Introduction

Tumorigenic driver mutations, common in other cancers, have not been detected in primary prostate tumors [1,2,3]. Indeed, localized PCa contains an overall low burden of single nucleotide variants, while metastatic PCa has multiple specific recurrently altered genes and pathways [3,4,5,6,7]. Instead, primary PCa displays nuclei with defects in size and DNA content, and enriched with large-scale genomic rearrangements, copy number variations (CNV), changes in DNA methylation patterns, chromothripsis, gene fusions, and other forms of genomic instability [8,9,10]. By analyzing the clonal status of genomic lesions within prostate tumors, a model has been proposed whereby oncogenic genomic alterations occur in a few successive bursts [1]. Mutations accumulate slowly at first, and a subset of driver mutations occurs late in PCa development [1, 11]. Thus, oncogenic events for localized PCa might include an unidentified mechanism, independent of specific driver mutations, to generate an early genomic crisis.

A genomic crisis could arise due to a change to centrosome numbers. Centrosomes are the major microtubule-organizing center in cells, giving rise to a variety of protein machines, including mitotic spindles [12]. At the centrosome core lies a centriole pair that serve as the duplicating elements of the organelle and recruit an organized shell of pericentriolar material possessing microtubule-nucleating activity [13, 14]. Cells exert tight control over centrosome copy number because centrosome overduplication (known as ‘amplification’) or centrosome loss produce cellular dysfunction and disease [15, 16].

Centrosome amplification is recognized as a general feature of tumors, seen in several solid tumors and blood cancers, and correlates with advanced tumor grade, recurrence, and poor survival [17]. Excess centrosomes lead to assembly of transient multipolar spindles with incorrect chromosome attachments [18, 19]. Consequently, anaphase lagging chromatids are produced, forming micronuclei susceptible to chromothripsis [20]. Induction of centrosome amplification in mice promotes tissue hyperplasia, and formation of solid tumors with focal CNV and whole chromosome aneuploidy [21,22,23,24].

In the opposite condition, vertebrate cells lacking centrosomes use an acentrosomal microtubule assembly pathway [25], and their proliferation is dependent on p53 status [26,27,28]. Acentrosomal spindles cause prolonged mitosis as well as mitotic errors including anaphase lagging chromatids, micronuclei, aneuploidy, and polyploidy [27,28,29,30]. Although centrosome amplification and loss have remarkably similar effects in promoting chromosomal instability (CIN), centrosome loss has not been reported in cancer.

Here, we show that human prostate tumors display significant centrosome loss. Moreover, centrosome loss in nontumorigenic prostate epithelial cells results in genomic alterations and malignant mouse xenograft tumors. Our results suggest that centrosome loss is a cellular crisis in prostate cancer (PCa), potentially explaining the presence of large-scale genomic defects without specific recurrent mutations.

Results

Localized prostate tumors display significant centrosome loss with increasing grade

Localized PCa was analyzed for centrosome detection and quantitation using an in situ immunofluorescence microscopy assay developed by us [31]. Since a pair of centrioles form the centrosome core, antibodies were used against centriole protein CEP135 and the PCM protein γ-Tubulin; colocalization of CEP135 and γ-Tubulin foci confirmed our ability to detect bona fide centrosomes (Fig. S1). E-Cadherin staining demarcated epithelial cell borders. Localized PCa shows elevated E-cadherin expression and, although no significant differences in E-Cadherin expression associate with higher Gleason scores [32], we used α6 integrin as an additional membrane marker.

Tissues were imaged in multiple z-sections, and although centrosomes are small, our approach reliably identified cell boundaries and maximized our ability to image entire cell volumes, allowing us to measure centrosomes within individual cells. Normal human prostate glands possess basal and luminal cell layers (Fig. 1a). CEP135 localizes to the proximal end of each centriole within the pair and, as expected, appeared as two distinct spots in the majority of basal and luminal cells (Fig. 1a, f). Each cell contained two CEP135 foci that could be positioned in different z-section planes or within the same focal section (Fig. 1a). Approximately 20% of cells in normal prostate glands contained no centrosomes (Fig. 1f), likely due to the known loss of cell content during tissue sectioning. Surprisingly, Gleason grade 3 PCa exhibited a significant increase in cells lacking centrosomes compared with cells of normal tissue. The proportion of cells devoid of centrosomes increased significantly with each increase of tumor grade: 52% for Gleason grade 4 and 73% for Gleason grade 5. This pattern held true when centrosome numbers are counted using CEP135 or γ-Tubulin foci (Fig. 1b, c, and f). Importantly, centrosomes were readily observed in surrounding normal stromal cells, distinguished as cells lacking E-Cadherin and serving as an internal control (Fig. 1c, yellow arrowheads). Detection of centrosomes in stromal cells demonstrated that our assay can identify centrosomes in cells with small volumes within the same section.

Fig. 1
figure 1

Primary prostate tumors display centrosome loss. a Normal human prostate tissue contains E-Cadherin (green) on cell borders and CEP135 (red) on centrosomes. DNA (blue). Centrosomes in individual cells were identified within a complete z-series of the tissue. (Left) A single z-section shows normal gland structure. Scale, 100 µm. (Middle) Maximum intensity projection of region (white box in left panel) showing centrosome distribution. Scale, 10 µm. (Right) Selected single z-sections (from box in middle panel) displaying distinct centrosomes. Labels identify specific z-sections. Scale, 5 µm; 1 µm (insets). b Centrosomes in a low grade (Gleason 3) PCa gland detected as in a. Scale, 10 µm. Insets are from boxed regions in larger image and are maximum intensity projections showing cells with two (left) and zero centrosomes (right) at higher magnification. Scale, 5 µm. c Region of high grade (Gleason 5) PCa. Centrosomes are absent in cancer cells (identified by positive E-Cadherin immunostaining; delineated by white-dashed lines) but are visible in the surrounding stroma (yellow arrowheads). Scale, 10 µm. d, e Tissue of normal human mammary gland (d) and triple negative breast cancer (TNBC) (e). Scale, 100 µm (left panels); 10 µm (right panels); 5 µm (large insets); 1 µm (small insets). f Number of centrosomes (CEP135 foci or γ-Tubulin foci) per cell in normal prostate, PCa, normal breast, and TNBC tissue cells. n > 100 cells from three independent experiments in a total of 16 different patient samples. Data are means ± SD. Cells with >2 centrosomes show amplification. Differences were tested for significance using unpaired t-tests; *p < 0.05; **p < 0.01

Full size image

As a further validation for our assay, we quantified centrosomes in triple-negative breast cancer (TNBC), which is known to display centrosome amplification [33]. Compared with normal mammary glands, we confirmed that TNBC contains a significantly higher fraction of cells with >2 CEP135 foci (24%) (Fig. 1d–f). Since TNBCs contain low levels of differentiation markers and lack basal-apical polarity [34], our assay is effective regardless of cell polarity or differentiation status. Thus, unlike most epithelial cancers described to date, centrosomes are lost in localized PCa and in prostate tumors with higher Gleason scores.

PCa cell lines show both centrosome amplification and loss

We next compared centrosome numbers in primary and immortalized nontumorigenic prostate epithelial cells as well as metastatic and neuroendocrine PCa-derived cell lines. Cells were immunostained for CEP135 and CENT3 (an additional centriole marker) or γ-Tubulin, and the numbers of foci per cell were measured. Most prostate epithelial cells (80%), including PEC (primary) as well as PrEC, iPEC37, and RWPE1 (immortalized), contained a normal centrosome count (two CEP135 foci per cell), while few displayed centrosome loss (zero CEP135 foci per cell) or amplification (>2 CEP135 foci per cell) (Fig. 2a–d, j). However, metastasis-derived PCa cell lines displayed either centrosome loss or amplification and could be subclassified by their androgen receptor expression status. Androgen-responsive cell lines, LNCaP and VCaP, showed significant centrosome loss compared with RWPE1 cells (Fig. 2e, f, j). Similar loss of centrosomes in LNCaP cells was previously reported [35]. In contrast, castration-resistant DU145 and PC3 cells, and neuroendocrine H660 cells, displayed centrosome amplification (Fig. 2g–j), as previously described [35, 36]. Thus, centrosome numbers during tumor progression may be dynamic. While a majority of primary prostate PCa in situ show centrosome loss, centrosome amplification appears in more clinically-aggressive variants of hormone-resistant metastatic cancer.

Fig. 2
figure 2

Centrosome numbers are altered in PCa lines. Immunostaining of centriole proteins CEP135 (red), CENT3 (green), or γ-tubulin (green) in primary (PEC) (a) and immortalized nontumorigenic prostate epithelial cells (b–d), and PCa cell lines (e–i). DNA, blue. Boxed regions are shown at higher magnification in insets. j PCa cell lines contain an altered centrosome number. Data are means ± SD (n = 100 cells, two independent experiments for PEC, and three independent experiments for all other cell lines). PCa lines displaying centriole loss (significantly larger % of cells with 0 centrosomes) and amplification (significantly larger % with >2 centrosomes) are indicated by black and red lines, respectively. Significant differences were evaluated with a two-tailed, unpaired t-test; *p < 0.05; **p < 0.01; ***p < 0.001

Full size image

Centrosome loss in prostate epithelial cells results in mitotic errors, chromosome fragmentation, and micronuclei

We next sought to determine whether the centrosome loss in localized PCa could account for the abundant genomic alterations reported [5]. To test this, we used two immortalized prostate epithelial cell lines, RWPE1 and PrEC; both have been used as stable, benign in vitro models [37,38,39,40,41] and contain normal centrosome numbers (Fig. 2j). We first induced centrosome loss chemically by treating cells with centrinone, a Polo-like kinase 4 (Plk4) inhibitor [28]. Plk4 is the master regulator of centriole duplication [42, 43], and its inhibition causes progressive centrosome depletion in proliferating cell cultures [28]. Over 90% of cells contained either one or zero centrosomes after a 10-day treatment but reassembled centrosomes within 2 days after centrinone removal (Fig. 3a). As a control for off-targeting effects, we generated a centrinone-resistant RWPE1 line, designated RWPE1-Plk4 (G95L), by substituting Leu for Gly95 (G95L) within the endogenous Plk4 gene using CRISPR/Cas9 [28]. Centrinone had no effect on centrosome number in RWPE1-Plk4 (G95L) cells after 10 days of drug treatment (Fig. 3b).

Fig. 3
figure 3

Centrosome loss increases the frequency of mitotic errors, chromosome fragmentation, and micronuclei. a, b Percentages of RWPE1 or PrEC cells with the indicated centrosome number after treatment with centrinone (300 nM) for the indicated number of days, followed by washout (WO). b Centrinone treatment does not change centrosome number in RWPE1 cells expressing centrinone-resistant mutant Plk4(G95L). Data are means ± SD (n = 300 cells, three independent experiments for each treatment or time point). Centrosome numbers in Plk4(G95L) expressing cells at day 0 and day 10 of centrinone treatment are not significantly different. c Selected frames from time-lapse videos of representative mock and centrinone-treated mitotic RWPE1 and PrEC cells stably expressing histone H2B–GFP and CETN1-tdTomato (not shown) to label chromosomes and centrioles, respectively. Anaphase lagging chromosomes (red arrowheads) and small chromosome fragments (yellow arrowheads) are indicated. Numbers indicate time (mins) prior to anaphase onset. See Supplementary Videos 1–4. d Measurements of mitotic duration (nuclear envelope breakdown to telophase) in mock, DMSO, and centrinone-treated RWPE1 (top) and PrEC (bottom) cells. Centrinone treatment significantly slows mitotic progression. Number of cells analyzed: RWPE1: WT = 87, DMSO = 77, centrinone = 88; PrEC: WT = 90, DMSO = 100, and centrinone = 100. Data are means ± SD. Significant differences tested using two-tailed, unpaired t-tests; ****p < 0.0001. e Mitotic RWPE1 cell treated with centrinone and stained with Alexa Fluor™ 647 Phalloidin (red) and expressing histone H2B–GFP (green). Arrowheads indicate chromosome fragments. Scale, 10 µm. f Mitotic RWPE1 cell treated with centrinone and immunostained for centromeric protein CENP-C (red) and microtubules (green). DNA, blue. Arrowheads mark chromosome fragments lacking centromeric protein CENP-C. Scale, 10 µm. See Supplementary Videos 7 and 8. g Percentage of RWPE1, PWPE1-Plk4(G95L), PrEC, and PrEC-Plk4KO cells containing micronuclei. Data are means ± SD (n > 300 cells from three independent experiments). Statistical significance evaluated using a two-tailed, unpaired t-test; **p < 0.01; ns not significant. h Centrinone-treated RWPE1 cells were immunostained for the centromeric protein CENP-C to establish the presence of centromeres within micronuclei. Data are means ± SD (n > 30 micronuclei from three independent experiments). i Spheroids of RWPE1 cells were treated with DMSO or centrinone from days 6 to 13, and stained on day 13 to label DNA. Arrowheads mark micronuclei. Scale, 10 µm

Full size image

We next determined how centrosome loss affected mitosis in prostate epithelial cells using cells stably expressing histone H2B–GFP and centriole marker CETN1-tdTomato. After a 10-day mock or centrinone treatment, we identified live cells with and without centrosomes (based on the CETN1 signal) and imaged chromosomes as cells progressed through mitosis by time-lapse microscopy (Fig. 3c and Supplementary Videos 1, 2). Whereas mock and DMSO-treated cells spent ~102 min in mitosis (defined here as nuclear envelope breakdown to telophase) (Fig. 3d, Supplementary Videos 3, 4), centrinone-treated cells lacking centrosomes spent 3–4 times longer in mitosis (Fig. 3d, Fig. S2a, Supplementary Videos 5, 6). Consequently, the mitotic index of centrosome-less cells increased fivefold but remained similar to control frequencies when centrinone was applied to RWPE1-Plk4 (G95L) cells (Fig. S2b), demonstrating that the rise in mitotic index was due to Plk4 inactivation.

In centrinone-treated RWPE1 and PrEC cells, chromosomes clustered on disorganized spindles during a prolonged prometaphase (Fig. 3e, f, Supplementary Video 7), but eventually separated into two chromosome masses (Fig. 3c, Supplementary Videos 5–8). Lagging chromatids were present during anaphase (Fig. 3c, red arrowheads, Supplementary Videos 5, 6), which have been shown to form from abnormal merotelic kinetochore attachments in acentrosomal vertebrate spindles [30, 44]. Strikingly, we detected several small chromosomes surrounding the central chromosome mass(es) (Fig. 3c, e) and that were inefficiently captured by spindles (Fig. 3f, yellow arrowheads). Immunostaining for the centromere protein CENP-C revealed these errant chromosomes lacked centromeres and, thus, are chromosome fragments (Fig. 3f, Supplementary Video 7). Missegregated chromosome fragments were enveloped into micronuclei during mitotic exit, causing an increase of micronuclei frequency that was not seen in RWPE1-Plk4 (G95L) cells (Fig. 3g and Fig. S2c). Cytokinesis failure was also frequent (30%; 20 of 66 cells) in centrinone-treated cells (Fig. 3c, RWPE1), and continuous imaging revealed that 80% of these cells died soon after entering G1-phase. As an alternative to centrinone, we generated two Plk4- null clonal lines using CRISPR (Plk4KO-1 and Plk4KO-2) (Fig. S2d, e); immunostaining for CEP135 and γ-Tubulin confirmed that both clones lacked centrosomes. Both Plk4KO-1 and KO-2 clones showed a significant increase in the frequency of micronuclei (Fig. 3g). Notably, 61% of micronuclei in centrinone-treated RWPE1 cells lacked CENP-C (Fig. 3h), suggesting that most micronuclei formed from chromosome fragments.

Lastly, we examined whether centrosome loss also generates micronuclei in 3-dimensional cultured spheroids that approximate normal prostate glands [45]. Similar to 2D cultures, centrinone-treatment produced spheroids with numerous micronuclei (Fig. 3i, arrowheads). Taken together, our findings suggest that centrosome loss in cultured prostate epithelial cells induces phenotypes known to promote CIN.

Centrosome loss produces abnormal nuclear envelopes and increased multinucleate cells in PCa tissue

We next imaged chromosomes of cultured cells exiting mitosis by time-lapse microscopy. Interestingly, we observed mitotic cells that formed multinucleated daughter cells as well as abnormally-shaped, multilobed nuclei in spite of having no apparent defects in anaphase chromosome segregation or cytokinesis (Fig. 4a, Supplementary Videos 9, 10).

Fig. 4
figure 4

Centrosome loss produces abnormal nuclear envelopes and multinucleate cells whose number significantly increases with Gleason grade. a Selected frames from time-lapse movies of mock and centrinone-treated mitotic RWPE1-H2B-GFP and PrEC-H2B-GFP cells. Arrowheads mark nuclear envelope defects including invaginated (red), multinucleated (green), and multilobed (orange) morphologies. Scale, 10 µm. b RWPE1 and PrEC cells were treated with DMSO or centrinone for 15 days and then stained with Alexa Fluor 647 Phalloidin (red) and nuclear envelope markers Lamin A/C or Emerin (green). DNA, blue. Arrowheads mark nuclear envelope invaginations. Scale, 10 µm. c, d Percentages of multinucleated cells after treatment with DMSO or centrinone for 15 days in RWPE1 and PrEC cells (c) or in the indicated cell line (d). Data are means ± SD (n > 300 cells from three independent experiments). Statistically analyzed using a two-tailed, unpaired t-test; ***p < 0.001; ****p < 0.0001. e Human prostate cancer tissue contains multinucleated cells. Tissue was stained for E-Cadherin (green) and nuclear envelope markers Lamin-A/C or Emerin (red). DNA, blue. Dashed lines trace the E-Cadherin-positive cell borders of individual PCa cells. All images are single sections selected from a z-stack. Boxed regions are shown at higher magnification and highlight individual multinucleated cells. Arrowhead marks a micronucleus. Scale, 100 μm in the upper left panel; 10 μm in all other panels. f Percentages of multinucleated cells of human PCa tissue. Data are means ± SD. n > 50 total cells from at least three different patients in a total 26 patient samples. Statistical differences tested using a two-tailed, unpaired t-test; *p < 0.05; **p < 0.01. g Human prostate cancer tissue contains multinucleated cells that lack centrosomes. Tissue was immunostained for E-Cadherin (green) and CEP135 (red). DNA, blue. This maximum intensity projection shows a binucleate cell (yellow arrowhead), and dashed lines trace the two nuclei. Note the absence of CEP135 foci. Scale, 10 μm. h Most multinucleated cells in human PCa tissue have a decreased number ofcentrosomes (<2 CEP135 foci per cell). Data are means ± SD. n = 18 patients with multinucleated cells. Statistical differences were tested using a two-tailed, unpaired t-test; *p < 0.05; **p < 0.01

Full size image

Since cells of early PCa contain nuclei with morphological defects [9], we analyzed whether centrosome loss results in defects in nuclear morphology by immunostaining cells for nuclear lamina proteins, Lamin A/C or Emerin. Plk4KO PrEC cells or centrinone-treated wild-type PrEC (or RWPE1) cells displayed a significant 3–9-fold increase in multinucleation frequency compared with control cells (Fig. 4b–d). Nuclear envelopes in acentrosomal cells frequently displayed invaginations (Fig. 4b, arrowheads), irregularities commonly seen in cancer cells [46].

Next, we examined whether similar nuclear defects were present in localized human PCa tissue by immunostaining for Lamin A/C or Emerin (Fig. 4e). Compared with cells from normal prostate glands, Gleason grade 3 PCa showed a higher average frequency of multinucleated cells but this was not significantly different (Fig. 4f). Gleason grade 4 PCa had an almost fourfold increase of multinucleated cells (Figs. 4e, 1 and 2, and f) and micronuclei were detected (Figs. 4e, 2′ arrowhead). Costaining for CEP135 and DNA revealed that ~70% of multinucleated cells displayed complete centrosome loss (Fig. 4g, h). Taken together, our findings suggest that centrosome loss promotes multinucleation and defects in nuclear envelope morphology in cultured cells, and these defects are also present in higher grade PCa. Although it is unclear how these defects arose, it is noteworthy that normal nuclear envelope shape and the efficiency of nuclear reformation rely on the suppression of microtubule growth near chromatin, which happens to be the primary site of microtubule assembly in mitotic cells lacking centrosomes [25, 47].

Centrosome loss results in aneuploidy and chromosomal rearrangements

Localized PCa displays numerical and structural CIN [48]. Since centrosome loss in dividing prostate epithelial cells promotes mitotic errors, we evaluated if centrosome loss could induce aneuploidy by measuring chromosome numbers from chromosome spreads of mitotic RWPE1 cells (Fig. 5a, b). Whereas mock and DMSO-treated cells had a near diploid average chromosome number (mock cells: average = 50; DMSO cells: 48) as previously reported [38], centrinone treatment induced aneuploidy, significantly increasing the average chromosome number to 69 (Fig. 5b). Chromosome fragments were also detected (Fig. 5a, arrowheads), consistent with our live-cell imaging data and the chromosome fragmentation reported in cells with micronuclei [20]. Flow cytometry confirmed that centrinone induced aneuploidy, displaying a sixfold increase in the >4N population (Fig. 5c). DNA profiles of RWPE1-Plk4 (G95L) cells (with or without centrinone) remained similar to DMSO-control cells (Fig. 5c), suggesting that changes in chromosome number were due to centrosome loss.

Fig. 5
figure 5

Centrosome loss results in aneuploidy and chromosomal rearrangements. a Metaphase spreads of RWPE1 cells treated with DMSO or centrinone for 15 days. Arrowheads indicate chromosome fragments. Scale, 10 μm. b Chromosome numbers per cell in mock, DMSO, or centrinone-treated RWPE1 cells. Data are means ± SD (n = 40 mock, 30 DMSO, and 79 centrinone-treated cells). Statistical differences were tested using a two-tailed, unpaired t-test with Welch’s correction; ***p < 0.001; ****p < 0.0001. c Cell cycle profiles of 15-day DMSO or centrinone treated RWPE1 and RWPE1-Plk4(G95L) cells analyzed by flow cytometry. Profiles are gated to show the percentages of cells with 2N (diploid) and >4N (aneuploidy) DNA content. Note that the frequency of aneuploidy increases >sixfold after centrinone treatment but not in centrinone-resistant Plk4 (G95L) mutant RWPE1 cells. d, e Representative spectral karyotyping (SKY) images of passage nine RWPE1 cells grown in keratinocyte serum-free medium (d) and passage 18 RWPE1 cells grown with 10% FBS (e). Cells were mock or centrinone treated for 15 days. f, g Heatmaps of chromosome copy numbers determined by SKY from RWPE1 cells in d and e. Twenty cells were analyzed for each condition. Heat scales indicate chromosome copy number. “X”, no detection

Full size image

Using spectral karyotyping (SKY), we next examined which specific chromosomes exceeded diploidy in aneuploid, acentrosomal RWPE1 cells. Cells without centrosomes have altered chromosome copy numbers (gains and losses) compared with control cells, but primarily showed gains not specific to any particular chromosome (Fig. 5d–g). Cells that were stimulated to grow (with FBS) and centrinone-treated to deplete their centrosomes had a sharply increased chromosome number compared with FBS-mock treated cells whose karyotype was consistent with a previous report (Fig. 5e) [38]. The frequency of near tetraploid karyotypes (>80 chromosomes) also increased by 30% after centrinone treatment, suggesting that cytokinesis failure likely contributed to some karyotypes. Furthermore, both RWPE1 cells and PrEC cells showed chromosomal structural rearrangements after centrinone treatment (Fig. S3a, b). Thus, instead of specific mutations, centrosome loss appears to generate gross genomic abnormalities.

Centrosome loss can stimulate malignant xenograft tumor formation

Our findings suggest that centrosome loss is a relatively early occurrence in localized PCa, producing CIN. We next determined if centrosome depleted nontumorigenic prostate cells could generate xenograft tumors. PrEC-Plk4KO clones (clones which proliferate while chronically acentrosomal) were injected subcutaneously into male immunodeficient NSG mice and examined for tumor formation weekly for ~16 weeks. Parental PrEC cells did not form tumors (Fig. 6a), consistent with a previous study [37]. Tumors were not detected by palpation of live mice injected with the PrEC-Plk4KO-1 or two cell lines. However, after these mice were sacrificed on day 112, tumors were found at 25% of the Plk4KO injection sites (Fig. 6a). The tumors were poorly differentiated and malignant as determined by histological examination. Indeed, tumors from both clones invaded into the surrounding adipose and muscle tissue (Fig. 6b). Notably, these tumors contained mitotic figures (Fig. 6c, red arrowhead), enlarged nuclei (green arrowhead), and multinucleated cells (yellow arrowheads), suggesting that tumor cells experience CIN and had histological characteristics of aggressive cancer.

Fig. 6
figure 6

Chronic or transient centrosome loss in nontumorigenic cells increases their potential to form malignant xenograft tumors. a Number of xenograft tumors formed from parental PrEC (4 injections), PrEC-Plk4KO-1 (8 injections), and PrEC-Plk4KO-2 (8 injections) cell lines. b, c Sections of PrEC-Plk4KO tumor stained with hematoxylin and eosin. b Low magnification of tumor section shows streams of tumor cells invading surrounding smooth muscle and fat layers. c Tumor section at low (left) and high (right) magnification shows proliferative cells and nuclear atypia. Mitotic cells (red arrowhead), enlarged nuclei (green arrowhead), and multinucleated cells (yellow arrowheads) are indicated. Scale, 100 μm. d Scheme used to select PrEC clones following transient centrosome depletion. PrEC cells were treated with centrinone (300 nM) for 15 days and cultured in normal growth medium after monoclonal expansion. Clones containing MYC and ERG protein levels consistent with PCa progression were selected (CTN1, 6, and 9) and subcutaneously injected into NOD/SCID male mice. e Tumor volume growth curves from CTN-1, 6, and 9 injections (12 injections for each CTN clone are plotted). Number of tumors formed is indicated above. f Kaplan–Meier tumor-free survival as evaluated by the presence of xenograft tumors. Data represent n = 12 injection sites per clone (CTN-1, 6, and 9), total of nine mice. g Centrinone-derived PrEC tumor stained with hematoxylin and eosin (H&E). Boxed region is shown at higher magnification (right). Scale, 100 μm. h An H&E-stained tissue section of a centrinone-derived PrEC tumor showing invasion through the adjacent muscle layer. Arrowheads indicate strands of tumor cells invading muscle. Scale, 100 μm. i CTN-9 tumor section immunostained for centrosome marker CEP135 (red). DNA, blue. Individual cells are shown at higher magnification in the boxed regions. Scale, 10 µm; 1 µm right panels. j Mitotic cells in CTN-9 tumors containing centrosome-associated (left) and acentrosomal (right) spindles. CEP135 (red), microtubules (green), and DNA (blue). Scale, 10 µm. k Centrosome numbers in CTN tumors. n = 100 number of cells counted per tumor

Full size image

Although centrosome depletion was sufficient to generate tumors, Plk4KO cells likely experience a continuous, high rate of CIN due to their acentrosomal state, which may explain the slow growth and hence small size of these tumors. To test this, we used an alternative to chronic centrosome absence by transiently depleting centrosomes with centrinone for several days, thereby potentially inducing a burst of CIN in the parental PrEC line. After removal of the drug, centrosomes reappeared within the cell population, with the expectation that the potentially altered genomes would then be relatively stably maintained and passed to progeny cells (outlined in Fig. 6d). Accordingly, PrEC cells were treated with centrinone for 15 days (>90% of cells contained one or zero centrioles by day 10; Fig. 3a), followed by monoclonal expansion in normal culture media without centrinone. Using this approach, three of the ten generated clones (CTN-1, 6, and 9) displayed elevated MYC and ERG levels compared with the original PrEC cells (Fig. S4a, b). We selected these three lines for xenograft testing because elevated levels of MYC and ERG correlate with PCa progression [49, 50]. CTN clones were genomically verified to be derived from PrEC using a STR profile. All CTN clones reestablished normal centrosome numbers after removal of the drug, similar to parental PrEC cells (Fig. S4c). Importantly, although the three CTN clones had normal centrosome numbers, they continued to display hallmarks of CIN and showed a significant, elevated frequency of micronuclei (Fig. S4d). All three strains inherited CIN from centrinone treatment, displaying dramatic deviations in chromosome number compared to the parental line; CTN-1 and CTN-6 showed chromosome loss, while CTN-9 had chromosome gain (Fig. S4e). Chromosomal rearrangements were also observed in CTN-9 by SKY analysis (Fig. S3b and S5a, b).

Next, the CTN clones were subcutaneously injected into male immunodeficient NSG mice to evaluate their tumorigenicity. As observed with the parental line, mice injected with CTN-1 did not form tumors (0 of 12) (Fig. 6e, f). Strikingly however, two CTN clones (CTN-6, 3 of 12; CTN-9, 9 of 12) formed large, visible tumors at the injection sites; the CTN-9 tumors developed much sooner than CTN-6 (80 days vs. ~100 days, respectively) (Fig. 6e, f). Histological examination of all tumors revealed that they were similar regardless of the specific injected clone, forming highly proliferative malignant tumors with poorly differentiated epithelial glands (Fig. 6g). Similar to PrEC-Plk4KO clones, both CTN clones formed tumors that invaded the surrounding smooth muscle and adipose tissue (Fig. 6h, arrowheads). Thus, transient centrosome elimination can transform cells, enabling them to form malignant tumors.

We further characterized the CTN tumors by immunostaining tissue sections with a panel of epithelial, mesenchymal, and secretory markers (Fig. S5c, d). CTN tumors maintained epithelial proteins including CD49f and CD44 but were negative for E-cadherin, CK5/14, CK8/18, CK14-16, 19, as well as luminal (FOXA1), neuroendocrine (synaptophysin), and mesenchymal (vimentin, desmin) markers (Fig. S5c, d). Tumors were highly secretory and stained positive for mucin and hyaluronic acid (HA) (Fig. S5c, d), a feature of poorly differentiated malignant PCa [51].

Lastly, we measured centrosome numbers within all 12 tumors using our in situ centrosome-counting assay. As in PCa tissue, we identified individual cells that clearly contained CEP135-labeled centrosomes (Fig. 6i). Dividing cells were also observed and formed mitotic spindles regardless of their centrosome status (Fig. 6j). Surprisingly, many tumor cells displayed centrosome loss, in some CTN-9 tumors, the percentage of cells with zero centrosomes exceeded 70% (Fig. 6k). Thus, transient removal of centrosomes was sufficient to induce tumor formation in a subset of immortalized prostate epithelial PrEC cells. During tumorigenesis, centrosomes were lost in a large population of cells resident at the initial injection site, resembling the phenomenon we observed in primary prostate carcinoma.

Discussion

In lieu of signature driver mutations that distinguish many cancers, genomic instability itself may provide the primary oncogenic force behind prostate tumorigenesis. Localized PCa show widespread genomic rearrangements, including chromothripsis (20% of tumors) [5]. Moreover, genomic alterations occur in a few successive bursts, likely initiated by a genomic crisis, leading to a proposed punctuated-progression model of prostate tumor evolution [1, 11]. Extensive studies have been performed to understand the complexity of genomic instability in prostate tumors and our findings point to centrosome loss as a potential origin of genomic crisis.

Centrosome loss is prevalent in localized primary prostate carcinoma, becoming more pronounced in higher Gleason grade tumors. Centrosome loss also promotes aneuploidy, chromosome fragmentation, micronuclei, and other forms of CIN. Thus, centrosome loss, in principle, may account for the wide range of genomic hallmarks assigned to localized PCa. Transient centrosome loss produced much larger solid tumors, suggesting that bursts of genomic crisis may be more suitable for sustained proliferation. Centrosome amplification has been recognized by many as a cancer hallmark with the capacity to promote oncogenesis [15]. It is surprising that the opposite situation, the absence of centrosomes in proliferating tumor cells, has not been described before in human cancer. Similar to multicentrosomal/multipolar spindles, acentrosomal spindles display errors in positioning chromosomes and high rates of CIN [30]. We observed that centrosome depletion in human epithelial prostate cells is transformative, enabling the previously nontumorigenic cells to form malignant tumors in mice. Future studies should determine whether centrosome loss is a phenomenon restricted to PCa.

If centrosome loss is a driver of genomic instability in early prostate tumors, what then causes this organelle to disappear? In general, little is known about the mechanisms of centrosome elimination. However, centrosomes are normally inactivated or lost during specific developmental stages in different animals [52]. For example, in the Drosophila female germline, centrioles disassemble when PCM is downregulated; this mechanism contributes to centriole elimination in the oocyte [53]. In contrast, pericentrin, a PCM component, accumulates in prostate carcinoma [54], suggesting that changes in PCM levels may not be a mechanism underlying centrosome disappearance in primary prostate tumors. Centrosome loss may be due to other factors such as age, inflammation, hypoxia, other environmental influences, or a combination of events. Determining the root cause of centrosome disappearance will be an important line of investigation. Indeed, an unidentified environmental factor(s) might be responsible for the loss of centrosomes from the xenograft tumors that arose from injections of cells that had been transiently centrosome depleted but contained a normal centrosome number at the time of injection.

Although centrosome loss occurs in primary prostate tumors, TNBC displays centrosome amplification [34], as do metastatic-derived androgen unresponsive PCa cell lines and neuroendocrine H660 PCa cells. Possibly, centrosome numbers fluctuate during prostate tumor evolution. Indeed, centrosome loss may be an early ‘driver’ event and, through further genomic instability, gives rise to cells with genotypes that trigger centrosome amplification, paving the way to malignancy. Notably, centrosome amplification promotes an invasive cellular behavior [55, 56]. Additional studies will be needed to determine when centrosome loss occurs in PCa. Moreover, we found a possible relationship between centrosome duplication and hormone receptor status. Specifically, clinically-aggressive variants of hormone-resistant metastatic cancer show centrosome amplification, raising the possibility that androgen receptor signaling might regulate centrosome duplication, which should be explored. For PCa, this is significant because loss of androgen signaling and gain of neuorendrocine features are indicative of increased lethality. In addition, changes in centrosome number could be used as a predictive biomarker for aggressive (high risk) disease and patient survival if it were established that centrosome numbers vary in tumors of different malignancies, particular those that are metastatic and recurrent.

Materials and methods

Cell culture

RWPE1 (ATCC #CRL- 11609TM), PEC, and iPEC37 (C. Miranti, University of Arizona) cells were grown in Keratinocyte Serum Free Medium (K-SFM; ThermoFisher) with 0.05 mg/ml BPE, 5 ng/ml EGF, and penicillin/streptomycin/amphotericin (5% CO2 at 37 °C). PEC was derived from prostatectomy specimens and verified to be free of stromal contamination [57, 58]. Immortalized human PrEC cells (W. Hahn, MIT) [37] were cultured in Iscove’s Modified Dulbecco’s Medium (IMDM; ThermoFisher) with 10% FBS. iPEC37 are PEC cells immortalized with HPV E6/E7 and hTERT [37]. LNCaP (ATCC C#RL-1740), VCaP (ATCC #CRL-2876), and PC3 (ATCC #CRL-1435) were cultured in RPMI 1640 (ThermoFisher) with 10% FBS, DU145 (ATCC #HTB-81) in IMDM, and NCI-H660 (ATCC #CRL-5813) in RPMI-1640 media (supplemented with 0.005 mg/ml insulin, 0.01 mg/ml transferrin, 30 nM sodium selenite, 10 nM hydrocortisone, 10 nM beta-estradiol, 2 mM L-glutamine, and 5% FBS. Cell identities were verified using genomic probes [59]. For RWPE1 cultured in KSFM, centrinone was used at 100 nM, but 300 nM for all cells cultured in FBS. Cells were serum starved for 48 h before addition of 100 nM DHT.

Stable lines and CRISPR/Cas9

Lentiviral plasmid (pLenti6) and packaging vectors (pMD2.G/psPAX2; Addgene #12259 and 12260) were used to generate H2B-GFP stable expressing cells. Virus was harvested from HEK-293 cells at 24–48 h post transfection, filtered, and added to media with 8 μg/ml polybrene. Stable lines were selected with 500 ng/ml puromycin after 2 weeks. RWPE1 lines were then transduced with centrin1/CETN1)-tdTomato lentivirus (pLentiLox CMV; V. Bautch, University of North Carolina). The Alt-R CRISPR/Cas9 system (IDT) with Plk4 gRNA (GGAAGCTGAGTGTTAAGTTC) was used to generate Plk4(G95L) [57]. The RNP complex of Plk4-crRNA, CRISPR-Cas9 tracrRNA (IDT #1072532), and Cas9 were transfected into cells using a Nucleofector 2b (Amaxa). Repair template plasmid (pY181) (K. Oegema, University of California San Diego) and packaging plasmids (pHelper/pAAV-RC) were transfected into HEK293T cells using Lipofectamine300. Virus was harvested 48 h post transfection, filtered, and added to media for 16 h. Cells were selected in 96-well plates for 2 weeks in 0.5 mg/ml G418, and homozygous G95L knockin clones were identified by PCR and sequencing of genomic DNA. UACC Genome Editing Facility generated Plk4 homozygous knockout PrEC cells. Double-strand break sites flanking the Plk4 transcriptional unit (4:127,879,990 and 4:127,909,266) were targeted with guide sequences 5-AACTACGGAATAAGCGGTAG-3 and 3-TCTCGCCAACTTAATTAAGG-5. Cells were transfected with Cas9, crRNAs, and tracrRNA using Lipofectamine. Two days post transfection, cutting efficiency was estimated using a T7 endonuclease assay employing PCR primers flanking the predicted ligation-junction product. Single cells were deposited in 96-well plates using the UACC Flow Cytometry Shared Resource and colonies were screened by PCR. Clones that were negative for two fragments internal to the targeted deletion but positive for ligation-junction fragment were potentially homozygous for the deletion. Absence of Plk4 RNA was confirmed by rt-PCR.

Microscopy

For IF, cells were cultured on coverslips, washed in PBS, and fixed in ice-cold methanol for 10 mins. After a 5 min rehydration in PBS, cells were washed in Buffer-A (PBS, 0.5% Triton X-100) for 5 min, blocking Buffer-B (PBS, 5% NGS, 0.1% Triton X-100) for 30 min, and incubated with primary antibodies in Buffer-B for 1 h at room temperature. Next, cells were washed 3 times in Buffer-A, incubated with secondary antibodies and Hoechst33342 for 1 h at room temperature, washed 3 times in Buffer-A (5 min), and mounted in 0.1 M n-propyl gallate, PBS, 90% (by volume) glycerol. For centromere immunostaining, cells were rinsed in PHEM (60 mM PIPES, 25 mM HEPES, 10 mM EGTA, 4 mM MgSO4·7H20), permeabilized with 0.5% Triton X-100 in PHEM for 4 min, and fixed with 4% paraformaldehyde in PHEM for 20 min at room temperature.

For metaphase spreads, cells were treated with 0.06 μg/ml colcemid overnight, trypsinized, resuspended in Buffer-M (40 mM KCl, 20 mM HEPES, 0.5 mM EGTA, and 9 mM NaOH), and incubated for 50 min at 37 °C. Next, cells were fixed in methanol:acetic acid (3:1), pelleted, and washed in fixative before dropped onto a precleaned coverglass. Cells were dried on slides for 3 days and stained with Hoechst33342. SKY analysis was performed by the Roswell Park Cancer Institute Cytogenetics SKY Core Laboratory. Formalin-fixed deidentified human cancer tissues were processed for detection of centrosomes using our published procedure [31]. Specimens were imaged using a DeltaVision Core equipped with an Olympus IX71 microscope, ×60 objective (1.42NA), and a cooled charge-coupled CoolSNAP-HQ2 camera. For time-lapse microscopy, cells were grown in 8-well Lab-Tek chambered coverglass coated with anti-β1 integrin/AIIB2 antibody to inhibit cell migration; images were acquired every 5 min with 4 μm z-section thickness for 24 h. Images acquired with SoftWoRx v1.2.

Antibodies/stains used include: anti-CEP135 (cells-1:500; tissue-1:100; Abcam ab75005), anti-γ-tubulin (cells-1:500; tissue-1:100; Sigma Aldrich GTU-88), anti-E-Cadherin/M168 (1:100; Abcam ab76055), anti-Centrin-3/3E6 (1:500; Abnova H00001070-M01), anti-phospho-Histone H3 (1:1000; Cell Signaling Technology #9706 S), 7-AAD Staining Solution (1:50; BioLegend #420404), Alexa Fluor™ 647 Phalloidin (1:40; ThermoFisher A22287), anti-α-tubulin (1:1500; Sigma Aldrich DM1A), anti-CENP-C (1:1000; MBL International PD030), anti-LaminA/C (1:100; Santa Cruz Biotechnology sc-6215), anti-Emerin (1:100; Sigma-Aldrich HPA000609), anti-β1 integrin (1:1000, DSHB AIIB2), anti-α6 integrin/CD49f (1:100; produced in our lab), anti-CD44 (1:100; ThermoFisher #156-3C11), anti-N-Cadherin (1:100; Santa Cruz Biotechnology sc-8424), anti-CK5/14 (1:100; produced in our lab), anti-CK18 (1:100; Abcam EPR1626), anti-Vimentin (1:100; Cell Signaling Technology D21H3), anti-Desmin (1:100; Atlas Antibodies HPA018803), Hyaluronic Acid-Binding Protein (1:100; Sigma-Aldrich #385911), anti-cMyc (1:50-100; Cell Signaling Technology D3N8F), anti-FOXA1(1:100; Abcam EPR10881), anti-Synaptophysin (1:100; Sigma-Aldrich MRQ-40), anti-laminin-332 (1:200; Abcam ab14509), anti-ITGB4 (1:100; Abcam ab110167), and anti-p63 (1:100; biobyt orb214808). Secondary antibodies (Jackson Immunoresearch) were used at 1:1500.

Animal studies

Mouse experiments were conducted with animal care and committee approval using the Experimental Mouse Shared Service. A total of 10 × 106 prostate epithelial cells were mixed with an equal number of nontumorigenic fibroblasts (WPMY-1) in 0.1 ml growth factor reduced matrigel and injected subcutaneously into male NOD/SCID/IL2Rγnull (NSG) mice (Jackson Laboratories); three mice per treatment group. Tumor growth was measured 2 times/week and tumor volume estimated according to the formula: [(width)2 × length]/2. Animals were terminated by CO2 when total tumor reached 2000 mm3.

Statistical analysis

The statistical significance of differences in average measurements was evaluated using unpaired t-tests (GraphPad Prism 6.0). Means are significantly different if P < 0.05. In figures, “*” indicates 0.05 > P ≥ 0.01; **0.01 > P ≥ 0.001; ***0.001 > P ≥ 0.0001; ****0.0001 > P, and not significant “ns” indicates P ≥ 0.05 for indicated pairwise comparisons. Error bars in all figures indicate standard deviation (SD).