Background

Prostate cancer is a multifocal, highly heterogeneous disease [1, 2] that is the most common cancer diagnosed in men in the world, with an estimated 50% of men over 60 having cancer present in the prostate [3]. The phenomenon of field cancerization was first described by Slaughter et al. [4] after observing the presence of multiple independent tumours in 11% of patients with oral squamous cell carcinomas. It was proposed that the areas surrounding these lesions were acting as a “field”, a preconditioned epithelium that could lead to cancer development. This theory suggests that tissue with a histomorphologically normal appearance can harbour a significant burden of mutations, early clonal expansions, distinct expression profiles and methylation changes that could potentially lead to tumour development. Numerous reports of somatic mutations and clonal expansions in aging individuals are in agreement with this theory [5,6,7,8]: there is clear evidence that somatic mutations are present in morphologically normal skin [5, 9], brain [10], liver [11], oesophagus [6, 12], and colorectum [13] – in some cases affecting cancer-associated driver genes. Comparable findings have been reported in blood, where the detection of clonal expansions in healthy patients over 65 has been associated with a significant increase in the risk of leukemia [14,15,3A). We observe that mutation clusters in normal tissue are all subclonal (Additional file 3), with a shared N1/N2 subclone in case 0007, two subclones (N1 and N3) in case 0006, and one clone in N2 in case 0008. These results show that multiple clonal expansions of morphologically normal cells are present in the prostate of some men with prostate cancer. There is no shared trunk between tumour clones and normal clones, indicating that they arise independently.

Fig. 3
figure 3

Phylogenies of patients with multiple samples. Phylogenies revealing the relationships between clones for each case. A patients where we have collected multiple tumours and normal. B patients where there was data from a tumour, non-BPH normal tissue, and BPH normal tissue. Each coloured line represents a clone/subclone detected in a particular sample. When two or more coloured lines are together, they represent a clone that is found in all the samples represented. The length of the line is proportional to the weighted number of single nucleotide variants present in each clone; the thickness represents the clonal cell fraction associated with that clone (more detail in Additional file 3). For example, case 0077 contains a shared subclone with 8% N, 33% BPH and 2% T (Tb) supported by 113 SNVs and 4 indels. Dotted lines are associated with samples that have no evidence of a unique sample specific clone. The very low fraction tumour subclone (< 4%) shared with normal and BPH tissue in case 0077 and between normal and tumour in case 0072 suggests cancer targeted tissue contained some of the N/BPH cells. Additional phylogenies can be found in Supplementary Fig. 3

Full size image

BPH and non-BPH morphologically normal tissue taken from the same prostate shared a subclone in all three cases examined (0065, 0073 and 0077, Fig. 3B). Generally, mutations present in morphologically normal tissue (BPH or non-BPH) and cancer were distinct but in case 0077 a subclone was observed with 2% contribution in the tumour sample, 8% in the morphologically normal sample and 33% in the BPH sample, consistent with a model in which the tumour sample contains a small proportion of the non-BPH/BPH subclone.

In the remaining 11 patients, where morphologically normal (either BPH or non-BPH) and tumour samples were taken, two patterns were present. The first pattern (Cases 0066, 0074, 0115, 0149 Supplementary Fig. 3) was characterised by separate cancer and non-BPH morphologically normal lineage. In the second pattern (Cases 0072, 0076, 0120, 0146, 0156, 0159, 0162) there was evidence of a subclone found in the normal cells also being present in the cancer sample at a low CCF (< 13%, median of 3, IQR of 2; Additional file 3). The minimum distance between cancer and normal samples for the prostates with independent lineages (median of 19 mm; IQR = 9) was on average larger than prostates where the cancer samples had a normal clone present (median 7.1 mm; IQR = 5) (Additional file 1), but this was not statistically significant (P = 0.18, Wilcoxon rank sum test).

In patients with at least one clonal expansion under selective pressure the association between the number of clones and the minimum proximity of the normal samples to the matched tumour was not statistically significant (P = 0.307, Wilcoxon rank sum test). Similarly, there was no evidence of an association between the matched tumour being multifocal and the presence of at least one clonal expansion (P = 0.79, Wilcoxon rank sum test).

Mutational signatures

Mutational signatures were inferred for each sample using SigProfiler [41] using the set of signatures defined by Alexandrov et al. [47] (Additional file 6). The cosine similarity between the reference signatures and the reconstructed profiles was high for all samples but higher in tumour compared to normal samples (median of 0.97 for tumour vs 0.88 normal), likely the result of a lower number of SNVs in normal tissues. Mutational signatures 1, 5, 8, 18 and 40 were detected both in tumour and in morphologically normal tissue/BPH samples (Fig. 4). All of these signatures have been previously been identified in prostate cancer samples [47]. Signature 1 was overrepresented in tumour samples (P = 4.89 × 10–03, Fisher’s exact test). This signature is thought to result from an endogenous mutational process started by the deamination of 5-methylcytosine and has been associated with ageing. Because of this we would expect a similar representation of this signature in both normal and tumour samples. The aetiologies of signatures 5, 8 and signature 40 are unknown [47]. Three signatures (3, 4, and 28) were unique to morphologically normal tissue. Signatures 4 and 28 were present in only one sample, whereas signature 3 is present in 10 samples. Signature 3 has been linked with defective homologous recombination-based repair, Signature 4 has been associated with tobacco smoking and the aetiology of signature 28 is unknown. There were no differences between non-BPH morphologically normal tissue and BPH.

Fig. 4
figure 4

Mutational spectra. Mutational signatures detected in tumour and matched morphologically normal tissue from prostate cancer patients and normal tissue from men without prostate cancer. The mutational spectra of each sample, as defined by the triplets of nucleotides around each SNV, were deconvoluted into mutational signatures (SigProfiler [41]) using the set of signatures defined by Alexandrov et al. [47]. The colour of the first row indicates patient when there is more than a normal-tumour (N-T) pair analysed. Six patients had more than two samples analysed and one patient had only a morphologically normal sample without a matched tumour

Full size image

Gene mutations with functional impact

In morphologically normal, fibroblasts and BPH samples a total of 281 SNVs and indel mutations were observed in coding regions of 165 genes. 110 of the 281 mutations show a potential functional impact according to wANNOVAR [48] and eight of these occurred in known cancer-related genes (PPARG, BRCA1, GATA1, ACR, WHSC1, FAT1, POLE and HOXD11) as reported in the cancer gene census [49] (Additional file 7). Of these, mutations in GATA1, WHSC1, ACR, and POLE were observed in at least one sample from a primary prostate fibroblast culture (WHSC1 and ACR occurring in the same sample). Mutations with predicted functional impact were observed in 11 genes that are designated prognostic markers of poor outcome in The Cancer Genome Atlas Research Network (TCGA) RNAseq dataset [50, 51]: FAT1, SOBP, CTHRC1, IQGAP1, FOXJ3, ATP1A3, PHF12, BCAT1, GMPR2, ADAM28, DHX32, DSG3, DDX19A, KIAA1217, PPARG, PTK2B, RPL18, DONSON, CHPF2 and XKRX. All apart from 4 of the 110 mutations were detected in a single sample: mutations affecting genes GYPA and NACAD were present in multiple samples from different patients, and mutations in genes BCAT1 and FAT2 were present in two samples from the same patient (Additional file 7). Of all the genes identified, only BRCA2 and ADAM28 have been previously classified as recurrently mutated drivers in prostate cancer [52, 53]. A previously described dN/dS driver detection method [5] was performed but no significant hits were found, possible due to the limited number of mutations and samples. From the 110 genes with a predicted functional impact, 13 were also observed to be mutated in at least one tumour sample (Additional file 7). However, there was only one instance where a potentially functionally important mutation occurred in both a normal sample and the matched tumour from the same patient (gene ACOT1 in patient 0122).

We conclude that some of the observed mutations had the potential to generate driver genes but there was an absence of evidence for recurrent mutations in cancer driver genes.

Comparison with tumours

When comparing the morphologically normal samples to their respective tumours, both the number of SNVs (median 421 vs 2560.5) and structural rearrangements (median 0 vs 40) was significantly higher in tumours (P = 3.73 × 10–09, P = 2.70 × 10–06, respectively, paired Wilcoxon signed rank test; Fig. 5A; Additional file 1). In total 17,370 indels (median of 445) were identified in morphologically normal samples whereas tumour samples harboured 11,087 indels (median of 265). The absence of copy number alterations is a notable characteristic of the normal samples, and the number is significantly less than in cancer tissue (median of 42 for cancer vs 0 for morphologically normal, P = 2.68 × 10–06, paired Wilcoxon signed rank test).

Fig. 5
figure 5

Tumours show a distinct mutation profile to normal tissue. A The difference between the number of single nucleotide variants (SNVs) detected in normal tissue compared to tumour tissue. Where multiple samples of either type were present the median number was used. B The distribution of the number of SNVs detected in morphologically normal tissue, tumour tissue with low CNAs (percentage genome altered (PGA) < 6%) and tumour tissue with high CNAs (PGA > 6%). Data from these last two categories came from Wedge et al. [52]

Full size image

We analysed a total of 91 of the 112 tumours examined by Wedge et al. [52] (removing the metastatic samples; Additional file 8). A group of 23 samples with less than 6% of the genome affected by copy number alterations were identified as “quiet tumours” (Supplementary Table 3). The numbers of SNVs (median = 2250 vs 2796) and structural rearrangements (median = 32 vs 56) were significantly lower in the quiet tumours than their high CNAs counterparts (P = 7.59 × 10–04 and P = 5.27 × 10–03, respectively, Wilcoxon rank sum test). The number of SNVs was significantly higher in “quiet tumours” when compared to samples from morphologically normal tissue (P = 1.88 × 10–10, Wilcoxon rank sum test, median = 421 vs 2250; Fig. 5B).

Discussion

Our study demonstrates several critical and recurrent features of the mutations present in non-neoplastic (BPH and non-BPH) tissue taken from cancerous prostates. Primarily, morphologically normal tissue from patients with prostate cancer had a high number of single nucleotide variants (SNVs) and indels and generally a clonal expansion under selective pressure was present. This contrasted with samples from prostates lacking cancer which had a significantly lower number of mutations and a lack of clonal expansions under selective pressure. Our results indicate that the presence of the clonal expansions in non-neoplastic tissue is a feature associated with development of cancer, a finding previously reported in leukemia [14,15,16,17].

We also show that there is evidence that clonal expansions from non-neoplastic tissue originates from stromal cells. This is highlighted by the finding that the clonal cell fraction of clonal expansions of morphologically normal tissue was always higher than the proportion of the prostate estimated as epithelial. This is supported by the relationship we observe between non-BPH and BPH normal tissue, with BPH in some cases thought to be associated with hyper-proliferation of stromal tissue [54] (although we found no evidence of an association between stromal content and mutation burden). Firstly, our constructed phylogenies reveal non-BPH morphologically normal and BPH samples within the same prostate can have a shared lineage. Secondly, high mutation rates were observed in five primary cell cultures of stromal cells prepared from BPH specimens; four of the cultures exhibiting evidence of selective clonal expansion; and three samples containing potential driver genes. Thirdly, higher mutation rates were observed in stroma-dominated BPH compared to non-BPH morphologically normal tissue. Finally, the cystoprostatectomy sample 0239 – which exhibited BPH – had the highest number of SNVs observed in non-cancer patients and had evidence of a clonal expansion under selection. The importance of stroma in prostate differentiation was established in mouse studies by Cunha et al. [55]. These studies have been extended into human cells [56,57,58] and Maitland et al. have studied prostate stromal influences for more than 20 years, exploiting primary clinical material and cultured cells [59, 60]. Foster et al. have reported clonal expansions in cancer-associated fibroblasts (CAFs) [61] and shown that stromal cells from BPH, unlike stromal cells from normal prostatic tissue, have capability of inducing growth of prostatic epithelia in vivo. Taken together, these findings indicate a model for cancer development wherein the presence of clonal expansions of stromal cells supports cancer development and contributes to the field effect. This theory is in agreement with previous reports of an association between BPH and prostate cancer [62,63,64], although a causal link has not previously been established. If this model is correct, it cannot exclude a role for stroma in non-BPH normal tissue since prostates without BPH also exhibit multifocal disease. Examining the estimated cellular composition of the stroma, derived from single cell sequencing data, in both PC and non-PC donors would further elucidate the differences we observe.

We found only very limited evidence that in normal tissue known genetic drivers were affected by mutations with potential functional impact – only PPARG, BRCA1, GATA1, HOXD11, WHSC1, FAT1 and POLE were identified. These genes have been associated with tumour suppression (BRCA1 and FAT1) [65, 66], DNA repair (POLE) [67], morphogenesis (HOXD11), epigenetic regulation (WHSC1) [68], lipid metabolism (PPARG) [69] and red blood cell development (GATA1) [70]. They have been previously linked with leukemia [71, 72], breast [73,74,75,76], bladder [73, 77] colon [78, 79], kidney [80], endometrial [81], head and neck carcinoma [82,83,84], pancreatic [73] and prostate [85,86,87] cancers. The low detection of mutations in potential driver genes agrees with a cross tissue study performed by Moore et al. in participants without detected cancer [7] and raises the possible importance of epigenetic alterations in driving clonal expansion. This involvement of epigenetic changes is supported by the reported high hypermethylation levels in genes such as APC, GTSP1 and RASSF1 in morphologically normal tissue in the prostate [88,89,90,91], that have also been shown to be good predictors of cancer development [88,89,90]. For example, hypermethylation in genes APC and GTSP1 was reported in 95% and 43% respectively in patients with an initial negative biopsy that later developed prostate cancer [88].

Clonal expansions identified in non-neoplastic tissue have a distinct unrelated pattern to those in malignant tissue but are driven by the same processes. Known prostate cancer associated mutational signatures [47, 92] were present in both morphologically normal and tumour tissue, suggesting that the same mutational processes are driving the clonal expansions in both cases. This is consistent with our own study in a smaller dataset [22] and studies at other cancer sites [93, 94]. Despite this, our constructed phylogenies reveal that clones in morphologically normal samples are of a distinct lineage from those in the tumour and their mutational characteristics are different: normal samples have significantly fewer SNVs, have very few rearrangements and a complete lack of copy number alterations. We observed this difference both with samples from the same prostate in this study and in comparison with “quiet” tumours studied by Wedge et al. [52]. Copy number alterations are an important driving feature of prostate cancer and copy number burden has been associated with a poor prognosis [95,96,97]. Homologous recombination, non-allelic homologous recombination, non-homologous end joining and microhomology-mediated break-induced replication are double stranded break (DSB) repair mechanisms that could result in CNAs, rearrangements and hypermutation [98]. The absence of these three types of genetic alterations in normal samples suggest that this type of DNA damage by DSB and errors in the repairing mechanisms (or both) occur at a lower rate in normal samples and supports the potential increase of replication errors and non-DSB DNA damage produced by endogenous or exogenous environmental factors.

In summary, these results provide further evidence that the whole prostate environment, in particular stromal cells, are involved in the development of prostate cancer and insights into potential genomic evolution mechanisms at very early stages of development. Our findings have implications for treatment (focal therapy) and early detection approaches.