Performance of an ancillary test for cervical cancer that measures miRNAs and cytokines in serum and cervical mucus.

Takuma Fujii: Conceptualization; funding acquisition; project administration; resources; supervision; writing – original draft. Eiji Nishio: Resources. Tetsuya Tsukamoto: Methodology. Iwao Kukimoto: Methodology. Aya Iwata: Data curation; formal analysis; validation; visualization; writing – review and editing.

Approval of the research protocol by an Institutional Reviewer Board: The study protocol was approved by the Ethics Committees of Fujita Health University (HM22‐516) and the National Institute of Infectious Diseases. Informed Consent: Written informed consent was obtained from each patient. All procedures were performed in accordance with the approved guidelines and regulations. Registry and the Registration No. of the study/trial: N/A. Animal Studies: N/A.

This work was partly supported by KAKENHI from the Ministry of Education, Culture, Sports, Science and Technology, Japan (Grant No. 23K08812) and a Fujita Health University Research Grant‐in‐Aid.

The identification of aberrant miRNA expression in cervical cancer in contrast with normal was achieved using miRNA microarray of total RNA extracted from serum or cervical mucus. Amongst 2588 miRNAs tested, three miRNAs (miR‐16‐5p, −223‐3p, and ‐451a) from serum were commonly downregulated in Group A and Group B (Table  S5 ), whereas five miRNAs (miR‐20b‐5p, ‐155‐5p, ‐144‐3p, ‐126‐3p, and ‐451a) from mucus were upregulated (Table  1 ). Association between microarray data and real‐time RT‐PCR in serum and the cervical mucus. Note : Comparison of fold change in expression of selected genes by microarray and real‐time PCR. Group A and B independently selected groups of patients to check the reproducibility of the microarray. Cancer: AD and SCC. Mann–Whitney U : Mann–Whitney U ‐tests with a Bonferroni correction. Abbreviations: AD, adenocarcinoma; CIN, cervical intraepithelial neoplasia; SCC, squamous cell carcinoma. p  < 0.05 was statistically significant. We analyzed paired serum and cervical mucus samples from 455 patients using real‐time RT‐PCR. Scatter plots of the fold changes in miRNA levels for each sample relative to the mean ΔC t value for normal patients are shown in Figure  2 , and the median values of fold change are shown in Table  1 . In the serum, three miRNAs (miR‐16‐5p, ‐223‐3p, and ‐451a) adjusted by an external control (cel‐miR‐39) were significantly downregulated in each disease category. Of note, the level of miR‐451a was downregulated regardless of disease severity. miRNA levels measured by real‐time RT‐PCR correlate with disease severity. miRNAs levels in (A) serum and (B) cervical mucus were analyzed by real‐time RT‐PCR. The paired serum and cervical mucus samples were collected from individual patients. Fold changes were determined relative to the average of the Δ C t value for normal. Box plots represent the median, and 25th and 75th percentiles. “ x ” indicates the mean value. Significant differences were observed between normal and cancer determined using the Mann–Whitney U ‐test with Bonferroni correction. * p  < 0.05 versus normal, † p  < 0.05 versus CIN2, 3. In the mucus samples, five miRNAs (miR‐144‐3p, ‐451a, ‐155‐5p, ‐20b‐5p, and ‐126‐3p) adjusted by internal controls were significantly upregulated compared with normal in each category. Of note, the levels of miR‐451a as well as miR‐144‐3p were significantly upregulated for all CIN grades and cancer. To select candidate cytokines, we used 41 serum samples and determined the expression levels of 18 cytokines within detectable limits (%) (Table  S3 ). Six cytokines, MIP‐1α, G‐CSF, IL‐8, MCP‐1, Eotaxin and IL‐6, were detected in more than 30% of samples and were analyzed in 455 samples. Levels of MIP‐1α between the normal and cervical disease groups were similar; therefore, it was omitted from further study. The levels of Eotaxin, MCP‐1, IL‐8, and IL‐6 were significantly upregulated between CIN2/3 and cancer (Table  2 , Figure  3A ). In contrast, the level of G‐CSF was not significantly downregulated. Association cytokine profiles with histology in 455 samples. Note : Cancer: AD and SCC, Mann–Whitney U : Mann–Whitney U ‐ tests with a Bonferroni correction. Abbreviations: CIN, cervical intraepithelial neoplasia. p  < 0.05 was statistically significant. Comparison of cytokine levels with the severity of cervical disease. Scattered box plots show the distribution of the cytokine levels in the (A) serum (pg/mL) and (B) cervical mucus (ng/mL). Significantly different by Mann–Whitney U ‐test with Bonferroni correction. * p  < 0.05 versus normal, † p  < 0.05 versus CIN2, 3. In the mucus, the levels of MCP‐1, IL‐8, IL‐6, Eotaxin, RANTES, and IFN‐γ were significantly upregulated between CIN2/3 and cancer (Table  2 , Figure  3B ). Of note, G‐CSF was significantly decreased in the order of normal, CIN2/3, and cancer. The diagnostic performance of the miRNAs and cytokines in serum and cervical mucus as markers for cervical cancer screening was compared between 215 cervical cancer cases and 48 normal cases from 455 samples. ROC curves were constructed, and the AUC was obtained to determine diagnostic performance (Figure  4A ). miRNAs from serum and G‐CSF from mucus were inversely correlated to disease severity; thus, values multiplied by minus 1 were plotted. The AUC showed that G‐CSF in the mucus was the most accurate marker (AUC = 0.944), followed by miR‐451a (AUC = 0.939), and miR‐144‐3p (AUC = 0.929). Which factors in the top eight marker combinations with the highest AUC (cervical mucus miRNA: miR‐155‐5p, ‐20b‐5p, ‐144‐3p, ‐451a, and ‐126‐3p; cervical mucus cytokines: Eotaxin, RANTES, and G‐CSF) should be used in the final model were determined by the AIC (Table  S4 ). For all combinations, the miRNAs from cervical mucus (miR‐20b‐5p, ‐451a, and ‐126‐3p) and cytokines from cervical mucus (Eotaxin and G‐CSF) had the lowest AIC value (50.4). Predicted probabilities from logistic regression were calculated and ROC curves were generated for the five combined molecules. The combination of five molecules had the largest AUC of 0.989 (95% confidence interval, 0.979–0.999) for discriminating between patients with cervical cancer and normal (Figure  4B ). In contrast, miRNAs or cytokines in the serum did not have good diagnostic performance. Diagnostic values of individual and combined miRNAs and cytokines as markers in serum and cervical mucus. ROC analyses were used discriminate cervical cancer. (A) AUC and (B) ROC curves. The best AUC combination consisted of five molecules, including cervical mucus miRNAs (miR‐20b‐5p, −451a, and‐126‐3p) and cervical mucus cytokines (Eotaxin and G‐CSF) as determined by AIC (details in Table  S4 ). Potential correlations between the levels of miRNAs and cytokines in serum and mucus were analyzed by Spearman's rank correlation test and shown in Figure  5 . We found that miRNAs in serum were positively correlated with each other. Similarly, all miRNAs in mucus were positively correlated with each other. Regarding cytokines in the mucus, RANTES had a strong positive correlation with miRNAs in the mucus of patients with cancer but not from the normal population. This trend was observed in patients with CIN2 or worse (Figure  S1 ). Associations between serum and mucus, and levels of cytokines and miRNAs were determined using Spearman's rank correlation for multiple comparisons. (A) All cases, (B) normal, and (C) cancer. Color and shading represent the degree of positive/negative correlations. Dark orange shading indicates strong positive correlations (correlation coefficient 0.6–1.0). Pale orange shading represents weak positive correlations (correlation coefficient 0.2–0.6). Pale blue shading represents weak negative correlations (correlation coefficient 0.2–0.6), and dark blue shading represents strong negative correlations (correlation coefficient 0.6–1.0). Correlation significance: * p  = 0.01–0.05, ** p  = 0.001–0.01, *** p  < 0.001.

Numerous previous studies have explored the development of diagnostic markers for cervical cancer and CIN based on miRNA expression levels. 11 , 18 , 29 However, there is a lack of consensus regarding the expression levels of miRNAs in serum, as different investigators have reported varying trends, including increased and decreased expressions. 20 , 30 Reis et al. reported elevated expressions of miR‐16‐5p and miR‐451a in serum, which is in contrast with our results. 31 These discrepancies have hindered the establishment of conclusive results. Furthermore, variations in sample types, such as serum versus tissue, were also reported to affect the observed expression levels, suggesting that differences in sample type and detection methods may contribute to the divergent outcome. 32 Regarding a detailed analysis of the miRNA expression levels, our current data showed that miR‐16‐5p and miR‐451a expression levels were increased in the mucus, but decreased in serum from the same patient. Decreased expressions of miRNAs, specifically miR‐218, in serum were reported in cervical cancer. 30 , 33 Moreover, miR‐223‐3p and miR‐451a were decreased in bladder cancer and liver cancer, 34 , 35 whereas miR‐16‐5p and miR‐451a were reduced in osteosarcoma but increased in esophageal cancer. 36 , 37 Interestingly, in patients with rheumatoid arthritis, miR‐223‐3p and miR‐16‐5p were decreased, 38 and in gout, miR‐223‐3p and miR‐451a were decreased. 39 The development of diagnostic markers based on the decreased expression levels of miRNAs in serum holds promise for various diseases. However, the accuracy of detection systems utilizing serum remains limited for cervical neoplasia. 40 , 41 Regarding cytokine studies, the expression levels of IL‐6 and IL‐8 were increased with cervical disease severity using tissue or serum samples. 42 , 43 , 44 , 45 , 46 Another study reported elevated TNF‐α, IL‐1β, IFN‐γ, and IL‐6 levels in the serum of patients with cervical cancer. 47 However, our findings indicated that, except for IL‐6, the expression levels of these cytokines were not significantly higher in patients with cervical cancer (Table  S3 ). Therefore, we did not analyze TNF‐α, IL‐1β, or IFN‐γ levels in 455 samples. Furthermore, the expression levels at local sites were reported to be higher than those in the serum. 48 This is consistent with our findings showing that the local expression level was approximately 100–1000‐fold higher than that in the serum. MCP‐1 and RANTES secreted by immune cells in patients with cervical cancer are involved in the immune attack on cancer cells. 49 Therefore, it is reasonable that they can be detected at local sites. Although MCP‐1 was measured in the serum, 50 , 51 its usefulness as a diagnostic marker is not anticipated due to its low sensitivity. The expression level of G‐CSF in the serum was reported to increase with disease severity, 52 which contradicts our findings using our serum samples. We cannot explain this discrepancy. Notably, the level of G‐CSF was significantly decreased at local sites in cancer patients compared with healthy individuals, indicating it is a useful biomarker, even when used alone. Previous studies have reported associations between miRNAs and cytokines. During influenza infection, IL‐6 increased the expression level of miR‐451a. 53 In endometriosis, IL‐6 and miR‐451a were upregulated in the peritoneal fluid. 54 In breast cancer, miR‐144‐3p and miR‐20b‐5p downregulated the expression level of G‐CSF via mTOR, whereas miR‐155‐5p did so via HIF1/2, and miR‐144‐3p, miR‐20b‐5p, and miR‐155‐5p downregulated G‐CSF via ERK2. 55 Furthermore, in immune cells, miR‐126‐3p, miR‐144‐3p, and miR‐20b‐5p upregulated the expression level of RANTES by controlling Runt‐related transcription factor (RUNX) 1 and RUNX3, and miR‐144‐3p, miR‐20b‐5p, and miR‐155‐5p controlled Core‐binding factor (CBF)‐β, which led to an increase in RANTES expression. In a study of breast cancer, patients with high miR‐155‐5p expression had high RANTES expression. 56 These data are consistent with the results obtained in our study. We propose a regulation cascade of RANTES gene expression with miRNAs in cervical neoplasia (Figure  S2 ). Briefly, increased expressions of miR‐144‐3p, miR‐20b‐5p, and miR‐155‐5p led to the regulation of CBF‐β, whereas an increase in the expressions of miR‐126‐3p, miR‐144‐3p, and miR‐20b‐5p resulted in the regulation of RUNX. As a consequence, the expression of RANTES was upregulated (Figure  S2 ). The aim of our study was to determine the potential usefulness of these molecules as an ancillary screening test for cervical cancer. Our findings revealed that an increase in the miRNA expression levels in mucus had higher diagnostic accuracy compared with those in the serum. Similarly, we observed comparable results with the expression levels of cytokines. Notably, the combination of decreased G‐CSF expression levels in mucus, along with the high expression of miRNAs, had the highest diagnostic accuracy. Whereas previous reports highlighted the utility of changes in miRNA and cytokine expression levels in serum as screening markers, our results suggest that utilizing changes in local expression levels compared with serum levels may be superior. One weakness of this study was its inability to explain the contradictory results observed in the expression levels of miR‐451a and miR‐16‐5p in the mucus and serum. It is intriguing to note the phenomenon of opposite expression levels within the same patient, and further analysis is eagerly anticipated to shed light on this mechanism in the future. In addition, we only analyzed limited numbers of cytokines and it would be valuable to explore combinations of other cytokines. For samples collected locally, inflammation may be induced by factors such as patient age, menstrual status, and microbial infections including chlamydia and viruses, potentially leading to different outcomes. In the future, aiming for clinical application, it is important to confirm the reproducibility of experimental results by conducting analyses at research institutions different from ours, and using different cohort populations and different laboratories to confirm the validity of the study, such as whether similar results can be obtained. In the process of developing future screening methods, if self‐collection methods are to be incorporated into screenings, research that takes this into account will also be necessary. We investigated the target molecules of the miRNAs we extracted using methods such as TargetScan. For instance, we found that miR‐126‐3p was associated with the PI3K‐Akt pathway, which is relevant to cervical cancer. We previously conducted an analysis of its biological significance. 57 Interestingly, we observed that the forced expression of miR‐126‐3p in cultured cells suppressed their proliferation. In the current study, increased expression levels of miR‐126‐3p were observed at stages from premalignant lesions to cancer, suggesting that its elevated expression level might suppress the progression of the lesions toward cancer. More functional studies are needed to understand the roles of these miRNAs and cytokines in cervical cancer and to identify potential therapeutic targets. As shown in Figure  5 , miR‐20b‐5p is correlated with G‐CSF and RANTES, but it is not a target molecule of miRNA. HPV tests and P16 immunostaining are used for screening and triage tests for cervical cancer. 6 Regarding HPV tests in the current study, we categorized the study population into HPV‐negative and HPV‐positive cases for the analysis. This can be considered a limitation because it does not represent the general population. Therefore, comparing our results with the HPV tests currently in general use, as the study population, should be performed in the future. Furthermore, our findings should be compared with p16 immunostaining. One of the advantages of our approach in this study was that we selected the optimal combination of biomarkers using previously reported methods, specifically the AUC and AIC, without the use of artificial intelligence or machine learning. One issue with machine learning is the lack of transparency in the calculation process. We considered it unnecessary to perform an artificial intelligence‐based analysis because the candidate markers were not chosen to be complex or competitive with each other. In conclusion, this paper reports for the first time that the analysis method using mucus samples could distinguish cervical tumors from normal tissues with higher accuracy compared with serum samples, when analyzing the expression levels of miRNA and cytokines in serum and mucus collected from the same patient. This method suggests potential usefulness as a third new molecular marker in addition to existing HPV testing and cytology‐based screening methods for cervical cancer screening. However, for clinical implementation consideration, further analysis using larger cohorts is needed in the future.

Globally, approximately 500,000 women are diagnosed annually with cervical cancer and the number of patients diagnosed with high‐grade cervical intraepithelial neoplasia (CIN), precursor lesions of the cervix, was estimated to be 20 times higher. 1 In England, the number of cervical cancer cases was 2700, and over 60,000 women are treated for CIN annually. 2 Therefore, it is necessary to develop a screening system that can efficiently detect high‐grade intraepithelial lesions in patients who would be eligible for treatment, as well as invasive cervical cancer. Even though organized cytology screening has helped reduce cervical cancer rates in developed countries, cytology has a low sensitivity for the detection of CIN. 3 Recently, human papillomavirus (HPV) tests were added to the screening system because they are highly sensitive compared with cytology. However, their specificity is poorer than that of cytology because HPV infections do not always lead to cervical lesions because they are transient. Thus, there is a need to identify more specific biomarkers. 4 , 5 , 6 MicroRNAs (miRNAs), non‐coding RNAs of 19–25 nucleotides, partially pair with 3′ untranslated regions of target mRNAs to modulate gene expression. 7 Aberrant miRNA expression was reported in cervical cancer and lesions that might be precursors to cervical cancer. 8 , 9 , 10 , 11 The ability to determine miRNA profiles might be a useful ancillary test for the screening of cervical cancer. 11 , 12 , 13 , 14 Our group identified miRNAs in cervical mucus that might be potential diagnostic markers for cervical neoplasia. 15 , 16 Regarding the samples used for screening, the use of routinely‐obtained serum for screening tests is simple and the benefit of serum miRNA‐based biomarkers in cancer and other diseases has been reported. 17 Other studies reported the use of diagnostic markers for cervical cancer and lesions that might be precursors to cervical cancer based on miRNA expressions in serum. 18 , 19 , 20 Several cytokines, molecules involved in immunity, are expressed in the cervical mucus. 21 We reported that the expression levels of cytokines varied with the progression of CIN. 21 , 22 Thus, it is important to examine whether they are useful as diagnostic markers. Finding a good correlation between local and serum cytokine expression levels will aid the development of serum‐based diagnostic markers. In this study, we utilized serum and mucus samples from the same patient to measure the levels of miRNAs and cytokines, respectively. The results were analyzed to determine which molecules and expression levels, obtained from serum or mucus samples, were the most promising screening markers.

The authors declare no conflict of interest.

This study compared the miRNA and cytokine profiles of paired serum and cervical mucus samples from patients with cervical cancer and precursors to cervical cancer. Demonstrating the benefit of a biomarker is dependent upon the disease category of patients and normal controls who are HPV negative. Patients underwent routine gynecological examinations at Fujita Health University Hospital, Aichi, Japan, from October 2014 to August 2022. The disease category was determined on the basis of HPV genotype, histology, and cytology, of individual patients. Exclusion criteria included age <20 years, pregnant, previously treated with radiation, chemotherapy, or had undergone surgery for any cancer or CIN. miRNAs collected from cervical mucus were obtained with a 1‐cm diameter cotton swab. These were then stored in a freezer at −80°C. Ectocervical and endocervical cells for the analysis of HPV genotypes were collected using a cervical brush that was placed into the cervical canal and stored in a freezer at −80°C. Cervical mucus was collected for cytokine analysis by Merocel cervical sponges (Medtronic Xomed, Inc., Jacksonville, FL, USA) and stored in a freezer at −80°C. Samples of serum and cervical mucus, tissues for histology, and exfoliated cells for cytology and HPV genotyping were collected simultaneously but paired samples were collected at different times within 2 months in some cases. Serum and cervical mucus pairs were collected from individual patients. The candidate miRNAs in sera commonly extracted from the experimental results using serum from 63 samples (Group A) and 33 (Group B) were used in subsequent experiments. The microarray analysis of 86 cervical mucus samples identified candidate targets and reference miRNAs (Figure  1 ). Study design of the population and miRNA/cytokine candidates. We initially screened a small number of samples to identify candidate miRNAs and cytokines, and subsequently measured miRNA and cytokine levels in a population of N  = 455. *Twenty‐four of the cervical mucus samples in N  = 455 were collected incorrectly, and therefore 431 samples were analyzed for cytokine measurement. We also evaluated 18 commercially available cytokines in 41 serum samples and 201 mucus samples to select candidate cytokines. Disease details of the samples used for candidate miRNA/cytokine selection are shown in Table  S1 . Finally, cervical mucus and serum samples from 455 patients were examined for expression of selected candidate miRNAs and cytokines (Table  S2 ). The miRNAs identified as candidates through microarray analysis were confirmed for expression using real‐time reverse transcription polymerase chain reaction (RT‐PCR). Ethically, it is difficult to obtain a biopsy from healthy volunteers. Therefore, “normal” was defined as patients who were HPV negative based on an analysis of cervical exfoliated cells ( N  = 48). The histology of biopsy samples was classified as CIN1 ( N  = 9), CIN2 ( N  = 75), CIN3 ( N  = 108), and cancer ( N  = 215). Three cases that were atypical squamous cells (ASC) and atypical glandular cells (AGC) for cytology, but were diagnosed as chronic cervicitis using colposcopically directed biopsy, were included in this study as CIN1. With reference to the International Federation of Gynecology and Obstetrics 2018 clinical staging criteria, four (1.9%) patients with cervical cancer were stage 1a and 209 (98.1%) were stage 1b or higher (Table  S2 ). Consecutive patients had a median age of 43 years (range 21–94) in 455 patients. PCR with PGMY primers was used for HPV genotyping, followed by reverse line blot hybridization, 23 which detects 31 HPV genotypes including HPV 6, 11, 16, 18, 26, 31, 33, 34, 35, 39, 40, 42, 44, 45, 51, 52, 53, 54, 55, 56, 57, 58, 59, 66, 68, 69, 70, 73, 82, 83, and 84. Total RNA was extracted from cotton swabs using an miRNeasy Mini Kit (Qiagen GmbH, Hilden, Germany). Briefly, each cotton swab was soaked in 1000 μL of QIAzol Lysis Reagent. The median recovery yield of total RNA was 12.1 μg (range, 1.6–84.5) from each patient. The normalization of serum RNA extraction efficiency was achieved using synthesized cel‐miR‐39 as a synthetic spike‐in control RNA oligonucleotide because it lacks a mammalian homolog. Ten μL of 0.1 fmol/μL cel‐miR‐39 was added to 200 μL of each serum sample, followed by the extraction of total RNA from 200 μL of serum using an miRNeasy Mini Kit. The final total RNA elute was 30 μL. RNA samples were stored at −80°C until further processing. RNA was extracted from pooled serum using 3D‐Gene RNA extraction reagent from liquid samples (Toray, Kamakura, Japan) as per the manufacturer's instructions. Then, 2 μg extracted pooled total RNA from the cervical mucus or total RNA from serum were prepared, and a 3D‐Gene miRNA labeling kit (Toray) was used to label 250 ng followed by hybridization onto 3D‐Gene Human miRNA Oligo chips (Toray). miRbase miRNA database Release 21 (miRbase; https://www.mirbase.org ) was used to confirm oligonucleotide sequences of the probes as well as their annotations. After stringent washing, a 3D‐Gene Scanner (Toray) was used to scan the fluorescent signals, which were analyzed with 3D‐Gene Extraction software (Toray). Normalization of raw data from individual spots was achieved by substituting the mean background intensity assessed using signal intensities of all blank spots and 95% confidence intervals. Spots with a signal intensity >2 standard deviations of the background signal intensity were considered valid. Relative miRNA levels were determined by the comparison of signal intensities between valid spots in microarray experiments. Normalized data per array underwent global normalization whereby the median signal intensity was set to 25. The value for each gene was normalized, resulting in a median disease category:normal ratio of 1. The verification of candidate miRNAs was achieved using real‐time RT‐PCR. TaqMan™ MicroRNA Assays (Thermo Fisher Scientific, Waltham, MA, USA) were used to quantify miRNAs for the cervical mucus (hsa‐miR‐7109‐5p, 466424_mat; hsa‐miR‐155‐5p, 002623; hsa‐miR‐3180, 463043_mat; hsa‐miR‐451a, 001141; hsa‐miR‐20b‐5p, 001014; hsa‐miR‐126‐3p, 002228; hsa‐miR‐144‐3p, 002676) and serum (cel‐miR‐39, 000200; miR‐223‐3p, 002295; miR‐451 a, 001141 has‐miR‐16‐5p, 000391). Reverse transcription was performed with 350 ng of cervical mucus total RNA or 9.41 μL of serum total RNA with the TaqMan™ MicroRNA Reverse Transcription Kit (Thermo Fisher Scientific) according to the User Bulletin (protocol for ‘Creating Custom RT and Pre‐amplification Pools using TaqMan MicroRNA Assays’ #4465407, Thermo Fisher Scientific). A 7900 Real‐Time PCR system (Thermo Fisher Scientific) was used to perform quantitative real‐time PCR. To individual 10‐μL PCR reactions in 384‐well plates, 0.5 μL of 20× TaqMan MicroRNA Assays with PCR primers and probes (5′‐FAM and 3′‐TAMRA), 1 μL diluted RT product, and 5 μL 2× TaqMan Fast Advanced Master Mix (Thermo Fisher Scientific) were added and mixed, then incubated at 95°C for 20 s. The following PCR conditions were 50 cycles at 95°C for 1 s, then 60°C for 20 s. Data analysis was performed with RQ Manager 1.2 (Thermo Fisher Scientific) and the fractional cycle number (the quantity of amplified target reaching a fixed threshold) was determined. The expression levels were normalized to the average signals of miR‐7109‐5p and miR‐3180 for the cervical mucus 16 and cel‐miR‐39 for serum, and presented as a Δ C t value as previously described. 24 , 25 The quantity of miRNA in each disease category relative to that for normal was calculated from the relative ratios of 2 − ΔΔ C t between the two conditions. Proteins were extracted from Merocel cervical sponges as previously reported 26 and used to analyze cytokines. Wet weights of sponges were recorded and they were placed in 2‐mL Spin‐X centrifuge filter tubes (Corning, NY, USA) containing 300 μL extraction buffer (phosphate‐buffered saline, Sigma‐Aldrich, St. Louis, MO, USA), containing 256 mM NaCl, and 100 μg/mL aprotinin (Wako, Amagasaki, Japan) were added slowly. Next, the sponges were incubated at 4°C for 2 h followed by centrifugation at 14,000 rpm for 15 min at 4°C. Next, 30 μL of fetal bovine serum was added to 270 μL of extract and vortexed briefly before being frozen at −80°C. CBA, a multiplexed bead‐based immunoassay (BD Biosciences, Franklin Lakes, NJ, USA), was used to measure cytokines, chemokines, and growth factors according to the manufacturer's protocol. Measured factors included interleukin (IL)‐1α (Cat# 560153), IL‐1β (Cat# 558279), IL‐2 (Cat# 558270), IL‐4 (Cat# 558272), IL‐6 (Cat# 558276), interferon (IFN)‐α (Cat #560379), IFN‐γ (Cat# 558269), tumor necrosis factor (TNF)‐α (Cat# 558273), granulocyte–macrophage‐colony‐stimulating factor (GM‐CSF) (Cat# 558335), granulocyte‐colony‐stimulating factor (G‐CSF) (Cat# 558326), IL‐10 (Cat# 558274), IL‐8 (Cat# 558277), IL‐17A (Cat# 560383), IL‐21 (Cat# 560358), monocyte chemoattractant protein (MCP)‐1 (Cat# 558287), macrophage inflammatory protein (MIP)‐1α (Cat# 558325), Regulated upon Activation, Normal T Cell Expressed and Presumably Secreted (RANTES; Cat# 558324), and Eotaxin (Cat# 558329). Depending on the levels of cytokines in the samples, thawed extracts of cervical mucus were diluted between 1:1 and 1:1000 using extraction buffer. However, undiluted serum was used for measurements. Then, a 10‐point standard curve (between 0 and 2500 pg/mL) was established for individual cytokines using cytokine standards provided in the kit. Samples and cytokine standards were incubated in the presence of capture beads for 1 h and then phycoerythrin‐conjugated antibodies specific for each cytokine were incubated at room temperature for 2 h. Buffers were provided in the CBA human soluble protein master buffer kit (Cat# 558265, BD Biosciences). After washing, the capture beads were analyzed on a BD FACSCalibur flow cytometer (BD Biosciences). Then, the 10‐point standard curve using FCAP Array™ software (BD version 3.0.1) was used to convert the mean fluorescence intensity of each bead cluster into cytokine concentrations. Levels of cytokines in samples of cervical mucus were adjusted using the weighted volume method as previously described. 22 , 27 To compare variances in the weight of sponges, a dilution factor was established as follows: [( x  −  y ) + 300 mg buffer]/( x  −  y ); x  = sponge weight after collection, y  = dry sponge weight. To obtain weight‐normalized values, the levels of cytokines measured were multiplied by the dilution factor. SPSS for Windows (ver. 22.0.0.0; IBM Corp, Armonk, NY, USA) was used to perform all statistical analyses. Significant trends between disease category groups were identified using the Jonckheere–Terpstra trend test, and a comparison of overall differences between disease category groups was obtained using the Kruskal–Wallis one‐way analysis of variance by ranks. Data were analyzed using two‐tailed Mann–Whitney U ‐tests with Bonferroni correction. A receiver operating characteristic (ROC) curve was generated, and the area under the ROC curve (AUC) was calculated to evaluate the diagnostic value. Then, the Akaike information criterion (AIC) was utilized for comparisons of goodness‐of‐fit for each model consisting of combined miRNAs. 28 We unified five different molecules to determine the predictive probability using logistic regression and then generated ROC curves according to the probability. We estimated Spearman's rank correlation for multiple comparisons for any associations between serum and mucus and any associations between cytokine and miRNA levels. p‐ values <0.05 were considered statistically significant.

Figure S1. Figure S2. Table S1. Table S2. Table S3. Table S4. Table S5.