Associations of P16INK4a promoter hypermethylation with squamous intra-epithelial lesion, cervical cancer and their clinicopathological features: a meta-analysis

To assess the associations of P16INK4a methylation status with low-grade squamous intra-epithelial lesion (LSIL), high-grade squamous intra-epithelial lesion (HSIL), cervical cancer (CC) and their clinicopathological features, a meta-analysis with 29 eligible studies was conducted. Pooled odds ratios (ORs) with their 95% confidence intervals (CIs) were estimated to assess the strength of the associations. Heterogeneity, sensitivity of pooled results and publication bias were also evaluated. Overall, there was an increasing trend of P16INK4a hypermethylation rates among LSIL (21.4%), HSIL (30.9%) and CC (35.0%) specimens. P16INK4a hypermethylation was significantly associated with the increased risk of LSIL, HSIL and CC, with the pooled ORs of 3.26 (95% CI: 1.86-5.71), 5.80 (95% CI: 3.80-8.84) and 12.17 (95% CI: 5.86-25.27), respectively. A significant association was also found between P16INK4a hypermethylation and smoking habit (OR = 3.88, 95% CI: 2.13-7.08). Taken together, meta-analysis results support P16INK4a hypermethylation as an epigenetic marker for the progression of cervical carcinogenesis.


INTRODUCTION
Cervical cancer (CC) is one of the most common gynecologic cancers worldwide [1], with an estimated 527,600 new cases and 265,700 deaths each year [2]. The development of CC is considered as a continuous process from normal epithelium to squamous intra-epithelial lesion (SIL) and ultimately to invasive carcinoma [3]. SIL, the precursor lesions of CC, can be further divided into lowgrade SIL (LSIL) and high-grade SIL (HSIL) depending on the risk of cancer progression [4]. Although infection with human papillomavirus (HPV) is a widely accepted risk factor for SIL and CC [5], the evidence that only a small subset of HPV-induced lesions progress to CC [6], suggests that HPV infection is essential but insufficient for cervical carcinogenesis [4].
DNA hypermethylation, the major epigenetic event in humans, can occur at CPG islands within promoter regions of tumor suppressor genes (TSGs), and consequently silence the TSGs' transcription [7]. P16 INK4a gene, a well known TSG, has been widely investigated in cervical cancer due to its downregulation in cell cycle [8]. Impaired P16 INK4a gene function caused by promoter hypermethylation could result in uncontrolled cell proliferation and eventually oncogenesis [9][10][11]. In 1999, Wong et al. first reported that P16 INK4a promoter hypermethylation was correlated with the advanced stage of CC [11]. Thereafter, numerous studies were carried out to assess the associations of P16 INK4a hypermethylation with the development of SIL and CC. However, most of these studies only included relatively small sample size, leading to inconsistent results and a broad range of P16 INK4a hypermethylation rates (from 2% to 93%) in cancer tissues [12,13]. Moreover, the effect of P16 INK4a promoter hypermethylation on different phases of cervical carcinogenesis (from LSIL to CC) is less summarized. Thus, a meta-analysis was conducted to systematically appraise the associations of P16 INK4a methylation status with LSIL, HSIL, CC and their clinicopathological features.

Association of P16 INK4a methylation status with CC risk
Eighteen studies [13, 17, 19, 21-26, 30, 31, 33-38] with 950 CC patients and 732 controls were included to appraise the effect of P16 INK4a promoter hypermethylation on CC risk. There was a significant association between P16 INK4a promoter hypermethylation and increased CC risk, with an OR of 12.17 (95% CI: 5.86-25.27) and a p value of < 0.001 ( Figure 4 and Table 5). Consistent with the increasing rates of P16 INK4a hypermethylation in LSIL, HSIL and CC specimens, we also found an increasing trend (p < 0.001) in effects of P16 INK4a promoter hypermethylation on the risk of LSIL (OR = 3.26), HSIL (OR = 5.80) and CC (OR = 12.17).
Since moderate heterogeneity was observed in the overall comparison (I 2 = 58%), subgroup, meta-regression  and Galbraith plot analyses were performed to seek the potential sources of heterogeneity. In subgroup analyses, P16 INK4a promoter hypermethylation was consistently associated with increased CC risk in all subgroups (Table  5). However, moderate heterogeneity remained in most of the subgroups, except for the subgroups involving highquality studies (I 2 = 0%), Asians (I 2 = 19%) and healthy controls (I 2 = 44%). The results of meta-regression analyses indicated that ethnicity (p = 0.668), source of controls (p = 0.678) and quality of studies (p = 0.289) were not major sources of heterogeneity (Supplementary Table  1). The subsequent Galbraith plot depicted three outliers [13,17,30] as the potential origins of heterogeneity (Supplementary Figure 1). When we excluded these three studies, the association between P16 INK4a methylation status and CC risk remained significant (OR = 17.36, 95% CI: 10.61-28.42, p < 0.001), followed by an effective reduction in I 2 value from 58% to 12%.

Association of P16 INK4a methylation status with clinicopathological features of SIL/CC
We first evaluated the associations of P16 INK4a methylation status with several risk factors for SIL/CC, including HPV infection (Positive vs Negative), smoking habit (Smoker vs Nonsmoker) and early age at diagnosis ( < 50 vs ≥ 50) (Table 6), and observed that P16 INK4a promoter hypermethylation was significantly associated with smoking habit, (OR = 3.88, 95% CI: 2.13-7.08, P < 0.001) ( Figure 5), but was not correlated with HPV infection and early age at diagnosis (Supplementary Figure 2 and 3). In meta-analyses for the effects of P16 INK4a methylation status on histological types (SCC vs AdC), clinical stages (FIGO stage: III + IV vs I + II) and tumor grades (Grade 2 + 3 vs Grade 1) in CC patients, no significant association was found (Table 6 and Supplementary Figure 4

Evidence grading
Because all eligible studies were observational, the Grading of Recommendations Assessment, Development and Evaluation (GRADE) process for all comparisons began as "low quality" [40]. For the comparisons of CC risk, HPV infection, early age at diagnosis, tumor type and clinical stage, the quality of evidence was further downgraded to "very low quality", due to study limitations, inconsistency or imprecision (Supplementary Table 2).

Sensitivity analyses for assessing the stability of pooled results
In all comparisons, sensitivity analyses by sequentially removing each study did not significantly change the pooled results, suggesting the stability of our meta-analyses (Supplementary Figure 7)

Analyses for publication bias
In all comparisons, funnel plots did not reveal obvious asymmetry (Supplementary Figure 8). These observations, combined with the results of Egger's test (p Egger > 0.05 for all comparisons), suggested that no significant publication bias was found.

DISCUSSION
Previous studies have long aimed to seek methylation biomarkers associated with diagnosis, progression or prognosis of cervical neoplasia. Particularly, a bi-marker panel consisting of CADM1-M18 and MAL-M1 has been considered as a stable triage tool, which could be equally discriminatory for CIN3 + as cytology or cytology with HPV16/18 genotyping in HPV-positive women [41]. In contrast, although P16 INK4a promoter hypermethylation has been linked to CC and SIL, the relatively small sample size of independent studies led to inconsistent results and a broad range of hypermethylation rates in cancer tissues. In this metaanalysis, on the basis of data from over 3000 subjects, we found that the hypermethylation rates in LSIL, HSIL and CC specimens were gradually increased, resulting in a growing trend in effects of P16 INK4a hypermethylation on susceptibility to LSIL, HSIL and CC. These results, combined with the previous epidemiological evidence that P16 INK4a hypermethylation was correlated with the progression of LSIL to HSIL [39,42], suggest that P16 INK4a promoter hypermethylation may be an epigenetic marker for the progression of cervical carcinogenesis. Hence, detecting P16 INK4a hypermethylation may help clinicians to determine whether patients with cervical neoplasia are in disease regression, persistence or progression. Especially in patients with an initial diagnosis of LSIL, once P16 INK4a hypermethylation is found, more effective clinical management for these patients are encouraged to conduct.
However, the existing evidence provides limited information on the prognostic value of P16 INK4a hypermethylation in cervical neoplasia. In a case-series study from China, Yang et al. found no significant association between P16 INK4a hypermethylation and overall survival [29]. In contrast, Blanco-Luquin et al. suggested that P16 INK4a hypermethylation was correlated with improved disease-free survival [38]. Considering that these two studies involved relatively small sample sizes and inconsistent follow-up times, better designed studies are required to address this issue.
The interaction of P16 INK4a hypermethylation with HPV infection is controversial in various HPV-related cancers. For HPV-related oral and oropharyngeal cancer (OSCC) [43], Schlecht et al. found four P16 INK4a -specific CPG loci associated with HPV infection in OSCC tissues  [44], while another study from Chile failed to replicate this association [45]. For cervical carcinoma, previous functional studies have suggested that P16 INK4a promoter hypermethylation mainly occurred at early cervical tumor cell populations without HPV's E7 transcription [46]. In this meta-analysis, HPV infection was not associated with P16 INK4a hypermethylation in patients with SIL/CC. P16 INK4a hypermethylation was associated with a 3.26-fold increased risk of LSIL, suggesting the effect of P16 INK4a hypermethylation on early stage of cervical oncogenesis. All these findings may suggest that P16 INK4a hypermethylation is an early event in cervical carcinogenesis, independent of HPV infection,.
In this meta-analysis, smoking habit was associated with increased P16 INK4a hypermethylation rates in patients with SIL/CC. The correlation between smoking habit and P16 INK4a hypermethylation has been revealed in several cancers, including non-small cell lung cancer (NSCLC) and esophageal squamous cell carcinoma (ESCC) [47,48]. In a longitudinal study, Ma et al. [49] reported that smoking initiation was associated with a 3.76-fold increased risk of the appearance of P16 INK4a hypermethylation in normal cervical smears, providing direct evidence for the relationship between smoke exposure and subsequent acquisition of P16 INK4a hypermethylation in cervix. As a well known risk factor for CC [50], exposure to tobacco smoke, or to its key ingredients (such as nicotine or its derivative), is followed by overexpression of DNA methyltransferases 1, 3A or 3B [51,52], which has been reported to cause hypermethylation of P16 INK4a promoter in mice and cancer patients [53]. Considering that our pooled results were based on the data from relatively few studies, more studies with large sample size are required to repeat this finding.
Moderate heterogeneity was found in our meta-analysis for the association between P16 INK4a hypermethylation and CC risk. Therefore, the results were first pooled by using the random-effects model, which cautiously estimates the study weights after accounting for the inter-study differences [54]. Then, by depicting the Galbraith plot, we found that three studies might be the major contributors to the existence of heterogeneity [13,17,30]. Notably, the hypermethylation rates of CC tissues enormously varied across these three studies (from 5% [17] to 23% [30] and to 95% [13]), suggesting the existence of inter-study differences. By appraising these three studies using our quality scoring system, we found some common flaws for these studies, including lack of biospecimen information [13,17,30], lack of information on conventional risk factors [17,30], and lack of quality controls for methylation detection [13,17]. Otherwise, two of three studies collected non-healthy samples (autologous tissues and samples with atypical squamous cells) as their controls [17,30]. All these issues may lead to the heterogeneous results. Thus, to increase the stability of results, subsequent association analyses for P16 INK4a hypermethylation and CC risk should collect healthy controls, and provide adequate information on related confounding factors.
The following limitations merit consideration. First, most of included studies used the MSP method to detect P16 INK4a methylation status. As a qualitative method, MSP mainly relies on primer designs to guarantee its accuracy [55]. However, the included studies applied different primers to detect methylation status, causing the potential bias that the promoter regions detected by MSP might not always be uniform. Second, lack of clinical data for each participant limited our ability to adjust for other covariates, such as age at primiparity and menopausal status. Finally, most of included studies adopted casecontrol or case-only design. This might lead to some selection bias due to inherent drawback of retrospective studies. Therefore, large prospective studies should be carried out with consistent primer designs, quantitative methylation analyses and multiple clinical data.
In this meta-analysis, P16 INK4a hypermethylation rates showed an increasing trend from LSIL to HSIL and ultimately to CC, causing the increasing effects of P16 INK4a hypermethylation on susceptibility to LSIL, HSIL and CC. Moreover, P16 INK4a hypermethylation was also correlated with smoking habit in patients with CC/SIL. Future studies are warranted to repeat these findings and elucidate the underlying mechanism.

Literature search
This meta-analysis was reported based on the PRISMA statement [56]. Electronic databases, including Pubmed, EMBASE and Web of Science (up to April 19, 2016), were searched by using the combinations of following terms: (P16 INK4a or P16 or CDKN2A) and (methylation or promoter methylation or DNA methylation) and (cervical cancer/cervical tumor/cervical neoplasia or SIL/LSIL/HSIL/ or cervical dysplasia/CIN/ CIS). Reference lists in reviews and retrieved articles were also checked for other relevant studies.

Eligibility criteria
Eligible studies were required to meet the following criteria: (1) an observational design (cohort, case-control, case-only or cross-sectional studies); (2) studies assessing the associations of P16 INK4a methylation status with LSIL, HSIL, CC or their clinicopathological features; (3) studies with sufficient data to calculate the hypermethylation rates, ORs and their 95% CI; (4) written in English.
Exclusion criteria were as follows: (1) reviews, letters, abstracts and case reports; (2) reports with