Persistent infection of high-risk human papillomavirus (hr-HPV) is an important risk factor for the development of cervical intraepithelial neoplasia (CIN) and cervical cancer. In 2020, the World Health Organization (WHO) classification of female genital tumours was updated from the original three-level classification of cervical intraepithelial neoplasia (CIN1, CIN2, CIN3) to a two-level classification that included low-grade squamous intraepithelial lesions (LSIL/CIN1) and high-grade squamous intraepithelial lesions (HSIL/CIN2 and CIN3) (1). HSIL is recognized as the true precancer with a higher risk of progression. However, it is difficult to determine whether or when a patient with HSIL will progress to invasive cervical cancer from an individual perspective, in fact, some patients may already have occult cervical cancer when diagnosed with HSIL. HSIL are primarily treated with cervical conization, including cervical cold knife conization (CKC) or loop electrode excision procedure (LEEP). On the other hand, most CIN2 lesions (60%), particularly in young women (<30 years), regress spontaneously, indicating that active surveillance, rather than immediate intervention, is justified, especially if patients adhere to monitoring (2). However, there are some limitations in the consistency between pathological assessment via colposcopy-directed biopsy(CDB) and final pathological diagnosis after conization, with an upgrade rate of 23.1% and degrade rate of 33.6% post-conization pathology (3).

To date, there are no accurate tests to determine whether CIN lesions have a tendency to regress or progress. The HPV genotype present in affected patients could not provide additional information to predict high-grade disease progression (4). Although the proportion of severe lesions caused by HPV16/18 has increased over time, its potential for progression remains uncertain (5, 6). Gene methylation is a kind of epigenetic modification that can contribute to the accumulation of mutated genes over time and methylation may play an important role in tumor genesis and progression. As such, using methylation as a marker due to its high sensitivity for cancer has potential as a primary screening tool. It may also be used for the management of women with CIN lesions to prevent overtreatment of CIN2/CIN3 lesions (7).

In particular, the efficacy of paired boxed gene 1 (PAX1) methylation (PAX1^m) as a biomarker for the detection of CIN3 or worse (CIN3+) has been demonstrated in various studies (8–10). PAX1^m can be used as a triage method for women with atypical squamous cells of undetermined significance (ASCUS) and has shown better diagnostic performance than HPV-DNA in predicting CIN2+ (11). Besides, PAX1^m has a comparable clinical performance to cytology and better accuracy and specificity than HPV16/18 as a triage tool for detecting CIN3+ in women with hr-HPV (12). PAX1^m has also been reported to predict the efficacy of concurrent chemo-radiotherapy in cervical cancer (13), and is a potential biomarker for monitoring the prognosis of cervical adenocarcinoma (14). However, few studies have evaluated PAX1 gene methylation before conization, it has been previously reported that PAX1^m would be a suitable alternative method to conventional options and it has the ability to predict the outcome of conization in CIN3 cases (15). However, the role of PAX1^m in predicting the pathological upgrade of CIN2 is unclear. In this study, we aim to investigate the predictive value of PAX1^m status in determining the upgrade tendency of CIN2 and CIN3. This information would help patients and doctors make more individualized treatment decisions.

Among the 218 women with CDB-confirmed HSIL, 100 cases were CIN2_CDB and 118 cases were CIN3_CDB. The mean age was 40.7 ± 10.8 years (22-69). The median △Cp_PAX1 was 19.3 (10.5-20.9) and 8.9 (6.0-14.0) in CIN2_CBD and CIN3 _CBD, respectively, which was significantly different ( Figure 1A ). In the CIN2_CBD group, 36% of CIN2 was pathologically upgraded to CIN3 and 30% regressed to LSIL and below, none of the CIN2 cases upgraded to SCC after CKC, and the positive margin rate was only 2%. However, in the CIN3_CBD group, 19.5% (23/118) of CDB-confirmed CIN3 were pathologically upgraded to ESCC after CKC. The detailed characteristics are presented in Table 1 .

After conization, the△Cp_PAX1 level of ESCC_CKC was significantly lower than that of CIN3_CKC, and that of CIN3_CKC was lower than CIN2_CKC, with statistically significant differences (p<0.001) ( Figure 1B ). Within the CIN2_CDB group, there was no difference in △Cp_PAX1 between CIN2_CDB that had degraded to ≤CIN1_CKC and those that had maintained at CIN2_CKC, but there was a significant difference between those that had maintained at CIN2_CKC and those that upgraded to CIN3_CKC. Among the CIN3_CDB group, △Cp_PAX1levels were significantly lower in those that had upgraded to ESCC_CKC than those that had maintained at CIN3_CKC and downgraded to ≤CIN2_CKC (p<0.001) ( Figure 1C ). Regardless of CIN2 or CIN3 status, △Cp_PAX1 level of women with upgraded pathology after CKC was significantly lower than that of women with degraded pathology (detailed information see Supplementary Table 1 ).

When analyzing PAX1^m at different thresholds, it was found that the risk of CIN3+ increased significantly when ΔCp_PAX1 < 7, while the trend turns flat when ΔCp_PAX1 > 15 ( Supplementary Figure 1 ). We divided the different threshold levels of PAX1^m according to high-, medium-, and low-risk for CIN2+, CIN3+, and ESCC. When ΔCp_PAX1 was 6.4 (1.1-9.0), the OR values of CIN2+, CIN3+, and ESCC were 12.52 (2.85-55.00), 20.61 (8.08-52.57), and 34.07 (4.45-261.08), respectively. In order to adjust variables that would have effects on PAX1^m, we established PAX1^m _{Model 1}(shown in the “Methods” section), and further confirmed that the risk of ESCC, CIN3+, and CIN2+ was still high, and the OR values were 24.85 (3.17-194.67), 19.27 (7.39-50.22), and 11.98 (2.68-53.65), respectively ( Table 2 ), indicating that if ΔCp_PAX1 less than 6.4, it should be alert for the occurrence of high-grade lesions or even cervical cancer.

The optimal ΔCp_PAX1 cut-off value in predicting whether CIN3 would upgrade to ESCC after CKC was 6.360 and the area under the curve (AUC) was 0.814 (95% CI: 0.714–0.915), with similar sensitivity (78.3%) but higher specificity (84.2%) than cytology≥ LSIL (Se:78.3%;Sp:58.9%) and HPV16/18 positive (Se:73.9%;Sp:46.3%). The optimal ΔCp_PAX1 cut-off value in predicting whether CIN2 would upgrade to CIN3 was 10.830 and the AUC was 0.636 (95% CI: 0.515–0.756), with lower sensitivity (44.4%) but higher specificity (82.8%), compared with cytology>ASCUS (Se:44.4%;Sp:45.3%) and HPV positive (Se:97.2%;Sp:9.4%) ( Figure 2 and Table 3 ).

In our study, CDB-confirmed CIN2 and CIN3 were stratified to predict pathological progression after CKC. We found that 19.5% (23/118) of the CDB-confirmed CIN3 cases were pathologically upgraded to ESCC after CKC, which was higher than previously reported (15), indicating that CDB alone is insufficient for the diagnosis of microinvasive cervical cancer (21). In the CDB-confirmed CIN2 group, 36% of the CIN2 cases were pathologically upgraded to CIN3 and 30% regressed to LSIL and below, however, none of the CIN2 cases were pathologically upgraded to SCC after CKC. Using accurate tests to determine whether CIN lesions have a tendency to regress or progress is crucial for subsequent disease management.

Prognostic testing for CIN could dramatically alter the treatment algorithm. Underdiagnosis leads to multiple follow-up visits, and either delayed or progressed the lesion, exacerbating the potential harm to patients. Alternatively, overdiagnosis can result in unnecessary or premature treatment, especially in younger women, as inappropriate treatment significantly increases the risk of adverse outcomes in subsequent pregnancies (22). The majority of HSILs require surgery for the purpose of completely removing lesions, as well as prevent cancer, a small number of special conditions or periods (such as young or pregnant women) can be given a short-term close follow-up (23). As recommended by the ASCCP2019 guidelines, CIN2 and CIN3 should be managed separately, and for patients with CIN 2 that are more concerned about the effects of treatment on a future pregnancy outweigh their concerns about cancer, observation without treatment is acceptable (24).

However, CIN2 and CIN3 diagnosed by CDB have limitations to some extent. For example, the diagnosis of CIN2 has historically been a gray area in pathology and it is difficult for pathologists to reproduce which might be overcalled CIN1 or under-called CIN3 (25). Some pathologists even use “CIN1–2 or CIN2–3” to equivocate the classification. Based on difficulties associated with receiving an accurate diagnosis, it is challenging to determine whether or when a patient with CIN3 will progress to invasive cervical cancer from an individual perspective. In fact, some patients may already have occult cervical cancer when diagnosed with HSIL. The rate of progression to invasive cancer after conization have been reported to be about 0.3%-15% (26–28). In our study, although none of the CIN2 cases progressed to invasive cancer, 36% of the women within the afore mentioned group did progress to CIN3, and progression to invasive cancer in the CIN3 group was as high as 20%. Relying on HPV testing alone cannot accurately predict the progression of cancer satisfactorily. A better prognostic risk evaluation for CIN2 and CIN3 is needed. The integration of molecular markers in cervical cancer screening, such as DNA methylation, might help avoid unnecessary referrals and repeatedly performing diagnostic procedures, which is a waste medical resources and generate needless worry for the patient and her family (29).

The PAX1 gene is located on chromosome 20p11 and consists of a paired domain (PD) and an octapeptide domain (OP). The expression PAX1 is associated with embryogenesis, especially the development of the skeleton, thymus, and the parathyroid glands (30, 31). In 2008, Lai et al. first reported that abnormal methylation of PAX1 was associated with cervical cancer, and that the PAX1 gene was found to be silenced by hyper-methylation and under-expressed in cervical cancer biopsies (8). PAX1 can regulate cell division and differentiation, and methylation and silencing of PAX1 is closely related to the progression of precancerous lesions into cervical cancer (32). It has been reported that the disruption between kinases and phosphatases caused by PAX1 methylation is involved in cervical carcinogenesis (33). An increasing number of studies have confirmed PAX1 methylation as a promising biomarker for cervical cancer based on its ability to discriminate between high‐grade cervical lesions and normal tissues, resulting in a reduced necessity for colposcopy referral and biopsy (9, 10, 34). The current study demonstrated that the ΔCp_PAX1 level of CIN3 determined by CDB was lower than that of CIN2 and the ΔCp_PAX1 level of ESCC was lower than that of both CIN2 and CIN3. Regardless of CIN2 or CIN3 status, the ΔCp_PAX1 level of women with upgraded pathology after conization was significantly lower than that of women with degraded pathology. We further stratified the PAX1^m level by different thresholds and found that the risk of CIN3+ increased significantly when ΔCp_PAX1 < 7. The optimal ΔCp_PAX1 cut-off value in predicting whether CIN3 would be upgraded to ESCC, and whether CIN2 would be upgraded to CIN3 after CKC was 6.360 and 10.830, respectively, and is more specific than using with cytology or HPV abnormalities. This concept implies that if biopsy pathology indicates CIN2 status with a ΔCp_PAX1 less than 10.83, then the actual pathology is more likely to be upgraded to CIN3, so it is necessary to be more careful if observation without treatment was selected. If biopsy pathology indicates CIN3 with a ΔCp_PAX1 less than 6.36, then cervical conization is inevitable because of the increased risk of pathological upgrade to early-stage cervical cancer. On the other hand, women with negative PAX1 methylation do not need immediate colposcopy or conization because of there being a relatively low short-term progression risk for cancer. For young women with CIN2 who have fertility requirements, this approach seems to be particularly important, since only hypermethylated lesions require treatment and the risk of preterm abortion due to treatment could be reduced.

This study has several limitations. First, since our research has not reached the follow-up endpoint, residual and recurrence of lesions have not been discussed here, and continued follow-up is needed in the future. Secondly, the sample size was not large enough and a larger longitudinal study is necessary to validate the natural history of CIN2 and CIN3 progression in relation to DNA methylation. Thirdly, only hospitalized patients with CKC were included in this study who can be followed up well, however, those patients for LEEP from outpatient were not included due to unstable follow-up. In addition, further studies are needed to explore PAX1^m levels after treatment and to compare PAX1^m changes before and after CKC. At last, further studies are also needed to determine the ideal interval of monitoring using PAX1^m to avoid underdiagnosis and overdiagnosis.

In this exploratory study, we found that PAX1 methylation could be a promising auxiliary marker in the prediction of pathological upgrade risk in patients with CIN2 or CIN3 before conization, especially if △Cp_PAX1 cut-off value is lower than 6.360, as we found this to be highly suggestive of invasive cervical cancer. Using PAX1 methylation as a monitoring tool could help prevent inappropriate conservative observation or ablation therapy. Further validation and prospective clinical trials are needed to confirm these findings in the future.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by Peking University People’s Hospital Ethics Committee. The patients/participants provided their written informed consent to participate in this study.

Data collection, study conception, and study design was performed by ML. Personnel organization, data collection, and the cold knife conization procedure was performed by CZ. Sample collection was performed by YZ, JL, HL, ZT, JW and YG. Supervision of the research program and manuscript review and guidance during the study was provided by LW. The first draft of the manuscript was written by ML and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.