Biomarkers for Early Detection of Pancreatic Cancer Open Access

In 2024, all major cancers in the EU expected an improved survival trend, except for lung and pancreatic cancer [1]. Pancreatic ductal adenocarcinoma (PDAC) is the third leading cause of cancer-related death, with a majority of cases diagnosed at advanced stages [2]. Acinar-to-ductal metaplasia is implicated in the development of pancreatic intraepithelial neoplasia (PanIN), the most frequent precursor to PDAC. While acinar-to-ductal metaplasia may serve a regenerative function following pancreatic injury or inflammation [3], its persistence can lead to PanIN formation, particularly in the presence of oncogenic KRAS mutations. The progression of PanIN to invasive PDAC is driven by sequential loss of tumor suppressor genes such as p16/CDKN2A, TP53, and SMAD4 [4]. Despite the low incidence of PDAC, PanIN structures in the population exceed 70% in autopsy series [5]. Oncotherapeutic advancements have conferred short-term survival benefits over the past two decades, but the actual chance for long-term survival has remained stagnant since start of systematic data collection [6]. Consequently, more resources have shifted toward the development of biomarker-based screening tools for early detection. This review will focus on current techniques, advances, and important considerations in each area of PDAC early detection, focusing on minimally invasive molecular tests. As such, the review will cover molecular and physiologic biomarkers but leave out histologic and imaging studies. This strategy for biomarker development is summarized in a flowchart adapted from the phased approach provided by the NCI’s early detection research network (Fig. 1) [7].

Since PDAC is largely asymptomatic in earlier stages, any biomarker aiding in early detection would be considered a screening marker [8]. Regardless if the biomarker is tissue-based, blood-based, radiologic, or defined by clinical characteristics, key considerations include sufficient clinical annotation, the inclusion of both non-cancer cases and non-operable stages, and a clear interconnection between the bioinformatics and the specimen itself [9]. To aid researchers in acquiring or utilizing biospecimens, several best practice guidelines exist. The Consortium for the study of Chronic Pancreatitis, Diabetes, and Pancreatic Cancer (CPDPC) has developed comprehensive guidelines covering most specimen types and sampling methods (blood, urine, saliva, stool, pancreatic/duodenal juice, and tissue) [10], but standard operating procedures for biobanking of cyst fluid acquisition of PDAC precursor lesions are lacking. Pancreatic cancer tissue is good at maintaining DNA and RNA integrity after prolonged ischemia during handling, while protein and posttranslational modifications are more sensitive [11]. To enhance the reporting of biospecimen-related research and diagnostic accuracy studies, adherence to the Biospecimen Reporting for Improved Study Quality (BRISQ) [12] and the Standards for Reporting of Diagnostic Accuracy Studies (STARD) [13] criteria is recommended.

The two main configurations for mass spectrometry are liquid chromatography-tandem MS and gas-chromatography MS, with liquid chromatography-tandem MS being the current primary modality for proteomic and metabolomic discovery in PDAC [14]. Residing on mucins of cancer cells, the glycoprotein complex CA 19-9 or sialylated Lewis antigen remains the only blood-based biomarker used in clinical practice, mainly as indication for surgery in pancreatic cystic lesions and for disease monitoring [15]. Baseline levels vary wildly in the general population [16], and elevated levels are seen in many benign conditions, with 10% of patients unable to synthesize it [17]. Preclinically, CA 19-9 functions as an anchor marker in protein panel studies and a reference for overall biomarker performance in PDAC [18, 19]. Mucins are glycoproteins where aberrant and de novo production may closely follow pancreatic tumor progression. Secretory MUC5AC [20] and MUC4 [21] mucins retain high expression through PanIN-3 to invasive disease, but their detection is highly dependent on antigen retrieval methods, and serum concentrations are lower in precancerous disease [22]. Other well-performing glycoproteins are thrombospondins and intercellular adhesion molecules, THBS1 and ICAM-1, oncosuppressive proteins that are downregulated in PDAC [23, 24]. THBS1 and ICAM-1 have good discriminatory power compared to both benign and healthy controls [25]. Furthermore, matrix metalloproteinases are often implicated in PDAC tumor initiation [26], particularly MMP7 which is also expressed in preneoplastic disease [27] and has good diagnostic performance [28]. Recently, MMP2 was reported as a key circulating protease in PDAC and a protease activity-based assay using a magnetic nanosensor was developed, termed PAC-MANN [29]. The assay combined with CA 19-9 was able to detect stage 1 PDAC with 85% sensitivity at 96% specificity. There have also been reports on oncogenic proteases such as cathepsins and urokinase plasminogen activator receptors [30, 31]. In addition, metabolic signatures have demonstrated improved performance compared to CA 19-9 alone in excluding PDAC in a prospective real-world cohort (METAPAC study) [32]. Furthermore, studies of metabolic pathways have generated lipid panels that, combined with CA 19-9, show good ability to differ between PDAC, healthy controls and chronic pancreatitis [33].

A challenge with urinary biomarkers is high sample variability due to time of day and degradation, information that may not be readily available in the biobank registry [34]. Promisingly, well-performing micro-RNA markers are equivalent between serum and urine [35], and comprehensive predictive tools such as PancRISK have been developed from urinary biomarkers [36].

Direct tumor constructs, termed “liquid biopsies,” include cell-free plasma DNA/RNA (cfDNA/RNA), circulating tumor cells (CTCs), circulating tumor DNA [37], exosomes [38‒40], and noncoding small RNA, such as circular RNAs [41], microRNAs [42], and transfer RNA-derived RNAs [43]. cfDNA demonstrates low concentrations of point-driver-gene mutations in early PDAC, and clonal hematopoiesis creates a high biological noise [44]. Instead, detection of tumor-specific methylation patterns of cfDNA may be more sensitive, also in PanIN lesions and early PDAC [45]. A recent meta-analysis demonstrated a correlation between peripheral blood CTCs with PDAC stage and survival [46], which could enable detection in earlier invasive disease. Still, the detection rate of CTCs is far lower in precancerous lesions [47].

Next-generation sequencing (NGS) is used for biomarker exploration of tumor constructs in PDAC. Due to low tumor cellularity, employing NGS to PDAC blood samples requires highly sensitive methods. Current NGS methods applied to PDAC include Illumina’s MiSeq, Ion Torrent, Pacific Biosciences, and Oxford Nanopore Technologies (ONT). The long turnaround time and high cost of Illumina sequencing would suggest Ion Torrent methods to be more suited for developing clinical tests. However, few studies targeting PDAC are conducted with Ion Torrent and have mainly been further developments of the platform for pancreatic circulating tumor DNA detection [48]. ONT, as with Ion Torrent, is also hampered by reduced multiplexing capabilities. Still, as ONT provides a shorter gap between the bench and clinic, recent attempts at adequate miRNA detection using ONT show promise for developing more portable liquid biopsy-based tests [35]. Pacific Biosciences has primarily been used in gut microbiome studies of PDAC [49], and MiSeq suffers from high DNA input requirement [50]. Newer third-generation PCR amplification has been given attention, such as multiplexed digital-droplet PCR. It has low turnaround time, highly multiplexed detection, and point-of-care testing capabilities [51], and is frequently applied in miRNA studies for PDAC early detection [52, 53].

Researchers might use external high-throughput affinity-based biomarker platforms by combining sequencing technologies with proteomic discovery. The Olink platform (Olink Bioscience, Uppsala, Sweden) utilizes proximity extension assays with antibody-conjugated oligonucleotides that hybridize upon target binding and are extended by DNA polymerase for protein quantification [54]. Based on preexisting libraries, thousands of proteins can be detected simultaneously, exceeding the discovery rates of the largest microarray platforms [55]. Published PDAC studies have utilized the “Immuno-oncology” [56], the “Oncology II” [57] and the “Explore 3072” panels [58]. As antibodies are bioreceptors with long half-lives, they create high background noise, lowering the discovery rate. The SomaScan (SomaLogic, Boulder, CR, USA) utilizes short, enriched oligonucleotides (slow-off rate-modified aptamers) treated with protecting groups such as benzene or 2-naphthyl for enhanced plasma stability. SOMAs functionally resemble antibodies but have shorter half-lives and combine with fluorophores for photometric analysis in complex protein samples. Multiplatform studies combining MS, SomaScan, and Olink may expand protein discovery rates even further [59, 60].

While these studies are more focused on etiology of significantly expressed biomarkers, they also give the possibility to conduct purely register-based discovery studies. For example, a proteome-wide association study utilized data from the ARIC and the INTERVAL studies (which in turn are based on the SomaLogic assay) to infer 16 novel proteins not previously reported [61].

Upon biomarker discovery, researchers must determine which combinations carry the best performance. The method, often termed “feature selection,” can be done via Monte Carlo optimization [62] or LASSO regression [63]. Consensus regarding PDAC biomarker performance is the need for high specificity, specifically over 98% [64], given the low pretest probability.

Once a candidate biomarker or biomarker signature is identified, a targeted assay platform is developed based on the intended use. Analytical validation involves evaluation of performance metrics of the assay, including sensitivity, specificity, reproducibility, linearity, and dynamic range. Standard operating procedures need to be established. Reference samples from PDAC patients and controls should be analyzed to benchmark results.

The clinical validation step assesses the clinical relevance and utility of the biomarker or biomarker signature.

The initial validation phase typically involves case-control analysis of established cases and controls, and retrospective evaluation in a cohort from prospectively collected samples. Controls should include confounding disease states, such as pancreatitis and benign biliary inflammation. Performance results are usually better on the initial patient sample, so cross-validation is used to finalize the prediction model and prevent over-fitting. Traditional ELISA is commonly used for absolute protein quantification after the discovery phase [18, 65]. For liquid biopsy approaches, researchers with tissue-based sequencing have to ensure that the intended sequencing method also is able to detect tumor constructs in noninvasive samples if transitioning to the validation phase [66].

For assistance in study design, the protocol from the prospective-specimen-collection retrospective-blinded evaluation (PRoBE) method is preferred. It involves prespecifying the clinical setting, benchmarks, the population at risk, target lesions, and how to ascertain the outcome [8]. Biomarker panel performance is often measured in sensitivity at 98% specificity in conjunction with area under the curve (AUC) metrics to report overall performance, but there have been proposals to use precision-recall curve (AUC_PR) to account for class imbalances between PDAC and the control groups. However, recent evidence maintains that regular AUC is the preferred performance metric [67]. Benchmark PDAC performance values from EDRN include an AUC >0.85 and a negative predictive value at or above 99% [68]. Importantly, time-dependent performance metrics are biomarker-dependent in PDAC [69] and could be misleading if used as benchmarks in this phase of development.

For pancreatic cancer, five distinct population groups exist for targeting by early detection: (1) patients with pathogenic germline variants, (2) familial pancreatic cancer, (3) symptomatic individuals reporting weight loss combined with new-onset diabetes and other worrisome features (according to NG12 NICE guidelines) [70], (4) patients with chronic pancreatitis, and (5) patients with mucinous cystic neoplasms [71]. Groups 1 and 2 termed high-risk individuals (HRIs) in the literature [16]. In 2019, the US Preventive Services Task Force maintained its recommendation against general population screening, but only in relation to the current performance of the radiologic surveillance strategy for HRIs [16]. Interestingly, sporadic cases with similar radiologic features harbor lower risk lesions upon resection and histological evaluation, while new-onset diabetes bears less relative impact on the likelihood of developing PDAC in familial cases [72]. Hence, from a biological standpoint, HRI surveillance studies do not translate well to sporadic PDAC.

Imaging is used to confirm the presence of the target lesion. The International Cancer of the Pancreas Screening (CAPS) Consortium defined successful target lesions in HRIs as PanIN3, PanIN with high-grade dysplasia and stage I PDAC [73]. Existing strategies for HRIs yield 5 to 8 target lesions per 1,000 person-years [16]. Concerningly, many prospective biomarker studies fail to include PanIN lesions [16, 74].

Researchers should utilize cohorts that recruit PDAC target populations in the clinical setting specified in the PRoBE protocol. This entails pre-diagnostic samples from large cohorts that are largely serum-based, and these could include the European Prospective Investigation into Cancer and Nutrition (EPIC) cohort [75], or ERDN’s own reference set [76], among others.

A challenge with pancreatic cancer is the slow progression, modeled to take 7–12 years until metastatic disease [4, 77]. Some PanINs also progress rapidly and undergo multiple simultaneous genomic alterations, mainly by chromosome fragmentation [78]. The sudden appearance of high-stage tumors between the 12-month image study intervals in HRI studies has been attributed to this phenomenon [79]. Ideally, the performance of the biomarkers should not be tied to tumor progression outside of the target lesion range. To evaluate this, researchers should conduct subgroup analyses in the pre-diagnostic cohort based on the time between specimen acquisition and target lesion diagnosis. This is to avoid having to include time-dependent outcome metrics further down the line, for which benchmark values are unclear.

In transitioning to prospective evaluation in high-risk groups, several important questions must be answered. What should the screening interval be? Is the intervention cost-effective? And most importantly, is there clinical utility? These questions may only fully be answered as the prospective trial is underway, but many prior studies do provide indices [80]. In fact, estimation of the clinically detectable time, e.g., sojourn time, has been invaluable in guiding screening intervals for breast cancer in different age groups [81]. A few studies have calculated average sojourn times for PDAC by hyperglycemic periods and counterfactual analysis, and researchers should complement their own prospective study with their target population’s estimated sojourn times [82]. In the following “penultimate period,” large-scale population studies are set up to evaluate the actual impact of the screening measure. To determine if the “benefit outweighs the cost” in the population, researchers must decide on how many nontarget lesions need to be worked up to detect one target lesion. These threshold values will vary considerably and not only with the disease prevalence of the population at risk, but also with geographic resources, and should not come into play until the last development phases of the biomarker test.

Several ongoing clinical trials are investigating biomarkers for the early detection of PDAC. Table 1 provides a summary of selected studies.

The final phase of biomarker development is clinical implementation, which requires regulatory approval from authorities such as US Food and Drug Administration (FDA) or the European Medicines Agency (EMA). Guideline committees such as NCCN and ASCO and key opinion leaders must be engaged for the inclusion of the biomarker in clinical practice recommendations. Post-market surveillance and cost-effectiveness analysis of the biomarker should be performed.

Out of thousands of biomarkers entering the validation phase, very few get approved for clinical use. Biomarkers intended for clinical use must demonstrate not only analytical and clinical validity, but also clinical utility, which means that their use leads to improved clinical decision-making and better results for the patients. Here, we have outlined what we believe to be the most important considerations for molecular biomarkers from bench to clinical practice for early detection of PDAC. To improve clinical translation, researchers should clear several tangible hurdles before moving onto the next phase of testing.

The authors have no conflicts of interest to declare.

This work was supported by the Swedish Cancer Society, the Swedish Research Council, the Crafoord Foundation, the Ingrid and Sverker Persson Foundation, and Regional Research Support as well as ALF funding from Region Skåne.

Axel Bengtsson: conceptualization, investigation, and writing – original draft. Roland Andersson, Johan Linders, and Aiste Gulla: writing – review and editing. Daniel Ansari: conceptualization, writing – review and editing, and supervision.