Mice deleted for cell division cycle 73 gene develop parathyroid and uterine tumours: model for the hyperparathyroidism-jaw tumour syndrome

The hyperparathyroidism-jaw tumour (HPT-JT) syndrome is an autosomal dominant disorder characterized by occurrence of parathyroid tumours, often atypical adenomas and carcinomas, ossifying jaw fibromas, renal tumours and uterine benign and malignant neoplasms. HPT-JT is caused by mutations of the cell division cycle 73 (CDC73) gene, located on chromosome 1q31.2 and encodes a 531 amino acid protein, parafibromin. To facilitate in vivo studies of Cdc73 in tumourigenesis we generated conventional (Cdc73+/−) and conditional parathyroid-specific (Cdc73+/L/PTH-Cre and Cdc73L/L/PTH-Cre) mouse models. Mice were aged to 18-21 months and studied for survival, tumour development and proliferation, and serum biochemistry, and compared to age-matched wild-type (Cdc73+/+ and Cdc73+/+/PTH-Cre) littermates. Survival of Cdc73+/− mice, when compared to Cdc73+/+ mice was reduced (Cdc73+/−=80% Cdc73+/+=90% at 18 months of age, P<0.05). Cdc73+/−, Cdc73+/L/PTH-Cre and Cdc73L/L/PTH-Cre mice developed parathyroid tumours, which had nuclear pleomorphism, fibrous septation and increased galectin-3 expression, consistent with atypical parathyroid adenomas, from 9 months of age. Parathyroid tumours in Cdc73+/−, Cdc73+/L/PTH-Cre and Cdc73L/L/PTH-Cre mice had significantly increased proliferation, with rates >fourfold higher than that in parathyroid glands of wild-type littermates (P<0.0001). Cdc73+/−, Cdc73+/L/PTH-Cre and Cdc73L/L/PTH-Cre mice had higher mean serum calcium concentrations than wild-type littermates, and Cdc73+/− mice also had increased mean serum parathyroid hormone (PTH) concentrations. Parathyroid tumour development, and elevations in serum calcium and PTH, were similar in males and females. Cdc73+/− mice did not develop bone or renal tumours but female Cdc73+/− mice, at 18 months of age, had uterine neoplasms comprising squamous metaplasia, adenofibroma and adenomyoma. Uterine neoplasms, myometria and jaw bones of Cdc73+/− mice had increased proliferation rates that were 2-fold higher than in Cdc73+/+ mice (P<0.05). Thus, our studies, which have established mouse models for parathyroid tumours and uterine neoplasms that develop in the HPT-JT syndrome, provide in vivo models for future studies of these tumours.


INTRODUCTION
Mutations of the cell division cycle 73 (CDC73) gene (OMIM #607393), which is located on chromosome 1q31.2 and encodes a 531 amino acid protein called parafibromin, are associated with hereditary and non-hereditary forms of parathyroid carcinomas and the hyperparathyroidism-jaw tumour (HPT-JT) syndrome (OMIM #145001). 1-3 HPT-JT, an autosomal dominant disorder, is characterized by the occurrence of parathyroid tumours, and ossifying fibromas of the jaw which occur in~30% of HPT-JT patients (Table 1). 1,[4][5][6] The parathyroid tumours are usually parathyroid adenomas (PAs) but may be atypical parathyroid adenomas (APAs) or parathyroid carcinomas (PCs) in 415% of HPT-JT patients. In addition, 415% of HPT-JT patients may also develop renal tumours, which include Wilms' tumours, hamartomas and carcinomas, and~75% of women with HPT-JT will develop, at an early age, benign and malignant neoplasms of the uterus that may be associated with recurrent miscarriages and severe menorrhagia, requiring hysterectomy. 4,5,[7][8][9] These uterine neoplasms which all arise in cells derived from the embryonic sequence identity and 47% similarity to the yeast Cdc73 protein, which is a component of the polymerase-associated factor-1 (Paf1) complex, 13 a key transcriptional regulatory complex that interacts directly with RNA polymerase II. The crystal structure of the yeast Cdc73 C-domain has been reported to adopt a Ras-like fold that participates in histone ubiquitination and methylation steps through both promoter and coding regions, and studies have shown that human homologues of the yeast Paf1 complex are associated with parafibromin. [14][15][16] Moreover, parafibromin and its Drosophila homologue, Hyrax, which is a component of the Wnt1 wingless pathway and has an essential role in normal embryonic development, have a high degree of sequence similarity in their C-terminal portions. 17,18 This suggested that parafibromin may have a role in embryonic development and studies of mice deleted for Cdc73 have also shown that parafibromin has key roles in mammalian embryonic development. 19 Thus, Cdc73 null mice were embryonic lethal by 6.5 day post-coitum, which is the stage when implantation occurs. 19 Parafibromin and Hyrax also have a high degree of sequence similarity in their N-terminal domains, which directly interact with β-catenin/Armadillo in the context of the PAF1 complex. 17 The role of parafibromin as a mediator of Wnt signalling is supported by studies in human HEK293 cells, which has shown that Wnt target gene expression is directly correlated with parafibromin expression. 17 Moreover, parafibromin overexpression in HEK293 and NIH3T3 cells strongly inhibits proliferation, and in HeLa cells it increases G1 phase arrest and apoptosis with a concomitant reduction in S-phase entry, and a resulting downregulation of the cell cycle regulator cyclin D1, which is an oncogene known to be upregulated in parathyroid tumours. [19][20][21] Furthermore, underexpression of parafibromin, induced by RNAi, has been reported to increase the proportion of HeLa cells in S-phase, and to reduce basal apoptosis. 13 These in vitro findings indicate that parafibromin is a likely tumour suppressor in mammalian cells, and to explore further the role of parafibromin as an in vivo tumour suppressor we studied mice deleted for Cdc73 for the development of tumours.
Examination for development of HPT-JT associated tumours. Development of HPT-JT associated tumours was assessed in 69 mice (21 Cdc73 +/+ mice (9 males and 12 females) and 48 Cdc73 +/ − conventional knockout mice (12 males and 36 females)) between the ages of 47 to o24 months (Table 1). Parathyroid tumours were found to occur in 68% of Cdc73 +/ − mice, and 25% of these were adenomas and 75% were APAs (defined by having collagenous fibrous septa, 25 and/or immunostaining for galectin-3 but lacking evidence of invasion or metastasis). We used immunostaining for galectin-3, which is an anti-apoptotic lectin that regulates cyclin D1 and C-Jun N-terminal kinase 1 (JNK1) expression and promotes tumour growth and metastasis, 26,27 as it has been reported to have a sensitivity of 495% and specificity of 90% for pathological diagnosis of PC in man. 28,29 Uterine neoplasms were found in~33% of Cdc73 +/ − females (Table 1). Parathyroid tumours or uterine neoplasms were not found to occur in Cdc73 +/+ mice. Ossifying fibromas of the jaw, and tumours of the kidneys, thyroid, pancreas or testis (Table 1) were not found in any of the Cdc73 +/ − or Cdc73 +/+ mice. Parathyroid tumour development was also found to occur in 440% of the 32 conditional knockout mice, which comprised 15 Cdc73 +/L /PTH-Cre mice (9 males and 6 females) and 17 Cdc73 L/L /PTH-Cre mice (8 males and 9 females), but not in any of 20 Cdc73 +/+ /PTH-Cre mice (11 males and 9 females), aged 20-21 months of age. The parathyroid and uterine neoplasms developing in the mutant mice were further studied.
Analysis of parathyroid tumours Parathyroid glands were identified in 74% of mice (n = 109) and the remaining 26% of mice in whom parathyroid glands were not identified were evenly distributed among all the genotypes. Parathyroid tumours were found in 465% of Cdc73 +/ − mice (n = 15/22) but in none of 16 Cdc73 +/+ littermates (P o 0.0001, twotailed Fisher's exact test, Table 1), between 9 and 18 months of age; and in ⩾ 50% of ⩾ 18-month-old Cdc73 +/L /PTH-Cre (n = 7/12) and Cdc73 L/L /PTH-Cre (n = 6/12) mice, but not in any of 19 Cdc73 +/+ /PTH-Cre littermates (P o0.005). The parathyroids in wild-type mice were~500 μm in length (Supplementary   75 which results in the 531 amino acid parafibromin. The 37 kDa bands, which do not correspond to any translated proteins from the other transcripts, are present with similar intensities in Cdc73 +/ − and Cdc73 +/+ mice (P = 0.493, n = 4), and are likely to represent non-specific bands. Such non-specific bands, which were not altered by the use of CDC73 siRNA (siCDC73), were also detected in HeLa cells (Supplementary Figure 1). B, blank; F, female; M, male; RTase, reverse transcriptase; S, size marker.
Assessment of proliferation rates, using BrdU incorporation, revealed the myometria from Cdc73 +/ − mice to have a~2-fold increase in proliferation rates when compared to that of wild-type littermates ( Table 2, Po 0.05). This was confirmed by immunostaining for Ki-67, which revealed a significantly increased proliferation in the endometria and myometria of Cdc73 +/ − mice with uterine tumours, by 1.5-and 2.5-fold, respectively, when compared to equivalent uterine tissues of wild-type littermates (Figures 5j-l).
Other tumours Tumours of the bones, kidneys, thyroid, pancreas, or testes, which may develop in HPT-JT patients, were not detected macroscopically or microscopically in Cdc73 +/ − mice, aged up to 21 months of age. Jaw bones of Cdc73 +/ − mice had 2-fold increased proliferation rates compared to wild-type littermates ( Table 2, Po 0.05), but renal and pancreatic proliferation rates were similar in Cdc73 +/ − and wild-type littermates ( Table 2).

DISCUSSION
The results of our study, which established mice deleted for Cdc73 alleles, reveal that these mice are representative of HPT-JT in man. Thus, Cdc73 +/ − mice develop: parathyroid tumours in association with increased mean serum calcium concentrations and increased mean serum PTH concentrations, consistent with primary hyperparathyroidism; and uterine neoplasms, which comprised endometrial hyperplasia and cysts, adenofibroma and adenomyoma. Moreover, 75% of the parathyroid tumours were APAs, and thus these Cdc73 +/ − mice with the conditional Cdc73 +/L /PTH-Cre and Cdc73 L/L /PTH-Cre mice provide important in vivo models for this rare but difficult to treat human neoplasm. These parathyroid tumours and uterine neoplasms had a lack (or reduction) of nuclear expresison of parafibromin, consistent with a tumour suppressor role for Cdc73, and similar to the findings reported in HPT-JT associated tumours in man. 36 However, there are some differences in the frequency of tumours that develop in the Cdc73 +/ − mice and patients with HPT-JT, and five of these are as follows: (1) PAs are the commonest manifestation occuring in 480% of HPT-JT patients, whereas they occurred in only 25% of Cdc73 +/ − mice, the majority of which instead had APAs with features similar to PCs that occur in~15% of HPT-JT patients ( Table 1); (2) ossifying jaw fibromas occur in~33% of HPT-JT patients, but were not found in Cdc73 +/ − mice, which did however have an increased mandibular cell proliferation rate; (3) uterine adenomyosis and adenofibroma which are the commonest manifestations, occurring in 53% and 33% of women with HPT-JT respectively, were rare (or not found) in Cdc73 +/ − mice, which instead had endometrial cysts that were not observed in women with HPT-JT; (4) renal hamartomas and Wilm's tumour, which occur in 415% and o2% of HPT-JT patients, were not found in Cdc73 +/ − mice; and (5) thyroid cancer, pancreatic adenocarcinoma and testicular germ cell tumours, which occur in o 2% of HPT-JT patients, were not found in Cdc73 +/ − mice. The basis of these inter-species differences remain to be defined but they may be partly due to: the methods of detection, which may have missed detecting the small tumours in the Cdc73 +/ − mice; the possible later onset of tumours, as suggested by the finding of the increased mandibular cell proliferation rate which may represent a pre-malignant or early neoplastic phase of tumourigenesis; the possible functional redundancy of parafibromin for tumourigenesis in kidneys, pancreas, testes and jaw of mice; and the effects of species-specific genetic modifiers that might alter the phenotypic expression of the Cdc73 mutation in a species. However, it is important to note that the Cdc73 +/ − mice developed two of the most common tumours, namely parathyroid tumours and uterine neoplasms that are observed in HPT-JT patients. Thus, Cdc73 +/ − mice provide an in vivo model for the study of APAs and uterine neoplasms. The development of these tumours resulted in Cdc73 +/ − mice having a reduced survival (Figure 2a). Moreover, survival of Cdc73 +/ − males was significantly less than Cdc73 +/ − females (Figure 2b), even though there were no differences in mean serum calcium or PTH concentrations, or the development of parathyroid tumours, between the genders. The basis of the decreased survival in Cdc73 +/ − males remains unknown, but a reduction in signalling via insulin-like growth factor-1 (IGF-1), which is reported to favour female longevity in mice, 37,38 may be contributing, especially as parafibromin in murine embryonic fibroblasts has been reported to bind the IGF-1 promoter, and the loss of parafibromin has been observed to decrease expression of Igf-1. 19 Finally, our results, which demonstrate that both conventional (Cdc73 +/ − ) and conditional (Cdc73 +/L /PTH-Cre and Cdc73 L/L /PTH-Cre) knockout mice develop parathyroid tumours, indicate that Cdc73 has a critical role in parathyroid tumourigenesis.
PCs, which have an incidence from 0.5% to 5% of primary hyperparathyroidism cases, may metastasize to regional lymph nodes or distant sites such as lungs, liver, bone or pancreas, and patients will generally die from complications of the associated hypercalcaemia. 39 The only curative treatment for PC is en bloc resection of the primary tumour. 3 However, PC cannot be easily distinguished from APA or PA pre-operatively or intra-operatively, in the absence of macroscopic tumour invasion or metastasis, and thus most patients with PC do not receive curative surgery and require long-term medical management. 39 Medical therapies, including chemotherapy and radiotherapy, are ineffective with the exception of cinacalcet, an allosteric modulator of the calcium-sensing receptor, which is effective in correcting the hypercalcaemia. 39 Thus, improved medical therapies for PC are required. Parafibromin immunostaining may represent an important prognostic marker, as loss of parafibromin immunostaining has been reported to be associated with decreased disease-free survival and tumour recurrence in patients with PCs. 40,41 Moreover, APAs with loss of parafibromin immunostaining are considered tumours of uncertain malignant potential as their recurrence rate is higher at 20% when compared to a 0% recurrence rate in APAs that express parafibromin. 40,41 Thus, our conventional and conditional Cdc73 knockout mouse models that develop APAs lacking parafibromin expression will facilitate studies aimed at understanding the molecular pathogenesis of APAs, PCs and PAs, and in providing pre-clinical models for evaluating drugs.
CDC73 mutations occur in~70% of patients with sporadic, nonsyndromic PCs, and in 475% of patients with HPT-JT. Indeed, CDC73 abnormalities, either due to mutations or LOH are the major driver for PCs in humans, although expression of a mutated parafibromin protein rather than complete loss of parafibromin expression has also been reported in some PCs. 2 In addition copy number gain of mutant CDC73 alleles, with loss of the wild-type CDC73 allele through focal deletion or loss of the chromosomal arm have also been reported, 42 and the roles and mechanims of such selections of mutated CDC73 alleles in parathyroid tumourigenesis remains to be elucidated. Mutations involving other genes are rare, and to date 6 multiple endocrine neoplasia type 1 (MEN1) germline mutations and 2 rearranged during transfection (RET) germline mutations have been reported in patients with PCs occurring in association with MEN1 and MEN2A, respectively; [43][44][45] and 5 Prune Homolog 2 (PRUNE2) mutations (1 germline and 4 somatic) have been reported in PCs. 42 Other genetic abnormalities that have been detected in human PCs include: retinoblastoma (RB) loss of heterozygocity (LOH) and loss of expression (LOE) in 485% of PCs; 46 cyclin D1 (CCND1) overexpression in 490% of PCs; 30 adenomatous polyposis coli (APC) LOH and LOE in~75% of PCs; 47 tumour protein 53 (TP53) LOH and LOE in 33% of PCs; 48 glycogen synthase kinase 3-β (GSK3β) LOE in 33% of PCs; 47 and enhancer of zeste homolog 2 (EZH2) gene amplification in 60% of PCs. 49 Abnormalities of these genes are not necessarily associated with PCs in mice. For example Men1 +/ − mice develop PAs but not carcinomas; 50 transgenic mice overexpressing cyclin D1 develop adenomas but not carcinomas; 51 Rb +/ − mice develop medullary thyroid carcinomas and pituitary adenocarcinomas but not PCs; 52 and Men1 +/ − /Rb +/ − mice developed pituitary, thyroid and pancreatic islet hyperplasia, but not PCs. 52 These findings indicate that loss of RB expression and increase of cyclin D1 expression may not be required for PC development in the mouse, and are consistent with our observations that RB and cyclin D1 expression were not altered in the APAs of Cdc73 +/ − mice. Moreover, these finding indicate that Cdc73 abnormalities represent the major driver for PCs in humans and APAs in mice.
Uterine corpus tumours, are common, occurring in 430% of women 440 years, and may be benign or malignant. 53 Uterine tumours may originate from: the epithelial layer for example, endometrial hyperplasia or carcinoma; the mesenchymal layers,  for example, leiomyomas (uterine fibroids), which are benign smooth muscle tumours that develop in the myometrium; or both (that is, mixed epithelial and mesenchymal) layers, for example, carcinosarcomas which have malignant epithelial and mesenchymal components, and adenosarcomas which are neoplasms composed of benign epithelium but malignant stroma. [54][55][56][57] The uterine tumours that develop in women with HPT-JT include benign tumours such as endometrial hyperplasia, adenomyosis, adenofibromas and leiomyosis, and malignant tumours, such as adenosarcomas. 13,14 Cdc73 +/ − female mice devloped uterine tumours, that were representative of those in women with HPT-JT and these included endometrial hyperplasia, adenomyoma and adenofibroma. These uterine neoplasms developed in~33% of female Cdc73 +/ − mice ( Figure 5), whilst spontaneous uterine lesions were not observed in wild-type mice in our study and are also reported to be exceedingly rare in normal wild-type mice. 58 Thus, these Cdc73 +/ − female mice provide a model to investigate the molecular basis of uterine tumourigenesis. A previous study of human uterine tumorigenesis has reported 470% of Mullerian adenosarcomas to have: copy number gain for MYB protooncogene like 1 (MYBL1), mouse double minute 2 proto-oncogene (MDM2) and cyclin dependent kinase 4 (CDK4); copy number loss for cyclin dependent kinase inhibitor 2A (CDKN2A), breast cancer type 1 susceptibility protein (BRCA1)-associated protein 1 (BAP1) and RB1; single nucleotide variations including nonsense mutations for TP53 and alpha thalassaemia/mental retardation syndrome X-linked (ATRX); and mutations in signalling pathways notably PI3K-AKT/PTEN. 59 In addition, 490% of leiomyomas (fibroids) have upregulation of G-protein coupled receptor 10 (GPR10) resulting in activation of the PI3K/AKT-mTOR pathway; 54 while 70% of leiomyomas have a mutation of the mediator complex subunit (MED12) gene 60 that encodes a scaffold protein which interacts with proteins that include β-catenin. It is interesting to note that parafibromin also directly interacts with β-catenin in the PAF complex to mediate Wnt signalling, 24 whose dysregulation has been reported to be associated with development of intestinal and colon cancers, and it may be that similar pathways are involved in uterine tumourigenesis. [61][62][63] Analysis of mouse embryonic fibroblasts from Cdc73 +/+ and Cdc73 -/mice revealed that the parafibromin/PAF complex regulated genes involved in cell growth and survival including the chromatin remodelling genes high mobility group AT-hook 1 (Hmga1) and 2 (Hmga2) to which parafibromin and PAF directly bind. 19 Moreover, parafibromin may also act indirectly via HMGA1 which is a downstream mediator of aberrant Wnt signalling. 64 Thus, loss of parafibromin expression in the mouse embryonic fibroblasts of Cdc73 -/mice has been reported to lead to downregulation of Hmga1. 19 However, this role of parafibromin in uterine tumourigenesis requires cautious extrapolation, as Hmga1 overexpression in transgenic female mice with 1-28 copies of Hmga1a, is associated with development of uterine tumours resembling human uterine adenosarcomas. 65,66 Finally, Wilms Tumour 1 protein (WT-1), which is often expressed in Mullerian adenosarcoma, has been reported to bind to the CDC73 promoter and to repress CDC73 expression in oral squamous cell carcinoma. 67 The roles of these interactions of parafibromin in the aetiology of uterine tumourigenesis remain to be explored and our establishment of the Cdc73 +/ − mice, which develop uterine tumours, will help to provide an important resource in these studies.
In summary, we have established a conventional Cdc73 +/ − mouse, in which males and females develop PAs and APAs, and females develop uterine tumours; thus this Cdc73 +/ − mouse model is representative of the human HPT-JT syndrome. [13][14][15] Moreover, we have developed parathyroid-specific Cdc73 knockout mouse models, which develop APAs and PAs. These mouse models will facilitate further in vivo investigations of the role of parafibromin in parathyroid and uterine tumourigenesis.

Mouse studies
The generation of the conventional and conditional Cdc73 knockout mouse models has been previously described. 19,68 Conventional Cdc73 +/ − mice established using the embryonic stem cell line RRE190, 19 were maintained on a C57BL/6 background for 10 generations to obtain congenic Cdc73 +/ − mice. Cdc73 L/L mice 19 were mated with parathyroidspecific Cre-expressing, PTH-Cre mice, 24 to generate heterozygous Cdc73 +/L / PTH-Cre mice. These Cdc73 +/L /PTH-Cre mice were intercrossed to generate three genotypes expressing the Cre-recombinase: Cdc73 L/L /PTH-Cre, Cdc73 +/L /PTH-Cre, and Cdc73 +/+ /PTH-Cre. All mice were fed a standard diet (RM1 expanded diet, Special Diet Services Ltd., Witham, UK) and kept in accordance with national welfare guidelines and project license restrictions. Specifically, the animal studies were approved by the University of Oxford Ethical Review Committee and were licenced under the Animal (Scientific Procedures) Act 1986, issued by the United Kingdon Home Office Department (PLL 30/2914), and the Instituitional Animal Care and Use Committee of the Van Andel Research Institute.
Cdc73 +/+ and Cdc73 +/ − mice underwent a full post mortem at~7 and 417-21 months of age, together with collection of blood samples for serum analysis and collecting of tissues for histological analysis. Cdc73 +/+ / PTH-Cre, Cdc73 +/L /PTH-Cre and Cdc73 L/L /PTH-Cre mice were studied at 7-12 months and~20 months of age. Macroscopic and microscopic examinations for HPT-JT associated tumours was undertaken.

Histology and immunohistochemistry
Tissues were fixed overnight in neutral buffered 4% paraformaldehyde before embedding and sectioning for immunohistochemical analysis. Haematoxylin and eosin staining was performed, using previously described methods. 70 Commercially available antibodies were obtained and used according to the manufacturer's instructions (rabbit antiparafibromin A300-171A and anti-parafibromin IHC-00379 (Bethyl, Montgomery, TX, USA), rabbit anti-galectin-3 ab53082 (Abcam, Cambridge, UK), rat anti-Ki-67 M7249 (Dako, Glostrup, Denmark), rabbit anti-cyclin D1 clone SP4 (Thermo, Waltham, MA, USA), and rabbit anti-retinoblastoma sc-7905 (Santa Cruz, Heidelberg, Germany)). Colour reaction was developed using secondary goat anti-rabbit antibody conjugated with horseradish peroxidase (HRP) (Dako, Glostrup, Denmark) or biotinylated rabbit anti-rat antibody and streptavidin/HRP (DakoCytomation, Glostrup, Denmark) and 3,3'-diaminobenzidine chromogen (DAB) solution (Vectashield, Peterborough, UK), and nuclei were counterstained with haematoxylin, as previously described. 70 Masson's trichrome staining was used to assess for collagen in parathyroid tissue, such that collagen fibres were stained blue, nuclei were stained black and muscle, cytoplasm or keratin background was stained red/purple. Proliferation analysis was performed, as described, 31 using continuous long-term administration of 1 mg/ml BrdU in drinking water, which specifically incorporates into the DNA of dividing cells, and BrdU was visualized utilizing commercially available antibodies (sheep anti-BrdU ab2285 (Abcam), Cy3-conjugated donkey antisheep (Jackson, West Grove, PA, USA)), as previously reported. 31 The proportion of BrdU and Ki-67 containing nuclei was calculated from a minimum of six slides per specimen and four animals per group. Daily proliferation rates were expressed as the percentage of BrdU-containing nuclei divided by the number of days of BrdU exposure, as described. 31 The Ki-67 proliferation index was calculated by dividing the number of Ki-67 labelled nuclei by the total number of nuclei, multiplied by 100, in randomly selected fields of view (200x magnification). ApopTag in situ apoptosis detection kit S7110 (Millipore, Billerica, MA, USA) was used according to manufacturer's instructions to assess for apoptotic cells utilizing terminal deoxynucleotidyl transferase (TdT) for detection of free 3′OH DNA termini formed by DNA fragmentation, as described. 31 Established criteria from the World Health Organization and other reports, [71][72][73] were used to distinguish between PCs, APAs and PAs. Histologically, the diagnosis of PC requires demonstration of either capsular, vascular, and/or perineural tumour invasion, tumour growth into adjacent tissues, lymph node metastasis, local recurrence or distant metastasis, while APAs have features of PCs that lack unequivocal evidence for invasive growth. 71,74 Moreover, the presence of ⩾ 4 associated features of malignancy that include: capsular invasion without extension to surrounding soft tissue; mitosis 45/10 high power fields; broad intratumoural fibrous bands; coagulative tumour necrosis; diffuse sheetlike monotonous small cells with high nucleus:cytoplasmic ratio; diffuse cellular atypia; and presence of macronuclei in many tumour cells, qualifies for a diagnosis of PC, whereas the presence of only 1-3 of these features, qualifies for a diagnosis of APA. [71][72][73] PAs, which are benign neoplasms, do not have any of the features and also do not have evidence of invasive growth.

Statistical analysis
Normally distributed data were analysed by Student's t-test or ANOVA followed by Tukey's multiple comparison post-hoc test. A two-tailed Fisher's exact test was used for 2x2 contingency tables, and Kaplan-Meyer analysis was performed using a two-tailed Log-rank test. 50 P-values o0.05 were considered statistically significant. Sample sizes are stated in the results section and the figure legends. The sample size for the survival study was selected based on a power of 80% to detect a 5% significance level (two-tailed) using equal numbers per group and a hazard ratio of 0.5; no animals were excluded, randomization was not required and blinding was not performed.