Clinical diagnostic value of long non-coding RNAs in Colorectal Cancer: A systematic review and meta-analysis

Background: Histopathological diagnosis remains the gold standard for the diagnosis of cancer, including colorectal cancer, but it is infeasible when tumor tissue is not available. With the recognition of long non-coding RNAs (lncRNAs), the expression of lncRNAs in serum or tissue samples has been reported as a diagnosis method for some cancers, however, the diagnostic value of lncRNAs for colorectal cancer remains unclear. Methods: A systematic review and meta-analysis were conducted. Eligible studies were identified through a comprehensive literature search in PubMed, PubMed Central, Web of Science, Embase, and Cochrane Library (up to May 05, 2020) according to the selection criteria. Meta-DiSc, Review Manager and STATA were used to analyze the association between lncRNAs expression and the diagnosis of colorectal cancer. Results: Fifteen studies that analyzed the expression of 15 lncRNAs in 1434 CRC patients were included. The summary area under the curve (AUC) of lncRNA for the diagnosis efficacy between patients with and without CRC was estimated to be 0.8629, corresponding to a weighted sensitivity of 0.75 (95% CI: 0.72 - 0.77), specificity of 0.80 (95%CI: 0.78 - 0.82). Subgroup analysis illustrated that the AUC of blood-based detection of lncRNA showed 0.8820, pooled DOR: 18.57, while tissue-based analysis showed 0.8203, pooled DOR: 10.47. Blood-based tests were then divided into two categories, plasma-based and serum-based lncRNA testing. Results revealed that the AUC of serum-based detection was 0.9077, pooled DOR: 26.64, and plasma-based detection was 0.5000, pooled DOR: 11.80. Conclusions: This meta-analysis indicates that the aberrantly expressed lncRNAs might serve as potential diagnostic biomarkers for CRC patients and blood-based lncRNA analysis is of higher diagnostic accuracy than tissue-based testing. Moreover, serum-based lncRNA testing achieved higher diagnostic efficacy than plasma-based analysis.


Introduction
Based on the GLOBOCAN2018 evaluation criteria for cancer morbidity and mortality by IARC, colorectal cancer (CRC) is still the third most commonly diagnosed malignancy with the second most lethal cancer rate worldwide [1]. According to the National Cancer Statistics released by the National Cancer Center in 2019, the incidence of colorectal cancers (CRC) ranks fourth among malignant tumors and has become the fifth leading cause of cancer mortality in China [2]. Previous studies have indicated that the colorectal cancer patients in the early stage would gain longer survival time with standard treatment [3][4][5]. The prognosis of CRC can be improved when the patients are identified at their Ivyspring International Publisher early stages. Therefore, we are supposed to improve CRC diagnostic strategies, which is as important as discovering new treatments.
Currently, the main biomarkers of colorectal cancer in clinical practice consist of carcinoembryonic antigen (CEA), cancer antigen 19-9 (CA19-9) and cancer antigen 242 (CA242). But the sensitivity and specificity are far from satisfying clinical needs. A previous study has demonstrated that the sensitivity of CEA was 46.59% and its specificity was 80% [6]. Another study using ROC curves to comparing the specificity and sensitivity of CEA and CA19-9 in the early diagnosis of CRC, the AUC of CEA is 0.797 (P < 0.001) and the AUC of CA19-9 is 0.664 (P = 0.001) [7]. There is a pressing need for more ideal biomarkers with higher sensitivity and specificity for early diagnosis of CRC.
Long non-coding RNAs (lncRNAs) are defined as capped transcripts > 200 nucleotides [8]. It is reported that several lncRNAs are aberrantly expressed in tissue or serum from colorectal cancer patients. Some studies have reported that MEG3 [9], lnc-ATB [10] and BLACAT1 [11] were up-expressed in colorectal cancer plasma, while NKILA [7] and HOTAIRM1 [12] were down-regulated in colorectal cancer tissue. A sensitivity of 37.16% and a specificity of 88.75% for CEA were significantly lower than that for exosomal CRNDE-h (P < 0.001) [13]. NKILA exhibited relatively higher sensitivity and specificity compared with CEA and CA19-9 in the early diagnosis of CRC [7]. And LINC02418 was also more sensitive than those existing detection methods [14].
Studies based on LncRNAs detection are constantly springing up; however, the diagnostic value of lncRNAs for colorectal cancer remains unclear. So we performed a meta-analysis to evaluate the relationship between clinical outcomes (clinicopathological parameters, diagnosis) and the expression of lncRNAs in patients with colorectal cancer, hoping to provide a theoretical basis for clinical application.

Literature Search
Two of the authors (Bi Chen and Ruonan Zhang) each searched several databases, including PubMed, PubMed Central (PMC), Web of Science, Embase, and Cochrane Library. The publication data used in the literature search was from database inception to May 05, 2020. The search strategies were based on combinations of the following keywords in titles or abstracts: (lncRNA OR long ncRNA OR lincRNA OR long non-coding RNA OR long non-translated RNA OR long untranslated RNA OR long non-protein-coding RNA OR long intergenic non-protein coding RNA) and (colorectal cancer OR colorectal neoplasm OR colorectal tumor OR colorectal carcinoma) and ("diagnose" or "diagnosis").

Selection Criteria
All included studies met the following inclusion criteria: 1) patients were diagnosed with CRC by histopathology and without other tumors; 2) volunteers were defined as no tumor and other diseases; 3) the expression level of lncRNAs was identified and analyzed for the diagnosis of CRC, and; 4) studies contained sufficient data, including sensitivity (SEN), specificity (SPE), area under the curve (AUC) or receiver operating characteristic (ROC). If unavailable, related data obtained by contacting the corresponding authors.

Data Extraction
Two independent authors (Bi Chen and Ruonan Zhang) collected the following information from the literature for each publication: (1) basic information: first author, publication year, type of lncRNA, expression, detection sample, test method, cut-off value, number of patients; (2) clinicopathological information: P values of age, gender, tumor location, tumor size, differentiation, lymphatic metastasis, distal metastasis, TNM stage, CEA level, CA 19-9 level and depth of invasion; (3) diagnosis information: SEN, SPE, AUC, ROC; (4) Engauge Digitizer software was utilized to obtain the SEN and SPE from ROC when data was not reported directly.

Quality Assessment
Study quality was assessed with the Quality Assessment of Diagnostic Accuracy Studies II (QUADAS-2) checklist, which is recommended for use in systematic reviews and applicability of primary diagnostic accuracy studies, consisting of four key domains covering patient selection, index test, reference standard, flow and timing. Each is assessed in terms of risk of bias and the first three in terms of concerns regarding applicability [15].

Statistical Analysis
The Meta-DiSc software (Version 1.4) was used to calculate the combined sensitivity, combined specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), overall diagnostic odds ratio (DOR), diagnostic advantage and 95% confidence interval for each study [16]. Review Manager (Version 5.3. Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014) and STATA 15.0 software (StataCorp, College Station, TX 77845, USA) were used to analyze study data and construct the forest plot. Heterogeneity among studies was assessed by the Cochran's Q and the I 2 statistic. The fixed-effects model was used if there was no substantial heterogeneity (P > 0.10 or I 2 < 50%). Inversely, the random-effects model was chosen when significant heterogeneity was observed [17][18][19]. Associations between lncRNA expression and clinicopathologic parameters were determined using the P values combined with Fisher's test. Publication bias was quantitatively judged by Deeks' funnel plot asymmetry test [20]. P < 0.05 was considered to be statistically significant.

Study Selection
We searched and captured 200 records in PubMed, PMC, EMBASE, Web of Science, and the Cochrane Library. Of these, 52 duplicate studies were excluded. We excluded 123 records after reading the titles and abstracts. Subsequently, we assessed the remaining 25 full-text articles and excluded 10 studies based on the exclusion criteria, including 4 reviews & updates, 5 lacking key data, and 1 uncorrected proof version. A total of 15 studies were ultimately included in this study. A flow diagram of the selection process for this study is presented in Figure 1.

Study parameters and study quality
Eighty-five percent of the selected studies were from China, with 11/15 (73.3%) being published between 2018 and 2020. A total of 1434 CRC patients were included and the number of patients ranged from 34 to 174. All patients were diagnosed based on the histopathological diagnostic criteria. The tissue samples and blood samples (serum and plasma samples) were obtained prior to clinical treatment. The expression levels of LncRNAs were determined using quantitative real-time polymerase chain reaction (qRT-PCR). Seven types of lncRNAs were recognized as tumor promoters [10,11,13,14,[21][22][23][24][25] and six were tumor suppressors [7,9,[26][27][28][29] in CRC patients. The associations between lncRNAs and clinicopathologic parameters in CRC patients are shown in Table 1. Diagnostic accuracy differed greatly between different lncRNAs. We found that ENST00000455974 detection had the highest sensitivity (95.6%) with a specificity of 81.2% [23] and CRNDE-h detection had the highest specificity (91.5%) with a sensitivity of 70.3% [13].

QUADAS-2
checklist was used to systematically assess the quality of all the included studies. The results indicated that all included studies were of high methodological quality. The results are shown in Figures 2-3.

Meta-analysis of clinicopathological parameters
The P values between different lncRNAs and clinicopathological parameters in CRC patients of all included studies are summarized in Table 2. Some clinicopathological parameters were not observed or reported in some studies, so the relevant data were not available. Altered expression of lncRNAs were significantly associated with some clinicopathological parameters we collected (tumor size: pooled P < 0.0001; lymphatic metastasis: pooled P < 0.0001; TNM stage: pooled P < 0.0001; levels of CEA: pooled P = 0.0007). But, it was not correlated with age (pooled P = 0.5507), gender (pooled P = 0.2558), or tumor location (pooled P = 0.7519).

Diagnostic performance
Forest plots of the pooled sensitivity, pooled specificity, pooled DOR and sROC curve of lncRNAs in diagnosing CRC are shown in

Meta-regression analysis
Due to the significant heterogeneity between these studies observed in sensitivity and specificity data (I 2 = 87.50% and I 2 = 85.80%, respectively), we firstly performed an analysis of diagnostic threshold showed in Table 3 to reveal no threshold effect in the studies (Spearman correlation coefficient: 0.314, P = 0.254). Then we constructed a meta-regression analysis in terms of the specified covariates including sample size, sample types and sample expression of lncRNAs. According to the P value from large to small, "size" and "expression" were eliminated one by one, and meta-regression analysis was performed. The results in Tables 4-6 showed that the heterogeneity might be associated with the sample type (RDOR = 1.69, 95%CI: 1.05 -2.72, P = 0.0339). The detection accuracy of blood samples was 1.18 times higher than that of non-blood samples, as shown in Table 6.      Therefore, we further analyzed the effect of sample type on diagnosis ( Figure 5). A random-effects model was applied because there was significant heterogeneity (I 2 = 63.6% and I 2 = 80.6%, respectively). The stratified analysis showed that the performance of blood-based detection was significantly superior to that of tissue-based detection (AUC: 0.8820 vs. 0.8203; pooled DOR: 18.57 vs. 10.47). Blood-based tests were then divided into two categories, plasma-based and serum-based lncRNA testing. The results in Figure 6 revealed that the AUC of serum-based detection was 0.9077, pooled DOR: 26.64, and plasma-based detection was 0.5000, pooled DOR: 11.80.

Publication bias
Deeks' funnel plot asymmetry test was performed to check publication bias in this meta-analysis. The result presented in Fig. 7 showed no evidence of publication bias (P = 0.95) existed for diagnostic analyses.

Discussion
Colorectal cancer is a major cause of cancer mortality worldwide. Early diagnosis greatly increases the chances of successful treatment and improves the cancer outcomes, potentially reducing mortality from cancer [30]. LncRNAs are characterized as a group of endogenous RNAs that have no protein-encoding function [31,32]. Recent studies have suggested that the expression of lncRNAs in CRC might be involved in cancer development, invasion, metastasis, prognosis [33][34][35][36]. Statistically significant differences in the expression of lncRNAs between precancerous lesions and early-stage cancer have been found in previous studies [37][38][39]. However, there is no relevant systematic review and meta-analyses focused on the expression of lncRNAs in CRC diagnosis. In this meta-analysis, we systematically analyzed the relationship between the expression of lncRNAs and the early diagnosis of CRC.    The included studies were mainly published between 2016 to 2020 and deemed to be of high quality according to the QUADAS-2. The results of our meta-analysis showed that the expression of LncRNAs was significantly associated with tumor size, differentiation, TNM stage, metastasis, levels of CA19-9 and CEA. It indicates that the dysregulated expression of lncRNAs is implicated in the progression of CRC and might act as potential biomarkers for the early diagnosis of CRC.
The ROC curve is a comprehensive index, which reflects the sensitivity and specificity of continuous variables. The pooled AUC of lncRNAs indicated that 86.29% of randomly chosen CRC patients had higher or lower levels of lncRNAs than normal controls. The pooled DOR is also an important indicator that facilitates formal meta-analysis of studies on diagnostic test performance. In the present study, a pooled DOR of 14.20 (higher than 1.0) was obtained, suggesting that detection of lncRNA is a powerful predictive biomarker for CRC diagnosis.
Fifteen types of lncRNAs with a different expression status in CRC were included in this review. And the meta-regression test further showed that different kind of samples for lncRNA detection was probably the source of heterogeneity (P = 0.0339, RDOR = 1.69, 95%CI: 1.05 -2.72). Therefore we conducted subgroup analyses to compare the difference. Stratified analyses based on sample type showed that blood-based lncRNA analysis was of higher diagnostic efficacy than tissue-based analysis (AUC: 0.8820 vs. 0.8203; pooled DOR: 18.57 vs. 10.47). Moreover, serum-based lncRNA testing achieved higher diagnostic efficacy than plasma-based analysis (AUC: 0.9077 vs. 0.5000; pooled DOR: 26.64 vs.11.80), suggesting that serum-based detection is more recommended.
The Deeks' funnel plot asymmetry test didn't reveal obvious publication bias for the diagnostic meta-analyses, suggesting that results from pooled data analysis were reliable. When taking into consideration the stable structures and easy detection, lncRNAs have the potentials to serve as novel, easily attainable biomarkers for the diagnosis of CRC.
Several limitations should be acknowledged in this systematic review and meta-analysis. Firstly, only 15 types of lncRNAs were included and they were analyzed based on different quality control standards, which might have generated potential heterogeneity. Secondly, the small sample size in sub-group analyses led to low statistical power and undermined the strength of the evidence of this systematic review. Thirdly, the majority of patients included in our study were Asians, potential ethnic-related differences in the expression of lncRNAs might restrict the applicability of our findings to other races. Finally, the clinical value of lncRNAs in CRC patients might have been exaggerated because the studies with positive results were more likely to be published than those with negative results.

Conclusion
In summary, this is the first meta-analysis to evaluate the clinical value of lncRNAs in the diagnosis of colorectal cancer. The results of our meta-analysis reveal that lncRNAs are promising diagnostic biomarkers in patients with CRC. Blood-based lncRNA analysis is of higher diagnostic accuracy than tissue-based testing. Moreover, serum-based lncRNA testing shows higher diagnostic efficacy than plasmabased analysis. However, considering the mentioned limitations above, a larger number of clinical trials are still needed to further verify the findings and confirm the clinical application of lncRNAs in the diagnosis of CRC. Further prospective studies on the role of each specific lncRNA in the early diagnosis of CRC are also warranted in the future.