*3.7. Inclusion of PDGFR*β *Improves the Diagnostic Power of the miRFIB-Score*

We previously reported the diagnostic utility of circulating PDGFRβ protein levels to detect significant liver fibrosis [17]. Thus, we performed logistic regression analysis on the derivation cohort, combining PDGFRβ levels with our five-miRNA panel. This generated the miRFIBp-score, which was calculated as follows:

$$\text{mirRFIB}^{\text{p}}\text{-score} = (0.97229 \times \text{miRFIB}\text{-score}) + (0.00021150 \times \text{PDGFR}\\$ \text{ (pg/mL)} - 1.8678.4)$$

The score had an increased diagnostic value for the identification of significant liver fibrosis in the derivation cohort (AUC = 0.7912; sensitivity = 80.82%; specificity = 70.37%), validation cohort (AUC = 0.8009; sensitivity = 68.75%; sensitivity = 81.48%), and total cohort (AUC = 0.7970; sensitivity = 79.05%; sensitivity = 69.51%) (Table 3 and Figure 5). The inclusion of PDGFRβ into the miRFIB-score further improved the correlation with fibrosis severity (r = 0.4847) (Supplementary Table S4), with a persistent possibility to differentiate patients with specific stage F2 liver fibrosis from patients with stage F0–1 fibrosis (*p* < 0.0001) (Supplementary Figure S2).


**Table 3.** Performance of the miRFIB- and miRFIBp-score, as compared to the AST/ALT, APRI, Fib-4, and PRTA scoring algorithms, for the detection of significant liver fibrosis (F ≥ 2).

AST/ALT ratio: aspartate aminotransferase/alanine aminotransferase ratio; APRI: AST to platelet ratio index; Fib-4: Fibrosis-4; PRTA-score: PDGFRβ-thrombocytes-albumin score; AUC: area under the curve; CI: confidence interval; PPV: positive predictive value; NPV: negative predictive value.

**Figure 5.** Diagnostic performance of the miRFIB- and the miRFIBp-score for significant liver fibrosis. Performance comparison of the miRFIB-, the miRFIBp-score, and commonly used validated diagnostic algorithms for the diagnosis of significant liver fibrosis (F2–4) in (**A**) the derivation, (**B**) validation, and (**C**) total patient cohort.

### **4. Discussion**

Studies concerning the identification of novel non-invasive diagnostic tools suitable for the screening and monitoring of liver fibrosis in a general patient population remain limited. However, they are highly needed, as an accurate diagnosis of liver fibrosis has been shown to be a critical determinant of patient outcome [23,24]. Serological scoring tools, such as Fib-4, APRI, and AST/ALT, have been integrated into clinical practice, but lack accuracy for the identification of early stages of, and minor changes in, liver fibrosis [25–27]. Therefore, they are often only used as an indicator of the need for liver biopsy. The need for liver biopsy in current clinical practice thus remains. The identification of an adequate, sensitive, and specific serological marker for liver fibrosis would be of great value, as it could be used as an efficient first-line diagnostic step in screening at-risk patients [28], provide an easy tool for monitoring patients with fibrosis, and be of use in clinical trials evaluating fibrosis.

In the present study, we assessed the diagnostic utility of miRNAs differentially expressed during the activation of in vitro cultured primary HSCs, to identify significant liver fibrosis in a heterogeneous patient cohort with chronic viral infection, chronic alcohol abuse, and NAFLD. We identified an increased expression of miRNA-451a and miRNA-142-5p in the plasma of patients with significant liver fibrosis versus patients with no or mild fibrosis, whereas Let-7f-5p expression was decreased (Figure 4A). The observed down-regulation of miRNA-451a in total liver tissue of CCl4-injected mice (Figure 3C) mimics their down-regulated expression in livers of NASH patients, compared to patients with simple steatosis [29]. Additionally, its enhanced circulating expression, observed in our patient cohort with significant liver fibrosis (Figure 4A), was also seen when comparing the serum of NAFLD patients with healthy controls [30]. In contrast, one study identified a down-regulation of miRNA-451a in the plasma of cirrhotic HBV patients, compared to healthy controls [31]. These results could not be confirmed in our patient cohort with chronic viral infection. Differences in miRNA-451a expression between HBV and HCV patients should further be investigated. Su et al [32] identified the inflammatory signals IL4 and IL13 to increase miRNA-142-5p expression in macrophages, activating them towards a pro-fibrogenic character. Furthermore, they identify its up-regulation in liver tissue of six-weeks CCl4-treated C57BL/6J mice. This is in contrast to the down-regulation we observed in the liver tissue of four-weeks CCl4-treated Balb/c mice (Figure 3C). Differences in fibrosis progression or in miRNA expression between mouse strains could be the cause of such discrepancy. Our results concerning liver miRNA-142-5p expression seem to mimic the down-regulation seen in the liver tissue of cirrhotic HCV patients [33]. While decreasing levels of circulating Let-7a-5p, Let-7c-5p, and Let-7d-5p are correlated with fibrosis severity in patients with chronic HCV infection [34], the diagnostic utility of Let-7f-5p was not investigated in these studies. We show that circulating Let-7f-5p has the highest diagnostic value of all candidate miRNAs (Figure 4A and Supplementary Table S3) for the identification of significant liver fibrosis. Additionally, from the candidate miRNA panel, Let-7f-5p was the only miRNA to be correlated with the Fib-4, APRI, and PRTA-scoring tools (Table 2), what further underlines its diagnostic potential.

In the search for the ideal diagnostic miRNA-based algorithm, we supplemented our candidate miRNA panel with the well-studied miRNA-122-5p and miRNA-29a-3p. miRNA-122-5p is the most abundant miRNA in the liver, with dominant expression in the hepatocytes (Supplementary Figure S3) [35], where it is involved in cholesterol synthesis [36]. The elevated levels of circulating miRNA-122-5p in patients with early stage fibrosis (F1–2) as compared to healthy controls are suggested to represent miRNA-release from injured hepatocytes. On the other hand, the decreased circulating miRNA-122-5p levels during later stages of fibrosis (F3–4) would be caused by the progressive loss of functional hepatocytes in the injured liver [37]. Its diagnostic utility for late-stage liver fibrosis and cirrhosis has already been suggested in various liver disease aetiologies [38–43]. Our study shows that while miRNA-122-5p has little diagnostic value on its own for significant fibrosis (Figure 4B), its contribution is essential to the miRFIB-score (Figure 5). In contrast to the hepatocyte-specificity of miRNA-122-5p, miRNA-29a-3p shows the highest expression in HSCs (Supplementary Figure S3), undergoing down-regulation upon in vitro and in vivo activation [22]. Its potential use as a therapeutic

target has been elaboratively reported by the group of Y.H. Huang et al. using various mouse models of liver disease [44–47]. The negative correlation of circulating miRNA-29a-3p levels with fibrosis/cirrhosis severity has been shown in patients with NAFLD [48], HBV [49], HCV, alcohol abuse, and biliary disease [22].

Individually, all of the analyzed significantly dysregulated miRNAs had low predictive values for the diagnosis of significant liver fibrosis, with AUC values ranging from 0.59 to 0.64 (Figure 4A–C and Supplementary Table S3). However, using logistic regression analysis, we generated an algorithm, the miRFIB-score, consisting of miRNA-142-5p, miRNA-451a, Let-7f-5p, miRNA-122-5p, and miRNA-29a-3p, with a predictive value superior to the clinical scoring systems Fib-4, APRI, and AST/ALT (Figure 5 and Table 3). As we recently reported the highly discriminative potential of circulating PDGFRβ-levels for significant liver fibrosis in patients with various aetiologies of liver disease [17], we generated a second diagnostic algorithm combining the miRFIB-score with such circulating PDGFRβ-levels, the miRFIBp-score. A marked improvement in diagnostic values was observed for this combinatory algorithm (Figure 5 and Table 3).

An unexpected finding was the discrepancy between the enhanced expression of miRNA-122-5p and miRNA-29a-3p in the plasma of mice treated with CCl4 (Figure 3D), and their lowered expression levels in the plasma of patients with significant fibrosis (Figure 4C). All other tested miRNAs seem to have an overlapping expression pattern between mouse and human subjects. In our experiments, Balb/c mice underwent only four weeks of CCl4-injections, which is thought to represent early-stage fibrosis. To induce late-stage fibrosis, or cirrhosis, 8–20 weeks of CCl4-injections should be performed [20]. However, the enhanced expression of miRNA-122-5p in the plasma of CCl4-injected mice is in line with our previous results comparing the plasma of patients with early-stage fibrosis to healthy subjects [14]. Alternatively, the enhanced expression of miRNA-122-5p in the plasma of mice treated with CCl4 could reflect the hepatocyte damage caused by the last CCl4-injection, since samples were taken only 24 h later. For miRNA-29a-3p, this is likely not the case, since this miRNA has thus far not been associated with hepatocyte damage. To further investigate this, healthy individuals should be included in future studies.

miRNA-451a, miRNA-142a-5p, Let-7f-5p, and miRNA-378a-3p were found to be significantly dysregulated in activated HSCs, as compared to quiescent controls (Figure 1D), which suggests a role in the HSC activation process. To further investigate this, target prediction was performed, and we focused on the target genes regulated by all four HSC-activation linked miRNAs. Of the predicted 13 overlapping miRNA targets (Figure 2A), three genes (*Ankrd52*, *Clcn5*, and *Peg10*), were found to be significantly up-regulated upon HSC activation (Figure 2C). As miRNAs negatively regulate gene expression by mRNA decay or inhibition of translation [50], and due to the dominantly down-regulated expression of the selected miRNAs, we hypothesize *Ankrd52*, *Clcn5*, and *Peg10* to be regulated by the selected miRNAs. However, to confirm this, further functional studies should be performed. While the roles of *Ankrd52* and *Clcn5* in liver disease remain unclear, *Peg10* has been widely studied in hepatocellular carcinoma (HCC) pathology. More specifically, *Peg10* expression levels are elevated in HCC [51], where it is found to inhibit the pro-apoptotic mediator *Siah1* [52], and stimulate cell proliferation by association with c-MYC [53]. Additionally, *Peg10* tissue mRNA levels mark HCC progression and poor survival [54,55]. Due to its high expression in activated HSCs (Figure 2C), and its important functionality in HCC, it would be of interest to investigate its role in liver fibrogenesis. As it cannot be excluded that a target gene is dominantly regulated by just one specific miRNA, it is possible that the increasing Let-7f-5p levels found upon HSC activation lead to the identified down-regulation of multiple predicted target genes (Figure 2D). Among these, *Cpeb3* and *Gnai3* have proven functionality as tumor suppressors in HCC. Both genes are found to undergo negative regulation by miRNAs, *Cpeb3* by miRNA-452-3p and miRNA-107 [56,57], and *Gnai3* by miRNA-222 [58]. However, their role during liver fibrosis remains to be determined.

Whether a miRFIB- or miRFIBp-score can be integrated into the clinical practice will depend on future technical developments. Currently, a combination of protein and miRNA detection from the same plasma sample is not standard practice in hospital settings. Due to this more complex character, we expect the miRFIBp-score to have a more important financial cost, compared to the miRFIB-score. However, our analysis identified the miRFIBp-score to have superior diagnostic value for the identification of significant liver fibrosis and an in-depth cost–benefit analysis should determine if this diagnostic superiority outweighs the additional costs. Furthermore, the manipulation of blood samples should be performed with great care, since hemolysis of red blood cells may result in the release of their cytoplasmic miRNAs, such as miRNA-451a [59,60], and thus influence the results of the miRNA-based diagnostic algorithms. Additionally, the time interval between blood sampling and plasma storage should be kept as short as possible, as specific miRNAs, including miRNA-122-5p, can undergo a time-dependent decline in stability when the sample is kept at room temperature [61].

There were a number of limitations with the current study. The patient cohort size was relatively small and contained an imbalance in the presence of liver disease aetiologies. Although recent research has reported substantial differences in miRNA expression values in the plasma of patients with HBV infection versus patients with HCV infection [62], due to insufficient patient numbers we were unable to make any claims regarding such differences in our patient cohort. Furthermore, due to the cross-sectional character of the study, we did not possess any clinical follow-up material of the included patients. We were thus unable to test the prognostic ability of the miRFIB- and miRFIBp-score. Finally, all included patients were staged for liver fibrosis by use of elastography. Future studies should focus on the validation of our results using plasma obtained from biopsy-staged patients. Moreover, the score should be tested during treatment to study if the scores can be used to evaluate early changes in fibrosis, and possibly predict the outcome.

In conclusion, we have identified five miRNAs that, when combined into the predictive miRFIB-signature, had high diagnostic values for significant liver fibrosis in a heterogeneous patient population with chronic alcohol abuse, viral infection, and NAFLD. Combining the miRFIB-score with circulating PDGFRβ-levels increased its diagnostic utility. Although these proposed scores require further validation, they may provide crucial information regarding liver fibrosis severity and evolution. Thanks to their non-invasive character, the scores would allow repeated measures and objective interpretation, at a relatively low financial cost. Additionally, functional studies could unravel the importance of the selected miRNAs during fibrogenesis and fibrolysis, and their potential utility as therapeutic targets.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2073-4409/8/9/1003/s1; **Supplementary Materials and Methods**: Extracellular vesicle isolation and RNA extraction, Nanostring miRNA analysis, Simultaneous isolation of different liver cell types. **Supplementary Figures and Tables**: Figure S1: Top enriched miRNAs in EVs derived from activating HSCs, Figure S2: Fibrosis-stage specific presence of circulating miRNAs, Figure S3: miRNA expression analysis in liver cell types, Table S1: miRNA primers, Table S2: mRNA primers, Table S3: Performance of individual plasma miRNAs, as compared to the AST/ALT, APRI, Fib-4, and PRTA scoring algorithms, for the detection of significant liver fibrosis (F ≥ 2), Table S4: Correlation of circulating miRNA expression levels with fibrosis stage.

**Author Contributions:** J.L. study concept and design; acquisition of data; analysis and interpretation of data; statistical analysis; writing of the manuscript. S.V. analysis and interpretation of data. H.R. provision of samples; interpretation of data; critical revision of the manuscript. L.A.v.G. study concept and design; interpretation of data; critical revision of the manuscript.

**Funding:** This research received funding by the Vrije Universiteit Brussel (VUB), the Institute for the Promotion of Innovation through Science and Technology in Flanders (IWT-Flanders) (HILIM-3D; SBO140045), the Fund of Scientific Research Flanders (FWO), and the Gilead Sciences BeLux Fellowship Programme 2018 (awarded to H.R.).

**Acknowledgments:** We would like to acknowledge Daniella Blyweert, Nathalie Eysackers, Aneta Kozyra, Iona De Mol, and Christella Ukunda for technical support.

**Conflicts of Interest:** The authors declare no conflict of interest.
