Implementation of a prostate cancer-specific targeted sequencing panel for credentialing of patient-derived cell lines and genomic characterization of patient samples

Abstract

Background:

Primary and metastatic prostate cancers have low mutation rates and recurrent alterations in a small set of genes, enabling targeted sequencing of prostate cancer-associated genes as an efficient approach to characterizing patient samples (compared to whole-exome and whole-genome sequencing). For example, targeted sequencing provides a flexible, rapid, and cost-effective method for genomic assessment of patient-derived cell lines to evaluate fidelity to initial patient tumor samples.

Methods:

We developed a prostate cancer-specific targeted next-generation sequencing (NGS) panel to detect alterations in 62 prostate cancer-associated genes as well as recurring gene fusions with ETS family members, representing the majority of common alterations in prostate cancer. We tested this panel on primary prostate cancer tissues and blood biopsies from patients with metastatic prostate cancer. We generated patient-derived cell lines from primary prostate cancers using conditional reprogramming methods and applied targeted sequencing to evaluate the fidelity of these cell lines to the original patient tumors.

Results:

The prostate cancer-specific panel identified biologically and clinically relevant alterations, including point mutations in driver oncogenes and ETS family fusion genes, in tumor tissues from 29 radical prostatectomy samples. The targeted panel also identified genomic alterations in cell-free DNA and circulating tumor cells from patients with metastatic prostate cancer, and in standard prostate cancer cell lines. We used the targeted panel to sequence our set of patient-derived cell lines; however, no prostate cancer-specific mutations were identified in the tumor-derived cell lines, suggesting preferential outgrowth of normal prostate epithelial cells.

Conclusions:

We evaluated a prostate cancer-specific targeted NGS panel to detect common and clinically relevant alterations (including ETS family gene fusions) in prostate cancer. The panel detected driver mutations in a diverse set of clinical samples of prostate cancer, including fresh-frozen tumors, cell-free DNA, circulating tumor cells, and cell lines. Targeted sequencing of patient-derived cell lines highlights the challenge of deriving cell lines from primary prostate cancers and the importance of genomic characterization to credential candidate cell lines. Our study supports that a prostate cancer-specific targeted sequencing panel provides an efficient, clinically feasible approach to identify genetic alterations across a spectrum of prostate cancer samples and cell lines.

Introduction

The genomic landscape of primary and metastatic prostate cancers consists of a relatively limited number of recurrent somatic alterations, indicating that most common alterations in prostate cancer may be captured by sequencing a limited number of genes. Whole-exome sequencing (WES) and whole-genome sequencing (WGS) studies of large prostate cancer cohorts have identified recurrent somatic mutations including single nucleotide variants (SNVs), copy number changes, and structural rearrangements ^1–7. Primary prostate cancers have a lower somatic mutation frequency (~1 mutation/Mb) compared to primary tumors of other cancer types (1–10 mutations/Mb) ⁸. Among primary prostate cancers, recurrent mutations commonly occur in TP53, SPOP, and FOXA1. Deletions of RB1, PTEN, and CHD1 and amplifications of MYC are common, as are ETS family gene fusions such as TMPRSS2-ERG.

Comprehensive genomic sequencing has also elucidated the landscape of alterations in metastatic castration-resistant prostate cancer (mCRPC) ^6,7,9–13. Even within this group of metastatic tumors, a relatively low mutation rate was observed at 3–4 mutations/Mb. Somatic mutations in mCRPC tumors are found in several genes that are uncommonly altered in primary prostate cancers, including PIK3CA, PIK3CB, RSPO, BRAF, RAF1, APC, CTNNB1, and ZBTB16 (PLZF) ⁹. Alterations in the androgen receptor (AR) are common in mCRPC but are rarely seen in primary treatment-naïve prostate cancer. These alterations are associated with androgen therapy resistance and include SNVs and amplification of the AR gene locus; structural rearrangements of AR and amplifications affecting an AR enhancer are also seen ^10,11. Deletions and mutations in PTEN and TP53 are present in 40–50% of cases. A substantial fraction of patients harbor defects in DNA repair proteins, and such patients may exhibit responses to poly(ADP-ribose) polymerase (PARP) inhibitor therapy ¹⁴. Furthermore, patients with metastatic prostate cancer have a high frequency (12%) of germline DNA repair alterations, which may have implications for both therapy and prevention of prostate cancer ¹⁵.

ETS family gene fusions are present in about 50% of primary and metastatic prostate cancers. Detection of ETS rearrangements typically relies on fluorescence in situ hybridization, whole-genome sequencing, or RT-PCR (which may be complicated by variations in fusion architecture), but these methods are not part of standard clinical practice ¹⁶.

Despite progress in the genomic characterization of prostate cancers, insufficient cell line models are available to represent the genetic diversity elucidated in these sequencing studies. Human prostate cancer cells are difficult to grow in cell culture; thus, few prostate cancer cell lines are available ¹⁷, and do not encompass the full spectrum of genetic alterations observed in patients, such as TMPRSS2-ERG, SPOP mutations, and CHD1 deletions ¹⁸. New methods to generate human prostate cancer cell lines include organoid methods ^19–22 and a “conditional reprogramming” method with irradiated mouse fibroblast feeder cells and an inhibitor of Rho-associated kinase ^23–26. These methods support the growth of both normal and malignant prostate cells from patients; indeed, growth of normal prostate cells is robust in both conditional reprogramming and organoid systems ^26,27. Thus, accurate methods are required to distinguish tumor and normal cells in new patient-derived cultures.

To address these challenges, we designed a prostate cancer-specific targeted next-generation sequencing (NGS) panel and applied it to evaluate genomic features of patient-derived cell line models and patient samples. Our targeted panel incorporates exon sequences of 62 genes frequently altered in prostate cancer as well as tiling across the intronic regions of genes with frequent rearrangements and copy alterations in prostate cancer. This approach allows single-nucleotide variants (SNVs), amplifications/deletions, and rearrangements to be simultaneously detected in the sequencing process, and is flexible in design such that the selection of genes can change over time. The prostate cancer-specific targeted panel was applied to primary prostate tumors as well as other clinical specimen types including circulating tumor cells (CTCs) and cell-free DNA (cfDNA) from metastatic prostate cancer patients. The panel detected driver mutations and/or rearrangements of known prostate cancer-associated genes in the majority of patient samples and in commonly used prostate cancer cell lines. In parallel, we attempted to generate new patient-derived cell lines from normal and malignant prostate tissue from primary prostate cancers using conditional reprogramming. We applied the prostate cancer-specific panel to interrogate the genomic features of the patient tissue samples and patient-derived cell lines.

Methods

Patient samples and cell lines

A schematic overview of the study is shown in Figure 1. We received Institutional Review Board (IRB) and institutional approval from Dana-Farber Cancer Institute (DFCI) and Massachusetts General Hospital (MGH) to collect samples of primary prostate cancers and adjacent normal prostate tissue from prostatectomy surgeries. Genitourinary pathologists at DFCI or MGH selected prostate cancer tissue fragments and adjacent normal-appearing prostate tissue from prostatectomy specimens. Fresh tissues were dissociated, and cell lines were generated according to the conditional reprogramming protocol ²³, with modifications as detailed in Supplementary Methods. Briefly, cell suspensions were generated by mechanical dissociation and enzymatic digestion, followed by plating cells in F-medium on a layer of irradiated 3T3-J2 fibroblasts or with conditioned medium from these fibroblasts ^28–30. Additional samples from each specimen were snap-frozen or cryopreserved in FBS/10% DMSO. We also acquired DNA samples from 2 tumors previously collected in a whole-genome sequencing study ¹. We acquired CTC DNA and cfDNA samples from patients with metastatic prostate cancer that were previously isolated and sequenced under IRB-approved clinical protocols at DFCI ^31,32. Prostate cancer cell lines were obtained from collaborators at DFCI and cultured in standard medium.

Figure 1.

Schematic of patient samples. Prostate cancer tissue and adjacent normal tissue from primary prostatectomies were used to generate patient-derived cell lines; both tumors and cell lines were sequenced with a prostate cancer-specific targeted panel. The targeted panel was also applied to circulating tumor cells and cell-free DNA from patients with metastatic prostate cancer.

Prostate cancer targeted panel

The targeted panel used in this study includes 62 genes that are recurrently mutated in prostate cancer and/or relevant to prostate cancer biology, selected based on published studies at the time of design of the bait set (Table 1) ^1,7,33. We designed RNA bait sequences (Agilent SureDesign, Custom SureSelect XT platform) targeting the exons of these genes as well as intronic and intergenic regions of ERG, TMPRSS2, ETV1, ETV4, SLC45A3, RAF1, and PTEN; total bait size is 790 kb.

Table 1.

List of prostate cancer-associated genes in the targeted panel

AKT1

APC

AR

ARID1A

ATM

BRAF

BRCA2

C14orf49

CDK12

CDKN1B

CDKN2A

CHD1

CTNNB1

EGFR

ELAC2

EPHA7

ERG*

ETV1*

ETV4*

ETV5

EZH2

FGFR3

FOXA1

FOXP1

HOXB13

HRAS

IDH1

KDM6A

KRAS

MED12

MLH3

MLL

MLL2

MLL3

MSH6

MSR1

MUTYH

MYC

NCOA2

NCOR1

NIPA2

NKX3-1

NRAS

OR5L1

PIK3CA

PIK3R1

PTEN*

RAF1*

RB1

RNASEL

RYBP

SCN11A

SHQ1

SKP2

SLC45A3*

SPOP

SPTA1

THSD7B

TMPRSS2*

TP53

ZFHX3

ZNF595

The most commonly altered genes in primary and metastatic prostate cancer were selected for inclusion in the panel. The custom RNA bait set for targeted capture targets the exons of most genes. Selected genes frequently involved in somatic copy number variations and/or rearrangements (indicated with an asterisk) had additional bait tiling across intronic and intergenic regions.

Targeted sequencing with prostate-specific panel of custom RNA baits

Snap-frozen solid tissue pieces were dissociated using TissueRuptor (Qiagen). DNA from these tissue or cell line samples was isolated using Allprep or DNeasy Blood and Tissue kits (Qiagen) and quantitated using Qubit dsDNA HS assay kit. The SureSelectXT Target Enrichment System for Illumina Paired-End Sequencing (Agilent) was used for library preparation and hybrid capture according to the manufacturer’s protocol with 200 ng input. The enriched DNA libraries were PCR-amplified using individual indexing primers containing 8-bp indexes to enable pooling for sequencing and subsequent deconvolution, followed by bead purification. DNA quantity/quality of the final amplified, indexed libraries was assessed by Agilent Bioanalyzer. Samples were pooled for sequencing on an Illumina HiSeq 2500 instrument with 200 paired-end reads.

Sequencing data analysis

Most genomic analysis was conducted in the Firehose environment (http://archive.broadinstitute.org/cancer/cga/firehose). Somatic single-nucleotide variants (SNVs) were called using MuTect ³⁴, and short insertions and deletions (indels) were called using Indelocator (http://archive.broadinstitute.org/cancer/cga/indelocator), except in the case of whole-exome sequenced CTC and cfDNA samples which were called with Strelka ³⁵. Data filtering steps and quality control thresholds are described in Supplementary Methods. To detect gene fusions, somatic rearrangements were identified using dRanger (http://archive.broadinstitute.org/cancer/cga/dranger), and the candidate rearrangement breakpoints were passed into Breakpointer ³⁶. Only candidate gene fusions between TMPRSS2 or SLC45A3 and a defined list of fusion partners (ERG, ETV1, ETV3, ETV4, ETV5, ETV6, ELK4, or BRAF) were selected. Somatic mutations and fusion calls for the two WGS samples were obtained directly from a published WGS study ¹. The Cancer Cell Line Encyclopedia (CCLE) data used to compare established cell lines sequenced on the targeted panel was based on a panel of 1651 genes that was used to profile mutations in CCLE cell lines (http://www.broadinstitute.org/ccle) ¹⁷. Our prostate cancer-specific targeted panel and the sequencing panel used in CCLE had 39 genes in common. Somatic copy number alterations (SCNAs) in samples sequenced using the targeted panel were assessed through an ad hoc method described in Supplementary Methods. Somatic CNAs in whole-exome sequenced samples were determined using ReCapSeg (http://gatkforums.broadinstitute.org/gatk/categories/recapseg). Each detected somatic alteration was evaluated for significance by manually assessing the frequency in prostate cancers via cBioPortal (http://www.cbioportal.org/) and COSMIC (https://cancer.sanger.ac.uk/cosmic) databases (accessed 3/2018).

Rearrangement analysis from cell-free DNA

An independent set of cfDNA samples from patients with metastatic CRPC underwent sequencing using a modified version of the prostate cancer-specific targeted panel that was adapted to detect common rearrangements and tile across the AR locus. cfDNA was extracted using the Qiagen Circulating DNA kit on the QIAsymphony liquid handling system. Tumor fraction was estimated from ultra-low pass whole genome sequencing by ichorCNA ³². cfDNA samples were quantitated using PicoGreen and their size distribution was analyzed by BioAnalyzer (Qiagen). 20 ng of input cfDNA was used for library construction using KAPA HyperPrep kit. The cfDNA was not fragmented before library prep. All libraries were quantified by sequencing on an Illumina MiSeq and then normalized based on the resulting sequencing metrics. The libraries were then pooled in equal mass to 500 ng and enriched using the bait set with the Agilent SureSelectXT Hybrid Capture kit. Pooled sample reads were de-convoluted and sorted using Picard tools ³⁷, and aligned to the reference genome using BWA and the GATK tool for localized realignment around indel sites ^38,39. Translocations were detected using Breakmer ⁴⁰.

Results

Clinical cohort

Under IRB-approved protocols, 31 tumor specimens from primary prostatectomies were collected (Table 2), along with 29 paired normal prostate tissue specimens from the same surgeries, as assessed by a genitourinary pathologist. The tumor percentage in each tumor tissue piece was estimated by a genitourinary pathologist. One tumor/normal pair was excluded for low (<10%) tumor content in the tumor sample.

Table 2.

Clinical and pathologic data from primary prostatectomy samples

Sample	Age	Gleason score	Pathologic stage	Tumor %
PK01	66	4+5	pT2cN0	60%
PK02	66	3+4	pT3aN0	90%
PK03	59	3+4	pT2cN0	95%
PK04	52	3+4	pT3bN0	70%
PK05	63	4+4	pT3aN1	80%
PK06	67	4+4	pT3bN0	80%
PK07	69	4+3	pT3aN0	80%
PK08	56	4+5	pT3bN1	100%
PK09	62	3+4	pT2aN0	85%
PK010	57	4+4	pT3bN0	95%
PK011	54	4+5	pT3bN0	90%
PK012	47	3+4	pT2cNx	70%
PK013	64	3+4	pT2cN0	N/A
PK014	66	4+3	pT2cN0	N/A
PK015	71	3+4	pT2cN0	N/A
PK016	54	3+4	pT3aN0	N/A
PK018	71	4+5	pT3aN0	N/A
MGH1	52	3+3	pT2cN0	90%
MGH2	67	4+5	pT3aN0	70%
MGH3	71	4+3	pT3bN0	80%
MGH4	66	3+4	pT2cN0	90%
MGH5	70	3+4	pT3aN0	50%
MGH6	52	3+3+(4)	pT2cNX	40%
MGH7	58	4+3	pT2aNx	N/A
MGH9	65	4+4	pT3aN1	60%
MGH10	60	3+4	pT3aN0	85%
MGH11	62	3+4	pT2cN0	25%
MGH12	74	4+3	pT2aN0	80%
MGH13	58	4+4	pT3aN0	95%
MGH15	70	3+4	pT3aN0	90%

Clinical and pathologic data for primary prostatectomy samples. The fraction of tumor in each tumor tissue piece was estimated by a genitourinary pathologist and the maximum percent tumor is listed. Gleason score and pathologic stage were determined by a genitourinary pathologist. Age indicates the patient age at time of sample collection. N/A, not available

Prostate cancer-specific targeted sequencing panel samples and sequencing metrics

To validate the prostate cancer-specific targeted sequencing panel, we sequenced multiple prostate cancer sample types (Figure 1), including: 1) fresh-frozen tumor and normal tissue from 30 primary prostatectomies; 2) DNA from two samples of primary prostate cancer that had previously undergone WGS ¹; 3) DNA from four individual CTCs from each of two patients with metastatic prostate cancer (DNA from the same CTCs and from a biopsy of each patient’s metastatic tumor underwent WES separately ³¹); 4) cfDNA isolated from the blood of two patients with metastatic prostate cancer (WES of CTCs and metastatic tumor biopsy from each patient was performed previously ³²); and 5) eight prostate cell lines including immortalized prostate epithelial cells and tumor cell lines (Supplementary Table 1). In addition, we sequenced new patient-derived cell lines generated by conditional reprogramming from tumor and normal tissues (see below).

Across all samples, the mean target coverage was 118X (as expected, coverage varied depending on sample quality and number of samples pooled per lane). The distribution of coverage by sample and by gene is shown in Supplementary Figure 1. One tumor/normal pair was excluded due to poor coverage of the tumor sample. One normal tissue specimen failed library preparation. Two normal-tissue-derived cell lines were excluded due to poor coverage. Across all samples, the percent bases selected was approximately 20%. The percent selected is relatively low relative to optimized whole-exome sequencing panels, but is within the expected range for custom panels, and was also probably lower due to off-target effects related to the baits tiled across intronic regions.

Prostate cancer-specific targeted panel identifies genomic alterations in primary prostate tumors

Using the prostate cancer-specific panel, we obtained sequencing data from fresh-frozen tissue from 29 primary prostate cancers, along with paired normal tissue from the same individual in most instances. Data was analyzed for single nucleotide variants (SNVs), insertions/deletions, and copy number variations as described in Methods. In general, somatic variants were distinguished from germline by comparison with matched normal tissue. In cases where normal tissue data was not available, the tumor sequencing data was compared to a panel of normal individuals to distinguish somatic vs. likely germline events.

Somatic single nucleotide variants:

Of the 29 sequenced primary prostate cancers, 16 (55%) harbored detectable non-silent SNVs or coding indels in the prostate cancer-associated genes in the targeted sequencing panel (Table 3). Some samples had multiple somatic alterations. Several events were known “hotspot” mutations, including KRAS Q61R, TP53 R273G, and CTNNB1 K335I. Two SPOP mutations were identified – one (F102L) at a recurrent hotspot locus and one (F104C) in close proximity. Three alterations in FOXA1 were also identified: C227R, which has been previously observed in prostate cancer, a frameshift mutation at P205, and a small in-frame deletion at position 265. Alterations in the PI3 kinase pathway included PIK3CA H1047R, a known hotspot mutation, and PIK3R1 I20L, which has not been previously reported. NKX3–1 Y177C has also been observed previously in prostate cancer. Other alterations that have not been reported in prostate cancer to our knowledge based on COSMIC, cBioPortal, or Pubmed include nonsense mutations in KDM6A (C293*), EGFR (E759*), CDKN1B (Q147*) and NCOR1 (Q2170*); missense mutations in NCOA2 (R974P), ATM (F1767S and A2893V), APC (S2497L), and SCN11A (P882T); and deletions in PIK3CA (WGIHLMPP11del) and CDKN1B (GLA143del). Of these, CDKN1B Q147*, ATM A2893V, and APC S2497L have been observed in tumors of other cancer types. These alterations may represent novel events in prostate cancer and may warrant further investigation.

Table 3.

Somatic single nucleotide variants and small indels in primary prostate tumors

Patient ID	Somatic SNVs/Indels
PK01	PIK3R1 p.I20L (39%)
PK02	None detected
PK03	ZFHX3 p.3527_3528insG (14%)
PK04	None detected
PK05	CDKN1B p.Q147* (23%) EGFR p.E759* (19%)
PK06	NCOA2 p.R974P (30%)
PK07	CTNNB1 p.K335I (35%)
PK08	TP53 p.R273G (20%)
PK09	NKX3–1 p.Y177C (35%) PIK3CA p.WGIHLMPP11del (9%)
PK010	None detected
PK011	None detected
PK012	FOXA1 p.P205fs (10%)
PK013	None detected
PK014	None detected
PK015	CDKN1B p.GLA143del (10%) NCOR1 p.Q2170* (27%) SPOP p.F102L (27%)
PK016	None detected
PK018	PIK3CA p.H1047R (26%) SCN11A p.P882T (9%)
MGH1	None detected
MGH2	None detected
MGH3	None detected
MGH4	FOXA1 p.RFK265del (8%) KDM6A p.C293* (8%)
MGH5	ATM p.A2893V (6%)
MGH6	FOXA1 p.C227R (23%)
MGH7	SPOP p.F104C (16%)
MGH9	KRAS p.Q61R (18%)
MGH10	None detected
MGH11	None detected
MGH13	None detected
MGH15	APC p.S2497L (54%) ATM p.F1767S (19%)

Somatic alterations in primary prostate cancer detected by targeted panel sequencing. Allelic fraction is an estimate of the fraction of the variant allele. None detected indicates no variants in the panel genes were identified, at the given mean target coverage, within the technical limits of the method.

ETS family transcription factor fusions:

The prostate cancer-specific targeted panel is designed to detect common fusion events with ETS family transcription factors. In our set of 29 evaluable primary prostatectomy specimens, ten individual samples (34%) had evidence of an ETS factor fusion event (Table 4). The structure of these fusions is in accordance with fusions reported in the literature for primary prostate tumors ¹⁶. Eight patients had a TMPRSS2-ERG fusion, one had a TMPRSS2-ETV4 fusion, and one had a SLC45A3-ETV1 fusion. Of these, four patients also had other SNVs/indels (Table 3) while for six patients the fusion was the sole detectable alteration among the genes in the panel. Taken together, we detected at least one somatic event—either an SNV/indel or an ETS factor fusion—in 22/29 (76%) of the primary prostate cancer samples.

Table 4.

Transcription factor fusion events in primary prostatectomy specimens

Patient ID	Gene 1	Gene 2	# reads in tumor
PK04	ETV1	SLC45A3	4
PK06	ERG	TMPRSS2	3
PK07	ERG	TMPRSS2	8
PK08	ERG	TMPRSS2	2
PK09	ERG	TMPRSS2	7
PK011	ETV4	TMPRSS2	2
PK013	ERG	TMPRSS2	2
MGH1	ERG	TMPRSS2	2
MGH10	ERG	TMPRSS2	2
MGH13	ERG	TMPRSS2	16

Samples harboring ETS family gene fusions are listed. The two genes involved in the fusion are listed as Gene 1 (5’) and Gene 2 (3’). The number of junctional reads across the fusion junction is listed. (Note: As expected, the number of junctional reads is small considering these reads have to cross the fusion junction which is likely a rare event; additional supporting reads were available for each fusion.)

We also performed targeted panel sequencing on DNA from two tumors which had previously undergone whole-genome sequencing and were known to contain ETS factor fusions ¹. The targeted panel detected a fusion in both tumors (Supplementary Table 2).

Copy number changes in prostate cancer genes:

We predicted the presence of amplifications and deep deletions in commonly altered genes within the prostate cancer-specific targeted panel using a combination of tools (see Methods). In our primary tumor sample set, some (though not all) of the primary prostate tumors showed aberrant copy number patterns (Figure 2). Several deletion events previously reported in prostate cancer were observed, including predicted copy loss at CHD1, PTEN, RB1, and BRCA2. Amplification of MYC was observed in one sample. In contrast, most of the normal tissues sequenced had minimal copy number alterations (Figure 2). The data for established prostate cancer cell lines are shown for comparison and discussed in the next section. Of note, compared to WES, SNP arrays, or other methods, a targeted panel cannot detect the full complement of copy number changes (especially low-level copy gains and losses) and can only include the targeted genes as opposed to genome-wide copy number alterations.

Figure 2.

Heatmap of predicted gene-level copy number amplifications (red) and deletions (blue) with log₂(CN) scale in normal tissues (left), tumor tissues (center), and cell lines (right). The computational approach for estimating copy number provides only estimated/predicted strong copy number variations, and only include genes in the targeted panel. Normal and tumor tissues from primary prostatectomies are shown in the left and center panels; the right panel shows commonly used prostate cancer cell lines.

Prostate cancer-specific targeted panel identifies mutations in blood biopsies from metastatic prostate cancer and in prostate cancer cell lines

To validate the prostate cancer-specific panel on independent cohorts of other prostate cancer sample types, we applied the panel to “blood biopsies” (circulating tumor cells and cell-free DNA) from patients with metastatic prostate cancer and to commonly used prostate cancer cell lines.

Prostate cancer circulating tumor cells:

WES can be performed on circulating tumor cells (CTCs) obtained from metastatic prostate cancer patients ³¹. We used the prostate cancer targeted panel on the same DNA from CTCs that had previously undergone WES (Supplementary Table 3). Nearly all alterations in the selected gene set in the targeted NGS panel were also detected by WES and vice-versa, except for a few mutations that lacked sequencing coverage by one method or the other, showing high concordance between targeted NGS and WES of DNA from CTCs in mCRPC.

Cell-free DNA:

WES of cfDNA from blood of metastatic prostate cancer patients has been reported ³². We used the prostate cancer targeted NGS panel on two samples of cfDNA that had undergone WES using this method (Supplementary Table 4). The prostate cancer targeted NGS panel detected 11 out of 17 SNVs identified by WES in the cfDNA; the non-concordant SNVs were either covered and not detected (2/6) or not covered (4/6) in the targeted panel. Of note, the targeted NGS panel identified a TMPRS22-ERG fusion gene in one patient that was not detected by WES.

To further test the ability of the prostate cancer targeted NGS panel to detect structural rearrangements in cfDNA, we sequenced an additional set of six cfDNA samples from patients with metastatic CRPC using a modified version of the panel that targets common rearrangements and additional baits covering the AR locus. Probes from this bait set were combined with a set of overlapping, tiled baits corresponding to non-repetitive regions of the androgen receptor (AR) locus as recently described in the literature ⁴¹ for analysis of circulating cfDNA. Plasma samples were subjected to targeted next-generation sequencing using this combined bait set, and the results were analyzed using BreaKmer ⁴⁰. Events determined through the presence of ≥2 split reads were detected in five of six samples, including one with tumor fraction of less than 10% as estimated through ichorCNA ³². A total of 23 events were detected – 13 indels and 10 rearrangements – including events in TMPRSS2-ERG, ETV1, and AR (Supplementary Table 5). This analysis demonstrates that a probe set comprised of baits tiled along loci frequently altered in prostate cancer can detect relevant translocation events from cfDNA that would otherwise not be detected through existing commercial platforms.

Prostate cancer cell lines:

We sequenced commonly used prostate cancer cell lines using the prostate cancer-specific targeted panel and compared the results to sequencing data from a 1651-gene targeted hybrid-capture panel in the Cancer Cell Line Encyclopedia project ¹⁷. We observed a high rate of concordance in events detected by both our custom prostate cancer-specific panel and the CCLE general cancer panel (Supplementary Table 6).

Collectively, these results indicate that the prostate cancer-specific targeted panel is able to capture the majority of driver events occurring in prostate cancer, in both clinical samples and cell lines, including gene fusions that are not typically detected by standard WES.

Generation of patient-derived cell lines from primary prostate tissue

We used the conditional reprogramming method ^23–26 to culture cells from fresh tissue samples of the prostate tumors and matched normal prostate tissues. We initiated cell lines from tumor and normal primary prostate specimens from 26 patients and observed robust cell growth for multiple passages (Supplementary Table 7). Of the samples for which serial passage was attempted, 20/20 normal tissues and 21/21 prostate cancer tissues generated cultures that could be serially passaged at least two times. Cell pellets were collected for DNA isolation and sequencing.

Targeted sequencing of patient-derived cell lines from primary prostatectomy samples

We used the prostate cancer-specific targeted panel to credential normal-tissue and tumor-tissue derived cell lines based on genomic features identified in the primary tumor using the targeted panel (Table 3, 4). From the 29 patients with targeted sequencing of the tumor tissue specimen, we sequenced 25 cell lines derived from primary prostate tumor tissue and 23 cell lines derived from matched normal tissue (of which 24 and 19 lines, respectively, gave rise to evaluable sequence data). In the cell lines derived from adjacent normal tissue, we did not observe any of the SNVs or fusion events that were present in the primary tumors, suggesting that on a genomic basis, these normal cell lines represent normal prostate cells. Although it is possible that adjacent normal prostate tissue might harbor precursor/premalignant cells related to the primary tumor, and thus these “normal” cell lines derived from prostate tumor tissue might harbor tumor-specific mutations, prostate cancer-associated mutations were not detected in any of the 21 sequenced normal cell lines, even at a low frequency, at least by the limits of detection of the targeted panel at the given sequencing coverage. Hence, the panel confirmed the expected absence of cancer-specific driver events in patient-derived cell lines from normal prostate.

Interestingly, for all 24 cell lines derived from tumor tissue that we sequenced, none of the SNVs or fusions observed in a patient’s primary tumor sample were observed (Figure 3). In essence, prostate cancer-associated mutations were not detected in cell lines derived from prostate cancer tissues. This suggests that the tumor-derived cell lines were probably composed mostly or completely of normal cells within the same tumor tissue specimen. It is possible that these normal cells had preferentially grown out in the conditional reprogramming conditions; growth of normal prostate cancer cells in conditional reprogramming medium has been previously reported ^26,30. Our experience thus highlights the importance of genomic credentialing of putative cancer-derived cell lines, and the utility of prostate cancer targeted sequencing to achieve this goal.

Figure 3.

Summary of mutations and gene fusions in prostate-cancer associated genes (columns) in the samples sequenced in this study (rows) including patient tumor samples, patient-derived cell lines, cell-free DNA, CTCs, and standard prostate cancer cell lines.

Discussion

To facilitate sequencing of relevant genomic alterations in prostate cancer clinical samples and patient-derived cell lines, we generated a prostate cancer-specific targeted NGS panel detecting somatic alterations in 62 prostate cancer-associated genes, including ETS family gene fusions. Targeted sequencing with the panel successfully detected single-nucleotide variants, insertions/deletions, and fusions in a cohort of primary prostate cancers, circulating tumor cells, and cell-free DNA from patients with metastatic prostate cancer.

In this study, we aimed to generate patient-derived cell lines from primary prostate cancers and to validate these cell lines using the prostate cancer-specific targeted panel. The targeted NGS panel facilitated genomic analysis to credential patient-derived cell lines by confirming whether the cell line DNA was consistent with the appropriate normal or malignant genome from its parent tissue. It was notable that none of 24 cell lines derived from primary prostate tumors that we sequenced harbored prostate cancer-specific mutations, using the same targeted NGS technology that reliably detected such mutations in the primary tumor tissue from which the cell line was derived. It is possible that the specific tissue piece used to generate the cell line was deficient in tumor cells, but a matched tissue piece was evaluated by a genitourinary pathologist and contained at least 25% tumor where this information was available. We suspect that, instead, a net outgrowth of normal prostate epithelial cells occurred during several passages in culture, as has also been previously observed in organoid systems of metastatic prostate cancers ²⁰, perhaps due a growth deficiency of the primary prostate tumor cells and/or a growth advantage of normal prostate epithelial cells. Further studies will be needed to determine the mechanisms underlying the inability to maintain primary prostate cancer cells in culture conditions that support other prostate epithelial cells, and this will be an important barrier to overcome in developing new prostate cancer patient-derived models. These findings also highlight the critical importance of validating new patient-derived cell lines to confirm maintenance of the tumor phenotype in culture, including genomic alterations consistent with the parent tumor from which the lines are derived.

In addition to genomic characterization of patient-derived cell lines, prostate cancer-specific targeted NGS panels may be advantageous for several applications in prostate cancer clinical care. In clinical oncology practice, targeted NGS is feasible and effective for identifying tumor-associated alterations and clinically actionable mutations ^42–44, typically using a panel of 50–400 genes commonly mutated across all cancer types. Several studies have reported targeted NGS of prostate cancers ^45–50 using such pan-cancer gene sets. However, the genes in these panels vary among institutions and commercial platforms, and the tumor-agnostic panels may omit key genes frequently mutated in prostate cancer but rare in other cancers. We note that FOXA1 and CHD1 are not included in the current versions of several of the commercial genomic panels (e.g. Foundation One, Guardant, Caris, Tempus). Compared to tumor-agnostic panels, selectively targeting genes altered in prostate cancer may increase the detection of biologically and clinically relevant alterations in prostate cancer. Our study is one of several to report a prostate cancer-specific targeted NGS panel covering genes recurrently altered in prostate cancer. For instance, sequencing of tumors and cfDNA from patients with mCRPC using a 72-gene panel of genes relevant to mCRPC showed that targeted sequencing of cfDNA could identify variants associated with clinical outcomes ^51–54.

As demonstrated in our patient cohort, targeted sequencing may be particularly valuable for sequencing blood biopsies (CTCs and cfDNA) of metastatic prostate cancer patients, which enable monitoring the diversity and evolution of metastatic tumors without invasive tumor biopsies. WES, WGS, and targeted sequencing are feasible from both CTCs and cfDNA from blood of prostate cancer patients ^{31,32,52–57}. However, compared to WES/WGS, targeted sequencing may be more clinically tractable for plasma samples, especially those with small amounts of DNA and low tumor fraction. Mayrhofer et al applied custom panel targeted sequencing to cfDNA in mCRPC and identified multiple alterations in DNA repair genes, among other findings, noting that targeted panel sequencing allowed cost-efficient deep sequencing to detect rare variants ⁵⁷. Noninvasive blood biopsies coupled with prostate cancer targeted sequencing could be an attractive approach to monitoring metastatic disease and selecting targeted therapies reflecting diverse metastatic lesions and resistance mechanisms, and is already showing promising results^52,53,58,59.

Identifying fusion genes within a targeted NGS assay is a feature that is not achieved with some tumor-agnostic NGS panels nor with standard WES, although such fusions can still be challenging to detect even when specifically targeted due to the requirement to target rare fusion reads. While our study used fresh-frozen tissue samples, targeted NGS of FFPE samples should be feasible using this method, especially given the low DNA input requirements. Including more intronic regions of genes involved in recurrent structural alterations in the panel would likely enable detection of additional fusion genes and structural rearrangements ¹⁰.

Although our study did have lower overall sequence coverage than some targeted NGS studies (averaging 118X for all samples; 143X for the tumor/normal tissues that passed with >10X coverage), the prostate cancer-specific panel still detected key alterations in 76% of primary prostate tumors; this percentage may increase with greater sequencing depth and/or more genes associated with prostate cancer added to the panel ⁶. Depth and accuracy may also be increased by using duplex sequencing methods which have become more technically feasible. Overall, the targeted panel detected many of the common alterations in prostate cancer reported from WES studies, yet can be performed at a lower cost and/or greater depth compared to WES due to the smaller total genomic size (790 kb) of the panel.

Prostate cancer-specific targeted NGS panels can also identify targetable alterations with therapeutic implications in primary and metastatic prostate cancer, including BRCA2 mutations or deletions that may predict sensitivity to PARP inhibitors; CDK12 alterations and mismatch repair deficiency that may suggest sensitivity to DNA repair targeting therapies and immunotherapies; PI3K pathway alterations that may be associated with sensitivity to PI3K/AKT inhibitors; and alterations in AR that may affect sensitivity to anti-androgen therapies ^11,14,60,61. Other future clinical applications of prostate cancer-specific targeted NGS panels could be envisioned, such as identifying cancer-associated mutations within a diagnostic algorithm for indeterminate prostate biopsies. Panels could also be modified to include estimates of tumor mutational burden and to identify microsatellite instability, both important biomarkers of response to immunotherapy. Given the flexibility of custom targeted NGS, prostate-cancer specific gene panels can be modified or expanded as further sequencing of large cohorts of prostate cancers reveal additional genes altered in the “long tail” of mutations, such as those described in metastatic prostate cancer ⁹.

From a research and discovery standpoint, our set of 29 primary prostate cancers harbored at least eight SNVs not previously reported, supporting that targeted panels can identify new somatic mutations in known prostate cancer-associated genes. Further functional analysis will be required to determine if these represent novel gain- or loss-of-function mutations that may have implications for prostate cancer biology.

In summary, a prostate cancer-specific targeted NGS panel can detect mutations, insertions/deletions, copy number changes, and gene fusions in prostate cancer-associated genes in cell lines, primary prostate tumors, and blood biopsies from metastatic prostate cancer patients. Targeted sequencing also revealed normal cellular outgrowth within patient-derived cell lines from primary prostate cancers, emphasizing the critical importance of genomic credentialing of patient-derived models. While the number of genes included in this panel is limited, application of a targeted NGS strategy specifically for prostate cancer can identify known and novel mutations in prostate cancer-associated genes and could be implemented clinically for diagnostic and/or therapeutic decision-making.

List of abbreviations

AR

androgen receptor

CCLE

Cancer Cell Line Encyclopedia

cfDNA

cell-free DNA

CTC

circulating tumor cell

FFPE

formalin-fixed paraffin-embedded

IRB

Institutional Review Board

mCRPC

metastatic castration-resistant prostate cancer

NGS

next generation sequencing

PARP

poly(ADP-ribose)

SCNA

somatic copy number alteration

SNVs

single nucleotide variants

WES

whole exome sequencing

WGS

whole genome sequencing

Data Availability Statement

The datasets supporting the conclusions of this article are included in the paper and sequencing data will be made available in the NCBI Sequence Read Archive repository.