Wisteria floribunda agglutinin-positive Mac-2-binding protein as a diagnostic biomarker in liver cirrhosis: an updated meta-analysis

Wisteria floribunda agglutinin-positive Mac-2-binding protein (WFA+-M2BP) had been suggested as a possible glycobiomarker for assessing liver fibrosis. Here, we conducted this updated meta-analysis to systematically investigate the predictive accuracy of WFA+-M2BP for diagnosing liver fibrosis and hepatocellular carcinoma (HCC) by comparing with multiple non-invasive indicators. We searched relevant literatures from Pubmed, Web of Science, EMBASE and Cochrane Library and enrolled 36 eligible studies involving 7,362 patients. Summary results were calculated using bivariate random effects model. The pooled sensitivities, specificities and areas under the summary receiver operating characteristic curves (AUSROCs) of WFA+-M2BP for identifying mild fibrosis, significant fibrosis, advanced fibrosis, cirrhosis, and HCC were 0.70/0.68/0.75, 0.71/0.75/0.79, 0.75/0.76/0.82, 0.77/0.86/0.88, and 0.77/0.80/0.85, respectively. The accuracy of WFA+-M2BP was strongly affected by etiology and it was not better than other non-invasive indicators for predicting early fibrosis. It showed similar diagnostic performance to hyaluronic acid and FibroScan for cirrhosis, but was equivalent to α-fetoprotein for HCC. In conclusion, WFA+-M2BP was suitable to diagnose late stage of liver fibrosis, especially cirrhosis. Individual cutoff value of WFA+-M2BP could be used to grade liver fibrosis in different etiology. Combined diagnostic model was suggested to improve its predictive accuracy for HCC.


Results
Basic characteristics of studies. As shown in Fig. 1, after excluding duplicates and non-experimental studies, 350 references were identified. Full-text review on 72 original articles eligible for detailed evaluation were conducted after we excluded non-relevant references. A total of 36 articles were further removed because of insufficient information to construct a 2 × 2 table. Ultimately, the remaining 36 articles were selected for metaanalysis.
We listed the main features of the included studies in Table 1 and 2. Overall, 7,362 participants were included. Among the 36 included articles, 29 articles studied the diagnostic accuracy of WFA+-M2BP on liver fibrosis and 8 articles were on HCC. For the studies on fibrosis, we noticed that 3 articles enrolled both training group and validation group [24][25][26] , and 2 articles recruited patients with 2 different etiologies 27,28 . Thus, we considered them as individual studies when the calculation of diagnostic accuracy was conducted. Overall, 7 kinds of etiologies of liver fibrosis that include HBV (n = 12), HCV (n = 10), NAFLD (n = 3), NASH (n = 3), AIH (n = 1), BA (n = 2), and PBC (n = 2), as well as mixed etiologies (n = 1) were discussed. HCC was mainly caused by 3 etiologies here: HBV, HCV and NAFLD. All studies employed retrospective design and used lectin-Ab sandwich immunoassay to detect serum WFA+-M2BP levels.
Quality assessment. On the basis of QUADAS-2 assessment, the overall quality of included studies was moderate. As shown in Supplementary Figs. 1, 2, in terms of patient selection, 13 studies had high risk of bias because of inappropriate exclusions or case-control designs. A total of 29 studies had high risk of bias in index test because of the awareness of reference standard result before conducting the index test. Five studies did not mention the use of blind method for index tests when explaining the reference standard results. Regarding flow and timing, 25 studies had high or unclear risk of bias because not all patients received the same reference standard or due to unclear interval between index test and reference standard. Moreover, we had significant concerns on 7 studies when evaluate the applicability of their patient selections.
Pooled predictive accuracy of WFA+-M2BP in liver fibrosis. Here, we summarized the predictive accuracy of WFA+-M2BP in each liver fibrosis. A total of 6 studies with 1,235 patients were evaluated for the performance of WFA+-M2BP on predicting mild fibrosis. The pooled sensitivity and specificity were 0.70 (95% CI 0.62-0.77) and 0.68 (95% CI 0.57-0.78), respectively ( Fig. 2A). Besides, the pooled AUSROC was 0.75 (95% CI 0.71-0.78). Twenty studies with 3,602 patients were included in significant fibrosis. The pooled sensitivity, specificity and AUSROC were 0.71 (95% CI 0.65-0.76), 0.75 (95% CI 0.69-0.81), and 0.79 (95% CI 0.75-0.82), respectively (Fig. 2B). For predicting advanced fibrosis, 28 studies involving 4,427 patients were assessed. The pooled sensitivity, specificity and AUSROC were 0.75 (95% CI 0.69-0.79), 0.76 (95% CI 0.72-0.80), and 0.82 (95% CI 0.78-0.85), respectively (Fig. 3A). For cirrhosis, 21 studies with 3,449 patients were identified. As displayed in Fig. 3B, the pooled sensitivity and specificity were 0.77 (95% CI 0. 69 Heterogeneity analysis, threshold effect and meta-regression. To investigate the heterogeneities of the included studies, threshold effect and overall heterogeneity were analyzed. Significant heterogeneities existed in each stage of liver fibrosis (Q = 23.11, I 2 = 91%, P < 0.001; Q = 94.75, I 2 = 98%, P < 0.001; Q = 50.32, I 2 = 96%, P < 0.001; Q = 64.79, I 2 = 97%, P < 0.001). However, no significant threshold effect was found. In four liver fibrosis stages from mild fibrosis to cirrhosis, the spearman correlation coefficients between sensitivities and specificities were − 0.94 (P = 0.88), − 0.01 (P < 0.01), − 0.02 (P < 0.01), and − 0.16 (P = 0.03), respectively. Meta-regression analysis (at least 10 studies were requested) was performed to further discuss the cause of heterogeneity in the studies reported for significant fibrosis, advanced fibrosis, and cirrhosis. We investigated 10 factors that might be the potential sources of heterogeneity: year of publication, region, median age, male proportion, number of patients, etiology, histological system, liver biopsy length, interval between biopsy and blood test, and blind method. For identifying significant fibrosis, the accuracy of WFA+-M2BP could be influenced by age, male proportion, etiology, and blind method (P < 0.01, P = 0.07, P = 0.05, and P < 0.01, respectively). For advanced fibrosis, the performance of WFA+-M2BP could be affected by age, male proportion, etiology, and region (P < 0.01, P < 0.01, P = 0.01, and P < 0.01, respectively). For cirrhosis, the heterogeneity of WFA+-M2BP for the detection might be due to the heterogeneity of age, male proportion, region, etiology, and blind method (P < 0.01, P < 0.01, P = 0.02, P = 0.07, and P = 0.08, respectively).  www.nature.com/scientificreports/ Predictive accuracy of WFA+-M2BP in liver fibrosis stratified by etiology. As etiology was one of the main reasons of heterogeneities based on our meta-regression analysis, we further analyzed the predictive accuracy of WFA+-M2BP in liver fibrosis caused by various etiologies. We combined studies related to NAFLD and NASH together, and combined studies related to AIH, BA, PBC and mixed etiologies into the "other etiologies" category because of limited number of references. Intriguingly, WFA+-M2BP showed diverse diagnostic accuracies in different etiology groups (Table 3). In general, for the prediction of significant fibrosis, advanced fibrosis, and cirrhosis, WFA+-M2BP owned the best diagnostic accuracies among patients with AIH, BA, PBC or mixed etiologies by reaching the highest pooled sensitivity, specificity, PLR, DOR, AUSROC and lowest NLR, when the results were compared with patients in other etiology groups. Besides, for advanced fibrosis, heterogeneities dramatically dropped in different etiology groups except for HBV and HCV. And for cirrhosis, no heterogeneity was found in the subgroup of NAFLD or NASH (Q = 3.11, I 2 = 36%, P = 0.106), indicating the accuracy of WFA+-M2BP was influenced by the etiology of disease. In Table 3, different weighted mean WFA+-M2BP values were observed in different etiologies, suggesting individual cutoff value of WFA+-M2BP should be applied to grade liver fibrosis in each etiology. In addition, we noticed that compared with significant fibrosis and advanced fibrosis, WFA+-M2BP possessed the highest AUSROCs in diagnosing cirrhosis regardless of the etiology.
Predictive accuracy of WFA+-M2BP versus non-invasive indicators for grading liver fibrosis. Due to limited number of studies containing the information of other non-invasive indicators in mild fibrosis, we compared WFA+-M2BP with other non-invasive indicators for predicting significant fibrosis, advanced fibrosis and cirrhosis. As shown in Table 4 Table 4. AUSROC values of seven non-invasive indicators for predicting significant fibrosis, advanced fibrosis and cirrhosis. WFA+-M2BP, wisteria floribunda agglutinin-positive Mac-2-binding protein; APRI, Aspartate aminotransferase-to-platelet ratio index; FIB-4, Fibrosis-4 index; AST/ALT, AST to ALT ratio; HA, hyaluronic acid; PLT, platelet count; AUSROC, area under the summary receiver operating characteristic curve; CI, confidence interval. Diagnostic accuracy of WFA+-M2BP for the prediction of HCC. For the prediction of HCC, a total of 8 studies with 2,240 participants were selected ( Table 2). Among them, 4 studies reported the occurrence of HCC after antiviral treatment or HBeAg seroconversion 49,50,55,56 , one study discussed the reoccurrence of HCC after curative resection 54 , and 3 studies focused on the development of HCC [51][52][53] . The WFA+-M2BP levels here were pretreatment or basal levels. As several studies described the diagnostic information of APRI, FIB-4, and AFP, we compared the diagnostic accuracies of WFA+-M2BP with these three indicators for HCC. There was no significant threshold effect in included studies (r = − 0.7, P = 0.49). However, significant heterogeneity was observed ( Publication bias and sensitivity analysis. As displayed in Supplementary Fig. 3, Deek's funnel plots were almost symmetric for studies that reported mild liver fibrosis, significant fibrosis, and advanced fibrosis (P values = 0.1, 0.33, and 0.09, respectively), suggesting no evidence of publication bias. However, a significant publication bias was observed in studies on cirrhosis (P = 0.03). For studies on the prediction of HCC, there was no publication bias (P = 0.83) (Supplementary Fig. 4). Through sensitivity analysis, we observed outlier studies existed in each stage of liver fibrosis ( Supplementary  Fig. 5). Surprisingly, after the removal of outlier studies, the heterogeneity in mild fibrosis disappeared (Supplementary Table 1) and the publication bias in studies on cirrhosis was diminished ( Supplementary Fig. 6). However, Supplementary Table 1 indicated that the summary results were not significantly affected by individual studies. Also, as displayed in Supplementary Fig. 7, no outlier study was found in HCC.

Discussion
WFA+-M2BP is a serum glycobiomarker that is receiving growing attentions. It had been reported that the elevated WFA+-M2BP level was associated with the risk of HCC [57][58][59] , the loss of HBeAg in chronic hepatitis B patients 60 , and the severity of liver fibrosis 61,62 . In our meta-analysis, we evaluated 36 articles in total, and explored the predictive accuracy of WFA+-M2BP for distinguishing liver fibrosis stages and HCC by comparing with diverse non-invasive indicators. Our results suggested WFA+-M2BP possessed satisfactory diagnostic accuracy for predicting cirrhosis and moderate diagnostic performance for detecting mild fibrosis, significant fibrosis, advanced fibrosis and HCC. The AUSROC of WFA+-M2BP was equivalent to HA and FibroScan for assessing cirrhosis, and similar to AFP for diagnosing HCC.
Previously, literatures on the diagnostic performance of WFA+-M2BP in different stages of liver fibrosis were controversial. Zou et al. 26 reported WFA+-M2BP was useful to assess early stages of liver fibrosis, and Ura et al. 36 showed WFA+-M2BP was more accurate in diagnosing significant fibrosis than advanced fibrosis. In contrast, several other studies indicated that WFA+-M2BP had the strongest ability to predict cirrhosis 37,62 . In our metaanalysis, we found that the overall AUSROCs of WFA+-M2BP for identifying mild fibrosis, significant fibrosis, advanced fibrosis and cirrhosis were 0.75, 0.79, 0.82 and 0.88, respectively. For cirrhosis, WFA+-M2BP reached the highest pooled sensitivity and specificity at 0.77 and 0.86. Since it is widely accepted that AUC between 0.85 and 0.90 is as good as liver biopsy for staging fibrosis 63 , our study underscores the notion that WFA+-M2BP could serve as a surrogate biomarker for biopsy when diagnosing cirrhosis.
Interestingly, WFA+-M2BP exhibited different predictive accuracies for staging liver fibrosis caused by different etiologies in CLDs. In our study, for patients with AIH, BA, PBC or mixed etiologies, WFA+-M2BP exhibited excellent performance for distinguishing significant fibrosis, advanced fibrosis and cirrhosis. In general, WFA+-M2BP had lower accuracy in HBV-infected patients than in patients with HCV infection, NAFLD or NASH. Our conclusion was consistent with a previous meta-analysis 23 . However, our study was more extensive, as we included more publications, had different subgroup setups, and used different models for the calculation of pooled results. WFA+-M2BP has the potential to reflect hepatic fibrosis as hepatic stellate cells (HSCs) are the source of WFA+-M2BP and its level is closely associated with α-smooth-muscle actin (αSMA) expression 19,21 . However, HBV-positive patients are more likely to experience quiescent hepatic inflammation, and HBV-related cirrhosis had large regenerative nodules and thin fibrous septa 23,27 . As a result, as shown in Table 3, minor changes Table 5. Meta-analyses results of four non-invasive markers for diagnosing HCC. HCC, hepatocellular carcinoma; WFA+-M2BP, wisteria floribunda agglutinin-positive Mac-2-binding protein; APRI, aspartate aminotransferase-to-platelet ratio index; FIB-4, fibrosis-4 index; AFP, α-fetoprotein; AUSROC, area under the summary receiver operating characteristic curve; CI, confidence interval; NA, not available. www.nature.com/scientificreports/ of WFA+-M2BP optimal cutoffs in each liver fibrosis stage of HBV-infected patients may lead to low diagnostic accuracy. Whether better predictive performance of WFA+-M2BP is related to more severe liver damage or inflammation response, or more activation of HSCs, will need to be investigated by further studies. Here, we suggest that individual cutoff value of WFA+-M2BP should be used to stage liver fibrosis of different etiology. Overall, WFA+-M2BP was not better than other non-invasive indicators for predicting significant fibrosis and advanced fibrosis. However, for assessing cirrhosis, the diagnostic accuracy of WFA+-M2BP was equivalent to HA and FibroScan and superior to four markers (i.e., APRI, FIB-4, AST/ALT, and PLT). First, our meta-analysis study revealed similar AUSROC values of APRI and FIB-4 in each fibrosis stage (significant fibrosis: 0.7407 and 0.7844, advanced fibrosis: 0.7347 and 0.8165, and cirrhosis: 0.7268 and 0.8448, respectively) 64 , indicating the reliability of our analysis. Second, we observed the similar result as a previous study, which reported that HA was more efficient than APRI and FIB-4 for fibrosis staging 63,65 . Although studies claimed that FibroScan could offer more promising results than HA for staging both early hepatic fibrosis and cirrhosis 66,67 , we were not able to tell the difference between FibroScan and HA in this study due to limited sample size. Since the accuracy of FibroScan is strongly influenced by disease etiology 68,69 , additional studies stratified by etiology will be necessary to draw the conclusion.

Number of patients
As M2BP has been shown to promote cancer progression, WFA+-M2BP may potentially be used to predict HCC development 18 . In our current study, WFA+-M2BP was superior to APRI and FIB-4 and equivalent to AFP for diagnosing HCC. Currently, whether AFP should be used for routine surveillance of HCC is controversial due to its limited sensitivity in early detection. In our study, WFA+-M2BP had a higher sensitivity (0.77) and a lower specificity (0.80) when it was compared with AFP (sensitivity = 0.61 and specificity = 0.94). This finding indicated the possibility of improving the diagnosis of HCC if WFA+-M2BP and AFP are used together. Besides, it had been reported the posttreatment level of WFA+-M2BP could nicely reflect HCC development for patients underwent anti-viral therapy by reaching a high sensitivity, specificity and AUROC (0.875, 0.939 and 0.973) 49 . As a result, combined diagnostic model or posttreatment detection of WFA+-M2BP might be helpful for the prediction of HCC.
It should be noted that there are limitations in our study: (1) Due to the limited study number, we could not evaluate the predictive accuracy of WFA+-M2BP in mild fibrosis stratified by etiology; (2) This meta-analysis is a pilot study which compared the performance of multiple indicators by analyzing results from literatures on WFA+-M2BP. We suggested future study could be performed to investigate the comparison between WFA+-M2BP and a certain non-invasive indicator for staging liver fibrosis or predicting HCC in a certain disease by analyzing more literatures from multiple databases; (3) More high-quality studies are needed for analysis. Our quality assessment demonstrated that the existence of moderate risk of bias was mainly due to the awareness of reference standard result before conducting the index test. Also, significant heterogeneities existed in studies regarding each stage of liver fibrosis and HCC. And we found that the age of the participants, male proportion, etiology, and certain outlier studies might be the source of these heterogeneities. Besides, we noticed obvious publication bias existed in studies on cirrhosis, although this bias could be diminished after the removal of three outlier studies. (4) We conducted this diagnostic meta-analysis and emphasized the diagnostic performance of WFA+-M2BP in liver fibrosis and HCC, without elaborating on the outcomes of patients with different basal levels of serum WFA+-M2BP.
In conclusion, our meta-analysis demonstrated that WFA+-M2BP could be used as a surrogate biomarker for liver biopsy to diagnose cirrhosis in chronic liver diseases. It showed modest accuracy for identifying early stage of liver fibrosis and HCC. Compared with other non-invasive indicators, the predictive performance of WFA+-M2BP was similar to HA and FibroScan in assessing cirrhosis, but was equivalent to AFP in HCC. As the accuracy of WFA+-M2BP was strongly influenced by the etiology of disease, individual cutoff value was suggested to be applied in certain etiology.

Methods
Literature search strategy. We performed this meta-analysis according to the Cochrane Handbook for Systematic Reviews of Diagnostic Test Accuracy 70 . Literatures published before September 22, 2019 in Pubmed, Web of Science, EMBASE, the Cochrane Library, and grey literature database including OpenGrey (https ://www. openg rey.eu) were searched. The search terms included "Wisteria floribunda agglutinin-positive Mac-2-binding protein or WFA+-M2BP or M2BPGi or Mac-2-binding protein glycosylation isomer" and "fibrosis or cirrhosis" or "hepatocellular carcinoma or liver cancer".
Inclusion and exclusion criteria. Inclusion criteria were: (1) study objects contained patients with liver fibrosis or hepatocellular carcinoma caused by various etiologies; (2) studies used liver biopsy as gold standard to measure the severity of liver fibrosis; (3) studies used typical imaging techniques and/or histopathology to diagnose HCC; (4) studies employed WFA+-M2BP and may also include APRI, FIB-4, aspartate aminotransferase to alanine aminotransferase ratio (AST/ALT), HA, PLT, FibroScan stiffness value or AFP to predict liver fibrosis or HCC; and (5) data on true-positive (TP), true-negative (TN), false-positive (FP), false-negative (FN) results were reported separately or could be calculated from the article.
Studies as follows were excluded: (1) duplicate studies; (2) non-experimental studies such as reviews, letters, clinical trials, correspondences, comments, case reports, and patents; (3) studies published in a language other than English; (4) non-human subjects; or (5) the relevant data were inaccessible or unclear.
Statistical analyses. We calculated 2 × 2 tables and performed QUADAS-2 quality assessment by using Review Manager 5.2 (The Nordic Cochrane Centre, Copenhagen, Denmark). Besides, we employed bivariate random effects model to conduct statistical analysis. Thus, "Midas" module in Stata version 14.2 (StataCorp, College Station, TX) was used to summarize test accuracy: pooled sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR), the area under the summary receiver operating characteristic curve (AUSROC). Z test was adopted to compare AUSROCs of WFA+-M2BP and other indicators. Diagnostic threshold effect would exist if the Spearman correlation coefficient > 0 and P < 0.05 72 . If inconsistency index (I 2 ) ≥ 50% or P < 0.05 was observed, it suggested significant heterogeneity 73 . In that case, we would conduct meta-regression analysis, which evaluated potential factors to determine covariates. In joint model, factors with P < 0.1 was considered the potential source of heterogeneity 74,75 . Furthermore, publication bias was determined by using Deeks's funnel plot, and P < 0.05 indicated possible bias 76 . Finally, we carried out sensitivity analysis to measure the robustness of the summary results.