A novel non-invasive mRNA-lncRNA biomarker panel for accurate prediction of cervical squamous cell carcinoma and adenocarcinoma

Abstract

Squamous cell carcinoma (SCC) and adenocarcinoma (ADC) represent predominant histological subtypes of cervical cancer. To improve screening efficacy, we leveraged RNA sequencing data from 4 cervical SCC samples, 4 cervical ADC samples, and 8 normal cervix samples and conducted a comprehensive mRNA and long noncoding RNA (lncRNA) profiling analysis followed with a multi-phase study comprising 556 samples. Validating the RNA sequencing data in a clinical sample set comprising 45 normal cervix tissues, 45 SCC tissues, and 45 ADC tissues, we identified 9 mRNAs (SMC1B, OTX1, GRP, CELSR3, HOXC6, ITGB6, WDR62, SEPT3, and KLHL34) and 4 lncRNAs (FEZF1-AS1, LINC01305, LINC00857, and LINC00673) differentially expressed in both SCC and ADC samples. Utilizing quantitative reverse transcription polymerase chain reaction analysis and receiver operating characteristic (ROC) curve analysis in a training set (45 normal, 126 SCC, and 82 ADC tissues), we refined a novel mRNA-lncRNA-based panel (SMC1B/CELSR3/FEZF1-AS1/LINC01305). Employing logistic regression model and ROC analysis, this panel exhibited significant distinctions and promising area under the curve (AUC) values in both SCC (AUC=0.9520, p<0.0001) and ADC (AUC=0.9748, p<0.0001) tissues. Subsequent validation in an independent set (11 normal, 32 SCC, and 20 ADC tissues) demonstrated its diagnostic accuracy in both SCC (AUC=0.9659, p<0.0001) and ADC (AUC=0.9636, p<0.0001) patients. Notably, this tissue-based biomarker panel robustly discriminated precancerous lesion and cervical cancer patients from non-disease controls in a blood-based validation set (30 normal, 25 HSIL and 50 cervical cancer) with an AUC value of 0.9320. This study presents a non-invasive, efficient diagnostic panel for cervical cancer screening.

Synopsis

A novel mRNA-lncRNA-based panel (SMC1B/CELSR3/FEZF1-AS1/LINC01305) was refined in our study. The biomarker panel we established can discriminate cervical cancer patients from non-disease controls both in tissue-based validation set and blood-based validation set.

INTRODUCTION

Cervical cancer ranks as the fourth most prevalent malignant tumor affecting women globally [1]. The incidence and mortality of cervical cancer varies widely with geography, with 85%–90% of new cases and deaths occurred in less developed regions [2]. In China alone, an estimated 111,820 new cases of cervical cancer were diagnosed, leading to approximately 61,579 deaths in 2022 [3]. Furthermore, the incidence and mortality rates of cervical cancer in China have exhibited an upward trajectory since 2000 [3]. Squamous cell carcinoma (SCC) is the predominant cervical neoplasm, while cervical adenocarcinoma (ADC) ranks as the second most common histologic subtype, constituting 10%–25% of all cases [4,5]. Although the introduction of human papilloma virus (HPV) vaccine and widespread screening practice successfully decreased the SCC burden, particularly in developed countries [1,6], developing countries like China still face a multitude of realistic issues such as financial constraints, limited medical resource and people’s fear of cancer detection [2]. Adapting a cost-effective and non-invasive approach of screening would expand the coverage of screening. Unlike squamous lesions, ADC more frequently originates in the endocervix, often with skip lesions, and displays a weaker association with HPV infection [7]. Consequently, ADC cases are less likely to be detected by current cervical cancer screening strategies, which primarily rely on HPV and cytology tests [5,7], leading to an increase in the relative prevalence of ADC compared to SCC [8,9]. In the United States, the incidence of SCC decreased by 61.1% over the past 40 years whereas the incidence of ADC increased by 32.2% in the same period [10]. Hence, there is a critical need to develop novel screening approaches capable of identifying both SCC and ADC patients to enhance the efficiency and accuracy for early cervical cancer detection.

Effective screening is recognized as a powerful tool to reduce the incidence and mortality of cervical cancer. The current available screening strategies include cytology, HPV test, co-testing, etc. As a widely used cytology detection, ThinPrep cytology test (TCT) showed a specificity over 90% but a relative low sensitivity of 53%–80% in detecting cervical intraepithelial neoplasia (CIN) 2+ lesions [11]. HPV-based testing offers a superior sensitivity compared with cytology [12]. However, HPV infection status is common among sexually active women, and 90% cases present as a transient infection. A positive screening result may generate anxiety and lead to overtreatment [13]. The triage strategies such as p16/Ki67 dual staining, DNA methylation further increase the diagnostic accuracy of a positive HPV screening result, while these detections require adequate resource and qualified laboratory facilities, restricting the implementation in massive screening [14]. The present research of novel screening strategies need to lay more emphasis on accessibility, precision and acceptance by populations.

Aberrant expressions of coding and non-coding genes, including mRNAs, long noncoding RNAs (lncRNAs), circular RNAs, and microRNAs (miRNAs), have been implicated in tumorigenesis, growth, and progression. LncRNAs, a novel class of RNA transcripts exceeding 200 nucleotides in length and devoid of protein-coding capacity, have emerged as key players in cancer development and progression at various molecular levels [15]. Moreover, the regulatory interplay between lncRNAs and mRNAs has been extensively investigated, offering insights into the potential of mRNA-lncRNA combinations as superior diagnostic or therapeutic targets in certain diseases [16]. For instance, Zhang et al. [17] demonstrated that a co-expression module comprising one lncRNA (RP11-847H18.2) and 3 mRNAs (KLHL28, SPRTN, and EPM2AIP1) exhibited superior predictive performance of myocardial infarction diagnosis and prognosis compared to individual molecule. Another study identified a multi-gene signature, including 8 mRNA and one lncRNA, which demonstrated improved predictive efficacy for laryngeal cancer [18]. In cervical cancer, a recent study established a lncRNA-mRNA (LINC00511-PGK1) co-expression model for predicting survival in HPV-driven cervical cancer patients through bioinformatics and systematic biological approaches [19]. Collectively, these findings underscore the potential of biomarker panels based on multiple differentially expressed RNAs to enhance diagnostic and prognostic accuracy. To date, there have been no established screening models for both SCC and ADC in cervical cancer.

In previous investigations, we conducted transcriptome analyses comparing 4 normal cervical tissues with 4 SCC tissues [20], and another 4 normal cervical tissues with 4 ADC tissues [21]. In the present study, we utilized published RNA-seq data to identify differentially expressed mRNAs and lncRNAs in both SCC and ADC samples, followed by preliminary verification. Then we systematically explored and validated a novel panel containing 2 mRNAs (SMC1B and CELSR3) and 2 lncRNAs (FEZF1-AS1 and LINC01305) based on quantitative real-time polymerase chain reaction (qPCR) and receiver operating characteristic (ROC) analysis in both tissue-based and serum-based model. In summary, our study underscores the potential of a non-invasive mRNA-lncRNA combined panel as a potent tool for cervical cancer screening.

MATERIALS AND METHODS

1. Patient cohorts and study design

In this study, 572 participants were enrolled from Women’s Hospital, School of Medicine, Zhejiang University under the approval of the Hospital Ethical Committee (IRB-2019062-R & IRB-20220353-R) from January 2018 to December 2023. Informed consent to use and publish clinical information for research purposes were obtained from all the patients in accordance with the Declaration of Helsinki.

The diagnosis of all samples was confirmed through pathology from surgical procedures or cervical biopsies in accordance with clinical guidelines. All tissue and blood samples were collected prior to any cancer treatment. All samples were stored at −80°C until use. Patients with a history of any cancer-related treatment, other malignant tumors, infectious diseases or immune deficiency syndromes were excluded.

The overall study design of this study is illustrated as Fig. 1. The study comprised of a systematic and comprehensive biomarker discovery, followed by multiple validation phases. For biomarker discovery phase, 8 normal cervical tissue samples, 4 SCC tissue samples and 4 ADC tissue samples were collected in 2018 and subjected for transcriptome sequencing [20,21]. A panel of cervical tissue samples including 45 normal, 45 SCC, and 45 ADC tissue samples were collected in 2020 and used for the preliminary validation by quantitative reverse transcription polymerase chain reaction (qRT-PCR) analysis. Another panel of cervical tissue samples containing 56 normal, 158 SCC and 102 ADC tissue samples were collected from 2019 to 2022 (Table S1). Theses cervical samples were then randomly divided into a training set (45 normal, 126 SCC and 82 ADC tissue samples) and a validation set (11 normal, 32 SCC and 20 ADC tissue samples, labeled as Validation phase 1) in an 8:2 ratio. For the translation of tissue-based biomarkers into a blood-based assay, peripheral blood samples of 25 high grade squamous intraepithelial lesion (HSIL) patients, 50 cervical cancer patients (33 SCC, 13 ADC and 4 other pathological type) and 30 non-disease controls were collected from August 2022 to December 2023 and analyzed as another validation set (labeled as Validation phase 2, Table S2). These peripheral blood samples were then centrifuged at 3,000 rpm for 10 minutes, and the supernatant was stored at −80°C until extraction of total RNA.

Fig. 1. Study design overview. Workflow depicting the exploration of potential biomarker panel for discriminating cervical squamous carcinoma and cervical adenocarcinoma.

ADC, adenocarcinoma; DE, differentially expressed; HSIL, high grade squamous intraepithelial lesion; lncRNA, long noncoding RNA; mRNA, messenger RNA; qPCR, quantitative real-time polymerase chain reaction; ROC, receiver operating characteristic; SCC, squamous cell carcinoma.

2. Transcriptome sequencing analysis

The transcriptome sequencing was performed as previously described, and the sequencing data has been published in our previous reports [20,21].

In the present study, the differential gene expression analysis was performed between the 4 SCC samples (labeled as SCC1–4) and 4 normal cervical tissues (labeled as Normal A, Normal 1–4) [20], and the 4ADC samples (labeled as ADC1–4) and another 4 normal cervical tissues (labeled as Normal B, Normal 5–8) [21] using the DESeq2 package in R. Genes that survived |fold change (FC)| ≥2 and false discovery rate (FDR) <0.05 were considered differentially expressed. Then the differentially expressed genes between SCC group and ADC group were separately plotted as clustered heatmaps. The raw data of our transcriptome sequencing have been uploaded to Gene Expression Omnibus (GSE:113942 and GSE: 145372).

3. RNA isolation and real-time quantitative reverse-transcription PCR

Total RNA was isolated from fresh frozen cervical tissues and 200 μL of serum specimens using Trizol reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s instructions. The isolated RNA was then reverse transcribed into cDNA using the Prime Script RT reagent kit (Takara, Osaka, Japan). To quantify the amount of mRNA and lncRNA expression level, qRT-PCR was performed using TB Green Premix Ex TaqII (Takara), and 18s was used as internal controls. Relative RNA expression levels were calculated using the 2^−ΔΔCT method. The primers used in this study are listed in Table S3.

4. ROC curve analysis

The diagnostic values of selected biomarkers were evaluated by calculating the area under the curve (AUC) of the ROC curves. The qPCR RNA expression validation results (2^−ΔΔCT) determined for the biomarker candidates in the training phase and validation phase were uploaded to IBM SPSS 26.0 (IBM Corp., Chicago, IL, USA) and applied for ROC curve analysis. An AUC above 0.7 denotes high significance of discrimination, whereas an AUC between 0.5 to 0.7 denotes a low significance of discrimination. The AUC under 0.5 is defined as a complete lack of discrimination.

5. Statistical analysis

All data were statistically analyzed with SPSS 26.0 (IBM Corp.) and visualized with GraphPad Prism version 9.0 (GraphPad Software Inc., San Diego, CA, USA). Data in this study are expressed as mean ± standard error of the mean.

Differences in RNA candidate expression between groups were determined using student’s t-test, Mann-Whitney t-test where appropriate. Differences were statistically significant at p<0.05, p<0.01, p<0.001 and p<0.0001.

RESULTS

1. Identification of mRNA and lncRNA expression signatures in SCC and ADC of cervical cancer

In our previous studies, transcriptome analyses were conducted on 4 normal cervical tissues (labeled as Normal A, Normal 1–4) compared to 4 cervical cancerous tissues (labeled as SCC1–4) [20], and another set of 4 normal cervical tissues (labeled as Normal B, Normal 5–8) compared to 4 cervical ADC tissues (labeled as ADC1–4) [21]. In this study, we aimed to characterize the expression profiles of both mRNAs and lncRNAs in these cervical samples. Our sequencing results revealed 1,918 differentially expressed mRNAs and 503 differentially expressed lncRNAs in the SCC tissues compared to the normal cervical tissues, with a cut-off criteria of |FC| ≥2 and FDR <0.05. Similarly, using the same criteria, we identified a total of 1,711 differentially expressed mRNAs and 671 differentially expressed lncRNAs in ADC compared to normal tissue samples. Subsequent analysis revealed that 510 differentially expressed mRNAs and 78 differentially expressed lncRNAs were shared between SCC and ADC tissues compared to normal cervical tissues (Fig. 2A and B). These overlapped differentially expressed mRNAs and lncRNAs are depicted as clustered heatmaps separately in Fig. 2C and D. Among the differentially expressed mRNAs and lncRNAs identifed, we selected 18 mRNAs and 8 lncRNAs for further investigation based on their statistical significance, fold-change thresholds, basal expression levels in the sequencing data, and their reported cancer-related functions in existing literature (Table S4).

Fig. 2. The transcriptomic signatures of DE mRNAs and lncRNAs in SCC and ADC samples. Venn diagram depicting the overlapped mRNAs (A) and lncRNAs (B) DE in both SCC samples and ADC samples compared to normal cervical tissues. Clustered heatmap showing DE mRNAs (C) and lncRNAs (D) in both SCC samples and ADC samples compared to normal cervical tissues. Red: up-regulation; Blue: down-regulation.

ADC, adenocarcinoma; DE, differentially expressed; lncRNA, long noncoding RNA; mRNA, messenger RNA; SCC, squamous cell carcinoma.

2. Verification of the RNA-Seq results in clinical cervical tissue samples

To validate the findings from RNA-Seq analysis in clinical cervical tissue samples, we conducted qRT-PCR on 135 cervical tissue samples, comprising 45 normal, 45 SCC, and 45 ADC tissues. Among the 18 differentially expressed mRNA and 8 lncRNA candidates, 15 mRNAs and 6 lncRNAs were significantly up-regulated, while 3 mRNAs and 2 lncRNAs were significant down-regulated in both SCC and ADC compared to normal cervical tissues. The qRT-PCR results confirmed the up-regulation of 13 mRNAs (SMC1B, OTX1, FOXD1, GRP, CELSR3, HOXC6, HOXC11, YBX2, ONECUT2, GRIN2D, ITGB6, WDR62 and SEPT3), and the down-regulation of 3 mRNAs (PCDH20, CALN1 and KLHL34) in both SCC and ADC tissues compared to normal cervical tissues, aligning with the sequencing data (Fig. 3A). Additionally, 5 lncRNAs (FEZF1-AS1, LINC01305, LINC00857, LINC000673 and LOC648987) were up-regulated, while LINC000284 were down-regulated in both SCC and ADC compared to normal cervical tissues (Fig. 3B). Further, ROC analyses were performed to assess their potential predictive ability in distinguishing SCC and ADC from normal cervix. Utilizing an AUC value >0.7 as a threshold for strong discriminative power in both SCC and ADC samples, we identified 9 mRNAs (SMC1B, OTX1, GRP, CELSR3, HOXC6, ITGB6, WDR62, SEPT3, and KLHL34), and 4 lncRNAs (FEZF1-AS1, LINC01305, LINC00857, and LINC000673) for subsequent investigation (Fig. 3C and D, Fig. S1).

Fig. 3. Experimental verification of selected mRNAs and lncRNAs in cervical tissue samples. The quantitative reverse transcription polymerase chain reaction validation of the 18 mRNA candidates (A) and the 8 lncRNA candidates (B) in 45 normal, 45 SCC and 45 ADC tissue samples. Receiver operating characteristic analysis of the individual mRNA candidates (C) and lncRNA candidates (D) with high discriminative power (AUC >0.7) in both SCC and ADC patients in 45 normal, 45 SCC and 45 ADC tissue samples.

ADC, adenocarcinoma; AUC, area under the curve; lncRNA, long noncoding RNA; mRNA, messenger RNA; NS, not significant; SCC, squamous cell carcinoma.

^*p<0.05; ^**p<0.01; ^***p<0.001; ^****p<0.0001.

3. Construction of prediction model for diagnosing both cervical SCC and ADC patients

Another panel of clinical cervical samples were gathered, comprising 56 normal, 158 SCC, and 102 ADC tissues. Subsequently, 80% of these cervical tissues (45 normal, 126 SCC, and 82 ADC tissues) were randomly allocated to a training set, while the remaining 20% (11 normal, 32 SCC, and 20 ADC tissues) constituted the validation set.

In the training set, the expression levels of 9 mRNA candidates (SMC1B, OXT1, GRP, CELSR3, HOXC6, ITGB6, WDR62, SEPT3, and KLHL34) and 4 lncRNA candidates (FEZF1-AS1, LINC01305, LINC00857, and LINC00673) were detected via qRT-PCR. Fig. 4A and B illustrate that 8 out of 9 mRNA candidates (SMC1B, OXT1, GRP, CELSR3, HOXC6, ITGB6, WDR62, and SEPT3) and all 4 lncRNAs were significantly up-regulated in both SCC and ADC samples. However, KLHL34 mRNA levels showed no significant difference in the expanded set of SCC samples compared to normal cervical tissues. ROC analysis was performed to evaluate the diagnostic value of these 8 mRNAs and 4 lncRNAs in the training set. The ROC curves and AUC values of each candidate in both subtype tissues were presented in Fig. S2 and Table S5. According to the ROC analysis, the AUC values of 6 mRNAs (SMC1B, OTX1, GRP, CELSR3, HOXC6, and SEPT3) and 4 lncRNAs (FEZF1-AS1, LINC01305, LINC00857, and LINC00673) ranged from 0.7 to 1.0 in SCC and ADC samples, indicating high diagnostic values in both pathological types of cervical cancer.

Fig. 4. Construction of prediction model for diagnosing both cervical squamous cell carcinoma and adenocarcinoma patients. The quantitative reverse transcription polymerase chain reaction validation of the 9 mRNA candidates (A) and the 4 lncRNA candidates (B) in the training set. Comparisons of the AUC values, sensitivity and specificity for prediction of SCC diagnosis (C) and ADC diagnosis (D) by SMC1B/CELSR3/FEZF1-AS1/LINC01305 combined panel and individual candidates in the training set.

ADC, adenocarcinoma; AUC, area under the curve; lncRNA, long noncoding RNA; mRNA, messenger RNA; NS, not significant; SCC, squamous cell carcinoma.

^*p<0.05; ^**p<0.01; ^***p<0.001; ^****p<0.0001.

To develop a prediction model for diagnosing both SCC and ADC, we employed stepwise logistic regression in the training set, starting with all 10 candidates and optimizing the panel to achieve the best AUC value with the fewest candidates. As a result, we found that the combination of SMC1B, CELSR3, FEZF1-AS1, and LINC01305 exhibited the most satisfactory discrimination between SCC, ADC and normal cervix tissues. The logit models were expressed as follows:

For SCC prediction: logit (p) = −1.98 + 0.07 × (SMC1B) + 0.157 × (CELSR3) + 0.137 × (FEZF1-AS1) + 0.066 × (LINC01305)

For ADC prediction: logit (p) = −3.495 + 0.158 × (SMC1B) + 0.114 × (CELSR3) + 0.176 × (FEZF-AS1) + 0.138 × (LINC01305)

Application of the models in the training set yielded an AUC value of 0.9520 with a cut-off value of 0.743, a sensitivity of 86.5%, and a specificity of 95.6% in SCC samples (Fig. 4C). For ADC samples, the AUC value for the SMC1B/CELSR3/FEZF1-AS1/LINC01305 panel was 0.9748 with a cut-off value of 0.675, a sensitivity of 89.0%, and a specificity of 93.3% (Fig. 4D). Moreover, the combined SMC1B/CELSR3/FEZF1-AS1/LINC01305 panel demonstrated higher diagnostic accuracy in predicting SCC and ADC patients compared to any single candidate.

4. Assessment of the diagnostic performance of the combined mRNA-lncRNA panel in SCC and ADC of cervical cancer

The diagnostic value of the developed prediction model was further evaluated in an independent validation set comprising 11 normal cervical tissues, 32 SCC, and 20 ADC tissues. The qRT-PCR analysis confirmed significant upregulation of SMC1B, CELSR3, FEZF1-AS1, and LINC01305 in both SCC and ADC tissues compared to normal cervical tissues (Fig. 5A). Similarly, ROC curves were constructed to assess the predictive performance. The AUC values and corresponding diagnostic parameters of each individual candidate and the combined panel are presented in Fig. 5B-D and Table S6. The combination of SMC1B/CELSR3/FEZF1-AS1/LINC01305 demonstrated superior AUC values (AUC: 0.9659 for SCC; AUC: 0.9636 for ADC) compared to any single candidate in both SCC and ADC tissues. In this validation set, the combined panel exhibited a sensitivity of 93.75% and 90% in diagnosing SCC patients and ADC patients, respectively. Additionally, the specificities reached 100% for both SCC and ADC patients, with cut-off values of 0.567 and 0.710, respectively. These findings suggest that the mRNA-lncRNA combined panel comprising SMC1B/CELSR3/FEZF1-AS1/LINC01305 serves as a robust predictor for diagnosing both SCC and ADC.

Fig. 5. Assessment of the diagnostic value performance of the combined mRNA-lncRNA panel in both squamous cell carcinoma and adenocarcinoma of cervical cancer. (A) The quantitative reverse transcription polymerase chain reaction validation of single candidate of the combined panel in the validation set. (B) Receiver operating characteristic analysis of the individual candidates in both SCC and ADC tissues of the validation set. Comparisons of the AUC values for prediction of SCC diagnosis (C) and ADC diagnosis (D) by SMC1B/CELSR3/FEZF1-AS1/LINC01305 combined panel and individual candidates in the validation set.

ADC, adenocarcinoma; AUC, area under the curve; lncRNA, long noncoding RNA; mRNA, messenger RNA; SCC, squamous cell carcinoma.

^*p<0.05; ^**p<0.01.

5. Diagnostic accuracy validation of the combined mRNA-lncRNA panel to distinguish precancerous lesions and cervical cancer via non-invasive method

To further expand the application of the combined panel, we applied the tissue-based panel in a serum-based validation set comprising 30 non-cervical disease controls, 25 HSIL patients and 50 cervical cancer (33 SCC, 13 ADC and 4 other pathological type) patients. We first conducted qPCR analysis in serum samples and found that all 4 candidates were significantly upregulated in both HSIL and cervical cancer patients (Fig. 6A). The ROC analysis was performed to access the diagnostic values. The combined biomarker panel yet again performed remarkably well and robustly distinguished patients with precancerous lesions and cervical cancer from controls, with an AUC value of 0.9320, a sensitivity of 84%, and a specificity of 100% (Fig. 6B). Of the 50 cervical cancer patients, 29 were in the early stage (International Federation of Gynecology and Obstetrics stage IA–IB2). The biomarker panel also showed an impressive AUC of 0.9380, sensitivity of 79.3%, and specificity of 100% to identify early-stage cervical cancer (Fig. 6C). To address the potential of the biomarker panel for early detection, we also analyzed the diagnostic accuracy in the combined group of 25 HSIL patients and 29 early-stage patients of cervical cancer, which was validated with an AUC of 0.9120, a sensitivity of 85.2%, and a specificity of 100% (Fig. 6D). These results highlight the ability of our combined mRNA-lncRNA panel for non-invasive screening in cervical cancer.

Fig. 6. Diagnostic accuracy validation of the combined mRNA-lncRNA panel to distinguish precancerous lesions and cervical cancer patients via non-invasive method. (A) The quantitative reverse transcription polymerase chain reaction validation of single candidates in serum specimens of an independent validation set comprising 30 normal controls, 25 HSIL patients and 50 cervical cancer patients. (B) Comparisons of the diagnostic performance of HSIL and cervical cancer diagnosis by SMC1B/CELSR3/FEZF1-AS1/LINC01305 combined panel and individual candidates in the serum-based validation set. (C) Comparisons of the diagnostic performance of early-stage cervical cancer diagnosis by SMC1B/CELSR3/FEZF1-AS1/LINC01305 combined panel and individual candidates in the serum-based validation set. (D) Comparisons of the diagnostic performance of early detection of cervical cancer by SMC1B/CELSR3/FEZF1-AS1/LINC01305 combined panel and individual candidates in the serum-based validation set.

AUC, area under the curve; HSIL, high grade squamous intraepithelial lesion; lncRNA, long noncoding RNA; mRNA, messenger RNA.

^**p<0.01; ^***p<0.001; ^****p<0.0001.

DISCUSSION

Effective and universal screening is one of the key strategies to eliminate cervical cancer [22]. Here, we conducted a multi-phase and comprehensive study and developed a novel mRNA-lncRNA combined panel which can robustly identify both SCC and ADC patients, even as a blood-based, non-invasive assay.

Numerous studies have proposed potential molecular biomarkers to differentiate between SCC and ADC. For example, Qiu et al. [23] preformed scRNA-seq and TCR-seq analyses on SCC and ADC samples, identifying diagnostic biomarkers unique to each histological subtype (KRT14 for SCC and CLDN3 for ADC). Despite promising findings, such as serum YKL-40’s potential as a biomarker for both subtypes, sensitivity levels remain insufficient for effective screening (75% for SCC and 78% for ADC) [24]. Validated biomarkers capable of efficiently identifying both SCC and ADC patients have yet to be established. RNA biomarkers have long been utilized in clinical practice [25], with recent studies indicating the potential for improved diagnostic accuracy through combined RNA types. Integrated mRNA-lncRNA signatures, as demonstrated in various cancers, have shown superior performance compared to individual markers [26]. Our previous research identified a miRNA-lncRNA-mRNA panel (miR-192-5p/HNF1A-AS1/VIL1) capable of accurately distinguishing ADC from normal cervical tissue [21], suggesting the promise of combined RNA panels for cancer diagnosis.

Expanding on our previous RNA-sequencing data, our study has identified a promising biomarker panel comprising SMC1B, CELSR3, FEZF1-AS1, and LINC01305, each previously implicated as oncogenes in cervical cancer [20,27,28,29]. SMC1B emerged as one of the predominant gene signatures during CIN progression [30]. Li et al. [27] also reported that SMC1B was significantly upregulated in cervical SCC and promoted EMT progression of cancer cells via the PI3K-AKT pathway. CELSR3 was identified in our prior research as a novel candidate gene associated with cervical lesion progression and carcinogenesis [20]. Expanded validation in our current study further supports CELSR’s potential as a solid biomarker for cervical cancer. However, the biological function and underlying mechanism of CELSR3 in cervical cancer require further exploration. Elevated FEZF1-AS1 expression in cervical cancer tissues was first observed by Chen et al. [28] and linked to aggressive behaviors and poor prognosis [31]. Some studies suggest FEZF1-AS1 may act as sponge for miRNAs such as miRNA-1254 and miR-367-3p, facilitating malignant behaviors of cervical cancer cells [32,33]. LINC01305 was reported as one of the top 10 up-regulated lncRNAs associated with cervical cancer progression [29]. Another study illustrated LINC01305’s pro-tumor role in cervical cancer by interacting with the RNA-binding protein KHSRP, activating NF-κB and STAT pathways [34]. While their individual roles in SCC have been established, their diagnostic value in ADC is novel. The combination of these markers demonstrated improved diagnostic ability compared to individual candidates in both cervical tissues and serum samples of SCC and ADC patients.

The sensitivity and specificity of screening methods are crucial for early detection in cancer screening. Currently, HPV testing is recommended as the primary method in many countries due to its superior sensitivity and negative predictive value compared to conventional cytology screening, as demonstrated by numerous randomized controlled trials [12,35]. However, it’s important to note HPV positivity may only indicate an HPV infection stage, potentially leading to over-diagnosis and over-treatment [12], and its high cost makes it impractical for large-scale screening programs. TCT is another important screening method in areas without access to HPV testing [12]. Clinical studies have shown TCT to have high specificity (over 90%) but relatively low sensitivity (53%–80%) [36,37], especially in detecting ADC, where its detection rate is unsatisfactory [38]. Although some studies suggest that co-testing (combining HPV testing with cytology) could improve ADC detection [39], the most effective screening strategy of both SCC and ADC remains unclear and requires further research [12]. In our study, a combined panel of biomarkers demonstrated promising sensitivity (86.5% for SCC, 89.0% for ADC) and specificity (95.6% for SCC, 93.3% for ADC) in the tissue-based training set. Validation using cervical tissues in a separate set showed even higher sensitivity (93.8% for SCC and 90.0% for ADC) with 100% specificity when all 4 candidates were combined. Further, we applied the tissue-based model to serum-based samples, offering a non-invasive method for populations reluctant to undergo cervical sampling. The non-invasive method used in another independent validation set demonstrated a high diagnostic value for cervical cancer patients with 84.0% sensitivity and 96.7% specificity, suggesting the potential clinical utility of the biomarker panel for the diagnosing major subtypes of cervical cancer. Moreover, the combined panel showed impressive performance in distinguishing HSIL and early-stage cervical cancer patients, highlighting its early detection potential for cervical cancer.

In conclusion, our study identified mRNA and lncRNA signatures in cervical SCC, ADC, and normal cervix tissues, proposing a novel SMC1B/CELSR3/FEZF1-AS1/LINC01305 panel as a potential non-invasive biomarker for both SCC and ADC subtypes of cervical cancer. These findings hold promise for enhancing screening efficiency, accuracy and coverage in cervical cancer detection.

SUPPLEMENTARY MATERIALS

Table S1

Patients’ information in the exploration, training, and validation phase

Table S2

Patients’ information in the serum-based validation set

Table S3

Sequences of primers

Table S4

Differentially expressed mRNA and lncRNA candidates

Table S5

AUC of mRNA and lncRNA candidates in training set

Table S6

Diagnostic parameters of each candidate and the combined panel in validation set

Fig. S1

Receiver operating characteristic analysis of the individual mRNA candidates (A) and lncRNA candidates (B) with low significance of discrimination (AUC <0.7) in SCC or ADC patients in 45 normal, 45 SCC and 45 ADC tissue samples.

Fig. S2

Receiver operating characteristic analysis of the individual mRNA candidates (A) and lncRNA candidates (B) in SCC or ADC patients in the training set.