastases in entire cohort. (A) CNA overview heatmap of 30 patients with zoomed-in regions of interest, showing typical PCa SGRs, such as chromosome 21 TMPRSS2-ERG focal deletions, chromosome 10 PTEN deletions, and chromosome 8q arm loss and 8p gain. Known blacklisted genomic loci were excluded, showing amplifications (red) and deletions (blue). (B) Aggregate frequency plot of all primary lesion CNAs, n = 70. (C) Aggregate frequency plot of all LN metastasis CNAs, n = 30.

Both a log LR and P value of clonality were calculated for pairs of P–N samples compared to a reference distribution representing the average profile of all samples (18). All intrapatient pairs were divided in 2 groups to separate index lesions (P₁–N pairs) from other primary lesions (P_2–n–N pairs). We defined pairs to be clonal for LR P values of less than 0.01 and observed more clonal P₁–N pairs than P_2–n–N pairs (Figure 3A). To assess the percentage of clonality for index lesions and LN metastases, we compared the similarity of breakpoints of CNA heatmaps for individual patients (Figure 3B). Interestingly, in 23 out of 30 patients (76.7%), P₁ was clonal with N (Figure 3B), and consequently in 23.3% of patients, the metastasis was not clonally related to the index lesion. For most cases, the other smaller foci studied were not clonally linked with the metastasis either, and only for 1 case we managed to connect a non-index lesion with the metastasis analyzed (Supplemental Figure 2). Based on these findings, we conclude that in most of these patients, the metastasis originated from an undetected, obscure primary lesion, as was reported earlier (12).

Figure 3

Clonality assessment of PCa lesions and LN metastases. (A) Density plot of log likelihood ratio (LR) of 3 pair groups: reference distribution, P₁–N pairs, and P_2–n–N pairs. A box plot representation of the density plot is shown in the corner. Dotted line represents LR P value of 0.01, which is considered significant. (B) Intrapatient CNA heatmap comparison of P₁ and N, indicating that 23.3% of LN metastases are not clonally related to the primary index lesion. Two CNA heatmaps from 2 patients are shown as examples. Patient 3 shows no clonality between P₁ and N. Patient 13 shows clonality between P₁ and N based on identical breakpoints in genomes.

In this study, we identified a clonal connection between primary tumors and a single LN metastasis, which could be considered a limited view on metastatic disease progression. The original paradigm of linear progressive metastatic seeding is increasingly giving way to an emerging view of more diverse patterns of subclonal metastatic seeding, such as (a)synchronous polyclonal seeding and cross-metastatic seeding (19, 20). LN metastases are not necessarily an intermediate step for the development of distant metastases but can also represent a genetic dead end in tumor phylogenies (21).

Robust comparative data on medium- to long-term oncological outcomes of FT versus whole–prostate gland treatments are currently lacking (10). Although FT is considered experimental, it is offered to patients with low to intermediate (<50%) risk of PCa recurrence (19). Implementation of FT without any risk of recurrence would require extensive preoperative biopsies (4), which would be impractical and can still easily miss small lesions.

However, important questions on the relation between primary lesions and lethal disease remain. Which primary lesion contributed to PCa recurrence after targeted ablation when any primary lesion theoretically has the potential to metastasize? Because we did not sample distant metastases, future studies should be aimed to yield more comprehensive insights into metastatic spread. Although we identified multiple tumor-positive LNs in some patients, successful FFPE DNA extraction requires both high tumor cellularity and sufficient size of the LN metastasis being analyzed. Due to this restriction, we focused our analysis on clonality between primary tumors and the biggest tumor-positive LN. However, PCa metastases were reported to share a monoclonal background (7), suggesting that assessing a single metastasis captures the early evolutionary history of metastatic PCa.

Interestingly, we found that in 23.3% of cases the index lesion was not the clonal origin of the LN metastasis in a high-risk cohort of patients who underwent a prostatectomy and a pelvic lymphadenectomy. Despite the difference in risk classification between patients offered FT and our cohort, it is highly likely that in a significant proportion of patients treated with FT, the primary source of metastatic lesions remains in situ. This suggests that a significant proportion of FT-treated patients are potentially undertreated and would consequently have a higher risk of PCa recurrence. Accordingly, our results urge for randomized trials comparing oncological outcomes of patients treated with RP or RT as compared to FT.

Patient selection. We selected 30 patients from the Netherlands Cancer Institute Antoni van Leeuwenhoek (NKI-AVL) biobank based on availability of a local LN metastasis size greater than 4 mm. Clinicopathological characteristics of this cohort are shown in Table 1.

Sample acquisition and DNA isolation. All tissues were processed into FFPE material by the molecular pathology division of the NKI-AVL using standard protocols. For determining tumor areas, 5-μm sections of FFPE prostate and local metastasis samples were prepared. H&E staining was performed according to standard protocols. Our expert pathologist delineated areas with a tumor cell percentage higher than 60% for all visible primary lesions and local metastases. Sample acquisition was performed by collecting 0.6-mm cores from the FFPE block in tumor areas. Core biopsies of primary tumors, metastases, and normal prostate tissue of all samples in the cohort were included in a tissue microarray (TMA). TMAs of the RP tissue were constructed using a manual tissue arrayer (Beecher Instruments). TMAs of the LN metastasis were constructed using the automated TMA Grand Master (3DHISTECH). Depending on availability, 2 or 3 different cassettes per tumor were used for TMA construction, and 3 cores of 0.6-mm diameter were put in the TMA. H&E slides were cut to assess the quality of the TMA.

An additional dropout TMA was constructed for tumor cores that had low core quality. Finally, DNA was isolated from TMA FFPE cores using standard protocols. High-throughput sequencing data were deposited in the NCBI’s Sequence Read Archive (SRP178079).

Sequencing. Isolated DNAs were preprocessed and obtained DNA libraries were analyzed on a 2100 Bioanalyzer using a 7500 chip (Agilent), diluted, and pooled equimolar into a 10-plex sequencing pool (Illumina Inc.). The libraries were sequenced with 65 base single reads on a HiSeq 2500 system using version 4 chemistry (Illumina Inc). Obtained reads were mapped to Homo sapiens reference genome hg19 using BWA-backtrack (23). In general, we achieved an average genome coverage of 0.2–1 times.

QDNAseq. Mapped sequence reads obtained with LC-WGS were analyzed using QDNAseq by R (24) with a 30-kb window and default settings. The median normalized log2 ratios from QDNAseq were segmented with circular binary segmentation (25). After segmentation, samples were mode normalized and chromosomal copy number changes defined using CGHcall by R (15).

Pearson correlation heatmap, principal component analysis, frequency plots, and CNA heatmaps. The Pearson correlation matrix was calculated based on the segmented log2 ratio values for patient 11 and plotted with heatmap.3 (26) using ward.D2 hierarchical clustering. Next, we performed a principal component analysis on segmented log2 ratio values to reduce dimensionality to 2 dimensions.

CGHcall data were plotted in both a genome-wide and zoomed-in CNA profile heatmap for the entire cohort. Moreover, we plotted aggregate CNA frequencies for 2 selected groups: primary tumors and metastases from combined CGHcall data.

Clonality R package. For each set of DNA copy number profiles from a single patient, we calculated the likelihood 2 samples were clonal. This was performed using the Clonality R package with default settings. CGHcall CNA data were downsampled 40-fold for optimal performance of the package as Ostrovnaya et al. described (18).

Statistics. Clonality package log LR ratio statistics were computed as described originally (18). One-tailed Student’s t test was used to compare the means among reference, P₁-N, and P_2-n-N groups.

Study approval. This study was performed in accordance with institutional medical ethical guidelines. The use of anonymous or coded leftover material for scientific purposes is part of the standard treatment agreement with patients, and therefore informed consent was not required from patients not actively opting out, according to Dutch law (22). The NKI-AVL has set up a secondary-use biobank under the conditions in “Human Tissue and Medical Research: Code of Conduct for Responsible Use” (Federa, 2011, http://www.federa.org/codes-conduct). The use of NKI-AVL biobank material requires review and approval by NKI-AVL’s Translational Research Board (TRB) according to the aforementioned Code of Conduct. Patient material was used only after approval by the TRB when otherwise disposed of or unnecessary for diagnostics.