Frequent alterations in p16/CDKN2A identified by immunohistochemistry and FISH in chordoma

Abstract The expression of p16/CDKN2A, the second most commonly inactivated tumour suppressor gene in cancer, is lost in the majority of chordomas. However, the mechanism(s) leading to its inactivation and contribution to disease progression have only been partially addressed using small patient cohorts. We studied 384 chordoma samples from 320 patients by immunohistochemistry and found that p16 protein was lost in 53% of chordomas and was heterogeneously expressed in these tumours. To determine if CDKN2A copy number loss could explain the absence of p16 protein expression we performed fluorescence in situ hybridisation (FISH) for CDKN2A on consecutive tissue sections. CDKN2A copy number status was altered in 168 of 274 (61%) of samples and copy number loss was the most frequent alteration acquired during clinical disease progression. CDKN2A homozygous deletion was always associated with p16 protein loss but only accounted for 33% of the p16‐negative cases. The remaining immunonegative cases were associated with disomy (27%), monosomy (12%), heterozygous loss (20%) and copy number gain (7%) of CDKN2A, supporting the hypothesis that loss of protein expression might be achieved via epigenetic or post‐transcriptional regulatory mechanisms. We identified that mRNA levels were comparable in tumours with and without p16 protein expression, but other events including DNA promoter hypermethylation, copy number neutral loss of heterozygosity and expression of candidate microRNAs previously implicated in the regulation of CDKN2A expression were not identified to explain the protein loss. The data argue that p16 loss in chordoma is commonly caused by a post‐transcriptional regulatory mechanism that is yet to be defined.


Introduction
Chordoma is a rare primary malignant bone tumour showing notochordal differentiation and affected individuals have a median survival of 7 years from presentation [1,2]. Chordoma is characterised by expression of the embryonic transcription factor TBXT, also known as brachyury and T, which plays a critical role in the development of the disease [3,4]. Genomic studies have failed to identify recurrent genetic driver alterations other than copy number gain of TBXT in 27% of cases [5] in addition to occasional sporadic chromosomal rearrangements and alterations involving RB1, TP53 and cyclin dependent kinase inhibitor 2A (CDKN2A) [6].
The CDKN2A gene (chromosome 9p21) encodes the proteins p14 ARF and p16 INK4a , also referred to as p16, generated through alternative exon usage [7]. p16 is transcribed using exons 1α, 2 and 3, whereas p14 ARF is transcribed using exon 1β and exon 2. Both proteins are involved in cell cycle control via the Rb and p53 pathways which are critical for self-renewal and ageing [8]. p14 ARF stabilises and activates the p53 pathway, whereas p16 blocks G1/S cell cycle progression by preventing phosphorylation of Rb: disruption of control of these pathways plays a pivotal role in the progression of a variety of cancers [9]. CDKN2A is part of a locus that also contains CDKN2B, which encodes p15 INK4b , a tumour suppressor that, like p16 INK4a , inhibits CDK4/CDK6 [10].
CDKN2A is the second most frequently inactivated tumour suppressor gene in cancer [9,11] and its inactivation is achieved in the majority of cases via homozygous deletion or promoter hypermethylation [11]. Germline mutations in CDKN2A confer susceptibility to melanoma and other tumours [12,13], and haploinsufficiency of p14 ARF has been implicated in genetic models of various cancers [12,14].
The CDKN2A gene locus is deleted and p16 protein expression is lost in a number of chordoma cell lines [15,16]. Loss of p16 protein expression has also been reported in up to 80% of chordomas [6,17,18]. The mechanism leading to its inactivation and the contribution of CDKN2A loss to disease progression have only been partially elucidated. Using small numbers of chordoma samples, it has previously been reported that 3-33% of chordoma cases harbour homozygous deletions of CDKN2A [6,17], single nucleotide variations are rare [5,18], and DNA promoter hypermethylation is an uncommon event [6]. The absence of therapeutic options for patients with chordoma makes this observation clinically significant as p16 loss has been shown to sensitise to CDK4/6 inhibitors [15,19,20], making expression of p16 a potential biomarker for patient stratification and prognosis. This prompted us to interrogate a large number of chordoma samples with the aim of increasing our understanding of the role of CDKN2A inactivation in the pathogenesis of chordoma.

Chordoma samples
Tumour diagnoses were made using the WHO classification [2]. Frozen tumour material was available for 35 chordomas: 10 were analysed by whole genome sequencing and RNA sequencing and 26 by whole exome sequencing, the results of which have been reported previously [5]. Formalin-fixed paraffin-embedded samples were obtained from the archive of the Royal National Orthopaedic Hospital and several other sites. The samples were used to construct tissue microarrays (TMAs), which were built as previously described [21].
Ethical approval for in-house chordoma samples was obtained from the Cambridgeshire 2 Research Ethics Service (reference 09/H0308/165) (HTA Licence 12198). Samples were also obtained through the Brain UK Biobank (reference 14/006 -Large scale genetic and epigenetic screen of chordoma).

Chordoma cell lines
UCH-1, UCH-2, MUG-Chor, UM-Chor, UCH-11, JHC7 (www.chordomafoundation.org) and UCH-7 [16] are well characterised human chordoma cell lines; all derived from sacral tumours except UM-Chor which was generated from a clival chordoma. U2OS (ATCC ® HTB96™, ATCC, Manassas, VA, USA), an osteosarcoma cell line that lacks expression of TBXT, used as a control, was cultured according to ATCC guidelines. Cell lines were quality controlled by shorttandem-repeat analysis (DNA Diagnostic Centre, London, UK) and were regularly tested to ensure that they were mycoplasma-free.

Fluorescence in situ hybridisation and immunohistochemistry
Fluorescence in situ hybridisation (FISH) was performed as described previously [22] using the p16/ CDKN2A (9p21) (Vysis, Abbott Molecular, Abbott Park, IL, USA) and the TBXT (Custom probe to chr6:166526346-166 623 395; Agilent Technologies, Santa Clara, CA, USA) probes. Assessment of CDKN2A and TBXT FISH was undertaken as previously reported [22]: for a probe signal to be counted as abnormal at least 15% of the nuclei analysed were required to reveal an aberrant signal on counting a minimum of 50 consecutive non-overlapping nuclei. The following categories were determined as follows (1) monosomy (one p16/CDKN2A and one centromeric signal); (2) heterozygous deletion (loss of one copy of p16/CDKN2A in the presence of two centromeric signals); (3) homozygous deletion (loss of two copies of p16/CDKN2A in the presence of one or two centromeric signals) and (4) amplification (p16/ CDKN2A centromeric ratio greater than 2). Immunohistochemistry (IHC) was performed on a Leica Bond 3 as previously described [21]. The p16 (JC8) antibody (Santa Cruz, USA, catalogue number SC-56330) was used at a dilution of 1 of 200. This

114
L Cottone et al antibody was previously validated by knock-down in vitro experiments [23]. As TMAs are not fully representative of heterogenous tumours, IHC was repeated and validated on full sections in samples where there was loss of immunoreactivity: this provided a high concordance (88%, 5 false negatives/43). For those cases for which the results obtained using TMAs was inconclusive, the IHC and FISH were repeated on full tissue sections.

Real-time quantitative PCR
Quantitative real-time PCR (qPCR) for mRNA and miRNA expression was performed as previously reported [16] (see supplementary material, Supplementary Materials and Methods).

DNA methylation analysis
DNA methylation array data (450K or EPIC array, Illumina, CA, USA) of 35 chordoma samples have been published previously [27] and EPIC array was performed on chordoma (UCH1, UM-Chor, UCH7, MUG-Chor) and U2OS cell lines. Nucleic acid was prepared as previously reported [5].

Statistical analysis
Continuous variables were compared via unpaired ttest, whereas categorical variables were compared via Fisher's exact test and Chi Squared test using Wizard 1.9.21. Data were considered statistically significant when p < 0.05. Q values were calculated using the R package q value. In gene expression studies, data were judged to be statistically significant when P value was calculated as less than 0.05 by two-tailed Student's ttest. Statistical analysis was performed in GraphPad PRISM 5.0 (GraphPad Software, La Jolla, CA, USA).

Data availability
Infinium Methylation array data of chordoma and U2OS cell lines have been deposited in the National Center for Biotechnology Information GEO database under GEO accession number (GSE139410).

Results
A total of 384 sporadic chordoma samples were collected from 320 patients from across Europe, including the UK, 286 of which were primary tumours, 86 local recurrences and 12 from metastatic disease. Samples (n = 2-6) from more than one time point were studied for 39 patients. The tumours were located in the skull base (n = 90), mobile spine (n = 48), sacrum/coccyx (n = 178) and extra-axial sites (n = 4). The age at presentation of the patients ranged from 6 to 91 (median 60) years: those with tumours located at the skull base, mobile spine, sacro-coccygeum and at extra-axial sites presented between the ages of 6-68, 14-79, 14-76 and 24-28 years of age respectively. The female to male ratio was 1:1.53 (83 females, 127 males).

Genome sequencing data
Whole genome sequencing previously reported [5] revealed copy number alterations at the CDKN2A locus in 7 of 10 chordomas studied (see supplementary material, Table S1 and Figure S1). The patterns of CDKN2A deletion were variable: three cases showed complete loss of chromosome 9, two cases showed loss of several megabases and one case harboured a smaller deletion in the range of kilobases. One case revealed CDKN2A copy number gain; none of the samples showed single nucleotide variants (SNVs) or indels of CDKN2A. FISH revealed TBXT amplification in three cases: this did not correlate with a specific CDKN2A copy number status. Whole genomes and whole exomes (n = 35) were also analysed for the presence of copy number neutral loss of heterozygosity (LOH), which was identified in only one case where it was associated with retention of p16 expression in the absence of mutations.

p16/CDKN2A IHC and FISH
The result of p16 IHC performed on 303 informative samples demonstrated loss of expression in 53% (162/303) of cases ( Figure 1A and Table 1), a finding not dissimilar to previous reports [6,15]. To determine if CDKN2A copy number loss could explain the absence of p16 protein expression we performed FISH 115 Frequent alterations in p16/CDKN2A in chordoma on consecutive tissue sections to those used for IHC ( Figure 1B and Table 1). Results from 274 of these samples were informative: the majority (167/274, 61%) of chordomas harboured CDKN2A copy number alterations, the most frequent event being copy number loss which was detected in 138/167 samples (83%) ( Table 1).
On aligning the FISH with IHC data ( Figure 1C), we found 100% correlation between homozygous loss of CDKN2A and p16 loss of expression. However, only 33% (48/147) of the p16 immunonegative samples revealed a CDKN2A homozygous deletion, leaving an explanation to be found for the loss of protein expression for the remaining 67% of samples. Notably, among p16 immunonegative cases, 27, 12 and 20% showed disomy, monosomy and heterozygous loss respectively ( Figure 1C). This apparent discordance between the FISH and IHC results raises the possibility that protein loss might be achieved via either epigenetic or post-transcriptional regulatory mechanisms.

DNA methylation
To determine if an epigenetic mechanism could explain the p16 protein loss in some chordomas, DNA promoter methylation status of CDKN2A was assessed in four chordoma cell lines (all p16-negative) and 35 chordoma samples, 15 of which were negative for p16 immunoreactivity. We found low levels of methylation in all samples, with the exception of the single clival INI-1-negative poorly differentiated chordoma analysed and the clival-derived INI-1-negative UM-Chor cell line: both of these showed higher levels of promoter methylation compared to all other samples,   Figure S2A). This demonstrates, particularly when previous reports are considered [6], that DNA methylation rarely accounts for p16 protein loss in chordoma.

Polymorphism in CDKN2A
CDKN2A gene expression has been reported to be influenced by the SNP rs11515, which is the most frequent CDKN2A polymorphism located in the 3 0 UTR of the gene and has been associated with various cancers [28]. Analysis of whole genomes and exomes revealed that, of 35 cases, 24 (68%) were homozygous (CC genotype) and 11 (32%) were heterozygous (CG genotype) for the major allele. This frequency is similar to that found in the general population [28] thereby excluding an association of the SNP rs11515 with chordoma.  [29]), were also not influenced by the SNP (see supplementary material, Figure S2B-D). Taken together these results argue against the SNP rs11515 influencing p16 expression in chordoma.

CDKN2A gene expression and IHC
To study further the loss of p16 protein expression in the absence of homozygous deletion we analysed the gene expression in 10 cases previously subjected to RNA sequencing [5] (see supplementary material, Table S1). Three of these cases, which showed loss of p16 protein expression, revealed monosomy in two of the cases and disomy in the third by FISH; however, the CDKN2A transcript levels were comparable to those that retained p16 immunoreactivity (Figure 2A). These data were supported by detectable levels of CDKN2A transcript associated with the loss of expression of p16 at the protein level in an additional 15 chordoma cases assessed by qPCR and in seven chordoma cell lines (three with homozygous deletions and four with monosomy) (Figure 2B-C). This further supports the concept that loss of p16 protein expression is achieved at the post-transcriptional level.
Post-transcriptional regulation of CDKN2A miRNAs including miR-24-2 [30,31], miR-10b-5p [32] and miR-125b [33] have been shown to regulate p16 expression at the post-transcriptional level in various cancers. We therefore tested the expression of these candidate miRNAs in monoallelic (n = 16) and disomic (n = 18) chordoma cases that had retained or lost p16 immunoreactivity ( Figure 2D-F), but we failed to identify a correlation with p16 expression by IHC, suggesting that other mechanisms yet to be identified control the expression of p16 in chordoma.

Patients with multiple samples
Next, we assessed tumour heterogeneity and evolution with respect to p16 IHC and CDKN2A copy number alteration from patients for whom there was more than one sample available including the primary tumour ( Table 2). Informative samples from 36 patients revealed that 26 (72%) showed no change in the immunoreactivity status over time. Tumours from five (14%) patients showed loss of p16 over time and, when correlating these results with FISH on consecutive tissue sections, we observed a change in copy number in three cases (Table 2). This included a case of a p16-positive primary tumour with CDKN2A disomy which revealed heterozygous loss in two local recurrences while retaining p16 positivity, and a homozygous deletion with p16 loss in a third local recurrence: this demonstrates a step-wise acquisition of CDKN2A inactivation in chordoma evolution (case 212). However, in another four cases (5, 54, 213 and 214), despite the loss of p16 immunoreactivity over time, there was no change in the copy number, which was represented in all cases by either heterozygous loss or monosomy (Table 2). Samples from another five patients (39, 88, 211, 217 and 219) revealed loss of p16 expression in the primary tumour but retention in one of the local recurrences (Table 2). We postulated that this could be explained by incomplete excision of an area of the primary tumour which was p16 immunoreactive. This was confirmed when we showed that p16 immunoreactivity was present focally in sections from one of two tissue blocks of the same specimen (case 39) (additional tissue blocks from the other four cases were not available for testing) (Figure 3 and see supplementary material, Table S2). We then tested for p16 expression in multiple blocks from another five cases (22, 56, 61, 90 and 95): three cases were consistently negative whereas two cases showed heterogeneous expression of p16 (see supplementary material, Figure S3 and Table S2). Tumour heterogeneity could 117 Frequent alterations in p16/CDKN2A in chordoma therefore explain the finding of the absence of p16 in the primary tumours and the retention in local recurrences (cases 39, 88, 211, 217 and 219). This could also explain the finding of CDKN2A monosomy in a primary tumour and disomy (cases 20) or polyploidy (cases 157 and 162) in local recurrences. However, these FISH results could also be accounted for by complex structural alterations developing over time including whole chromosomal doubling, but a detailed analysis using whole genome sequencing would be required to confirm such a mechanism.

p16/CDKN2A status and clinical data
Finally, we correlated CDKN2A FISH and IHC results with patients' clinical data (Tables 3 and 4). The anatomical location of the tumour did not influence CDKN2A copy number status or p16 protein expression. There was a trend towards CDKN2A copy number loss being more common in older patients (>40 years old, p = 0.020, q = 0.102) and in samples from local recurrences compared to primary tumours (p = 0.048, q = 0.154). Analysis of the 12 metastatic tumours available revealed that p16 protein expression was lost in all samples (p = 0.0005, q = 0.005, compared to primary tumours). CDKN2A copy number loss in metastatic disease was not significantly different form primary tumours (p = 0.225, q = 0.342).

Discussion
In this p16 IHC study of chordomas, the largest to date, we report that diffuse p16 loss is a frequent finding in this disease, occurring in at least 53% of cases, confirming the previously described frequent loss (66%) in 43 chordoma cases [15]. Much of the reported evidence available to date implies that this loss is associated with either a heterozygous or

118
L Cottone et al homozygous deletion of the region covering the CDKN2A locus but the number of chordoma samples studied using both markers is limited. Specifically, the array CGH findings by Hallor and colleagues [17] showed that 70% of cases displayed genetic alterations but the protein expression was not studied. Le et al [6] performed immunohistochemical analysis on 18 cases and showed loss of expression in 83% of cases; they also observed loss of 9p, either through entire chromosome 9 loss or partial 9p loss alone, in 15/20 cases (75%) by array CGH. Our FISH results also corroborate the published array CGH findings, with 50% of chordomas revealing copy number loss represented by homozygous or heterozygous deletion or monosomy. As expected, homozygous deletion was always associated with loss of p16 expression but otherwise we found little correlation between copy number and p16 immunoreactivity. Notably, in our study 27% of chordomas with loss of p16 protein expression exhibited a normal diploid CDKN2A copy number status, a finding also reported by Le et al [6] in two of 18 cases. Furthermore, 30 of 147 (20%) cases with complete loss of p16 protein expression showed heterozygous loss and 18 of 147 (12%) showed monosomy. This raises the question of the mechanism by which the loss of protein occurs. The frequent loss of p16 protein expression not associated with homozygous loss of CDKN2A could not be explained by a variety of mechanisms that we have investigated.
Specifically, promoter DNA hypermethylation, one of the most common mechanisms implicated in the silencing of CDKN2A in tumours, was not found to account for this, confirming the work of others who reported that only one of 15 chordoma cases tested had definitive evidence of CDKN2A promoter methylation [6]. Only one case, a clival chordoma, showed higher levels of DNA methylation for one of the probes at the promoter region. An interesting finding is that this case showed negativity for INI-1, a feature that is found in poorly differentiated chordomas, which are known to exhibit a methylation profile different from conventional chordomas [34]. Furthermore, with only one in 35 cases revealing copy number neutral LOH, a common copy number alteration caused by uniparental disomy and usually associated with homozygous mutations, homozygous deletions or alterations in cancer-promoting imprinted genes [35,36], this mechanism is unlikely to explain the loss of p16 protein in the absence of homozygous deletion. Indeed, the copy number neutral LOH identified in one chordoma was associated with p16 positivity in the absence of SNVs or indels, a finding rarely reported in other cancers [37].
We pursued two other major lines of enquiry in an attempt to explain the mechanism by which p16 protein could be lost. First, we analysed whether the SNP rs11515 was associated with CDKN2A expression because of the known association in melanoma, sporadic colorectal, skin, bladder, cervical, breast cancer and glioblastoma [28]; however, we failed to detect any association. Second, we tested the expression of miRNAs which were previously shown to control p16 protein expression. MiR-24-2, a negative regulator of p16, blocks p16 translation in keratinocytes and chondrocytes [30,31]. Increased expression of miR-10b-5p correlates with reduced expression of its target genes including CDKN2A in renal papillary carcinoma and glioma [32]. miR-125b is a known tumour suppressor gene that blocks translation of a number of transcripts involved in the control of cell proliferation in various cancers [33], but a convincing association of miR-125 and p16 expression has not been reported, with the exception of a study showing that hsa-miR-125b exhibited significant negative correlations with CDKN2A expression in glioblastoma multiforme [38].

L Cottone et al
In this study we excluded an association between p16 and the expression of these miRNAs.
In view of the difficulty in explaining the loss of p16 protein by promoter hypermethylation and other selected epigenetic events, even though this has only been undertaken using a candidate approach, a number of reasons argue for a post transcriptional regulatory mechanism to explain the loss of protein in the presence of a retained allele as determined by FISH. It is otherwise difficult to explain how comparable levels of CDKN2A could be detected in chordomas with and without protein expression. Although CDKN2A mRNA would be expressed by non-neoplastic cells in chordomas sample, this cannot be the case in the three chordoma cell lines in which mRNA was detected despite the loss of protein expression in the presence of monosomy for chromosome 9. Only one cell line (JHC7) was immunoreactive for p16 in the presence of monosomy, whereas the other three showed CDKN2A homozygous deletion and absence of protein expression.
The median life expectancy of patients with chordoma is 7 years [1]. However, for individuals, it can range from months to more than 25 years (case 143), hence having a prognostic biomarker such as p16 IHC, which could be undertaken easily, would be valuable for patients and clinicians. However, retention of p16 expression cannot be reliably employed to predict less aggressive behaviour in chordoma, because it may be heterogeneously expressed, and detection is dependent on tumour sampling. Furthermore, as immunoreactivity is lost in 53% of our samples and others have reported an even greater percentage, we do not consider that it is useful to employ p16 IHC as a prognostic marker. However, p16 protein expression could be valuable for stratification of patients for the purposes of a clinical trial using a CDK4/6 inhibitor, in which it would be advisable to assess multiple areas [15]. Finally, as much of the expression of p16 is controlled at a post transcriptional level, using FISH as a biomarker to determine protein expression provides limited information and is not recommended in a clinical setting.

SUPPLEMENTARY MATERIAL ONLINE
Supplementary materials and methods Figure S1. Patterns of CDKN2A copy number aberrations using whole genome sequencing data Figure S2. Lack of CDKN2A promoter DNA methylation in chordoma samples and expression of p16, p14 and ANRIL in chordoma cases dependent on rs11515 SNP genotype Figure S3. Heterogeneous expression of p16 in chordoma samples Table S1. p16/CDKN2A assessed by IHC, genomic alterations and DNA methylation in 10 chordoma cases also analysed by whole genome sequencing