Prognostic value of circulating tumor DNA in pancreatic cancer: a systematic review and meta-analysis

Increasing evidence has revealed the potential correlation between circulating tumor DNA (ctDNA) and the prognosis of pancreatic cancer, but inconsistent findings have been reported. Therefore, a meta-analysis was performed to evaluate the prognostic value of ctDNA in pancreatic cancer. The Embase, MEDLINE, and Web of Science databases were searched for relevant articles published until April 2020. Articles reporting the correlation between ctDNA and the prognosis of pancreatic cancer were identified through database searches. The pooled hazard ratios (HRs) for prognostic data were calculated and analyzed using Stata software. A total of 2326 patients pooled from 25 eligible studies were included in the meta-analysis to evaluate the prognostic value of ctDNA in pancreatic cancer. Patients with mutations detected or high concentrations of ctDNA had a significantly poorer overall survival (OS) (univariate: HR = 2.54; 95% CI, 2.05-3.14; multivariate: HR = 2.07; 95% CI, 1.69-2.54) and progression-free survival (PFS) (univariate: HR = 2.18; 95% CI, 1.41-3.37; multivariate: HR = 2.20; 95% CI, 1.38-3.52). In conclusion, the present meta-analysis indicates that mutations detected or high concentrations of ctDNA are significant predictors of OS and PFS in patients with pancreatic cancer.


INTRODUCTION
Pancreatic cancer (PC) is one of the most lethal malignancies [1], with 47,050 estimated deaths in the United States in 2020 [2]. It is predicted to become the second leading cause of cancer-related deaths within the next decade, while the mortalities of other gastrointestinal cancers have declined over time [2,3]. The poor outcomes of PC patients are attributed to diagnostic difficulty, early metastatic tendency and therapy resistance [4,5]. Although the diagnostic and therapeutic strategies are constantly advancing, the mortality of PC patients remains high. Hence, for PC patients, valid biomarkers with simple test methods are urgently needed to predict survival outcomes, monitor disease changes, assess treatment responses and optimize treatment plans.
Liquid biopsy is a new technology of noninvasive detection that obtains biomarkers from nonsolid biological tissues such as blood [6], and it has received increasing attention in cancer research due to its convenience and noninvasiveness [7,8]. Through the analysis of obtained markers, information from the source tissues can be reflected and cancers or other diseases can be monitored. For example, several recent studies have successfully applied liquid biopsy to achieve diagnosis and staging of PC [9,10]. One of the AGING biomarkers of liquid biopsy, circulating tumor DNA (ctDNA) arises from the apoptosis, necrosis, and lysis of circulating tumor cells, as well as active secretion from tumors [11,12]. With the improvement of detection methods, ctDNA has been proposed as a biomarker for early diagnosis, prognostic stratification, disease monitoring and treatment response assessment in a variety of cancers [6,[13][14][15].
A growing body of studies has revealed the potential prognostic value of ctDNA in PC in recent years [16,17]. The mutations and concentration levels of ctDNA may be associated with overall survival (OS) and progressionfree survival (PFS) in PC patients. Our previous studies demonstrated that K-ras mutations and ERBB2 exon17 mutations detected in ctDNA are significantly associated with OS [18][19][20]. However, inconsistent findings have been reported [21,22]. Systematic and significant evidence that validates the prognostic value of ctDNA in PC is lacking. Therefore, the present systematic review and meta-analysis was performed to clarify the relationship between ctDNA and the prognosis of PC patients to provide more reliable evidence that may improve the management and treatment of PC.

Literature screening and study characteristics
The flow chart of the study selection criteria is presented in Figure 1. We initially identified 1233 studies from the Embase, MEDLINE, and Web of Science databases. AGING Among these studies, 1175 were excluded after screening the titles and abstracts to remove duplicates or irrelevant studies. A total of 33 studies were excluded according to the inclusion and exclusion criteria. Thus, a total of 25 eligible studies that included 2326 patients were enrolled in this meta-analysis . The characteristics of the included studies are summarized in Table 1. All of the included studies had OS outcomes, while 5 of them had PFS prognostic data [21,28,32,35,36]. Seven studies included patients with PC of all stages (stage: I -IV) [23-26, 29, 32, 33], 8 studies included patients with resectable PC (stage: I -II) [21,22,27,31,35,37,38,42], and 12 studies included patients with advanced PC (stage: III -IV) [18-20, 22, 28, 30, 31, 34, 36, 39-41]. All of the included studies sampled at baseline, and 4 of them had postoperative data [21,31,35,38]. For the quality assessment according to the Newcastle-Ottawa Scale (NOS), all of the included studies obtained a score greater than 6 (Supplementary Table 1).

ctDNA as a prognostic factor of PC
A total of 24 studies that included 34 sets of samples were available for OS analyzed by univariate analysis and were pooled for meta-analysis. The meta-analysis results for ctDNA and the prognosis of PC patients are shown in Figure 2A. The results indicated that patients with mutations detected and high concentrations of ctDNA had a significantly poorer OS (HR = 2.54; 95% CI, 2.05-3.14). As shown in Figure 2B, for the multivariate analysis, 21 studies that included 28 sets of samples were analyzed and showed significant results (HR = 2.07; 95% CI, 1.69-2.54). Because of the significant heterogeneity (I 2 = 87.7% and I 2 = 85.8%), a random-effects model was applied. As shown in Figure 3A, for PFS, 5 studies that included 8 sets of samples were analyzed via univariate analysis, and the results showed significant associations between mutations detected and high concentrations of ctDNA and the PFS of PC patients (HR = 2.18; 95% CI, 1.41-3.37, I 2 = 65.9%). Similar results ( Figure 3B) were found when analyzed by multivariate analysis (HR = 2.20; 95% CI, 1.38-3.52, I 2 = 61.5%).

Publication bias
The publication bias of the included studies was assessed by Egger's and Begg's tests in this metaanalysis. Significant publication bias was found in both the univariate and multivariate OS analyses according to the results of Egger's and Begg's tests ( Figure 4A, 4B). Thus, we performed a trim and fill analysis, and the pooled HRs calculated by the trim and fill analysis still showed a significant effect of ctDNA in both the univariate OS analysis (HR = 2.05; 95% CI, 1.68-2.49) and multivariate OS analysis (HR = 1.74; 95% CI, 1.43-2.12). The funnel plot for the trim and fill analysis is shown in Figure 5A, 5B.

DISCUSSION
PC is well known to have a poor prognosis, of which the five-year survival rate is approximately 10%, and only 15-25% of PC patients are resectable [4,43]. Patients in different stages are suggested to undergo different treatments. For advanced PC, patients are usually suggested to undergo chemotherapy with combination regimens, but therapy resistance is a characteristic of PC [44,45]. For resectable PC, extensive surgery with adjunctive chemotherapy is recommended; however, many resectable PC patients experience recurrence [46]. Thus, many studies have largely focused on finding effective prognostic factors. Increasing evidence has revealed the potential prognostic value of ctDNA in PC, although many studies validated their results using small sample sizes. However, inconsistent findings have been reported. Therefore, this systematic review and metaanalysis was performed to assess the potential prognostic value of ctDNA in PC.
A total of 2326 patients pooled from 25 eligible studies were included in this meta-analysis. The sample sizes for In terms of heterogeneity, the I 2 value decreased to approximately 50.0% in most subgroups, which indicated that some of the sources of heterogeneity were the differences in tumor stages, sampling times, and detection markers.
As shown in the results of the subgroup analysis by detection methods, the I 2 value decreased to approximately 60% in the droplet digital PCR (ddPCR) subgroups, which indicated that the differences in detection methods might be another potential source of heterogeneity. The technologies for ctDNA analysis are based on next-generation sequencing (NGS) or polymerase chain reaction (PCR). The methods based on PCR, such as digital PCR (dPCR), ddPCR and peptide nucleic acid-clamp PCR (PNA-clamp PCR), are more sensitive, quick, and inexpensive, but they can analyze only some specific loci once a time [14]. The methods based on NGS can analyze more loci and even perform whole-exome sequencing, but their sensitivity is lower than that of PCR-based methods. Thus, some researchers coupled the two technologies to reduce this limitation [30]. Before detection, the procedures of sample collection and ctDNA extraction are equally important [47]. However, standardized protocols for the collection, extraction and detection of ctDNA are lacking, which is one of the most important reasons why ctDNA has not been widely used in the clinic.  Most of the studies included in this meta-analysis focused on the mutation detection of ctDNA, especially K-ras mutations. K-ras, which encodes a small GTPase that modulates downstream signaling from growth factor receptors, is one of the major driver genes in PC and plays an important role in tumor progression [48]. Although the work of Nakano et al. [21] and Watanabe et al. [22] reported inconsistent findings, our analysis still shows that the detection of K-ras mutations in ctDNA had significant prognostic value in PC. In our view, we speculate that the reasons for the inconsistent results may be related to the use of different sources of samples, including plasma and serum; the inclusion of patients with pancreatic cancer in different stages that have received neo-adjuvant chemoradiotherapy or not; and the inclusion of patients with demographic differences. Several studies have reported that a type of K-ras mutations, the K-ras G12V mutation, is an independent prognostic factor in PC [19,25,35]. Cheng et al. [20] and Guo et al. [42] showed that the K-ras G12D mutation also has a significant prognostic value, although Kinugasa et al. [25] suggested that there is no significant association between the K-ras G12D mutation and OS in PC patients. According to the results of this meta-analysis, we considered that the detection of the K-ras G12D mutation is a significant prognostic factor in PC (HR = 1.77; 95% CI, 1.22-2.58, I 2 = 55.6%), although more studies are needed to further confirm this finding. On the other hand, 4 studies focused on the concentration level of ctDNA [28,30,40,41].

AGING
Although only 4 studies were included in this subgroup analysis, the heterogeneity among these studies was significantly high. We hold the opinion that in terms of concentration detection, different detection methods and cut-off values will have a greater impact on the results, which leads to high heterogeneity. Thus, more highquality studies are needed to test the prognostic value of the concentration levels of ctDNA. In addition to these studies, Henriksen et al. reported that cell-free DNA promoter hypermethylation in plasma was associated with poorer OS [49]. We did not include this study in the meta-analysis because it studied cell-free DNA instead of ctDNA, and the detection methods between hypermethylation and mutation were considerably different. However, this study offers a potential idea and basis for further research on ctDNA. AGING In addition to one-point assessments of ctDNA, Watanabe et al. revealed that the longitudinal monitoring of ctDNA can also predict the prognosis of PC [22]. The emergence of ctDNA in longitudinal tests was associated with OS in both patients who underwent surgery (HR = 54.50; 95% CI, 6.64-447.60) and those who did not undergo surgery (HR = 10.40; 95% CI, 2.95-37.00). One of our recent studies reported that the detection of the K-ras G12V mutation in ctDNA correlated with circulating Treg cells can further stratify the prognosis in patients with advanced PC [20]. Patients with both the K-ras G12V mutation and high Treg cell levels had the worst survival. These studies suggested the great potential value in the innovative application of ctDNA for analyzing prognosis.

AGING
The prognostic value of ctDNA is not limited to OS and PFS. Lee et al. reported that ctDNA is associated with recurrence-free survival (RFS) [38]. For patients with resectable PC, postoperative ctDNA detection is a significant predictor of recurrence (HR = 6.30; 95% CI, 2. 40-16.20). Guo et al. also found that the detection of the K-ras G12D in ctDNA is an independent prognostic factor of recurrence (HR = 5.10; 95% CI, 2.20-11.80) [42]. The effective prediction of recurrence plays an important guiding role in the selection of clinical treatment strategies, which allows for more aggressive screening and treatment to prevent recurrence.
Several limitations in the current meta-analysis should be considered. First, the methodological and clinical characteristics of the eligible studies were not quite consistent, which may cause heterogeneity in the overall results. Second, the differences in patient characteristics (such as age or other variables) may be another potential source of heterogeneity. Third, only studies published in English were included in our meta-analysis, which may result in bias. Finally, significant publication bias was observed in our meta-analysis. The results of the trim and fill analysis revealed that unpublished studies with negative conclusions might not impact the results of our meta-analysis. However, to provide more reliable evidence, the findings of our meta-analysis need to be confirmed by further multicenter, prospective clinical trials.

CONCLUSIONS
In conclusion, the current meta-analysis indicates that there is a significant association between detectable ctDNA and PC prognosis. Mutations detected or high concentrations of ctDNA are significant predictors of the OS and PFS of PC patients. Our meta-analysis strongly supports that ctDNA is a useful predictive biomarker for PC, with advantages of convenience, safety and noninvasiveness.

Search strategy
A systematic literature review was conducted by searching Embase, MEDLINE, and Web of Science from inception to April 2020. The keywords used in our searches were "circulating tumor DNA/ctDNA/ cell free DNA/cfDNA", "pancreatic cancer/pancreatic neoplasm/pancreatic ductal adenocarcinoma/PDAC" and "prognosis/prognostic". The reference lists of the selected articles were also used to supplement the searches. Two reviewers (ZL Fang and QC Meng) independently assessed the eligibility of the potential studies following the guidelines of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement [50].

Inclusion and exclusion criteria
The inclusion criteria were as follows: (1) all PC patients included in the studies were confirmed by histopathology; (2) ctDNA was extracted from peripheral blood; (3) the time of sample collection was declared; and (4) the prognostic value of ctDNA was reported or able to be calculated. The exclusion criteria were as follows: (1) ctDNA was not extracted from peripheral blood; (2) the HRs of prognostic data were not reported and Kaplan-Meier curves were not provided to calculate them; and (3) literature published as reviews, conferences, case reports, letters or expert opinions.

Data extraction
The data were reviewed and extracted independently by two reviewers, and any disagreements were settled by discussion.  [52].

Quality assessment
The NOS was used to evaluate the quality of the included studies by two reviewers independently [53]. This scale consists of 8 items and is used to assess the quality of the selection, comparability, exposure (casecontrol studies) and outcomes (cohort studies) of the included studies. In the present study, we considered a study with a score of 6 or more as a high-quality study.

Statistical analysis
Pooled HRs of OS and PFS were calculated using Stata 12.0 (StataCorp, College Station, TX, USA). If the original HRs were analyzed by both univariate and multivariate analyses, the pooled HRs were calculated respectively. Heterogeneity was assessed by chi-square tests and I 2 values. For values of I 2 > 50% or p < 0.1, the random-effects model was used; otherwise, the fixed-effects model was applied. All p-values were two sided, and statistical significance was set at p < 0.05. Subgroup analyses were performed according to the tumor stage, sampling time, detection marker, detection AGING method, ctDNA positive rate, sample origin, ethnicity and patient size.

AUTHOR CONTRIBUTIONS
ZLF, QCM and BZ collected the related papers, performed the meta-analysis and drafted the manuscript. SS, JL, CL and JH participated in the design of the metaanalysis. WW, JX and XJY initiated the study and revised the manuscript. All authors have read and agreed to the published version of the manuscript.