NDN and CD1A are novel prognostic methylation markers in patients with head and neck squamous carcinomas

Recruitment

The University of Michigan’s Head and Neck Specialized Program of Research Excellence (SPORE) approaches every incident, previously untreated HNSCC patient to participate in longitudinal epidemiology studies. This unselected study population represents 28 % of incident HNSCC cases in the state of Michigan. From November 2008 through June 2012, subjects were screened for eligibility and 92 % (n = 513) of subjects approached signed a written, informed consent. Consented subjects completed a baseline questionnaire of demographics, epidemiologic characteristics, and behavior modules. Comorbidity data were abstracted from the medical record and graded by severity (none, mild, moderate, severe) using the Adult Comorbidity Evaluation of 27 conditions organized by 12 systems (ACE-27). Research assistants collected formalin-fixed, paraffin-embedded (FFPE) HNSCC tissue blocks and detailed pathophysiological and clinical data annually until death or the patient was lost to follow-up. This study was approved by the Institutional Review Board of the University of Michigan Medical School.

Tissue acquisition

The FFPE tissue blocks were collected pretreatment from three possible sources: (1) a biopsy obtained from an outside hospital, (2) a biopsy performed at the University of Michigan hospital, and/or (3) from surgery performed at the University of Michigan. Tissue acquired from the three sources yielded at least one sample for 88 % (n = 450) of the subjects.

Study population

An expert head and neck pathologist (JM) confirmed tumor histology and screened representative blocks for areas of >70 % cellularity and minimal necrosis. Seventy-two percent (n = 369) of all subjects had sufficient tissue and DNA to yield methylation results. Of these, 15 subjects were excluded for tumors arising from rare sites or non-squamous histology (e.g., unknown primary, nasopharynx, salivary gland, sinus) and 7 subjects were excluded for indeterminate HPV status. One subject was lost to follow up. This resulted in a total of 346 subjects used in the methylation analyses, representing 67 % of the 513 eligible participants screened.

Follow-up

All patients were followed prospectively at designated intervals by clinicians at the University of Michigan or through contact with referring physicians. The median follow-up period was 27 months for survival and 24 months for recurrence (range: 1-54 months). Number of patients alive and followed for OS at 1, 2 and 3 years were 307, 90 and 85 patients, respectively. Number of patients alive and followed for RPFT at 1, 2 and 3 years were 242, 129 and 53 patients, respectively. Deaths were captured through the Social Security Death Index, yearly survey updates, notification from family, and medical record reviews. Survival time and events were censored as of 4/30/13. Recurrence and persistent disease events were confirmed updated annually during a chart review at every subject’s yearly anniversary of their date of initial diagnosis.

Target gene selection

Our group recently completed a discovery-based study designed to identify novel prognostic epigenetic biomarkers for patients with HNSCC [22, 23]. CCNA1 (cyclin A1) was chosen for further testing based on its potential for clinical relevance and the discovery analysis that identified regions of their promoters to be significantly differentially methylated in head and neck cancer patients by HPV status [22]. NDN (necdin) and CD1a (cluster of differentiation 1a) were also differentially methylated in this discovery analysis, however they were not significant, potentially due to small sample size. NDN is an imprinted gene previously implicated in epithelial ovarian, bladder, breast, colorectal, and urothelial cancers, as well as premalignant lesions such as vulval intraepithelial neoplasia and Barrett’s oesophagus, although has not been studied in the context of HNSCC [22–29]. CD1A was the first immune gene found to be differentially methylated in the discovery analysis. CD1A methylation has not been previously studied in HNSCC, however significant hypermethylation of CD1B, CD1C, CD1D and CD1E has been found in HPV (+) HNSCC tumors compared to HPV(-) tumors [30]. DCC (deleted in colorectal carcinoma), GADD45 (growth arrest and DNA damage 45) and p16 (cyclin-dependent kinase inhibitor) were all previously found to be hypermethylated in HNSCC and were chosen for their role as tumor suppressors and potential involvement with HPV [9, 31–34].

Microdissection/DNA extraction/bisulfite conversion

Designated areas of FFPE tissue were microdissected from unstained slides and DNA was extracted using the QIAamp DNA FFPE Tissue Kit (Qiagen, Valencia, CA) according to the manufacturer’s protocol. DNA concentration and purity was measured with a NanoDrop spectrophotometer (Thermo Scientific, Waltham, MA). Sodium bisulfite treatment was performed on 250 ng of DNA using the Epitect Bisulfite Kit (Qiagen, Valencia, CA) according to the manufacturer’s recommended protocol.

HPV testing

HPV status was determined by an ultrasensitive method using real-time competitive polymerase chain reaction (PCR) and matrix-assisted laser desorption/ionization mass spectroscopy, as described in Tang et al. [35]. Multiplex PCR amplification of the E6 region of 15 discrete high risk HPV types (HPV 16, 18, 31, 33, 35, 39, 45, 51, 52, 56, 58, 59, 66, 68 and 73), and human GAPDH control included a competitor oligo identical to each natural amplicon except for a single nucleotide difference. Probes that identify unique sequences in the oncogenic E6 region of each type were used in multiplex single base extension reactions extending at the single base difference between wild-type and competitor HPV so that each HPV type and its competitor were distinguished by mass when analyzed on the MALDI-TOF mass spectrometer as described previously [32, 36–39].

Methylation analysis

Methylation assays for promoter regions of DCC, CD1A, GADD45 and NDN, were designed using PyroMark Assay Design 2.0 software across 5, 2, 5 and 3 CpG sites, respectively (Qiagen, Valencia, CA). The methylation assay for p16 was adapted from Shaw, et al. and covered 4 CpG sites [40]. The promoter region of CCNA1 was sequenced using the Sequenom EpiTyper, a MALDI-TOF mass spectrometry based platform across 4 CpG sites. Primer sets are shown in Additional file 1: Table S1. Location of each CpG site and distance to transcription start site are denoted in Additional file 1: Table S2. Bisulfite singleplex PCR amplification was performed using FastStart Taq Polymerase (Roche Diagnostics, Indiana, US) for CCNA1, and HotStar Taq® Master Mix Kit (Qiagen Valencia, CA) for all other genes, with a forward and reverse primer concentration of 0.2 mM and 30 ng of bisulfite-converted DNA. Confirmation of PCR product quality and absence of contamination was established from 2 % agarose gels with ethidium bromide staining. Fifteen microliters of each PCR product was combined with the respective sequencing primer and methylation analysis by pyrosequencing was conducted using the Pyromark™ MD System (Biotage) according to manufacturer’s protocol, including single strand binding protein (PyroGold reagents). Four bisulfite and four pyrosequencing controls were generated by mixing unmethylated and methylated control DNA (genomic: EpigenDX; bisulfite-converted: Epitect) to obtain controls with 0 %, 30 %, 60 % and 100 % methylation. Each sample plate was run with all controls. If methylation values of controls were incorrect, all samples on plate were re-run. Measurement of all samples for every methylation marker selected was not possible if there was insufficient quantity of total extracted DNA. Average methylation across all CpG sites measured for each gene were used in all statistical analyses as strong associations between CpG sites of each gene has been previously shown [41].

Statistical analysis

Overall survival time (OST) and recurrence/persistence-free time (RPFT) were calculated beginning at date of diagnosis. An OST event was defined as death from any cause. For RPFT, an event was defined as any recurrence (local, regional or distant) of the tumor or persistence of the tumor after definitive treatment. In the case of persistence, an RPFT of 1 day after diagnosis was assigned. In the case of death prior to recurrence, a subject was censored for RPFT at the last known date recurrence-free.

Univariate analyses, including Kruskal-Wallis and Wilcoxon-rank tests, were conducted to test for differences in methylation of each gene by clinico-pathological and epidemiological characteristics. The Kaplan-Meier method was employed to estimate survival time distributions and graphically visualize time-to-event outcomes for overall survival time (OST) and recurrence/persistence-free time (RPFT) by methylation. Methylation of each marker was categorized into quartiles for the Kaplan-Meier curves, with quartile 1 containing the lowest values and quartile 4 containing the highest (Additional file 1: Table S3). Statistical differences in curves were tested using the log-rank test.

Multivariable Cox proportional hazard models were used to test associations between each methylation marker and OST/RPFT after adjusting for known or identified prognostic factors. All mean methylation values were log-transformed after adding an offset value of 1. For each outcome, a model with only clinical predictors was developed using a backward selection algorithm (alpha criteria = 0.05) to arrive at a parsimonious model, with the stipulation that stage and disease site would remain in the model regardless of their significance. Variables introduced for potential inclusion were: age, HPV status, ACE comorbidity score (none, mild, moderate, severe), tobacco use (never, former, current within 12 months), and alcohol use (never, former, current within 12 months). The final clinical model for both OST and RPFT included age, tumor stage, disease site, and HPV; the final clinical model for OST included comorbidity score in addition. After data exploration, significant violation to the proportional hazards assumption was observed for HPV. Stratified Cox proportional hazard models were performed that allowed differing baseline hazard functions for HPV+ and HPV- groups accounting for the non-proportional hazards observed in our data. These stratified Cox proportional hazards models included the same adjustment covariates as the unstratified version, and included an additional interaction term between HPV and methylation marker to test for differences in the associations of the biomarker with OST or RPFT across HPV status. Finally, each methylation marker was added to both the stratified and un-stratified version of the clinical Cox models to assess associations between the marker and outcome after covariate adjustment. All methylation measurements were standardized to interquartile ranges (IQR) of each respective marker (Additional file 1: Table S3). Therefore, hazard ratios (HRs) are interpreted as a comparison between those with methylation in the 25th percentile compared to those with methylation in the 75th percentile.

Unadjusted p-values are presented, however authors advise that a significance finding near the threshold of p < 0.05 should be interpreted with caution. Due to multiple tests being performed (each outcome was modeled for 6 genes), a more conservative Bonferroni threshold for significance was calculated as p < 0.004 (0.05/12) and reflected in superscripts in the Cox model results (Table 4).

Statistical analyses were conducted in R 3.1.1 and SAS 9.3.

HNSCC patient characteristics

Tumor methylation differs by epidemiologic characteristics

Stratification by HPV status across these epidemiologic and epigenetic characteristics revealed HPV (+) tumors were hypermethylated across CCNA1, NDN, CD1A, and DCC, but hypomethylated in p16 as compared to HPV(-) tumors, regardless of clinical epidemiological traits (data not shown).

Tumor methylation and survival/recurrence

Kaplan-Meier curves for the association between methylation, categorized into quartiles, and OST and RPFT across all HNSCC patients, are shown in Additional file 1: Figure S2. Differences in methylation of both CD1A and NDN were significantly associated with OST in all HNSCC patients (p-value = 0.005; p-value = 0.001, respectively). The lowest quartile of methylation of CD1A (Q1) was associated with the lowest probability of OST (0.68 at 24 months) whereas higher methylation in Q2, Q3 and Q4 clustered above a probability of OST of 0.80 at 24 months. Methylation of the lowest quartile of NDN was associated with the lowest probability of OST. In terms of RPFT, there were significant associations with CCNA1, DCC and NDN methylation in all HNSCC patients. NDN revealed a trend of increasing probability of RPFT with increasing quartile of methylation. The lowest and highest quartiles of CCNA1 and DCC had the lowest probabilities of RPFT.

When patients were separated and analyzed by HPV status, distinct associations between methylation markers and OST/RPFT were identified (Additional file 1: Figures S3 and S4). In HPV (+) patients, higher quartiles of methylation of CD1A in Q3 and Q4 were associated with better OST compared to the lower quartiles (p-value = 0.03). Within HPV (-) patients, NDN and DCC were significantly associated with probability of OST. Low methylation of NDN in Q1 had the lowest probability of OST compared to methylation in all other quartiles (p -value = 0.04). DCC methylation in Q2 was associated with the highest probability of OST (p -value = 0.005). A similar pattern was observed for DCC methylation and RPFT, although with DCC, it was Q1 had much poorer probability of recurrence/persistence compared to other quartiles (p-value = 0.002). (Additional file 1: Figure S4)

NDN and CD1A methylation are novel markers of survival in HNSCC patients

To determine the extent to which HPV status may have played a role in these findings, stratified Cox models with interaction between methylation level and HPV status were used to measure associations separately within each group (Table 4). Divergent associations were found in patients based on their HPV status. The test for interaction indicated differences in OST for CD1A (p-value = 0.001) and in RPFT for CD1A (p-value = 0.04) and CCNA1 (p-value = 0.02). In HPV (+) patients, CD1A methylation in the 25^th percentile revealed a 3.34 times higher hazard of a death event than those in the 75^th percentile (95 % CI: 1.88-5.93). In addition, patients had twice the hazard of a tumor recurrence or persistence if they were at the 25^th percentile of CD1A methylation compared to the 75^th percentile. These results suggest that hypermethylation of CD1A is associated with better survival and lower recurrence/persistence in HPV (+) patients specifically. Conversely, for CCNA1, HPV (+) patients had a 0.31 times lower hazard of a recurrent event or persistence comparing those in the 25^th percentile to those in the 75^th percentile of methylation, indicating that hypomethylation of this gene is associated with lower risk of recurrence/persistence in HPV(+) patients.