Diagnostic value of the combination of circulating serum miRNAs and CA125 in endometriosis

Abstract

Circulating microRNAs (miRNAs) can serve as noninvasive biomarkers for endometriosis, but their diagnostic and prognostic values require investigation. This study evaluated the potential of 6 miRNAs in diagnosing endometriosis and predicting fertility. The study included patients with endometriosis (stages I–IV) and controls admitted to Sun Yat-Sen Memorial Hospital between May 2013 and March 2014. The serum expression levels of 6 miRNAs (miR-199a, miR-145*, miR-122, miR-9*, miR-141*, and miR-542-3p) were determined using qRT-PCR. Receiver operating characteristics curves were used to determine the diagnostic accuracy. The study included 155 patients with endometriosis and 77 controls. The model combining miR-199a, miR-122, miR-145*, and miR-141* with the carbohydrate antigen 125 (CA125) exhibited 81.8% sensitivity and 92.6% specificity and an area under the curve of 0.939 for diagnosing endometriosis. When combining miR-199a, miR-122, miR-145*, miR-542-3p, and CA125, the receiver operating characteristics curve showed an area under the curve of 0.759 and 79.6% sensitivity and 73.5% specificity for stage I/II versus III/IV endometriosis. Circulating miRNA levels were associated with pelvic adhesions (miR-199a, P < .05), lesion distribution (miR-9*, miR-145*, and miR-141*; all P < .05), and the presence of deep infiltrating endometriosis (miR-199a and miR-122; both P < .001). The expression levels of miR-199a, miR-122, and miR-542-3p decreased with an increasing endometriosis fertility index. The model combining circulating miRNAs (miR-199a, miR-122, miR-145*, and miR-141*) and CA125 is promising for diagnosing endometriosis and its severity. miR-199a, miR-122, and miR-542-3p were associated with the endometriosis fertility index and might be used to predict fertility preoperatively, but these results require confirmation.

1. Introduction

Endometriosis is a highly dynamic disease affecting 6% to 10% of reproductive-age women and characterized by the presence and growth of endometrium-like tissue located outside of the uterine cavity, resulting in dysmenorrhea, deep dyspareunia, intestinal disorders, and infertility.^[1–3] The disease is classified into 3 types: ovarian endometriosis, peritoneal endometriosis, and deep infiltrating endometriosis.^[4,5]

The gold standard for diagnosing endometriosis is diagnostic laparoscopy with histological confirmation, but it is invasive and associated with risks.^[1,6] As a result, there is a delay of approximately 6.7 years between the onset of symptoms and diagnosis.^[7] Thus, serum biomarkers with high sensitivity and specificity are needed for diagnosing endometriosis. Serum carbohydrate antigen 125 (CA125) can be used for diagnosing endometriosis, but it has low sensitivity, with 50% for stage III/IV endometriosis and even lower for stage I/II disease.^[8]

microRNAs (miRNAs) are small non-protein-coding RNAs (19–25 nucleotides in length), are stable in serum, and can be useful as noninvasive biomarkers for various diseases,^[9–11] including endometriosis.^[12,13] Six circulating miRNAs (miR-199a, miR-145*, miR-122, miR-9*, miR-141*, and miR-542-3p) were identified as potential biomarkers for endometriosis, with combined sensitivity and specificity of 93.2% and 96.0%, respectively.^[14]

The endometriosis fertility index (EFI) is a reliable clinical tool for predicting pregnancy rates in patients with endometriosis after surgery,^[15,16] but its ability to predict fertility preoperatively is limited. There are no data on the relationship between diagnostic biomarkers and the EFI. Given that infertility is one of the main complaints among endometriosis patients,^[1–3] whether abnormal levels of serum miRNAs could be used to predict fertility in patients with endometriosis preoperatively has to be examined.

Therefore, this study aimed to evaluate the potential of 6 miRNAs (miR-199a, miR-145*, miR-122, miR-9*, miR-141*, and miR-542-3p) in diagnosing endometriosis and predicting fertility in patients with endometriosis. The results could help improve the management of women with endometriosis.

2. Materials and methods

2.1. Study design and patients

In this study, 155 patients with endometriosis (stages I–IV) and 77 controls were enrolled from the Department of Obstetrics and Gynecology of Sun Yat-Sen Memorial Hospital between May 2013 and March 2014. All participants were nonsmokers, and none had received estrogen/progestin medications for at least 3 months before sampling. The patients with malignant or benign ovarian cysts (except endometrioma), metabolic or endocrine disease, evidence of severe inflammation, or undergoing surgical treatment/exploration were excluded. Moreover, each patient with endometriosis had a final histologically confirmed diagnosis. The endometriosis stage was defined according to the Revised American Fertility Society classification system.^[17] The controls included 77 patients confirmed with the absence of endometriosis by laparoscopy surgery; the primary diagnosis of these patients was infertility caused by tubal factors. An EFI score of 0 to 3 predicts the poorest cumulative prognosis for pregnancy, while a score of 6 to 10 suggests a better fertility prognosis. Therefore, the EFI score was divided into 0 to 3, 4 to 6, and 7 to 10.

The study was performed according to the protocols approved by the ethics committee of Sun Yat-Sen Memorial Hospital (Guangzhou, China). Written informed consent was obtained from all participants before this study.

2.2. Serum collection and RNA isolation

Venous blood samples (4 mL) were collected from all patients 1 to 3 days before surgery and left at room temperature for 1 hour before being centrifuged at 3000 rpm at 4ºC for 10 minutes. The serum samples were extracted and stored at −80ºC. Total RNA was extracted from approximately 1 mL of serum according to the kit instructions (mirVana PARIS kit, Ambion, Austin, TX, USA). All RNA samples were stored at −80ºC. The RNA concentration was detected using a NanoDrop1000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA).

2.3. Real-time PCR

cDNA was synthesized from 50 ng of total RNA using commercially available miR-specific primers (for miR-542-3p, miR-9*, miR-145*, and miR-141*, miR-199a, and miR-122) and endogenous control-specific (U6) reverse primers according to the ReverTra Ace qPCR RT Kit (Toyobo, Osaka, Japan). Real-time PCR was performed using SYBR Premix Ex Taq II-based (Takara, Kyoto, Japan) quantitative real-time PCR with Stepone plus (Applied Biosystems, Foster City, CA, USA). The expression of the 6 miRNAs was normalized to the levels of U6^[18,19] using the 2^−ΔΔCt method.^[20]

2.4. Statistical analysis

The continuous variables were tested for normal distribution using the Kolmogorov-Smirnov test. After confirmation of the normal distribution, the continuous variables were presented as means ± standard deviations and tested using Student t test (comparison of 2 groups) or one-way analysis of variance with the least significant difference post hoc test (comparison of 3 or more groups). Categorical variables were presented as n (%) and tested using the chi-square test. The diagnostic efficacy was evaluated by multivariable logistic regression models and receiver operating characteristic curves. The area under the curve (AUC) was calculated. Youden index was used to check the optimal cutoff point. All tests were 2-sided, and P values < .05 were considered statistically significant. All statistical analyses were performed using SPSS PASW Statistics 25.0 (IBM, Armonk, NY, USA). All figures were generated using Prism 9.0 (GraphPad Software Inc., San Diego, CA, USA).

3. Results

3.1. Characteristics of the patients

Table 1 presents the demographics and clinical features of the patients. The endometriosis group was a little older than the controls (32.9 ± 7.7 vs 29.5 ± 4.8 years, P < .001), and the CA125 levels were higher (89.9 ± 21.5 vs 21.1 ± 5.4 ng/mL, P < .001).

Table 1.

Characteristics of the patients.

	Endometriosis (n = 155)	Control (n = 77)	P
Age (yr)	32.9 ± 7.7	29.5 ± 4.8	<.001
Body mass index (kg/m2)	20.9 ± 2.9	20.9 ± 2.7	>.999
CA125 (ng/mL)	89.87 ± 21.48	21.14 ± 5.44	<.001
Infertility	60 (38.7)	69 (89.6)	<.001
Primary	31	34
Secondary	29	35
Duration of infertility (yr)	4.0 ± 3.7	3.5 ± 2.1	.347
Stage of the menstrual cycle			–
Follicular phase	122 (78.7)	67 (87.0)
Luteal phase	33 (21.3)	10 (23.0)
Pelvic adhesion caused by endometriosis	65 (41.9)	NA	–
Stages of endometriosis			–
Mild (stage I/II)	48 (31.0)	NA
Severe (stage III/IV)	107 (69.0)	NA
Distribution of endometriosis			–
Ovarian endometrioma	81 (52.3)	NA
Peritoneal lesion	41 (26.5)	NA
Combined ovarian and peritoneal	33 (21.3)	NA
Deep infiltrative endometriosis	29 (18.7)	NA

CA125 = carbohydrate antigen 125.

3.2. Assessment of circulating miRNAs for the early diagnosis of endometriosis

miR-199a, miR-122, miR-145*, miR-141*, miR-542-3p, and miR-9* showed significantly differential expression levels in endometriosis versus controls (Fig. 1A–F, all P < .05). miR-122, miR-9*, and miR-145* had a significant diagnostic performance. As shown in Table 2, the AUC was 0.715 (95%CI: 0.614–0.816, P < .001) for miR-122, 0.718 (95%CI: 0.627–0.810) for miR-9*, and 0.762 (95%CI: 0.675–0.849) for miR-145*. The sensitivity and specificity at the best cutoff points for each miRNA were 68.4% and 63.6% for miR-122, 91.7% and 49.3% for miR-9*, and 93.8% and 50.8% for miR-145* (Table 2, Fig. 2).

Figure 1.

Box plots showing the miRNA expression levels in serum from endometriosis patients (n = 155) and negative controls (n = 77). The expression levels of miR-199a (A) and miR-122 (B) were upregulated in patients with endometriosis (both P < .001), whereas the expression of miR-145* (C), miR-141* (D), miR-9* (E) and miR-542-3p (F) were downregulated in patients with endometriosis (all P < .05).

Table 2.

ROC analysis for the prediction of endometriosis.

miRNAs	Disease stage	P value	AUC	Sensitivity (%)	Specificity (%)	95%CI
miR-199a	Ctrl versus All	<.001	0.621	54.55	98.61	0.550–0.705
miR-199a	Ctrl versus III, IV	<.001	0.698	67.29	94.29	0.615–0.782
miR-122	Ctrl versus All	<.001	0.756	45.75	97.40	0.694–0.817
miR-122	Ctrl versus I, II	<.001	0.715	68.42	63.64	0.614–0.816
miR-122	Ctrl versus III, IV	<.001	0.758	46.15	98.70	0.693–0.824
miR-122	I, II versus III, IV	.042	0.611	40.17	91.89	0.520–0.703
miR-9*	Ctrl versus All	.017	0.603	77.08	41.54	0.517–0.690
miR-9*	Ctrl versus I, II	<.001	0.718	91.67	49.25	0.627–0.810
miR-145*	Ctrl versus All	.01	0.612	85.33	38.10	0.524–0.699
miR-145*	Ctrl versus I, II	<.001	0.762	93.75	50.79	0.675–0.849
miR-141*	Ctrl versus All	.041	0.588	81.05	37.5	0.505–0.671
miR-542-3p	Ctrl versus All	.018	0.603	82.88	36.51	0.520–0.687
miR-199a, miR-122, miR-145*	Ctrl versus All	<.001	0.869	77.37	83.33	0.807–0.929
miR-199a, miR-122, miR-145*	Ctrl versus I, II	<.001	0.836	79.41	75.93	0.751–0.921
miR-199a, miR-122, miR-145*	Ctrl versus III, IV	<.001	0.879	79.61	83.33	0.818–0.941
miR-199a, miR-122, miR-145*, CA125	Ctrl versus All	<.001	0.939	81.75	92.59	0.898–0.980
miR-199a, miR-122, miR-145*, CA125	Ctrl versus I, II	<.001	0.753	70.59	79.63	0.646–0.860
miR-199a, miR-122, miR-145*, CA125	Ctrl versus III, IV	<.001	0.937	97.09	79.63	0.893–0.981
miR-199a, miR-122, miR-145*, CA125	I, II versus III, IV	<.001	0.759	79.61	73.53	0.652–0.866

miRNAs = microRNAs, ROC = receiver operating characteristics.

Figure 2.

ROC analysis for the prediction of endometriosis. The AUC of single microRNA (A) miR-199a and (B) for diagnosing endomtriosis. The AUC of optimal combination (C) miR-199a, miR-122 and miR-145* and (D) miR-199a, miR-122, miR-145* and CA125 for diagnosing endomtriosis. AUC = area under the curve, CA125 = carbohydrate antigen 125, ROC = receiver operating characteristics.

An optimal model was built using the combination of 4 miRNAs (miR-199a, miR-122, miR-145*, and miR-542-3p) to predict stage I/II endometriosis from the controls, which achieved AUC = 0.836 (95%CI: 0.751–0.921) with 79.4% sensitivity and 75.9% specificity. These values were significantly higher than CA125, which only exhibited 28% to 50% sensitivity for diagnosing early endometriosis.

3.3. Combination of circulating miRNAs and CA125 for endometriosis and its severity

miR-199a showed the best sensitivity and specificity, with an AUC of 0.687 (95%CI: 0.595–0.779) and 64.1% sensitivity and 94.1% specificity for distinguishing stage I/II from stage III/IV endometriosis (Table 2). miR-122 has the potential to distinguish mild and severe endometriosis, with AUC = 0.611 (95%CI: 0.520–0.703) and 40.2% sensitivity and 91.9% specificity. The optimal combination of miR-199a, miR-122, miR-145*, and miR-542-3p achieved AUC = 0.693 (95%CI: 0.583–0.804, P < .01) and 62.4% sensitivity and 67.9% specificity (Table 2).

The model combining miR-199a, miR-122, miR-145*, and miR-141* with the CA125 exhibited 81.8% sensitivity and 92.6% specificity and AUC = 0.939 for diagnosing endometriosis (Table 2, Fig. 2). In addition, when combining miR-199a, miR-122, miR-145*, miR-542-3p, and CA125, the receiver operating characteristics curve showed AUC = 0.759 (95%CI: 0.652–0.866) and 79.6% sensitivity and 73.5% specificity for stage I/II versus III/IV endometriosis.

3.4. miRNA signature associated with endometriosis-associated pelvic adhesion and endometriosis distribution

miR-144* levels were higher in patients with pelvic adhesions than in non-pelvic adhesion patients (P < .05) (Fig. 3). Moreover, circulating miR-145*, miR-141*, and miR-9* were upregulated in endometriosis with ovarian involvement compared with patients without ovarian involvement (all P < .05). Among these 3 miRNAs, miR-9* showed the highest sensitivity and specificity (P < .01, AUC of 0.670, sensitivity of 66.7%, and specificity of 64.6%). When comparing deep infiltrating endometriosis (DIE) (n = 29) to non-DIE endometriosis (n = 126), the 6 miRNAs were upregulated, of which miR-199a and miR-122 were the most significantly dysregulated (P < .001). A model with the combination of the 6 miRNAs was constructed using logistic regression, achieving an AUC of 0.739 and 71.4% sensitivity and 68.6% specificity for discriminating DIE from non-DIE patients (Table 3).

Figure 3.

Box plots showing the miRNA expression levels in the sera of patients with endometriosis in different lesion distributions and pelvic adhesion. The expression levels of miR-199a (A) and miR-141* (A) were higher in patients with pelvic adhesion versus no pelvic adhesion (P < .05), whereas the expression levels of miR-145* (B), miR-141* (B) and miR-9* (B) were lower in patients with ovarian involvement versus endometriosis without ovarian involvement (P < .05). All 6 miRNAs (miR-199a, miR-122, miR-9*, miR-145*, miR-141* and miR-542-3p) were upregulated in deep-infiltrative endometriosis (DIE) versus non-DIE patients (C). miRNAs = microRNAs.

Table 3.

Diagnostic value of serum miRNA in endometriosis patients with different clinic features.

Clinical features	Biomarkers	P value	AUC	Sensitivity (%)	Specificity (%)	95%CI
Lesion distribution
	miR-9*	.002	0.670	66.67	64.55	0.582–0.758
	miR-145*	.018	0.629	55.26	66.37	0.534–0.724
	miR-141*	.027	0.618	75	49.09	0.522–0.715
Deep infiltrative endometriosis
	miR-199a	.002	0.690	53.57	80.16	0.581–0.799
	miR-122	.002	0.683	51.72	82.05	0.567–0.798
	miR-9*	.043	0.622	44.83	82.05	0.497–0.746
	miR-145*	.003	0.683	70.37	58.06	0.5790.787
	miR-141*	.007	0.666	51.85	74.60	0.560–0.772
	miR-542-3p	.014	0.651	70.37	55.65	0.547–0.755
	miR-199a, miR-122, miR-9*, miR-145*, miR-141*, miR542-3p	<.001	0.739	71.43	68.55	0.638–0.839

miRNAs = microRNA.

3.5. Circulating miRNAs and CA125 to predict the fertility of patients with endometriosis

The scores of most patients ranged from 4 to 9 (84.5%, 131/155) (Fig. 4A). The median EFI score was 7. The EFI was associated with the Revised American Fertility Society score (Fig. 4B). In the analysis of the relationships between the serum miRNAs and EFI, the expression of the 6 circulating miRNAs showed a progressive change along with the EFI scores (Fig. 4D–F). miR-122 serum levels increased with decreasing EFI. CA125 serum levels (Fig. 4C) were associated with the EFI.

Figure 4.

Histograms showing the distribution of the patients according to their endometriosis fertility index (EFI) (A). The EFI was divided into the following 3 parts: 1 to 3, 4 to 6, and 7 to 10, according to the prognosis of infertility. The revised American Fertility Society (r-AFS) score (B) and the expression levels of CA125 (C), miR-199a (D), miR-122 (E), and miR-542-3p (F) progressively changed along with EFI. CA125 = carbohydrate antigen 125.

4. Discussion

Previous studies reported that miRNA expression is aberrant in patients with endometriosis, but most studies focused on the expression of miRNAs in ectopic and eutopic endometrium tissues.^[21,22] Although tissue miRNAs can provide an accurate diagnosis, their application is limited because collecting tissue samples requires an invasive surgical procedure. Therefore, developing a noninvasive method for diagnosing endometriosis is important and necessary. Once released into the blood, circulating miRNAs can persist in circulation by residing in microvesicles or other lipoprotein complexes.^[23] A previous study by the authors showed that 6 circulating miRNAs (miR-199a, miR-145*, miR-122, miR-9*, miR-141*, and miR-542-3p) had the potential to diagnose endometriosis,^[14] but that original study had a small sample size, limiting the observation of the diagnostic efficacy and association with the clinical features of endometriosis. In this study, the diagnostic efficacy of the 6 miRNAs was investigated in a larger sample size. With the application of discriminant analysis, the 6 circulating miRNAs were examined for the early diagnosis of endometriosis and their association with other clinical features of the disease was analyzed. A model combining miR-199a, miR-122, miR-145*, and miR-542-3p was developed to detect the early stage of endometriosis versus controls with 79.4% sensitivity and 75.9% specificity. The sensitivity was significantly higher than that of CA125, which only exhibits 28% to 50% sensitivity^[24,25] for diagnosing mild endometriosis. The miRNA signature had the potential to distinguish endometriosis severity and was associated with other disease-related clinical features, such as pelvic adhesion and the distribution of endometriosis. More importantly, the expression profile of circulating miRNAs decreased with an increasing EFI, suggesting that these circulating miRNAs might be useful for determining fertility before surgery. It could have implications for the patient decision to undergo surgery or not.

Women with endometriosis experience delays of approximately 6.7 years before receiving an accurate diagnosis.^[7] Therefore, the early diagnosis of endometriosis is important for patients, especially those seeking fertility.^[26] More importantly, an early diagnosis of the disease using a noninvasive method helps improve fertility by guiding the decision to perform surgery.^[27] A previous study showed that it is possible to diagnose minimal-mild endometriosis with a combination of multiple biomarkers (IL-6, IL 8, TNF, hsCRP, CA-125, and CA-199) with a high sensitivity (87%–92%) and an acceptable specificity (60%–71%).^[28] On the other hand, the study also highlighted that these biomarkers are easily affected by stress factors; for example, the blood was drawn just prior to anesthesia.^[28] In the present study, 4 miRNAs (miR-199a, miR-122, miR-145*, and miR-542-3p) could clearly distinguish stage I/II patients from controls with 79.4% sensitivity and 75.9% specificity. With the advantage of easy detection and steady levels of miRNAs in serum, these circulating miRNAs might be useful biomarkers for earlier disease detection.

Studies also demonstrated that combining a panel of biomarkers can improve the sensitivity and specificity for a diagnosis of endometriosis. Vodolazkaia et al evaluated a panel of 28 biomarkers and analyzed 2 models (model 1: annexin V, VEGF, CA125, and glycodelin; Model 2: annexin V, VEGF, CA125, and sICAM-1) with a high sensitivity of 81% to 90% and an acceptable specificity of 63% to 81% in predicting endometriosis,^[29] which is better than the sensitivity of CA125 alone, at 20% to 50%.^[24,25] Consistent with previous studies, in this study, a model combining 4 miRNAs (miR-199a, miR-122, miR-145*, and miR-542-3p) could reach a high sensitivity and specificity for endometriosis diagnosis. Specifically, when combining the circulating miRNAs and CA125, a better performance was achieved, and the diagnostic power was improved to 81.8% sensitivity and 92.6% specificity for endometriosis. These values are higher than previous reports with a single miRNA molecule with an AUC of 0.74, 0.79, and 0.85 for miR-17-5p, miR-20a, and miR-22, respectively,^[30] suggesting that the combination of a set of circulating miRNAs or other serum markers might be more useful in clinical practice.

Another important finding of this study is that circulating miRNAs were potentially associated with specific features of endometriosis, i.e., pelvic adhesion, distribution of the disease, and DIE. miR-199a and miR-141* were upregulated in patients with pelvic adhesion compared with non-pelvic adhesion patients, while miR-145*, miR-141*, and miR-9* appear to have the potential for identifying ovarian involvement endometriosis from patients without ovarian involvement. All 6 miRNAs were upregulated in DIE compared with non-DIE. An early paper by Koninckx et al^[31] suggested that there might be different pathogenic mechanisms for the different clinical features of endometriosis. In the present study, the differences in abnormally expressed miRNAs among different endometriosis features might indicate different pathological mechanisms involved in developing different endometriosis lesions since miRNAs are regulators of specific genes. Ovarian endometriosis lesions are associated with markedly higher E2 levels than endometrium or peritoneal and deep endometriosis and suggest high local estrogen synthesis with ovarian endometriosis.^[32] Another study demonstrated that ovarian endometrioma, but not DIE, was associated with high serum levels of interleukin-8 and interleukin-6, supporting different pathological processes in these 2 endometriotic entities.^[33] Furthermore, clinically, it is important to detect or predict pelvic adhesions and DIE in women with endometriosis, especially for women seeking a pregnancy soon or who complain of dysmenorrhea or dyspareunia. Thus, a pre-surgical test for pelvic adhesion and DIE would be highly valuable for women. Given that a serum miRNA-based test has a low cost and sample management is easy, including sample collection and processing, it can be suggested that the miRNA signature could serve as a potential marker for these clinical manifestations, which have not been previously reported, suggesting different pathogenetic processes of endometriosis associated with them.

In this study, the expression of CA125 and 3 circulating miRNAs (miR-199a, miR-122, and miR-542-3p) exhibited progressive changes along with a decreasing EFI score. These circulating biomarkers may have the potential in predicting the fertility of patients with endometriosis.

This study has limitations. The patients were from a single institution over a short period, and the sample size is relatively small. In addition, even though the controls were confirmed to be without endometriosis, most of them suffered from tubal factor infertility, and whether it impacts, the circulating miRNAs is unknown. Since a single patient population was used, whether the cutoff values are generalizable or not is unknown. A large-cohort follow-up study with a validated test is required.

5. Conclusions

This study investigated the diagnostic value of circulating miRNAs in a large cohort of patients with endometriosis. Combining miRNAs with CA125 as clinical biomarkers for diagnosing endometriosis is promising. The miRNA signatures can serve as potential biomarkers for the clinical manifestations of endometriosis. More importantly, the expression of a set of miRNAs is associated with the EFI score, suggesting that these circulating miRNAs may be useful for determining fertility preoperatively.

Acknowledgment

We thanks for Sun Yat-Sen Memorial Hospital for providing serum samples; the School of Life Sciences, Sun Yat-Sen University for providing an experimental platform and the guidance of Shunjia Hong and Yanan Zhao. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Author contributions

Conceptualization: Guansheng Chen.

Data curation: Guansheng Chen, Jiayi Guo, Wenjing Li, Rong Zheng, Hongyu Shang, Yongjun Wang.

Formal analysis: Hongyu Shang, Yongjun Wang.

Methodology: Guansheng Chen.

Project administration: Jiayi Guo.

Resources: Wenjing Li, Rong Zheng.

Supervision: Rong Zheng.

Visualization: Jiayi Guo.

Writing – original draft: Guansheng Chen.

Writing – review & editing: Guansheng Chen, Yongjun Wang.