Genomic Heterogeneity Within Individual Prostate Cancer Foci Impacts Predictive Biomarkers of Targeted Therapy

Abstract

Background:

Most lethal prostate cancers progress from relapse of aggressive primary disease. Recently, the most significant advances in survival benefit from systemic therapy have come from moving the administration of therapy to an earlier disease state. There is movement toward using biomarkers from the intraprostatic index lesion to guide early systemic therapy.

Objective:

To determine the genomic heterogeneity, including the heterogeneity of predictive biomarkers, within the index focus of treatment-naïve prostate cancer.

Design, setting, and participants:

Ten patients with treatment-naïve prostate cancer underwent prostatectomy. DNA was extracted from 70 spatially distinct regions of the 10 index foci.

Outcome measurements and statistical analysis:

Single nucleotide mutations, small indels, and copy number changes were identified. Intrafocal genomic heterogeneity and heterogeneity of alterations that predict response to therapy was determined.

Results and limitations:

Exome sequencing and copy number estimates demonstrate branched evolution with >75% of point mutations being subclonal, including numerous pathways associated with castrate-resistant prostate cancer. Seven of 10 patients harbor alterations in one of five genes that predict response to targeted therapies with survival benefit in prostate cancer. Within biomarker-positive cases, 25% of intraprostatic regions are biomarker negative, with discordance between intraprostatic regions and lymph node metastases.

Conclusions:

Treatment-naïve, nonmetastatic prostate cancer has marked intrafocal heterogeneity. Numerous alterations in pathways associated with castration-resistant prostate cancer are present in subclonal populations, including biomarkers predictive of response to targeted therapy.

Patient summary:

Untreated patients’ tumors have alterations that predict response to targeted therapies, but the presence of a biomarker is dependent on what region of the tumor was evaluated.

1. Introduction

Thirty percent of patients receiving definitive therapy for newly diagnosed prostate cancer relapse, with higher rates of relapse for those with high-risk or locally advanced disease [1–3]. For over 7 decades, systemic therapy has focused on the androgen receptor, a disease-specific target, but marked improvement in overall survival has been achieved by aggressively treating androgen-sensitive disease. One goal of genomic characterizations of both untreated and advanced prostate cancer is to identify subtypes of disease that predict sensitivity to new therapies [4–6]. These studies have suggested that in advanced disease over 60% of patients have targetable altered pathways in addition to the androgen receptor signaling pathway [6], although there are far fewer targets in treatment-naïve disease [5]. The poly adenosine diphosphate ribose polymerase (PARP) inhibitor olaparib is the targeted therapy furthest in clinical development, for patients with tumors harboring mutations in DNA repair pathways, comprising approximately 20% of advanced disease [6–8]. Although not yet approved for advanced disease, several trials are evaluating the use of PARP inhibitors for disease naïve to systemic therapy (, ).

The efficacy and durability of response to targeted therapies depends largely on the heterogeneity of the tumor and the prevalence of the targeted alteration within the tumor cell population. Multifocal prostate cancer is known to have early divergence between foci, but there is often a dominant, high-grade index focus considered to represent clinically significant disease [9–13]. Index foci have been reported to have heterogeneous gene expression, although more homogeneous than across distinct foci [14]. Heterogeneity at the DNA level has been reported for a handful of intermediate risk foci [11], although the extent to which this affects predictive biomarkers and other clinically significant genes is not fully characterized.

We undertook multiregion genomic evaluation of an individual index focus from 10 patients with localized (lymph node negative) or locally advanced (lymph node positive) disease, representing those most likely to relapse. We found branched genomic evolution characterized by marked heterogeneity in both point mutations and copy number alterations. Significantly, alterations predictive of response to targeted therapy are heterogeneous within the prostate and between intraprostatic disease and lymph node metastases.

2. Materials and methods

2.1. Patients and consent

Prostatectomy specimens were collected from patients undergoing radical prostatectomy at the University of Chicago Medical Center. Protocols were approved prior to the study, and all individuals included in the study provided written informed consent. All cases were reviewed by a genitourinary pathologist (GP), and three-dimensional (3D) reconstructions of the index focus were created based on location and histologic appearance of tumor within each block. Up to nine tumor regions from a single contiguous focus were chosen from each case. Cases were categorized as locally advanced (lymph node involvement) or localized (no lymph nodes involved).

2.2. Genomic analysis

Five to nine tumor regions were sampled with a 0.6-mm or 1.0-mm punch, and DNA was extracted using PAXgene Tissue DNA kit (Qiagen, Hilden, Germany). Germline DNA was extracted from blood except for case 1027, where it was obtained from histologically benign prostate tissue. Exome and low coverage whole genome sequencing performed on a HiSeq 2000 (Illumina, San Diego, CA, USA). Reads were evaluated with fastqc and cutadapt, mapped with bwa mem, duplicate reads removed with samtools, and mapping quality analyzed with picard. The average duplicate read rate was 0.242 and average coverage was 83x. A custom script using Fisher’s exact test identified positions where call differed between matched germline and tumor, which were filtered by removing positions with variant reads >2% or >2% in germline, or <4% or <10% in tumor samples, variants identified on only one strand, and positions with >0.05 frequency in 1000 genomes or ESP5400.

A subset of variants was verified with ultradeep targeted resequencing on a MiSeq (Illunimina; Supplementary Table 1). Subsequently, 17 regions were excluded from further analysis due to low tumor cell purity (< 30%) based on variant allele frequency of identified mutations, leaving 70 regions with sufficient sequencing quality. Positions harboring mutations that passed quality filters were then evaluated for all regions in the corresponding case using samtools mpileup. Mutations were considered present if the variant allele frequency >1%, and variant allele frequency >10% in at least one region. Indels were manually reviewed for read alignment and genomic context using Integrative Genomics Viewer.

Copy number estimates were generated by BICseq2 [15] setting filters to a p value <0.01 and log2 ratio <–0.2 or >0.2 [15]. Whole genome sequencing data was unavailable for case 1009, and copy number alterations were determined using CNVkit [16] where log2 ratio was <–0.2 or >0.2 and confidence interval excluded one. To avoid falsely elevating the fraction of subclonal events due to lower tumor purity in some samples, if log2 ratios in all regions from the case were in the same direction (> 0 or < 0), an alteration was considered clonal. If not, it was considered subclonal.

Immunohistochemistry was performed for erythroblast transformation-specific-related gene (ERG) expression. Nuclear ERG expression intensity and staining distribution were evaluated by a genitourinary pathologist (GP). All ERG-negative regions had <1% cells with nuclear ERG expression.

2.3.3. D representations of prostatectomy specimens and network trees

Representations of prostatectomy specimens were created using Povray (PO-Team, Victoria, Australia) based on manual mapping of the tumor focus and regions sampled. Network analysis was constructed using Network [17,18] from features comprising all somatic mutations identified.

2.4. Predictive genetic biomarkers

Potential biomarkers were queried using The Drug Gene Interaction Database (dgidb.genome.wustl.edu) [19]. A biomarker was considered positive if it contained both a nonsynonymous mutation and deletion (log2 < –0.2) or a deep deletion (log2 < –0.4). To avoid falsely elevating heterogeneity of biomarkers a deletion of log2 ratio < –0.2 was also considered a biomarker without mutation if at least one region in the case had a deep deletion at that locus.

2.5. Statistics

A two-sided Student t test using R was used to analyze the difference between the number of mutations in castration-resistant prostate cancer pathways and the percent genome altered in localized versus locally advanced cases.

3. Results

3.3.1. Intrafocal heterogeneity of treatment-naïve, localized, and regional disease

To examine the heterogeneity of treatment-naïve, potentially lethal primary prostate cancer, multiregion sequencing was performed on an individual index focus from the prostatectomy specimen of 10 patients. Clinical characteristics are typical of patients with intermediate, high-risk, or very high-risk disease, with Gleason scores 7–9 and pathologic tumor stage pT2–T3 and pN0 or pN1 (Table 1). Three patients had persistent prostate-specific antigen following prostatectomy, and four underwent additional therapy. For each case, exome and low coverage whole genome sequencing was performed on five to nine regions of a single contiguous index focus. Lymph node metastases obtained at the time of prostatectomy and seminal vesicle regions involved through direct extension were included for two cases each (1010 and 1022, 1015 and 1022, respectively).

Table 1 –

Clinical information for cases evaluated

	1009	1010	1015	1022	1029	1003	1020	1024	1027	1034
Stage category	Loc adv	Loc adv	Loc adv	Loc adv	Loc adv	Localized	Localized	Localized	Localized	Localized
Age at surgery (yr)	57	66	57	61	51	56	62	60	60	64
Ethnicity	AA	White	AA	White	White	White	White	White	White	AA
PSA (ng/ml)	8.4	12.8	43.1	NA	8.0	4.9	4.3	14.2	9.2	5.0
Pathologic stage	T3bN1	T3bN1	T3bN1	T3bN1	T3aN1	T2N0	T2N0	T3aN0	T3bN0	T2N0
Gleason grade	4+5	4+5	3+4 (5)	4+5	4+3	4+5	3+4	3+4	4+3 (5)	4+3
Tumor volume (%)	40	60	65	45	25	15	25	25	25	20
Surgical margin	Positive	Negative	Positive	Positive	Negative	Negative	Negative	Positive	Negative	Negative
Nodes examined	22	20	18	14	15	5	4	22	22	13
Nodes involved	6	1	2	4	1	0	0	0	0	0
Germline sample	Blood	Blood	Blood	Blood	Blood	Blood	Blood	Blood	Benign prostate	Blood
Prostate or SV regions	7	8	8	6	8	5	9	8	7	6
Lymph node regions	0	1	0	1	0	0	0	0	0	0
Time to recurrence (mo)	1	2	3	NA	NA	NA	NA	NA	NA	NA
Additional therapy	Yes	Yes	Yes	Yes	No	No	No	No	No	No

AA = African-American; Loc adv = locally advanced; NA = not applicable (have not recurred); PSA = prostate-specific antigen; SV = seminal vesicle.

We identified 887 unique mutation events across 70 index focus regions (88.7 variants/case, ~1.3 nonsilent variants/Mb of the exome; Supplementary Table 2). A subset of mutations was verified with ultra-high coverage targeted sequencing (~40 000×). Of 107 mutations assessed, two (2%) were not able to be confirmed, and none were found in the germline (Supplementary Table 1). Allele frequencies measured by exome sequencing were similar to those measured by targeted resequencing (R² = 0.861; Supplementary Fig. 1).

Exome sequencing demonstrated marked heterogeneity of single nucleotide mutations and indels within the cancer genome of the index focus in all cases examined (Fig. 1A). In each focus trunk mutations (mutations identified in all regions) comprised the minority. Work from The Cancer Genome Atlas proposed seven alterations—four genes altered by translocation/overexpression, and three by mutation—define the molecular subtype of 75% of localized prostate cancer [5].

Fig. 1.

Heterogeneity of mutations within index foci. (A) Exome sequencing was performed on five to nine regions of a single contiguous index focus from 10 cases of treatment-naïve prostate cancer. Presence (green) or absence (white) of all mutations, both nonsynonymous and synonymous, within each region sampled is shown. Regions are represented by columns, mutations by rows. Previously identified prostate cancer driver genes are indicated in black. Upper panels, the level of the radical prostatectomy (RP) specimen from which the region was sampled, from white (apex) to black (base), with purple (seminal vesicles) and red (lymph node metastasis) indicating extraprostatic regions. Left panels specify mutations as trunk (found in all regions), branch (at least two regions and not all regions), or leaf (identified in one region) mutations. (B) The distribution of transitions/transversions across all index focus regions sampled is similar in trunk (old), branch (middle), and leaf (recent) mutations. (C) The distribution of types of mutation across all index focus regions sampled is similar in trunk, branch, and leaf mutations.

freq = frequency; LN = lymph node; NS = XXX; SNV = single nucleotide variants; SV = seminal vesicle; Syn = XXX.

We evaluated for four (ERG, SPOP, FOXA1, IDH1) drivers, and five of 10 cases presented here could be assigned one of these subtypes (3 ERG overexpression, 1 FOXA1 mutation, 1 SPOP mutation; Supplementary Fig. 2). As with a single region sequencing approach, subtypes were mutually exclusive. As expected, these alterations occurred early, with the alteration identified in every region sampled in all but one case (ERG in case 1029; Supplementary Table 3).

Five of 10 cases had a mutation in one or more of five genes reported to be significantly mutated in localized prostate cancer by MutSigCV [5,20]: FOXA1 and SPOP (which are subtype-defining), and PTEN, TP53, and ZMYM3 (Fig. 1A). Early events included mutations in FOXA1, SPOP, PTEN, and TP53, involving all or nearly all regions in involved cases. PTEN was also a late mutation in a separate case, and ZMYM3 was a late mutation. TP53 and PTEN mutations, which are associated with poor outcome, were found only in locally advanced, not localized, cases.

3.2. Stability of mutation processes over time

Several studies have presented lines of evidence for a field effect in prostate cancer [12,21], suggesting there is a long accumulation of mutations over the life of the patient. Given this, we examined the types of alterations contributing to early (trunk), late (leaf), and middle (branch) events. In contrast to the 60% or more clonal mutations shown for 10 other solid tumors [22], fewer than 25% of variants identified were trunk variants. Unlike other malignancies like lung and kidney cancer [22–24], we found no evidence for changes in mutational processes over time (Figs. 1B and 1C).

3.3. Heterogeneity among structural changes

Chromosomal events such as TMPRSS2:ERG fusion appear to be early events in prostate cancer, present even in a fraction of premalignant lesions [25,26]. To identify the regional pattern of chromosomal changes, we performed copy number estimates from low-coverage whole genome sequencing (Fig. 2A, Supplementary Table 4). There were many canonical copy number changes that were estimated, including high frequency of loss of chromosomes 8p, 13, and 16, focal deletion on chromosome 10, and gain of chromosomes 8q and 7. Consistent with previous reports, lymph node positive locally advanced cases had a higher mean percent of genome altered than those without lymph node involvement, which is associated with a worse prognosis (Fig. 2B) [27,28].

Fig. 2.

Heterogeneity of chromosomal events within index foci. (A) Plots of copy number estimates based on log2 ratio of normalized read depth between tumor and normal demonstrating regions of trunk, branch, and leaf copy number events. Copy number estimates for case 1009 are based on exome sequencing, all others are based on low coverage whole genome sequencing. Left panel, the level of the prostatectomy specimen from which the region was sampled, as in Figure 1. Center panel, heat map of copy number estimates. Right panel, percent genome altered by copy number variation. (B) Mean fraction of the genome altered in locally advanced (lymph node positive) and localized (lymph node negative) cases. Error bars represent standard deviation.

Loc adv = locally advanced.

3.4. Correlation between histology and genetic features

We constructed network trees for each case based on mutation patterns. These were overlaid on 3D representations of the prostatectomy specimens (Fig. 3, Supplementary Figs. 3–11). In cases with histologic heterogeneity, it appeared the relationship between regions was as likely to correlate with histologic similarity than spatial proximity. For example, regions C36B and C46B in case 1010 share high-grade features and late divergence of their cancer genome, and both are more similar histologically and diverge later from C50A than regions that are more proximal, C36A and C46A, respectively.

Fig. 3.

Correlation between genomic and histologic features. Network analysis of case 1010 based on shared mutations among eight regions of a contiguous index focus and one region from a lymph node metastasis is overlaid on a cartoon representation of the prostatectomy specimen. Corresponding hematoxylin and eosin images of each region that underwent exome sequencing are depicted on the left. More closely related regions have fewer edges separating them.

Met = lymph node metastasis.

We also observed early divergence of spatially proximal regions without obvious correlations with histology (early divergence of region B3A in case 1024; Supplementary Fig. 8). In most cases, however, there was consistent histologic appearance across the focus, and spatially proximal regions showed late divergence. This was most clearly seen in cases 1024 and 1034 (Supplementary Figs. 8 and 11).

3.5. Subclonal alterations in clinically significant pathways

A recent report of the genomic landscape of castration-resistant prostate cancer demonstrated alterations in statistically or clinically significant genes from several pathways that contribute to prostate cancer progression, including androgen receptor, cell cycle, PI3K, RAF/RAS, WNT, DNA repair, and chromatin modifier pathways [6]. Many of these genes are found to be altered at low frequency in treatment-naive disease [5]. We hypothesized these genes might be altered in more cases of treatment-naive disease than previously appreciated, but in subclonal populations below the level of detection in other studies. Indeed, all 10 cases have alterations in at least two of 41 castration-resistant prostate cancer-related genes (Fig. 4A). Those with lymph node involvement at the time of prostatectomy had a higher median number of alterations in castration-resistant prostate cancer pathways than those without lymph node involvement (p value < 0.05; Fig. 4B). The majority of the alterations we identified were chromosomal events, highlighting the frequency and importance of chromosome level events compared with single nucleotide mutations [29].

Fig. 4.

Heterogeneity in castration-resistant prostate cancer (CRPC)-related alterations and predictive biomarkers. (A) Alterations across all cases in a panel of clinically relevant genes found to be altered in advanced prostate cancer [6]. (B) Number of mutations in CRPC-related genes in locally advanced and localized cases, p value <0.05. (C) Fraction of intraprostatic regions within a case that harbor deletion and mutation or deep deletion in five predictive biomarker genes. The fraction of regions with alterations represented by size of the blue circle (deletion) or “M” (mutation). Two cases (1010, 1022) had lymph node metastases evaluated, represented by gray shading of the square when alteration present. AR = androgen receptor; LN = lymph node; Loc Adv = locally advanced; NS = XXX; PI3K = phosphatidylinositol-4,5-bisphosphate 3-kinase.

Given that many of the castration-resistant prostate cancer pathways altered are potential therapeutic targets, we evaluated the heterogeneity in genetic biomarkers that predict response to targeted therapy. Using concurrent copy loss and mutation or deep deletion we identified five genes that predict susceptibility to seven targeted therapies (Supplementary Table 5), including two therapies with demonstrated efficacy in castration-resistant prostate cancer (olaparib, ipatasertib) [8,30], with individual cases harboring biomarker alterations in up to three separate genes (Fig. 4C). One case (1009) had a germline mutation in the DNA-binding domain of BRCA2 that is of unknown significance (p.Ala2730Pro), with concurrent somatic copy loss of BRCA2. The remaining predictive biomarkers were heterogeneous among regions sampled such that, of biomarker-positive cases, 25% of regions were biomarker negative. Moreover, there was discordance between intraprostatic regions and lymph node metastases: in two cases with lymph node metastases available, only one of three biomarker alterations found in intraprostatic regions was also found in the lymph node, and one of two alterations identified in a lymph node was found in the prostate (Fig. 4C).

4. Discussion

We present a multiregion genomic evaluation of the index focus of untreated primary prostate cancer, demonstrating that fewer than 25% of mutations are found in all regions of the focus. Coincident foci are known to be divergent, leading to reliance on the dominant index focus for prognostication. These data demonstrate remarkable heterogeneity within the clinically relevant index focus.

The treatment of prostate cancer has recently entered the genomic era, with PARP inhibitor therapy based on alterations in DNA repair pathways [6–8] demonstrating a response in 88% of biomarker-positive patients, and an Akt inhibitor has also demonstrated benefit in biomarker-positive patients [30]. There is interest in moving these therapies up to an earlier disease state, with trials underway testing olaparib in the neoadjuvant or biochemically recurrent space (, ), where biomarker evaluation is necessarily based on the primary tumor. In the present study we show that while prostate driver alterations are early events, predictive biomarkers such as ATM, BRCA1, and BRCA2 can be altered in a small fraction of cells of the index focus. This spatial discordance of biomarkers was demonstrated by evaluating entire prostatectomy specimens but would also be expected to apply when evaluating biomarkers in prostate biopsies, when only limited tissue is available for evaluation. This is in line with studies demonstrating that intratumoral heterogeneity is linked with poor outcome and therapy resistance [23,31–33]. While some of the cases evaluated had other lower grade foci, each case contained a single high-grade focus, suggesting that discordance between lymph nodes and intraprostatic tumors was due to subclonal heterogeneity rather than independent tumor foci. Currently there is uncertainty about how to manage high-risk, nonmetastatic disease, in part due to uncertainty about its biology. The LATITIUDE and CHAARTED trials [34,35] indicate a remarkable survival benefit in moving life-prolonging therapies up to an earlier disease state, and the STAMPEDE trial [36,37] suggests this also applies to nonmetastatic disease, although there are fewer events for earlier disease states. The findings presented here indicate that untreated prostate cancer, especially locally advanced disease, is biologically similar to metastatic, castrate-resistant disease. It harbors a high percent of the genome altered by copy number alterations, and there are numerous alterations typically associated with metastatic castration-resistant prostate cancer.

Our data expand the growing body of work demonstrating heterogeneity of primary prostate cancer. Previous work demonstrated interfocal [9,10,38] or intrafocal and interfocal heterogeneity [11,12,14], although these have focused on lower-risk, lower-stage disease, a disease state that is likely to be cured by surgery and is more dissimilar to widely metastatic disease. Multiple expression-based assays have been developed to refine the prognosis of localized disease, and Wei et al [14] demonstrated intrafocal and, to a greater extent, interfocal heterogeneity in these assays. Two studies have compared primary and metastatic disease and demonstrated cases of biomarker-negative prostate tumors and biomarker-positive metastases due to the accumulation of additional alterations in metastatic disease, which would presumably lead to a false-negative primary tissue biomarker [39,40]. The present study now demonstrates positive prostate tissue biomarkers that are subclonal and thus, are not representative of all tumor cells and are predicted to lead to a poor response.

A limitation of this study is the relatively small number of cases examined, and enrichment for cases with large tumor foci, which made feasible the sampling of the index focus from multiple regions. Not unexpectedly, within the cohort, larger foci were more likely to be heterogeneous. In addition, we focused on subclonal heterogeneity as evidenced by differences in DNA. It is not clear how the differences in biomarkers we observed influence differences in transcriptional profiles, though others have examined transcriptional heterogeneity [14].

These data demonstrate that there can be marked heterogeneity among regions of an individual index focus in untreated primary prostate cancer. This heterogeneity also applies to genetic biomarkers, highlighting the risk of selecting biomarker-driven therapies based on primary tissue. The concern about the heterogeneity of castration-resistant prostate cancer has led to the proposal of novel therapeutic strategies to compensate for this [41]. Our data suggest concerns about tumor heterogeneity also apply to the management of androgen-sensitive disease.

5. Conclusions

Treatment-naïve, nonmetastatic prostate cancer can have marked genomic heterogeneity within the index focus. Alterations found in subclonal populations within an individual focus include alterations found in metastatic castration-resistant prostate cancer, as well as biomarkers predictive of response to targeted therapy.