Characterization of circulating miRNAs in the treatment of primary liver tumors

Abstract Background and Aim Circulating micro RNAs (miRNAs) indicate clinical pathologies such as inflammation and carcinogenesis. In this study, we aimed to investigate whether miRNA expression level patterns in could be used to diagnose hepatocellular carcinoma (HCC) and biliary tract cancer (BTC), and the relationship miRNA expression patterns and cancer etiology. Methods Patients with HCC and BTC with indications for surgery were selected for the study. Total RNA was extracted from the extracellular vesicle (EV)‐rich fraction of the serum and analyzed using Toray miRNA microarray. Samples were divided into two cohorts in order of collection, the first 85 HCC were analyzed using a microarray based on miRBase ver.2.0 (hereafter v20 cohort), and the second 177 HCC and 43 BTC were analyzed using a microarray based on miRBase ver.21 (hereafter v21 cohort). Results Using miRNA expression patterns, we found that HCC and BTC could be identified with an area under curve (AUC) 0.754 (v21 cohort). Patients with anti‐hepatitis C virus (HCV) treatment (SVR‐HCC) and without antiviral treatment (HCV‐HCC) could be distinguished by an AUC 0.811 (v20 cohort) and AUC 0.798 (v21 cohort), respectively. Conclusions In this study, we could diagnose primary hepatic malignant tumor using miRNA expression patterns. Moreover, the difference of miRNA expression in SVR‐HCC and HCV‐HCC can be important information for enclosing cases that are prone to carcinogenesis after being cured with antiviral agents, but also for uncovering the mechanism for some carcinogenic potential remains even after persistent virus infection has disappeared.


| INTRODUCTION
Hepatocellular carcinoma (HCC) and biliary tract cancer (BTC) are typical primary malignant tumors of the liver.They have different origins and are generally distinguished by tumor markers and diagnostic imaging.However, there are also similarities between the two: Hepatitis C Virus (HCV) and Hepatitis B Virus (HBV) infection is a risk factor not only for HCC, but also for BTC. 1 Among different forms of HCC, there is a pathologically mixed HCC in which cholangiocarcinoma (CCC) tissue is also present, known as combined cancer. 2It is difficult to distinguish between the two using tumor markers or diagnostic images from computed tomography and magnetic resonance imaging scans if not typical cases.Although no miRNA-based differentiation between HCC and BTC has been reported to date, changes in circulating N-linked glycosylation are associated with various cancers such as gastric cancer 3 and lung cancer. 4Recent attempts have been made to differentiate between HCC and BTC using six types of N-glycans. 5 far as HCC is concerned, the recent availability of direct-acting antiviral agents (DAAs) has enabled viral clearance in almost all patients with HCV infection.[6][7][8] The annual incidence of HCC in patients with chronic hepatitis is less than 1% with patients who had a sustained virologic response (SVR).9 However, the annual rate of hepatocarcinogenesis in patients with cirrhosis is 3%-7%.[10][11][12] Even if SVR is acquired in this way, the risk of carcinogenesis varies depending on the degree of the primary disease, and the risk of carcinogenesis is not low.There is a need for continuous follow-up for carcinogenesis even after acquiring SVR.Acquisition of SVR using antiviral drugs means that the viral infection is cured, and it also means that the virus does not reactivate. Howver, since the carcinogenic potential has not completely disappeared as described above, the characteristics of carcinogenic cases from cases that have acquired SVR and methods for predicting carcinogenesis are being actively developed.For example, machine learning was used create a prediction algorithm to predict cancer recurrence with an accuracy of 88.5% by analyzing the extracellular vesicle (hereafter EV) rich fraction miRNA expression pattern of patients with liver cirrhosis who acquired SVR using DAA as HCC treatment.In our previous works, the cancer prediction rate for HCC-naïve patients with chronic liver disease is 85.5%.13 Furthermore, when hepatocellular carcinogenesis prediction was performed using age, Alanine Aminotransferase (ALT) value, and alpha fetoprotein (αFP) value, the cases with subjects 65 years or younger, and ALT<30, αFP <5 are predicted to not develop cancer after 5 years of follow-up.14 Efforts are being made to predict HCC at an early stage and also assess the degree of liver fibrosis and inflammation [15][16][17][18] using circulating miRNA analysis.New biomarkers have been reported, such as post-treatment prediction methods.
Using information miRNA expression from EV fractions, we aimed to analyze a method for distinguishing between HCC and BTC, elucidation of the characteristics of carcinogenesis in HCC, and a method for predicting postoperative recurrence of HCC and BTC.

| Sample information
Serum samples before surgical treatment (pre) and after surgical treatment (post1 and post2) were obtained from 85, 74 to 47 patients with HCC, respectively.Post1 and post 2 samples refer to the serum samples taken after operation within 14 days, and beyond 14 days, respectively (Table 1).These samples were analyzed using 3D-Gene Human miRNA Oligo Chip based on miRbase ver.20 (Toray Industries, Inc., Kanagawa, Japan).Serum before surgical treatment was also obtained from 177 patients with HCC and 43 patients with BTC (41 CCC and 2 gaLL bladder cancer).These samples were analyzed using 3D-Gene Human miRNA Oligo Chip based on miRbase ver.21.

| RNA preparation and microarray analysis
RNA from the EV-rich fraction was prepared using ExoQuick (System Biosciences, Palo Alto, CA).RNA was extracted using a miRNeasy Mini Kit (Qiagen, Hilden, Germany).Sixty nanograms of total RNA were analyzed using the 3D-Gene miRNA microarray RNA extraction reagent from the liquid sample kit (Toray Industries, Inc., Kanagawa, Japan).Comprehensive miRNA expression analysis was performed by using two types of 3D-Gene miRNA Labeling Kit and 3D-Gene Human miRNA Oligo Chip (Toray Industries, Inc.).They correspond to miRbase release 20 (2555 miRNA published) and miRbase release 21 (2588 miRNA published).Since the microarray was upgraded from miRbase release 20 to 21 during the analysis plan, the cohort v20 (hereafter v20 cohort) was the group of which was analyzed with the miRbase release 20 microarray, and the cohort v21 (hereafter v21 cohort) was the group of which was analyzed with the miRbase release 21 microarray.All microarray data for this study conformed to the "Minimum Information about a Microarray Experiment guidelines" and are publicly available in the GEO database GSE212211.

| Statistical analysis
Tensor Decomposition Based Unsupervised Feature Extraction (TD based UFE) was used to extract unsupervised features of each group using the respective miRNA expression information obtained by microarray.
Suppose that x ij represents the expression of ith microRNA of jth sample.x ij is normalized as.
We apply principal component analysis to xij and we got principal component score u li and principal component loading v lj as In order to identify which js are associated with the distinction between two classes, we applied regression analysis to vlj as After identifying vlj, which is distinct between two classes, we attribute P-values to i as where P_(χ^2) [>x] is the cumulative χ2 distribution where argument is larger than x.P-values are corrected by BH criterion and is associated with adjusted P-values less than .01are selected.
After selecting ith, we recomputed vlj using the selected is.Using recomputed vlj (l Ωl), js are discriminated with liner discriminant analysis.

| Diagnostic analysis of cancer by miRNA
In the v21 cohort, analyzed using microarrays conforming to mirbase ver.1).Among these 10 miRNAs, there were 3 miRNAs (miR-204-3p, 4497, and 3960) with p < .05 or less between both groups, and other 7 miRNAs had no significant difference.The miRNAs selected here were not those with large differences between the two groups, but rather these 10 miRNAs were obtained in order to obtain the maximum discrimination ability between the two groups.

| Characteristics of miRNAs by carcinogenesis etiology
First, using an analysis of sera collected from 45 patients with HCC (v20 cohort analyzed using microarrays conforming to mirbase ver.20) before resection, miRNA expression patterns were compared between sera obtained from 36 patients with HCV-caused HCC (hereafter HCV-HCC) and 9 patients with HCC in HCV-treated cases (hereafter SVR-HCC).In v20 cohort, analysis was performed using before operation (pre), after operation within 14 days (post 1) and after operation beyond 14 days (post 2) specimens, and in v21, analysis was performed using pre specimens.3A, Supplementary Figure 1, and Table 1).
Looking at the 19 miRNAs used to differentiate between HCV-HCC and SVR-HCC individually, only miR-3665 showed p < .05,and the other 18 miRNAs have not achieved p < .05.Although it is not possible to distinguish between the two groups using information on individual miRNAs, for example miR-4488, when information on 19 miRNAs is combined using Tensor decomposition, the fine discrimination results are obtained.In univariate analysis of the two groups of HCV-HCC and SVR-HCC, the difference of expression in several miRNAs are shown p < .05 or less, however, the discriminative ability of this analysis is significantly lower than that of analysis using Tensor decomposition.
There are 68 HCV-HCC and 22 SVR-HCC cases in the v21 F I G U R E 3 Discrimination between HCV-HCC and SVR-HCC.(A) Using the 19 miRNAs shown in supplementary Figure 1, we distinguished HCV-HCC and SVR-HCC from cases used in cohort v20.The left table shows the prediction result.Using the expression information of 19 miRNAs, 24 cases were predicted to have the cause of carcinogenesis as HCV-HCC, 23 cases were HCV-HCC as expected, and one case was SVR-HCC contrary to expectations.Next, there were 21 cases in which the cause of carcinogenesis was predicted to be SVR-HCC, 8 cases in which the cause of carcinogenesis was SVR-HCC as expected, and 13 cases in which contrary to predictions the cause was HCV-HCC.Right panel shows AUC curve of this prediction.The number of cases in which predictions and results matched are shown in bold.(B) Using the 19 miRNAs shown in supplementary Figure 2, we distinguished HCV-HCC and SVR-HCC from cases used in cohort v21.The left table shows the prediction result.Using the expression information of 19 miRNAs, 51 cases were predicted to have the cause of carcinogenesis as HCV-HCC, 49 cases were HCV-HCC as expected, and 2 cases was SVR-HCC contrary to expectations.Next, there were 39 cases in which the cause of carcinogenesis was predicted to be SVR-HCC, 20 cases in which the cause of carcinogenesis was SVR-HCC as expected, and 19 cases in which contrary to predictions the cause was HCV-HCC.Right panel shows AUC curve of this prediction.
predicted.Furthermore, 39 cases were predicted to be SVR-HCC where 20 cases were correctly predicted (Figures 1B, 3B

| Recurrence prediction after cancer resection by miRNA expression pattern
Finally, we examined whether the v20 cohort could be used to predict recurrence time after resection of the HCC tumor.Eighty-five HCC samples collected before surgery (pre) were used.Twenty-five had recurrence within 1 year (<1), 12 had recurrence for more than 1 year and less than 2 years (1<, <2), and 48 had no recurrence for more than 2 years (<2).We then analyzed 71 HCC cases collected within 14 days after surgery (post 1).Twenty had recurrence within 1 year, 7 had recurrence over 1 year and less than 2 years, and 44 had no recurrence over 2 years.Furthermore, we analyzed 50 HCC cases collected 14 days or more after surgery (post 2).Fifteen had recurrence within 1 year, 5 had recurrence within 1 to 2 years, and 30 had no recurrence within 2 years or more.Using the v21 cohort, we analyzed 168 HCC cases collected before surgery (pre).40 had recurrence within 1 year, 42 had recurrence over 1 year and less than 2 years, and 86 had no recurrence over 2 years (Figure 1D).Using any cohorts, miRNA expression patterns could not predict recurrence prediction time.

| DISCUSSION
We analyzed the miRNA expression pattern in the blood serum of patients with HCC and BTC for primary liver tumors that are indicated for surgery, before and after surgery.We found that it was possible to distinguish between HCC and BTC cases using miRNA collected before surgery.We believe that it is important to accurately differentiate between BTC and HCC in a non-invasive manner.Typical cases can be easily differentiated based on images and histopathology, but there are cases in which it is difficult to differentiate even when histopathology is also evaluated.Differentiating between BTC and HCC may be clinically difficult.The infection of hepatitis virus (HBV and HCV) has been well known not risk factor only in HCC but also in BTC. 19Moreover, there are also cases in which there is a histopathological concept of a mixed carcinoma of both BTC and HCC. 20 addition, there are some cases where sampling is difficult under echo guidance because they exist near structures such as blood vessels and bile ducts.Given this situation, we believe that deriving a method to differentiate BTC and HCC using blood information would be clinically useful.
But to date, there has been no report on the differentiation between HCC and BTC using blood miRNA expression patterns.Therefore, this study is novel in reporting the differentiation between HCC and BTC using miRNA expression patterns.There are several reports on the prognosis and diagnosis of HCC using 8 miRNAs (miR-320b, 6724-5p, 6877-5p, 4448, 4749-5p, 663a, 4651, and 6885-5p) in the blood that can be used to diagnose HCC at its early stage. 18On the other hands, in diagnosing ICC, the combination of low blood miR-150 level and CA19-9 level improved the diagnostic ability. 21Low expression of serum miR-1281, miR-126, miR-26a, miR-30b and miR-122 compared to healthy subjects. 22miR-122 is associated with postoperative prognosis in ICC. 23Until now, studies have only compared cancer with normal or benign diseases, whether HCC or BTC, and direct comparison between HCC and BTC has not been performed.Therefore, we reviewed the results of this analysis based on the previous comparison of HCC and BTC in tumor tissues. 24Comprehensive transcriptome analysis and metabolomic analysis were reported for classifying four groups of not only HCC and BTC, but also corresponding noncancerous areas.17 miRNA expression patterns (let-7a-5p, let-7b-5p, miR-16-5p, 21-5p, 29a-3p, 122-5p, 451a, 642a-3p, 3917, 3960, 4286, 4454, 4459, 4516, 5100, 6087, and 6089) were used for classifying the four groups. 24miR-4454 is a common miRNA used in tissue and blood studies that are useful for differentiation of HCC and BTC.In particular, increased circulating miR-4454 expression positively correlated with prognosis post hepatocellular carcinoma treatment. 25MiR-4454 promotes carcinogenesis in liver tissues, 26 and its mode of involvement in carcinogenesis differs between blood and tissue.
With the development of HCV treatment, many patients with chronic liver disease with hepatitis C can achieve SVR.However, even if virus-related proteins are not continuously supply, hepatocarcinogenesis is observed with a certain probability.In this analysis, the expression patterns of miRNA in EV were clearly different between HCV-HCC and SVR-HCC samples.Analysis of miRNAs expression in cancer tissues with SVR-HCC and HCV-HCC, and whose expression are elevated in HCV carcinogenesis compared with SVR carcinogenesis include miR-130a, 30a-3p, 100, 134, 139-5p, 144, 150, 192, and 451a).Previous our reports showed that miR-130a, 27 and 134 28 are "so-called" oncomiRs involved in carcinogenesis promotion, when overexpressed.However, miR-30a-3p, 29 100, 30 139-5p, 31 144, 32 150, 33 451a, 34 192 35 have been reported as anti-oncomiRs that suppress carcinogenesis when overexpressed.The down-regulated miRNAs in HCV carcinogenesis compared with SVR carcinogenesis include miR-18a, 19a, 21, 30d, 93, 146a, 181, 494, 24, 26a, 26b, 27a, 92a, 127, 142-3p, and 145a. 36nce both HCV-HCC and SVR-HCC represent cancer cases, it is inferred that the miRNA expression patterns are different between HCV-treated and un-treated cases.The miRNA expression patterns in the HCV-HCC and SVR-HCC tumor tissues as previously reported 36 were not related to the circulating miRNA pattern analyzed in this study.
Furthermore, there were no specific features in miRNA expression patterns based on the antiviral treatments used in achieving to SVR (Table 2).Recurrence is a problem in HCC treatment.There is an important concept of eliminating HCV, which is the cause of HCC, and both AASLD and EASL have treatment guidelines for direct treatment with viral agents if HCV remains after HCC treatment.It has also been reported that PDL-1 treatment of HCC differs between viral HCC and NASH-derived HCC. 37These findings suggest that the HCV infection status may influence the selection of anticancer treatment and the recurrence of HCC.Based on the above, the differentiation between HCV-HCC and SVR-HCC is useful information for analyzing responsiveness to cancer treatment and recurrence mode, and is considered useful for analyzing carcinogenic mechanisms.
We investigated the relationship between the condition before cancer and after cancer.For prediction of HCC recurrence in patients with liver cirrhosis who had a history of cancer and who developed SVR using antiviral agents, the expression patterns of four miRNAs (hsa-miR-4718, 6511a-5p, 642a-5, and 4448) were used.Four miRNAs (hsa-miR-211-3p, 6826-3p, 1236-3p, 4448) were used to predict HCC recurrence in chronic hepatitis (CH) and liver cirrhosis (LC) patients with a history of cancer who developed SVR using antiviral drugs.Four miRNAs (hsa-miR-762, 8069, 7847-3p, 7846-493p) were used for predict HCC occurrence in CH and LC patients with no history of cancer and who developed SVR after using antiviral drugs. 13miR-6090 was used to predict carcinogenesis and to differentiate between HCV-HCC and SVR-HCC.The expression of miR-6090 is enhanced by ionizing irradiation 38 and is associated with stroke in cardiovascular disease. 39This suggests that abnormal expression of miR-6090 is induced as a result of acute or chronic inflammation.
In this study, Taguchi newly developed TD based UFE to perform unsupervised feature extraction.This method is applicable to gene expression, DNA methylation, and histone modification etc.It can perform multi-omics analysis.EV-rich fractions were collected before and after surgery for primary hepatic malignant tumors, and miRNA oncogenesis analysis was performed.HCV-HCC and SVR-HCC had different miRNA expression patterns.It is also potentially applicable to single cell omics data sets.We have already reported a method to predict SVR carcinogenesis as well as to identify miRNAs related to SVR carcinogenesis by performing miRNA expression analysis of EV collected before carcinogenesis in SVR cases. 13Together, these data reveal that miRNAs present in EV differ not only before and after carcinogenesis, but also according to the mode of carcinogenesis.We further identified miRNAs that distinguish between HCC and BTC.Regarding the treatment of HCC, it has been shown that the reactivity of the drugs used differs depending on the cause of carcinogenesis. 37The fact that peripheral blood miRNA profiles differ depending on the cause of carcinogenesis is expected to be useful in elucidating the mechanisms of carcinogenesis and therapeutic response.Analysis of miRNA expression in EV can be useful for follow-up observation of chronic liver disease in anticipation of carcinogenesis prevention.Finally, when comparing two groups, some differences in individual gene expression are not significant, and one of the limitations of this analysis is that the function of individual genes cannot be compared between groups.However, considering that pathway analysis is currently being used as a mainstream method, our analysis method uses multiple gene expression rather than individual gene expression, so it can be said to be appropriate.

s¼1b
ls δ js where al and bls are regression coefficients and δjs taken when j belongs to s th class otherwise 0. P-values are corrected with BH criterion.ls associated with adjusted P-values less than .05are selected.Then we found that l Ωl are associated with adjusted P-values less than .05.

F
I G U R E 1 Analysis design of this study.(A) Comparison of miRNA expression in EV between HCC and BTC.(B) Comparison of miRNA expression pattern in EV between HCV-HCC and SVR-HCC.(C) Comparison of miRNA expression pattern in EV among ALD-HCC, HBV-HCC, HCV-HCC, NBNC-HCC, and HBV + HCV-HCC.HBV + HCV-HCC analyzed only in the v21 cohort.(D) Comparison of miRNA expression pattern in EV among recurrence period of HCC resection.
Umezu, Shogo Tanaka and Yoshiki Murakami performed the experiments, intellectual input and discussions.Shoji Kubo, Masaru Enomoto and Akihiro Tamori helped write the manuscript, and provided discussions.Takahiro Ochiya, Y.-H.Taguchi and Masahiko Kuroda supervised the research and provided helpful discussion.Y.-H.Taguchi provided chemometric support and analyzed data.Yoshiki Murakami designed the experiments, analyzed and interpreted data, and wrote the manuscript.

1
Summary of clinical feature.