Non-Invasive Biomarkers for Earlier Detection of Pancreatic Cancer—A Comprehensive Review

Simple Summary Pancreatic ductal adenocarcinoma (PDAC), which represents approximately 90% of all pancreatic cancers, is an extremely aggressive and lethal disease. It is considered a silent killer due to a largely asymptomatic course and late clinical presentation. Earlier detection of the disease would likely have a great impact on changing the currently poor survival figures for this malignancy. In this comprehensive review, we assessed over 4000 reports on non-invasive PDAC biomarkers in the last decade. Applying the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool, we selected and reviewed in more detail 49 relevant studies reporting on the most promising candidate biomarkers. In addition, we also highlight the present challenges and complexities of translating novel biomarkers into clinical use. Abstract Pancreatic ductal adenocarcinoma (PDAC) carries a deadly diagnosis, due in large part to delayed presentation when the disease is already at an advanced stage. CA19-9 is currently the most commonly utilized biomarker for PDAC; however, it lacks the necessary accuracy to detect precursor lesions or stage I PDAC. Novel biomarkers that could detect this malignancy with improved sensitivity (SN) and specificity (SP) would likely result in more curative resections and more effective therapeutic interventions, changing thus the present dismal survival figures. The aim of this study was to systematically and comprehensively review the scientific literature on non-invasive biomarkers in biofluids such as blood, urine and saliva that were attempting earlier PDAC detection. The search performed covered a period of 10 years (January 2010—August 2020). Data were extracted using keywords search in the three databases: MEDLINE, Web of Science and Embase. The Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool was applied for study selection based on establishing the risk of bias and applicability concerns in Patient Selection, Index test (biomarker assay) and Reference Standard (standard-of-care diagnostic test). Out of initially over 4000 published reports, 49 relevant studies were selected and reviewed in more detail. In addition, we discuss the present challenges and complexities in the path of translating the discovered biomarkers into the clinical setting. Our systematic review highlighted several promising biomarkers that could, either alone or in combination with CA19-9, potentially improve earlier detection of PDAC. Overall, reviewed biomarker studies should aim to improve methodological and reporting quality, and novel candidate biomarkers should be investigated further in order to demonstrate their clinical usefulness. However, challenges and complexities in the path of translating the discovered biomarkers from the research laboratory to the clinical setting remain and would have to be addressed before a more realistic breakthrough in earlier detection of PDAC is achieved.


Introduction
Pancreatic ductal adenocarcinoma (PDAC), which represents approximately 90% of all pancreatic cancers, is an extremely aggressive disease and one of the most lethal cancers. Its incidence almost equals its mortality, and it has a 5-year survival of only around 9% [1,2]. It caused 7800 deaths in the United Kingdom (UK) in 2018 alone, and, according to GLOBOCAN, there were 458,918 new cases of pancreatic cancer and 432,242 pancreatic cancer-related deaths worldwide in the same year [3]. Furthermore, PDAC incidence is steadily increasing [4][5][6][7].
PDAC is considered a silent killer due to a largely asymptomatic course, late clinical presentation, and rapid progression [8], with fewer than 20% of PDAC patients being diagnosed at early, resectable stages I and II [9,10]. Importantly, if PDAC is detected when still a localized disease, the 5-year survival is around 32% [11], but it can approach 70% following resection of incidentally diagnosed stage I tumors [12][13][14]. It is estimated that it takes at least 10 years between initiating mutation and the birth of parental founder [15]; combined with an increased PDAC incidence with age [5], resectable PDAC has likely already been growing for a number of years. We therefore further on use the term "earlier" rather than "early" stage.
To date, there exists no reliable, non-invasive screening test, either molecular or imaging-based, that will allow accurate PDAC detection at an earlier stage in asymptomatic patients. The conventional methods such as computerized tomography, magnetic resonance imaging or endoscopic ultrasonography are expensive and have low sensitivity (SN) and specificity (SP) for detection of small premalignant lesions [16]. Furthermore, the deep anatomic location of the pancreas makes detection of small-localized tumors unlikely during a routine abdominal examination [17].
In contrast, testing for biomarkers in body fluids such as serum, plasma, urine or saliva is minimally invasive, relatively inexpensive and could allow for earlier PDAC diagnosis [18]. The current gold-standard serum biomarker for PDAC is carbohydrate antigen 19-9 (CA19-9), also known as sialylated Lewis (a) antigen. It is a product of exocrine epithelial cells and various mucins secreted by PDAC cells that are normally found on the surface of erythrocytes [19]. The test was approved by the U.S. Food and Drug Administration (FDA) in 2002 [20]. However, it is recommended to be used only for disease monitoring and prognosis, since the performance of CA19-9 as a diagnostic biomarker is limited by poor SN, false-negative results in the individuals with Lewis negative phenotype and increased false positivity in several benign and malignant biliary disorders [21]. Furthermore, CA19-9 is not recommended as a screening marker due to its low positive predictive value (0.5-0.9%) [22][23][24].
The accuracy of any tumor biomarkers assay, including CA19-9, is linked to the concept of threshold, or cut-off [25]. The CA19-9 cut-off values differ significantly between studies due to the lack of an international standard for CA19-9 and the difference in the assay design and instrument utilized [26]. When the two most frequently used assays for measuring CA19-9, Abbott Architect CA 19-9XR and Roche Elecsys CA19-9, were compared, despite being based on the same monoclonal antibody 1116-NS-19-9 [27], the Abbott assay measured significantly lower CA19-9 values than the Roche one in healthy and benign cases [28]. While the upper reference limit of 37 U/mL, reaching mean SN and SP of 81% and 90%, respectively, as first reported by Steinberg [29], is still able to sufficiently differentiate between controls and PDAC, the CA19-9 values obtained by different methods should not be used interchangeably, the method utilized should be clearly stated and the patients should be monitored using the same assay [28,30]. Thus, despite its role as the principle serological test for PDAC, the standardization of the laboratory testing and correct interpretation of high serum CA19-9 levels with the adequately established cut off has not yet been achieved [31]. The search for a new non-invasive PDAC biomarker with superior SN and SP to CA19-9, which will be easily identifiable and robustly measured in a fully standardized fashion in biofluids, is therefore being actively pursued.
Recent advances in molecular techniques have empowered the discovery of novel biomarkers for earlier detection of PDAC, resulting in a significant and constantly growing body of literature. Despite the discovery of thousands of biomarkers, relatively few have been adequately tested and validated, making the assessment of their potential challenging [32]. This sparked an EDRN (The Early Detection Research Network) workshop at National Cancer Institute (NCI) in 2016, where a selection of the most promising studies was reported [33]. A comprehensive update and critically appraised resource of information on potentially useful novel biomarkers for PDAC is, however, needed.
In the present study, we reviewed over 4000 articles and further assessed 49 published studies from the last decade describing biomarkers in serum, plasma, urine and saliva for earlier PDAC detection, in conjunction with serum CA 19-9, when available. We utilized the Quality Assessment of Diagnostic Accuracy Studies tool (QUADAS-2) to assess the quality of the selected studies and highlighted their potential clinical utility and limitations for earlier PDAC detection.

Methods
Two of the authors (G.B. and V.S.) independently conducted a systematic search of MEDLINE, Web of Science and Embase and assessed all the articles published between January 2010 and August 2020. A robust reproducible search strategy was created which utilized the MESH terms "Pancreas OR Pancreatic" AND "Ductal Adenocarcinoma OR PDAC OR Cancer OR Malignancy OR Adenocarcinoma OR Tumour OR Tumor" AND "Liquid Biopsy OR Biomarkers OR Screening OR Stool OR Markers OR Marker OR Urine OR Saliva OR Blood OR Non-invasive OR Non Invasive OR Fluid Biopsy OR Fluid Phase Biopsy OR Detection OR Diagnosis OR Diagnose OR Screen" to search for the potential biomarkers specific for this disease. The filters were adjusted to show the results for studies in the English language only. The article type was set to only include clinical and observational studies in Medline. In Medline and Embase, titles and abstracts were searched. In Web of Science, only titles were searched. PRISMA guidelines were used to assist in the structuring of this systematic review [34]. The inclusion and exclusion criteria for selecting the studies are listed in Table 1. Assessment of the methodological quality of reviewed articles was performed according to QUADAS-2, a validated quality assessment tool commonly used to assess the methodology of investigated studies, and as a measure to improve diagnostic accuracy [35]. With QUADAS-2, we were able to appraise three key domains, namely Patient Selection, Index Test and Reference Standard, for both risk of bias and applicability. For the Patients selection, we assessed if the sample groups were well described and if patient spectrum was representative of the disease. We also assessed if the selection of the controls was appropriate, i.e., if, in addition to healthy controls, samples collected from patients with benign pancreatic diseases at risk for progressing to PDAC were also included. Within the Applicability concerns, we assessed if the data for PDAC stages I/II were reported, as we were aiming to review biomarkers for earlier detection of PDAC. For the Index test (biomarker assay), we appraised how well the assay was described and executed and the results interpreted (SN/SP/AUC with 95% confidence intervals (CIs)) and if the data were split into training and validation sets. To increase the validity of the results, QUADAS-2 also recommends that researchers do not know which study group each patient belongs to ("blinding") in order to reduce the effects of any foreknowledge in the interpretation of the biomarker assay results [36]. QUADAS-2 includes the need to specify a threshold value for the Index test prior to obtaining the results; however, this is not suitable for newly discovered biomarkers, where one of the aims is to determine the appropriate cut-offs with regards to the SN, SP and AUC. For the Reference Standard, we assessed if the study specifies that PDAC diagnosis was confirmed by histology and if the staging was pathological or clinical. However, as biomarker discovery studies are typically performed on retrospectively collected samples with an already established diagnosis, we additionally included a comparison of the new biomarker with CA19-9. Flow and Timing were excluded from our QUADAS-2 scoring, as it was difficult to assess this parameter, i.e., in the majority of the studies, it was unclear in what time frame were the diagnostic biopsy or imaging performed in respect to the sample collection and the biomarker testing.

Results and Discussion
A PRISMA flow diagram demonstrating the search strategy and study selection criteria that were applied is shown in Figure 1.  The overall quality of the 49 selected and reviewed studies is illustrated in Figure 2. In Figure 2 it is evident that the majority of the studies (70%) had a low risk of bias in the Reference Standard domain, as the information on histological confirmation that samples originated from PDAC was specified. The unclear bias in this domain seen in approximately 30% of the studies was due to the absence of comparison of biomarkers with CA19-9. High risk of bias in the Index test (biomarker assay) was seen in 36% of studies and was associated with missing information on if the samples were analyzed in the blinded fashion. The lack of clarity, which was seen in the following 36% of the studies, was due to absence of robust statistical analysis of the data, i.e., splitting of the data into the training and the validation sets. Regarding patient selection criteria, around 20% of studies did not compare PDAC results to appropriate benign controls, leading to a high risk of bias. Around 20% of studies raised concerns about the applicability of the biomarkers for earlier PDAC detection, as they were not assessed in resectable stage I-II PDAC.
A graphical summary of the QUADAS-2 results with the descriptive score for each of the selected studies is presented in a tabular form in supplementary Appendix A, Table A1. Table 2 summarizes the key findings from all 49 studies included in this review.        The studies summarized in Table 2 were all aimed at earlier PDAC detection; however, around 20% of them did not analyze their biomarkers in resectable stage I-II samples. While the paucity of earliest stage I samples is a recognized problem in the PDAC field, this was surprising, considering the critical importance of diagnostic stage shift. Selection of the relevant control population is also of the essence, as PDAC is an uncommon cancer, which dictates the need to assess the putative biomarkers in high-risk groups [1,84,85], a criterion which was not satisfied in 20% of the studies. It also requires the high accuracy of any potential future biomarker assay with at least 88% SN and 85% SP if the test is to be cost-effective [86].
A third of the reviewed studies did not include a comparison of their biomarkers to CA19-9, which was unexpected considering that CA19-9 is the most commonly measured biomarker in PDAC patients. Very few biomarkers were conclusively superior to CA19-9 and in the majority of studies where their combination was tested, the accuracy of PDAC detection was superior to either biomarker alone. For example, Dong et al. [71] described the panel of serum biomarkers comprising CA19-9, POSTN and CA242, which was superior to CA19-9 alone in differentiating between PDAC and both healthy controls (AUC of 0.98, 92.1% SN and 97.3% SP) and benign pancreatic conditions (AUC of 0.93, 80% SN and 97.7% SP), albeit with somewhat lower AUC of 0.90 and a decrease in SP to 88% in resectable PDAC. In the study of Berger et al. [77], CA19-9 combined with THBS2 and cfDNA in PDAC vs. non-PDAC patients showed an increase in AUC from 0.70 to 0.78, with an increased SN and SP of 80% and 96%, respectively. A study by Fahrmann et al. [79] combined a five-metabolite panel (acetylspermidine, diacetylspermine, an indole-derivative and two lysophosphatidylcholines) with a previously validated protein panel comprising CA19-9, TIMP1 and LRG1, achieving AUC of 0.924 (95% CI 0.864-0.983). Recently, another panel (FGA, KRT19, HIST1H2BK, ITIH2, MARCH2, CLDN1, MAL2 and TIMP1) obtained by plasma extracellular vesicle long RNA profiling (d-signature), successfully differentiated PDAC from CP and healthy controls with AUC of 0.936, and both SN and SP > 90%; furthermore, when combined with CA19-9, the signature showed improved AUC of 0.964 (95% CI: 0.943-0.984) [80]. Traeger et al., reported that miRNA-205 combined with CA19-9 resulted in higher SN and SP in comparison of PDAC vs. benign specimens (86.7% SN and 93.3% SP) than either of the biomarkers alone [75].
The absolute SN and SP values for PDAC detection using glypican-1 (GPC1) in circulating exosomes when PDAC samples were compared to healthy and benign samples was reported by Melo et al. [54]. However, several subsequent studies failed to support these findings [62,73,87,88], while Zhou et al. [76] demonstrated that combining GPC1 and CA19-9 showed 92.31% SN at 65.83% SP for distinguishing healthy, chronic pancreatitis and benign pancreatic tumors from PDAC. Regardless of the underlying reasons, these conflicting results highlight the critical importance of independent validation of any proposed biomarkers.
The above study [54] and the more recent studies indicate the "expansion" of the biomarker field, moving from the "traditional" DNA, RNA and protein biomarkers to novel, versatile analytes and their combinations, such as non-coding RNAs, epigenetic markers, lipids and metabolites, as well as circulating tumor cells and extracellular vesicles (EV) [88]. For example, Yang et al. [83] performed a multi-parametric analysis of blood tumor-associated EV miRNA and mRNA, circulating cell-free DNA concentration and KRAS G12D/V/R mutations with CA19-9, and demonstrated their higher accuracy for PDAC staging (84%) than imaging alone (accuracy = 64%; P < 0.05). Furthermore, recent advances in computer science, machine learning and artificial intelligence (AI) have given rise to the emerging field of bioinformatics and computational biology which allow analyses of large collections of biological data [89]. While AI was mostly utilized for the prediction of risk/diagnosis based on health records or abdominal imaging [1,90,91], it also shows a great potential for in silico discovery of putative novel biomarkers. For example, Khatri and Bhasin [92] used transcriptomics-based meta-analysis of tissue and blood samples combined with machine learning to identify a nine-gene panel (IFI27, ITGB5, CTSD, EFNA4, GGH, PLBD1, HTATIP2, IL1R2 and CTSA) that differentiated PDAC and healthy controls with 92% SN and 90% SP. Moreover, this panel also discriminated PDAC from chronic pancreatitis and early precursor lesions in non-malignant tissue and peripheral blood. Chung et al. [93] used probe electrospray ionization mass spectrometry (PESI-MS) and a machine learning approach to analyze peripheral blood samples and achieved an accuracy of 92.1% in healthy vs. PDAC (90.8% SN, 91.7% SP). The combination of PESI-MS profiles with age and CA19-9 increased the accuracy for detection of stage I and II PDAC to 92.9% (81.2% SN, 96.8% SP) [40]. This demonstrates that the access to richly annotated patients' health records will enable the construction of novel PDAC prediction tools such as the one developed by Muhammad et al. [90]. Combining such prediction tools with biofluid biomarkers now opens up promising new avenues and likely a fruitful approach to identify asymptomatic patients at increased risk of developing PDAC.
Very few studies explored the effectiveness of their proposed biomarkers in prediagnostic samples from asymptomatic patients. Although such samples are not readily available, these studies are essential for determining how early in the latency period the biomarkers can be detected [40,47,[94][95][96][97][98][99][100][101]. Nolen et al. [47] interrogated the large prospective Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial serum samples to detect eight biomarkers: CEA-related cell adhesion molecule 1 (CEACAM1) and prolactin (PRL) were detectable up to 35 months before PDAC diagnosis; CA19-9, carcinoembryonic antigen (CEA), neuron-specific enolase (NSE) and beta-human chorionic gonadotropin (b-HCG) up to 24 months; and CA125 and interleukin 8 (IL-8) up to 12 months prior PDAC diagnosis. In the UK Collaborative Trial of Ovarian Cancer Screening (UKCTOCS) serum samples, a decrease in circulating TSP-1 levels up to 24 months prior to PDAC diagnosis was described by Jenkinson et al. [97], and O'Brien et al. demonstrated that CA19-9 could be detected in such pre-diagnostic PDAC samples with 95% SP and 68% SN up to 1 year and 53% SN up to 2 years prior to PDAC diagnosis [98]. A three-biomarker panel comprising Erb-B2 receptor tyrosine kinase 2 (ERBB2), estrogen receptor 1 (ESR1) and tenascin C (TNC) was analyzed in 87 pre-diagnostic plasma samples from Women's Health Initiative (WHI) collected up to 4 years before diagnosis and achieved AUC of 0.71 when combined with CA19-9 [100]. Recently, the analysis of 1196 proteins in longitudinal plasma samples from the Institute for Systems Biology's 100K Wellness project identified several markers, including three that were observed in PDAC: CEA cell adhesion molecule 5 (CEACAM5) was expressed at high levels 26.5 months pre-diagnosis, while calcitonin related polypeptide alpha (CALCA) and delta1 like homolog (DLK1) were predating PDAC diagnosis for around 17 months [101]. More exciting projects such as this are needed and should be actively encouraged.
While performing this review, we encountered a number of hurdles, particularly noticeable in the way the studies were reported. Easier reviewing of published resources would be facilitated, and the extraction of the relevant information simplified, if a consensus set of standards for the reporting of biomarker studies were utilized. In addition to QUADAS-2, several of such recommendations and guidelines exist: BRISQ (Biospecimen Reporting for Improved Study Quality) [102] provides recommendations for reporting details about the interrogated biospecimens; the Standards for Reporting of Diagnostic Accuracy Studies (STARD) [103] was developed to improve the completeness and transparency of the reports; and Reporting Recommendations for Tumour Marker Prognostic Studies (REMARK) [104] and Consolidated Standards of Reporting Trials (CONSORT Statement) [105] are used to improve the overall quality of tumor markers in the prognostic studies and randomized trials [104,105] . The latter is the final step in establishing the clinical utility of the biomarker, i.e., increased survival due to earlier diagnosis achieved by biomarker testing. Interestingly, only studies published in high-impact factors journals followed any of these recommendations [49,54,66,74,80].
In addition to the standardized way of reporting, a standardized pathway for early detection and diagnosis is also needed. Such pathway has recently been summarized in the updated Cancer Research UK (CRUK) roadmap, which aims to consolidate the current diverse and fragmented research activities, in line with the future vision to detect 75% of cancers at stage I-II by 2028 [106]. A roadmap also illustrates a long way from biomarker discovery to its implementation, with a number of hurdles encountered at every step in biomarker pipeline, from initial biomarker discovery, its quantification, verification and assay optimization to finally biomarker validation and commercialization, as comprehensively discussed previously [107][108][109]. While biomarker discovery is usually well funded and results straightforward to report, validation, which is a necessary step prior to commercialization, is time-consuming and expensive process, with scarce funding available. The transition from discovery to the clinical utility of any biomarker is therefore a cumbersome process, akin to the implementation of novel therapeutic [110].
It is not surprising thus that, despite a large number of candidate biomarkers, very few are close to clinical implementation. Based on the accuracy of 95% of serum biomarker signature published by Mellby et al. [74] and using their proprietary antibody platform, Immunovia is now developing the commercial assay IMMRay™ PanCan-d, which is currently being further validated in the prospective studies [111]. Our urinary biomarker panel comprising REG1B, LYVE1, TFF1 and the affiliated PancRISK score [112,113] are now being validated in a prospective study UroPanc [114] . A multi-analyte blood test that simultaneously evaluates the levels of eight cancer proteins and the presence of cancer gene mutations from circulating DNA, CancerSEEK ® , screens for eight common cancers, including PDAC [115], and has already been commercialized. GRAIL is advocating its own Galleri test, which is based on cfDNA methylation signal and can detect over 50 different cancers [116]. Both platforms, however, suffer from lower SN in detecting stage I cancers, and there is a legitimate concern that the tests might be less specific in real-world conditions, where comorbidities may increase the risk of a false-positive result [117].

Conclusions
In summary, our systematic review demonstrates the current progress in the biomarker field and highlights several promising non-invasive PDAC biomarkers. Most of these, however, still require further validation prior to being deemed appropriate for translation into the clinical setting. All future studies should aim to improve methodological quality and the way of reporting. Novel biomarkers that have shown promising results should be further investigated on a larger number of specimens, as well as in a pre-diagnostic and prospective setting, to further demonstrate their real clinical utility in earlier detection of pancreatic adenocarcinoma.

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