Genomic analyses of the metastasis-derived prostate cancer cell lines LNCaP, VCaP, and PC3-AR

Abstract

Background:

The lymph node metastasis-derived LNCaP, the bone metastasis-derived PC3 (skull), and VCaP (vertebral) cell lines are widely used as preclinical models of human prostate cancer (CaP) and have been described in more than 19,000 publications. Here, we report on short-read whole-genome sequencing and genomic analyses of LNCaP, VCaP, and PC3 cells stably transduced with WT AR (PC3-AR).

Methods:

LNCaP, VCaP, and PC3-AR cell lines were sequenced to an average depth of more than 30-fold using Illumina short-read sequencing. Using various computational methods, we identified and compared the single-nucleotide variants, copy-number profiles, and the structural variants observed in the three cell lines.

Results:

LNCaP cells are composed of multiple subpopulations, which results in nonintegral copy number states and a high mutational load when the data is analyzed in bulk. All three cell lines contain pathogenic mutations and homozygous deletions in genes involved in DNA mismatch repair, along with deleterious mutations in cell-cycle, Wnt signaling, and other critical cellular processes. PC3-AR cells have a truncating mutation in TP53 and do not express the p53 protein. The VCaP cells contain a homozygous gain-of-function mutation in TP53 (p.R248W) that promotes cancer invasion, metastasis, and progression and has also been observed in prostate adenocarcinomas. In addition, we detect the signatures of chromothripsis of the q arms of chromosome 5 in both PC3-AR and VCaP cells, strengthening the association of TP53 inactivation with chromothripsis reported in other systems.

Conclusions:

Our work provides a resource for genetic, genomic, and biological studies employing these commonly-used prostate cancer cell lines.

1 |. INTRODUCTION

Cancer lines, including the human-derived LNCaP, VCaP, and PC3 cells, are critical tools for prostate cancer (CaP) research. The androgen-sensitive LNCaP and VCaP cell lines were derived from a lymph node metastasis and a vertebral metastasis, respectively.^1,2 The androgen-independent PC3 cell line was derived from a skull metastasis.³ PC3 cells are AR-negative, but several groups including ours have reintroduced WT androgen receptor (AR) into these cells to study androgen signaling.^4–7 Thus, stable transduction of AR into PC3 cells has been used to generate “PC3-AR” cells that express full-length AR protein at levels comparable to endogenous AR in VCaP and LNCaP. Transcriptomic analysis of PC3-AR cells revealed the presence of a large number of genes that are androgen-induced (1268) and androgen-repressed (1313), and that these gene sets overlap the transcriptomes in LNCaP and VCaP cells.⁸ PC3-AR cells were also analyzed by AR-ChIP, which revealed androgen-dependent AR binding to target genes.⁹ We recently sequenced the LNCaP, VCaP, and PC3-AR cell lines to determine whether these prostate cancer models harbor deleterious mutations in the DNA repair machinery.⁸ Here, we report and compare the features of the whole-genome sequences from the three cell lines.

2 |. MATERIALS AND METHODS

2.1 |. Cell culture

LNCaP, VCaP, and PC3 cells were purchased from ATCC and they were mycoplasma-free upon purchase. VCaP cells were grown in DMEM supplemented with 10% FBS and 1% antibiotic/antimycotic. LNCaP and PC3 cells were grown in RPMI supplemented with 5% FBS and 1% antibiotic/antimycotic. PC3-AR cells, as described previously⁸ were generated by stable lentiviral infection of full-length AR using a pWPI-GFP-FLAG-AR plasmid in which the GFP portion was swapped with the antibiotic selectable hygromycin resistance gene. All cells were incubated at 5% CO₂ and 37°C. PC3 cells were again tested for mycoplasma after the stable introduction of AR. The passage number for the three lines at the time of harvest for DNA sequencing was not defined, but no passage-related changes in growth, morphology, or androgen-induced transcription were noted. As we have characterized previously, we were able to stably reintroduce AR into PC3 cells, which responded to androgen, activated endogenous gene expression, and is suitable as a model to study WT AR function in prostate cancer cells.⁸

2.2 |. Genome sequencing and alignment

For LNCaP, VCaP, and PC3-AR cell lines, genomic DNA was prepared using the Qiagen DNeasy kit. Libraries were prepared, and samples were sequenced by Hudson Alpha. DNA was aligned using BWA MEM¹⁰ available in v0.7.17 to the b37+decoy reference sequence. The putative PCR duplicates were flagged using SAMBLASTER v0.6.9,¹¹ and a sorted BAM file was generated using SAMtools v1.9.¹²

2.3 |. Variant calling

We used FreeBayes¹³ to identify variants (single-nucleotide variants and small indels) in the three cell lines and filtered to keep all variants with a Phred-scaled variant quality greater than 20. We used vt¹⁴ to decompose and normalize the variants, and VEP¹⁵ to annotate the variants. BCFtools¹⁶ was used to filter and summarize the resulting VCF file, and TAPES¹⁷ was used to identify the pathogenic mutations in the variant calls.

2.4 |. Determination of copy number states

HMMcopy¹⁸ was used to assign read counts to 10 Kbp bins, and the copy number segments were identified using circular binary segmentation¹⁹ of those bin counts.

2.5 |. Pathway enrichment

We used SLAPEnrich²⁰ to identify the significantly mutated KEGG pathways in the cell lines.

2.6 |. Structural variant (SV) detection

We used Meerkat²¹ to identify the SVs in the three cell lines. In addition to detecting the SVs, Meerkat uses sequence homology at the breakpoints to infer the mechanism that formed the SVs and has been applied to understand the landscape of SVs in several studies.^21,22 To differentiate detected SVs from nonunique regions, we required SVs to have support from both discordant read-pairs and split-reads, and we limited SVs to the ones with ≤40 bps of homology around the breakpoint.

2.7 |. Chromothripsis

We used ShatterSeek²² to determine if the cell lines exhibit known signatures of chromothripsis.

3 |. RESULTS

3.1 |. Copy number profiles and variant calls

Spectral karyotyping (SKY) of CaP cell lines has demonstrated aneuploid karyotypes with many chromosomal aberrations, including complex chromosomal rearrangements and a high degree of karyotype instability.²³ Figure 1A–C shows the distribution of read counts assigned to mappable bins of 10 Kbps for VCaP, PC3-AR, and LNCaP cells. The majority of the bins in VCaP cells can be assigned a copy number state of three, which agrees with previous reports that VCaP cells are near triploid.²⁴ The distribution of read counts in the PC3-AR cells shows similar separable peaks corresponding to integral copy number states. Previous reports suggest that PC3 cells are near triploid, though based on the assignment of copy number states, these PC3-AR cells appear to be tetraploid.²⁵ In contrast to VCaP and PC3-AR cells, the distribution of read counts in LNCaP cells does not show discrete integral peaks corresponding to copy number states. This non-separability of copy number states is in line with reports that LNCaP cells harbor multiple clones with naturally occurring differences in androgen sensitivity caused by spontaneously arising changes.¹

FIGURE 1.

Distribution of read counts in mappable bins of 10 Kbps for (A) VCaP, (B) PC3-AR, and (C) LNCaP cells. VCaP and PC3-AR cells show integral copy number states. (D) Rare putative somatic SNVs shared by the three cell lines. (E) Pathways with recurrent pathogenic mutations. The size of the nodes is proportional to the number of genes in the pathway, and the thickness of the edges between the two nodes reflects the overlap based on gene membership between them [Color figure can be viewed at wileyonlinelibrary.com]

Table 1 enumerates the protein-coding genes with zero copies in the three cell lines based on read count assignments in 10 Kbp bins. In LNCaP, we identified a ~300 Kbp homozygous deletion (chr2:47,677,007-47,986,422) that overlaps the region with the KCNK12 gene and multiple exons of MSH2. Consistent with this observation, the LNCaP parental strain is known to lack MSH2 expression. The MSH2 protein dimerizes with MSH6 to recognize mismatches in DNA, and MSH2 also dimerizes with MSH3 to identify large insertion-deletion loops.²⁶ The deletion of MSH2 in LNCaP likely affects the mismatch repair pathway in the cells. Two of the six genes deleted in VCaP cells (USP10 and PPP2R2A) have known links to androgen signaling. The USP10 gene facilitates epigenetic effects induced by AR. Loss of USP10 reduces the stability of the AR target gene product G3BP2 and also leads to poor outcomes in human CaP.²⁷ Similarly, PPP2R2A is hemizygously lost in ~42% of prostate adenocarcinomas, correlates with reduced expression, poorer prognosis, and an increased incidence of hemizygous loss (>75%) in metastatic disease.²⁸

TABLE 1.

Genes with zero copies in LNCaP, VCaP, and PC3-AR cell lines

Cell line	Genes completely deleted in cell line
LNCaP	MSH2, KCNK12
VCaP	PPP2R2A, BNIP3L, USP10, CRISPLD2, LCE3C, LCE3B
PC3-AR	CTNNA1, LRRTM2, SFTPA2, SFTPA1, MAT1A, DYDC1, DYDC2, PRXL2A, TSPAN14, PTEN, RNLS, LIPJ, LIPF, LIPK, RP11-473M10, KAT2A, CTD-2132N18.3, HSPB9, RAB5C, KCNH4, HCRT, GHDC, STAT5B, STAT5A, STAT3, CAVIN1, ATP6V0A1, OR4F17, PLPP2, MIER2, CIC, PAFAH1B3, LCE3C, LCE3B

Both PC3-AR and VCaP cells have deleted segments on chromosome 1 overlapping LCE3B and LCE3C genes. Even though these late cornified envelope genes are deleted in ~18.2% of prostate adenocarcinomas, their role in tumorigenesis is unclear.²⁹ The gene PTEN is entirely deleted in the PC3-AR cells. Deep deletion of PTEN is a known driver of CaP observed in ~40% of metastatic prostate adenocarcinoma patients and is associated with reduced expression of AR-regulated luminal epithelial genes.³⁰ Capicua, a known tumor suppressor and an evolutionarily conserved high-mobility group-box transcription factor, is also deleted in PC3-AR cells.^31,32 The gene CTNNA1, known to be truncated in patients with hereditary diffuse gastric cancer (HDGC), is another gene with a known oncogenic role deleted in PC3-AR cells. These deletions were confirmed in the RNA-seq data generated from the same cells (Figure S1).⁸

We next used FreeBayes¹³ to identify variants (positions with SNVs and small indels) in the three cell lines. FreeBayes identified 4,268,194 variants in PC3-AR, 4,433,670 variants in VCaP cells, and 5,699,585 variants in LNCaP cells. A significant fraction of these mutations are likely from the germline of the patients from whom these cell lines were derived. Since tumors are typically enriched for rare and deleterious mutations, we filtered the calls to remove the mutations with an allele frequency greater than 10⁻⁵ in the gnomAD project except for any cancer-recurrent variants defined by a minimum frequency of 3 in TCGA or 10 in the COSMIC database in an attempt to identify putative somatic mutations. This left us with 302,636 variants in PC3-AR, 273,109 variants in VCaP and 1,156,722 variants in LNCaP cells. Figure 1D shows the overlap of these mutations in the three cell lines, again highlighting that LNCaP cells have a high somatic burden due to the heterogeneity of the cells.

3.2 |. Pathogenic coding SNVs

We used TAPES¹⁷ to identify pathogenic single-nucleotide variants (SNVs) in the protein-coding regions of the three cell lines using the criterion as defined by the American College of Medical Genetics (ACMG).³³ We identified nine variants in the VCaP cell line, 17 variants in the PC3-AR cell line, and 351 pathogenic variants affecting 327 genes in the LNCaP cell line annotated as “Pathogenic” or “Likely Pathogenic” (Table S1).

Comparing the list of pathogenic variants (Figure S2) reveals that LNCaP and PC3-AR both harbor such variants in TP53. This known tumor suppressor plays a pivotal role in genomic stability, cell cycle arrest, and other critical signaling pathways. 18% of patients diagnosed with prostate adenocarcinoma have an alteration in TP53. PC3-AR cells are hemizygous for chromosome 17p, but the remaining TP53 allele has a frameshift deletion leading to premature termination of the protein product (TP53:NM_001126112:exon5:c.414delC:p.K139Rfs*31). Consistent with this result, PC3-AR does not express TP53. LNCaP cells, on the other hand, harbor a missense mutation in TP53 (TP53:NM_001126112:exon7:c.T700C:c.Y234H) that overlaps a known somatic locus in multiple carcinomas. Even though TAPES did not annotate any mutations as pathogenic in the TP53 gene of the VCaP cells, we identified a homozygous c. 742C > T (p.R248 W) mutation that has been observed in prostate adenocarcinomas and has been recently shown to cooperate with EZH2 to promote cancer growth and metastasis.³⁴ That mutation is annotated as a variant of uncertain significance (VUS) by TAPES which assigns it a probability of 0.675 of being pathogenic. We also investigated the presence of the TP53 variants in RNA-seq data sets generated for the three cell lines.⁸ We found evidence for the TP53 mutations in PC3-AR and VCaP cells that have been described previously,^34,35 but did not detect the mutated p53 for LNCaP cells, despite it being detected in the DNA sample with a high SNP quality (Figure S3). Because this TP53 mutant allele was not detected in RNA-seq data from LNCaP cells, it seems likely that it is either not expressed, or the transcript levels are too low for representation in the data set.

Besides TP53, LNCaP and PC3-AR also harbor pathogenic variants in CRB1 and CSMD3, but these genes are not expressed in the prostate. CRB1 encodes a protein that localizes to the inner segment of mammalian photoreceptors and controls the proper development of the polarity in the eye, and CSMD3 is involved in dendrite development. VCaP and PC3-AR harbor a deltaG allele in the IFNL4 gene, which encodes for interferon IFNλ4 protein and is associated with an impaired ability to clear the hepatitis-C virus.³⁶ Similar to CRB1 and CSMD3, we found that IFNL4 was also not expressed in these cell lines (Figure S4).

We also found three pathogenic mutations in LNCaP cells that are annotated as relevant to human prostate cancers. We found a previously reported stop-gain in CHEK2 (CHEK2:NM_001349956: exon5:c.G514T:p.E172X), a serine/threonine-protein kinase that is required for checkpoint mediated cell cycle arrest and activation of DNA repair and apoptosis in response to the presence of DNA double-strand breaks.³⁷ We also observed a loss-of-heterozygosity (LOH) and a frameshift deletion in PTEN, leading to a stop-gain (PTEN: NM_000314:exon1:c.16_17del:p.K6Rfs*4). PTEN is a known tumor suppressor that antagonizes the PI3K-AKT/PKB signaling pathway and is also entirely deleted in the PC3-AR cells.³⁸ Lastly, we found that LNCaP harbors a frameshift deletion in RNASEL (RNASEL: NM_021133:exon2:c.471_474del: p.K158Rfs*6), which is associated with familial predisposition to prostate cancer.³⁹

Using SLAPenrich,²⁰ we then identified the pathways that are recurrently modified in these cell lines. Figure 1E shows a network of overlapping KEGG pathways recurrently mutated in these cell lines with an FDR <0.05. The nodes represent the pathways, and the thickness of the edges reflects the overlap between the pathways based on gene membership. Besides the pathways annotated as being involved in cancers (pathways in cancer, colorectal cancer, bladder cancer, and thyroid cancer, and basal cell carcinoma), we observed pathogenic mutations in glutathione metabolism, cell cycle, Wnt signaling pathway, and RNA degradation pathway (HSPA9, PAPOLG). We also found recurrent dysregulation of the mismatch repair pathway in the three cell lines. In the VCaP cell line, we identified a heterozygous frameshift insertion in the MSH6 gene. In the LNCaP cell line, we found a homozygous deletion of exons 9-16 of MSH2, which results in a truncated protein, and a frameshift deletion in the MSH3 gene. Despite strong evidence of the MSH3 mutation in the DNA alignments (Figure S5), we did not observe the variant in the RNA-seq data. MSH3 heterodimerizes with MSH2 to form MutS-beta, which binds to the DNA mismatches and recognizes large insertion-deletion loops, whereas MSH6 heterodimerizes with MSH2 to form MutS-alpha, which binds to DNA mismatches and initiates repair. This analysis further demonstrates that even though the cell lines have accumulated independent passenger mutations, the pathogenic mutations are enriched for dysregulation of pathways with a known role in oncogenic processes.

3.3 |. SVs

We used Meerkat²¹ to identify the SVs in the three cell lines. The detected SVs include the germline SVs and the changes acquired by the cells during tumorigenesis and in vitro culture. In addition to identifying the SVs, Meerkat uses sequence homology at the breakpoints to infer the mechanism that formed the SVs and has been applied to characterize the landscape of SVs in several studies. Table 2 shows the counts of the various types of SVs in each sample after filtering. VCaP cells have the highest number of SVs among these cell lines, whereas LNCaP cells have the smallest.

TABLE 2.

Count of various types of structural variants identified in LNCaP, PC3-AR, and VCaP cells

SV type	LNCaP	PC3-AR	VCaP
Deletion	778	747	690
Deletion with insertion	151	223	244
Deletion with inversion	4	14	34
Insertion	30	28	29
Interchromosomal translocation	202	211	223
Inversion	88	159	276
Tandem duplication	97	176	141

Deletions are usually the consequence of defective DNA double-strand break repair. Transposable element insertion (TEI) is the dominant mechanism of deletion-like SV formation in the germline, whereas nonhomologous end joining (NHEJ) and alternative end-joining (alt-EJ) have been reported as the dominant mechanisms in somatic cells.²¹ We used Meerkat to predict the mechanism of formation for deletion-like fragment joints in these cells. Figure 2A shows the count of deletion-like joins in the three cell lines predicted to be created via various mechanisms, and Figure 2B shows the size distribution of these deletion-like joins. Based on these results, deletions generated via NHEJ and fork stalling and template switching/microhomology-mediated break-induced repair (FoSTeS) tend to be longer than those generated via TEI, which probably reflect germline variants. Looking at interchromosomal translocations and inversion-like joins in the cell lines, we can see that most breakpoints lack homology and are likely created via NHEJ/alt-EJ/FoSTeS mechanisms that are the dominant repair mechanisms in tumors.

FIGURE 2.

Characteristics of SVs in LNCaP, VCaP, and PC3-AR cells. (A) The number of deletions generated via various mechanisms. (B) Size distribution of deletions stratified by the putative mechanism that generated them. (C) Homology around breakpoints for interchromosomal translocations detected in three cell lines. A negative homology indicates the insertion of bases. (D) Size and homology for inversion-like structural variants [Color figure can be viewed at wileyonlinelibrary.com]

The TMPRSS2-ERG gene fusion is the most frequent genomic alteration in prostate cancer, resulting in overexpression of the transcription factor ERG. Several studies have reported the occurrence of the TMPRSS2-ERG fusion in VCaP cells, but the landscape of fusions in these commonly used CaP cell lines has not been reported. We identified 533 gene fusions in the VCaP cell line, 553 gene fusions in the PC3-AR cell line, and 476 gene fusions in the LNCaP cell line (Table S2). As expected, most of the gene fusions do not change the protein products (Table 3). In the VCaP cells, TMPRSS2 is involved in several rearrangements, including two intragenic rearrangements and two intergenic rearrangements with ERG and C16orf7. Based on the microhomology around the breakpoints, it appears that the TMPRSS2-ERG fusions are generated via fork stalling and template switching mechanisms. We also observed a ZBTB20-LSAMP gene fusion in the VCaP cell line, arising from a duplication that leads to a UTR swap.

TABLE 3.

Gene-gene fusions in LNCaP, VCaP, and PC3-AR cells

Fusion type	LNCaP	PC3-AR	VCaP
head-head/tail-tail	70	109	120
In frame	16	24	13
No impact	336	322	315
Out of frame	20	38	29
Unknown	6	9	9
UTR swap	28	51	47

In the LNCaP cells, we found evidence for a BLNK-TLL2 gene fusion that has been reported in cervical cancer and a BMPR2-FAM117B in-frame gene fusion that has been reported in breast adenocarcinoma.⁴⁰ In the PC3-AR cells, we found an IMMPL2L-DOCK4 gene fusion that has been reported in prostate adenocarcinoma (PRAD), stomach adenocarcinoma (STAD), and esophageal carcinoma (ESCA). We also identified a KDM5B intragenic rearrangement, which has been reported in colonic neoplasms, and a TFCP2-SMAGP gene fusion reported in lung squamous cell carcinoma (LUSC). We also detected repeated deletions in the PTPRD genes, all of which are predicted to have been generated via alt-EJ and NHEJ mechanisms.

3.4 |. Chromothripsis

Chromothripsis involves the shattering of one or a few chromosomes followed by fragment reassembly. It is defined based on (1) occurrence of genomic rearrangements in localized chromosomal regions, (2) copy number changes alternating a few copy number states, and (3) the alternation between regions where heterozygosity is preserved with regions displaying loss of heterozygosity. Chromothripsis has previously been reported in the VCaP cells on the q arm of chromosome 5.²⁴ A recent study analyzed the breakpoints involved in canonical chromothripsis events with interspersed LOH and concluded that NHEJ has a principal role in DNA repair, with partial contributions from MMBIR/FoSTeS or alt-EJ.²²

We used ShatterSeek²² to detect the presence of chromothripsis in the three cell lines. We did not find any evidence of chromothripsis in LNCaP cells. For VCaP cells, ShatterSeek reported high confidence of chromothripsis in chromosome 5, chromosome 9, chromosome 13, chromosome 14, chromosome 16. Figure 3A shows the fragment joins and copy numbers observed in chromosome 5 of VCaP cells. The affected region on 5q has 358 intrachromosomal SVs in 182 segments, with at least a contiguous oscillation between two copy number states for 12 of those segments. Of the 170 deletion-like fragment joins in the region, we find that 85 exhibit signatures consistent with being generated via fork stalling and template switching mechanisms (FoSTeS), whereas the majority of the remaining breakpoints were split between NHEJ (35) and alt-EJ (36) mechanisms. As reported by Alves et al.,²⁴ we did not find any evidence of positive selection of in-frame fusion transcripts in this region.

FIGURE 3.

Chromothripsis of chromosome 5, arm q is observed in both (A) VCaP and (B) PC3-AR cells [Color figure can be viewed at wileyonlinelibrary.com]

Surprisingly, we also found evidence of chromothripsis in chromosome 5 and chromosome 8 of the PC3-AR cells. Figure 3B shows the fragment joins, copy number states for chromosome 5 of the PC3-AR cells. The affected region on 5q has 55 intrachromosomal SVs in 105 segments, with at least one contiguous oscillation between two copy number states for 15 of those segments. Of the 57 deletion-like fragment joins in the region, we identified that 25 are generated via fork stalling and template switching mechanisms (FoSTeS), 11 are generated via alt-EJ, and six are generated via NHEJ.

3.5 |. COMPARISONS TO PUBLIC CELL LINE AND PATIENT DATA SETS

The Cancer Cell Line Encyclopedia (CCLE) project has characterized 1,746 cell lines including VCaP, PC3, and LNCaP-FGC cells.⁴¹ After filtering to remove putative germline variants, CCLE identified 241; 190; and 5,461 mutations in the VCaP, PC3, and LNCaP-FGC cell lines. Out of those, we identified 204 (84.65%), 133 (70%), and 3,252 (59.55%) of the variants in our call set. To investigate similar correspondence of gene copy-number, we determined the median number of reads assigned to 10 kb bins for each gene and calculated the Spearman correlation between the cell lines in this study to the copy-numbers calculated by CCLE. Similar to the mutations, we found the highest correlation for VCaP cells (r_s = 0.84, p < 2.2e–16), followed by PC3-AR to PC3 (r_s = 0.63, p < 2.2e–16), and LNCaP cells to LNCaP-FGC (r_s = 0.58, p < 2.2e–16). The lowest agreement for LNCaP cells is the result of the differences between the LNCaP strain used in this study and LNCaP-FGC, but it also highlights the heterogeneity in the LNCaP cellular population.

To compare the PC3-AR cells to the parental PC3 strain from ATCC further, we downloaded the whole-genome sequence data for the PC3 cells from the Short Read Archive (SRA, accession number SRX2512382). The distribution of read counts in mappable bins of 10 Kbps for the PC3 cells is shown in Figure 4A. Comparing it to the distribution in Figure 1B, PC3-AR cells sequenced in this study harbor a larger fraction of the genome at higher copy-number states. Figure S6 shows a karyogram plot of regions that are amplified or deleted in PC3-AR cells compared with the PC3 cells and Figure S7 shows the read count profile in the two cell lines showing the loss of chromosome 7p in PC3-AR cells.

FIGURE 4.

Comparison of PC3-AR cells to PC3 cells. (A) CNV profile of the PC3 cells shows a significant fraction of bins at lower copy-number states compared with PC3-AR cells in this study. (B) Similar to PC3-AR, we observe chromothripsis of chromosome 5 in PC3 cells [Color figure can be viewed at wileyonlinelibrary.com]

We also compared the SNVs and SVs in the two cell lines. Overall, FreeBayes identified 4,290,756 variant locations that were genotyped in both cell lines, and 4,031,949 (93.97%) of the variants were shared between them. 151,489 (3.53%) variants were found to be private to the PC3-AR cells. Using the same filters we use to identify putative somatic mutations (Figure 1D), we determined that 27,617 of the variants likely represent somatic mutations gained by PC3-AR cells as a consequence of genetic drift and selective pressures. Similar to SNVs, the counts and types of SVs in PC3-AR and PC3 also show a significant overlap (Table S3). Though some of the events such as the IMMPL2L-DOCK4 fusion were not observed in the PC3 cells, we found evidence of chromothripsis on chromosome 5 of the PC3 cells, similar to that observed in the PC3-AR cells (Figure 4B). The affected region on 5q has 81 intrachromosomal SVs in 173 segments, with at least one contiguous oscillation between two copy number states for 14 of those segments.

We also compared the variants observed in the cell lines to the exome sequencing data from 1013 prostate cancers, which includes 680 primary and 333 metastatic tumors from multiple cohorts.⁴² We first asked whether putative somatic mutations identified in the three cell lines were also identified in the primary tumors. We found 30 overlapping mutations, 14 of which resulted in a change in the protein. The 14 non-silent mutations were each observed in a single patient of the cohort. This included the p.R248W mutation in TP53 that was observed once in the TCGA-PRAD cohort but has been reported in several other prostate cancer cohorts as well as several other cancers, and the frameshift deletion in ZMYM3, which harbors non-silent mutations in 12 primary samples and 3 metastatic samples. We have included a list of the 30 variants in Table S4. Comparing the same variants to metastatic samples in the cohort, we found 183 overlapping mutations, 48 of which were non-silent. The p.T878A mutation in AR observed in LNCaP cells was found in 13 metastasis samples, and the p.P48fs mutation in the ERF gene was observed in 3 patients. ERF encodes a transcription factor and a proto-oncogene involved in development, apoptosis, and the regulation of telomerase. It is mutated in 18 patients of the cohort and is one of the recurrently mutated genes. A complete list of these mutations and their frequency in the CRPC patients are included in Table S5.

Changing the focus from individual mutations to genes, we next asked if the 97 recurrently mutated genes for Prostate Cancer identified by Armenia et al.⁴² harbor putative non-silent somatic mutations in the three cell lines. We found one mutation in TP53 in VCaP cells and mutations in three of the genes (IL6ST, TP53, and AR) in the PC3-AR cell line. The AR mutations in the PC3-AR cells are the non-frameshift deletions that are part of the AR sequence used in lentiviral infection of full-length AR. A complete of these mutations, including the mutations in 46 of the 97 genes in LNCaP are included in Table S6.

4 |. DISCUSSION

Whole-genome ploidy and modal chromosome number have been characterized for most cancer cell lines. Using mappable bins of 10 Kbps, we found that most bins in VCaP and PC3-AR can be assigned to integral copy number states. However, the same is not valid for LNCaP cells, which appear to consist of multiple cellular subpopulations. Other studies have reported that LNCaP cells harbor multiple clones with naturally occurring differences in androgen sensitivity caused by spontaneously arising changes. This instability and molecular heterogeneity in LNCaP cells could explain how multiple strains have been developed from the LNCaP parental strain, including cells under selection for growth under conditions of androgen depletion. As an example, one of the LNCaP derivative cell lines (LNCP-Abl) differentially displays resistance to enzalutamide.⁴³

Our study used TAPES¹² to identify the pathogenic coding mutations in the three cell lines. Even though this highlighted several essential pathways, such as the mismatch repair pathways that are recurrently mutated in the cell lines, this analysis could still have missed several variants relevant to the disease. For example, the homozygous c.742C>T (p.R248 W) mutation in the VCaP cells was annotated as a variant of uncertain significance by TAPES, even though it has been shown that the mutation inactivates TP53. This result underscores the challenges of variant prioritization and highlights the difficulties in determining somatic variants in a tumor-only sample.

In this report, we identified several SVs in these cell lines. Among them was a ZBTB20-LSAMP gene fusion in the VCaP cell line, arising from a duplication that leads to a UTR swap. In another study, LSAMP locus rearrangements were found to be associated with African American ethnicity, and LSAMP deletion was found to be correlated with rapid disease progression.⁴⁴ Our analysis suggests that rearrangements of LSAMP locus might not be specific to African Americans.

Deletion of chromosome 5q is common in prostate cancer, affecting between 6% to 19% of prostate cancers, and is linked to an aggressive course of disease.⁴⁵ We found evidence of chromothripsis in chr5q in PC3, PC3-AR, and VCaP cells. About half of the breakpoints exhibit evidence of generation via fork stalling and template switching mechanisms (FoSTeS). This contrasts with an analysis of various tumors, which concluded that more than half of the breakpoints appear to be generated via NHEJ or alt-EJ. We did not observe any complete gene loss in the regions affected by chromothripsis. Since VCaP and PC3-AR are polyploid cell lines, this suggests that the shattering event occurred after the duplication events.

We compared the genomic alterations in PC3-AR cells to the parental PC3 cells. The fact that certain genomic alterations were only detected in PC3-AR cells suggests, but does not prove, that AR signaling may have been a driver of changes in genome structure. However, It cannot be excluded that said changes arose as a consequence of growth in culture. Comparing the variants to patient samples, we observed a significant number of mutations in the cell lines that are also seen in patients. As expected, the overlap was higher with patients with metastatic disease.

We also found inactivating TP53 mutations in both PC3-AR and VCaP cells. TP53 malfunction has been reported as a predisposing factor for chromothripsis.²² LNCaP cells also have a pathogenic missense mutation in TP53. Still, they do not show any evidence of chromothripsis, suggesting that TP53 deficiency rather than malfunction has a role in predisposing cells to chromothripsis. It also remains to be determined whether chromosome-specific properties or the nonrandom localization of chromosomes play a role in the preferential target of specific chromosomes by chromothripsis.

DATA AVAILABILITY STATEMENT

The Raw sequence reads from the whole-genome sequencing of the cell lines in this study are openly available in the Short Read Archive (SRA) under accession PRJNA491402.