Prognostic value of Lin28A and Lin28B in various human malignancies: a systematic review and meta-analysis

Background The mammalian homologs of Lin-28, Lin28 (also called Lin28A) and Lin28B, are promising cancer biomarkers. This meta-analysis was performed to evaluate the prognostic values of Lin28A and Lin28B in multiple human malignancies. Methods Systematic searches of the PubMed, Web of Science and Embase were used to identify relevant studies. Pooled hazard ratios (HRs) with 95% confidence intervals (CI) for overall survival (OS), recurrence-free survival (RFS), disease-free survival (DFS), or progression-free survival (PFS) were respectively calculated. Results 3772 Lin28A-associated patients and 1730 Lin28B-related cases were ultimately enrolled in this meta-analysis. The elevated expression level of Lin28A was significantly associated with poor OS (HR = 1.60, P < 0.001) and poor RFS/DFS/PFS (HR = 1.62, P < 0.001) in patients with malignancies. Lin28B overexpression significantly correlated with unfavorable OS (HR = 1.72, P < 0.001) and RFS/DFS/PFS (HR = 2.35, P < 0.001) of human malignancies. Conclusions Lin28A and Lin28B possess significant prognostic values in various human malignancies. Overexpression of Lin28A or Lin28B suggests poor prognosis for cancer patients. Electronic supplementary material The online version of this article (10.1186/s12935-019-0788-z) contains supplementary material, which is available to authorized users.


Background
In recent years, cancer has become the primary cause of mortality in most countries and regions, and the incidence of human malignancies has increased substantially [1]. Given the low survival rate of multiple cancer types, credible biomarkers for cancer prognosis are urgently required. Recently, the mammalian homologs of Lin-28, Lin28 (also called Lin28A) and Lin28B, have been considered as promising biomarkers.
Lin28A and Lin28B are highly conserved RNA-binding and microRNA-regulated proteins. In general, they selectively block the expression of let-7 microRNA family members, which act as tumor suppressors by inhibiting the expression of oncogenes and key regulators of mitogenic pathways, including RAS, MYC, and HMGA2 [2]. Lin28A recruits TUTase to inhibit let-7 precursors to block Dicer processing in the cell cytoplasm, whereas Lin28B represses let-7 maturation through a TUTaseindependent mechanism [3]. Many studies indicated that both Lin28A and Lin28B show upregulated expression in human malignancies and that they function as oncogenes by promoting transformation and tumor progression [4,5].

Search strategy
We carefully performed a comprehensive search on Pub-Med, Web of Science, and Embase until June 2017 to identify relevant potential literatures. The following set of keywords was simultaneously applied for the online study search: "Lin28" or "Lin28" or "Lin28A" or "Lin28B" and "tumor" or "tumour" or "cancer" or "carcinoma" or "neoplasm" or "malignancies. "

Inclusion and exclusion criteria
This meta-analysis was strictly conducted in accordance with the guidelines of the preferred reporting items of systematic reviews and meta-analyses (PRISMA) statement [27]. Studies were considered eligible if they met the following criteria: (i) assayed the expression level of Lin28A or Lin28B from human tumor samples; (ii) stratified the expression of Lin28A or Lin28B; and (iii) investigated the association of Lin28A or Lin28B expression levels with cancer overall survival (OS) or progression, along with a corresponding hazard ratio (HR) or survival curves. If more than one article had been published on the same study cohort, only the most comprehensive study was included in this meta-analysis. Moreover, the following criteria were also considered: (i) publications in English, (ii) researches on human malignancies, and (iii) studies on the association of Lin28A or Lin28B expression with cancer prognosis. Furthermore, articles were excluded when they did not cover the points above. Letters, case reports, review articles, and experiments on animals alone were excluded. A flow diagram of the selection process with further details is shown in Fig. 1.

Data extraction
The extracted data included the following elements: (i) first author and publication year; (ii) characteristics of the studied population, including patients' nationality, region, mean or median age, study size, disease type, and examined type of sample; (iii) cut-off definition and method of sample analysis; (iv) source of HR, mean, or median follow-up duration; and (v) HRs of elevated Lin28A or Lin28B expression for OS, recurrence-free survival (RFS), disease-free survival (DFS), or progressionfree survival (PFS). If HRs were not directly reported, then the survival data were extracted from Kaplan-Meier plots by using Engauge Digitizer V.5.1 (license type: GPL; developed by: Mark Mitch; Category: C:\Science/CAD). Furthermore, HRs with 95% confidence intervals (CIs) were calculated using practical methods by Stata V.12.0 (StataCorp LP, College Station, Texas, USA) [28]. All data above are comprehensively detailed in Tables 1, 2, Additional file 1: Table S1 and Additional file 2: Table S2.

Statistical analysis
In this meta-analysis, HRs and corresponding 95% CIs were combined to estimate the prognostic values of high Lin28A or Lin28B expression for cancer prognosis. An individual or pooled HR of more than 1.0 indicated poor prognosis for patients with Lin28A or Lin28B overexpression, and an HR of less than 1.0 represented better prognosis. Moreover, a fixed-effects model or a randomeffects model was applied for meta-analysis on the basis of the heterogeneity among the pooled studies [29]. Heterogeneity among studies was evaluated by Chi-square test (assessing the P value) and Higgins I 2 statistic. Sensitivity analyses, including influence analysis and Galbraith plot, were implemented for individual studies to identify the source of heterogeneity. If statistically significant heterogeneity was observed (P < 0.10 or I 2 > 50%), the random-effects model was applied to estimate the pooled HR; otherwise, the fixed-effects model was used. If necessary, we also classified the included studies into subgroups on the basis of similar characteristics to minimize the influence of significant heterogeneity. In addition, publication bias was estimated using Egger's test with funnel plot. Stata version 12.0 (Stata Corporation, College Station, TX, USA) was used to calculate all statistical analyses. All P values were two sided, and a P value of less than 0.05 was considered statistically significant.
The main characteristics and data of the included studies are summarized in Tables 1, 2. All of 23 eligible studies were retrospective observations in the meta-analysis.

Correlation between Lin28A overexpression and OS
Statistically significant heterogeneity was found among 15 studies for Lin28A overexpression and OS (I 2 = 93.7%, P < 0.001) (Additional file 3: Figure S1A). Therefore, influence analysis and Galbraith plot of individual studies were applied for the source of heterogeneity, and the results suggested that two studies of relative low quality (The study of Faria et al. determined the Lin28 protein expression by LIN28 rabbit polyclonal antibody, which recognizes both Lin28A and Lin28B [23], meanwhile the study of Korshunov et al. enrolled ETMR minor patients aged under 18 years [26]) which might lead to the instability and heterogeneity of the pooled outcome (Additional file 4: Figure S2A, B). Meta-analysis and sensitivity analyses were repeated after excluding the two studies (Additional file 4: Figure S2C, D), and the results showed no heterogeneity (I 2 = 0.0%, P = 0.510). The pooled HR value of Lin28A overexpression for OS was 1.60 (95% CI 1.38, 1.86) (P < 0.001), as detected by the fixed-effects model (Fig. 2a). Thus, we considered that the elevated expression level of Lin28A was significantly associated with poor OS in patients with malignancies.

Correlation between Lin28B overexpression and OS
Among the 10 studies that evaluated the prognostic value of Lin28B overexpression for OS, statistically  Figure S4A). We applied influence analysis and Galbraith plot of individual studies to identify certain studies that might result in the instability of the pooled outcome (Additional file 7: Figure S5). No heterogeneity remained when three studies [5,8,20] were excluded (I 2 = 0.00%, P = 0.635), and the pooled HR detected by the fixed-effects model demonstrated that high Lin28B expression significantly correlated with poor OS of human malignancies (HR = 1.72, 95% CI 1.43, 2.08) (P < 0.001) (Fig. 3a).

Correlation between Lin28B overexpression and RFS/DFS/ PFS
Significant heterogeneity was evident for the seven studies on Lin28B overexpression and RFS/DFS/PFS (I 2 = 83.2%, P < 0.001) (Additional file 6: Figure S4B), and one low-quality study (No detailed information was shown about the anti-body for IHC, and no description of the case number for high Lin28B expression, and for low Lin28B expression was reported [8]) was excluded by influence analysis and Galbraith plot (Additional file 8: Figure S6). The pooled HR of Lin28B overexpression for RFS/DFS/PFS was 2.35 (95% CI 1.61, 3.43), as detected by the random-effects model, and significant heterogeneity was retained among the studies (I 2 = 69.2%, P = 0.006) (Fig. 3b). Therefore, subgroup analyses were correspondingly performed on the basis of similar characteristics, such as nationality, region, and malignant disease. No heterogeneity was found in the subgroups of Chinese Taiwan (I 2 = 0.0%, P = 0.667) and EOC (I 2 = 0.0%, P = 0.806) (Additional file 9: Figure S7). By contrast, significant heterogeneity was found in the subgroups of Asia (I 2 = 64.9%, P = 0.036) and North America (I 2 = 70.6%, P = 0.065) (Additional file 10: Figure S8).

Publication bias
Egger's test with funnel plots were used to evaluate the publication bias among the eligible studies. In the pooled analyses of studies on Lin28A expression level with OS and RFS/DFS/PFS, the P values for Egger's test were 0.069 and 0.239. Begg's test reached P values of 0.127 and 0.266, respectively, suggesting no significant publication bias (Fig. 4a, b). Because there were less than 10 studies in the pooled analyses of Lin28B expression level with OS and RFS/DFS/PFS, we did not assess the publication bias according to Cochrane Guidelines [30].

Discussion
Lin28A and its homolog Lin28B belong to highly conserved RNA-binding proteins family, which are found to involve in numerous biological processes, including cell development, pluripotency, reprogramming, and oncogenesis [31,32]. Several recent investigations have confirmed that Lin28A and Lin28B can regulate gene expression either by directly binding to messenger RNAs (mRNAs) or by blocking microRNA biogenesis, and the underlying mechanisms include Let-7-family-dependent and Let-7-family-independent modes of action [3]. Let-7 and lin28 (Lin28A and Lin28B) were first identified through mutagenesis screening as heterochronic genes in Caenorhabditis elegans [33,34], and the expression and regulation of lin28 and let-7 are highly conserved throughout evolution [35]. Recent studies have established the lin28/let-7 pathway as a central regulator of mammalian glucose metabolism [36]. Although the exact roles of the let-7 family in adult mammalian tissues have not been definitely characterized, let-7 is known to serve as a tumor suppressor. Considerable evidence demonstrated that let-7 expression level is downregulated in various cancers and that let-7 overexpression restrains the growth and metastatic potential of cancer cells [37][38][39][40]. As one targeting gene of let-7, lin28 expression is upregulated in various tumors, such as OSCC [6,7], colon cancer [8,9], EOC [4,10,11], gastric cancer [12][13][14], HCC [15][16][17], and breast cancer [18,19]. Although many major effects of lin28 are mediated by blocking let-7 microRNA biogenesis, lin28 can also directly bind to GGAGA (G, guanosine; A, adenosine) sequences enriched with loop structures in mRNA targets, and this activity is similar to its interaction with let-7 microRNA precursors [41]. Many of these targets function as oncogenes or tumor growth indicators, such as insulin-like growth factor 2 (IGF-2), a crucial growth and differentiation factor for muscle tissues [42], and IGF-2 mRNA-binding protein 2, which binds several mRNAs encoding mitochondrial respiratory chain complex subunits [43]. Therefore, Lin28A and Lin28B, along with their target genes, deserve further analysis in the future.
In this meta-analysis, we investigated the prognostic values of Lin28A and Lin28B for multiple human malignancies. Combining the outcomes of studies regarding the association of lin28 expression and tumor prognosis, we have successfully drawn many valuable results. First, increased expression level of Lin28A was considered to predict poor OS and RFS/DFS/PFS for cancer patients, with combined HR values of 1.60 (95% CI 1.38, 1.86) and 1.62 (95% CI 1.33, 1.97), respectively. Second, considering the pooled outcomes of studies on the relation between Lin28B expression and cancer prognosis, we found that elevated Lin28B level was significantly associated with poor OS and RFS/DFS/ PFS of malignant diseases, with pooled HR values of 1.72 (95% CI 1.43, 2.08) and 2.35 (95% CI 1.61, 3.43), correspondingly, which also exerted statistical significance. Our findings suggested that Lin28A and Lin28B are promising biomarkers, and the detection of Lin28A and Lin28B expression in cancer patients is of potential value for monitoring patients' survival.
However, significant heterogeneity was observed in the initial meta-analysis (Additional file 3: Figure S1 and Additional file 6: Figure S4). Therefore, influence analysis and Galbraith plot were applied for individual studies to investigate the source of heterogeneity. After excluding several studies of relatively low quality, no heterogeneity was found among the studies on cancer prognosis and lin28 expression, except for those concerning the association between Lin28B expression and RFS/DFS/ PFS. Furthermore, subgroup analyses were performed to minimize the influence of heterogeneity (I 2 = 69.2%, P = 0.006). In subgroup analyses, no heterogeneity was observed in the subtotals of Chinese Taiwan (I 2 = 0.0%, P = 0.667) and EOC (I 2 = 0.0%, P = 0.806). Also, no heterogeneity was shown in the subgroup analyses by different assay methods for Lin28A overexpression and OS, and RFS/DFS/PFS (Additional file 11: Figure S9A, B), and for Lin28B overexpression and OS (Additional file 11: Figure S9C). However, the subgroup of IHC still indicated of high heterogeneity for Lin28B overexpression and RFS/DFS/PFS (I 2 = 75.8%, P = 0.016) (Additional file 11: Figure S9D). Besides, the subgroups of Asia (I 2 = 64.9%, P = 0.036) and North America (I 2 = 70.6%, P = 0.065) showed significant heterogeneity (Additional file 10: Figure S8). Therefore, we consider that the source of heterogeneity might result from the influence of different populations and disease types of the patients, rather than different methods.
Despite the meta-analysis was performed with rigorous statistics, our conclusion still has several limitations for the following reasons. First, the amounts of included studies in the meta-analysis was not sufficiently enough for more powerful results, as well as the study numbers for each cancer type. Second, all eligible studies were retrospective for analysis, which might impair the credibility of meta-analysis. Third, the population diversity, disease type, source of tissue sample and antibody for IHC may cause heterogeneity to a certain extent. Moreover, many researchers used a median immunohistochemical score as a division value, but the median scores slightly differed among studies. Several studies also performed a ternary method to classify the expression levels of Lin28A into high, medium, and low categories, which might result in an undervalued HR for Lin28A [19]. All the disadvantages above might cause heterogeneity in the meta-analysis and produce deviation when evaluating the prognostic significances of Lin28A and Lin28B in human malignancies. Also, HR values of several studies were calculated using data extracted from the survival curves, which might unavoidably cause slight statistical errors.

Conclusion
To summarize, Lin28A and Lin28B provide significant prognostic values in various human malignancies. Overexpression of Lin28A or Lin28B suggests poor prognosis for cancer patients. In consideration of the complicated regulatory mechanism between lin28 and its target genes, further investigation and additional relevant studies are required to establish the clinical significance of Lin28A and Lin28B as ideal prognostic biomarkers.