Pancreatic cysts are commonly found incidentally during routine imaging for unrelated disorders. Advances in CT and MRI imaging have improved resolution such that more and smaller pancreatic cysts can now be identified, providing an opportunity for earlier diagnosis (19). However, this comes with the drawback that smaller cysts are more challenging to classify and produce less cyst fluid for biochemical analyses. The primary utility of imaging for classification relies on shape/size and location. It has been established over the years that cysts that form within the main pancreatic duct and those that have grown beyond 3 cm are more likely to progress to cancerous lesions than cysts that form in other regions of the pancreas or smaller cysts, however the risk remains highly variable (10). Although helpful, as with any image analysis, this approach comes with a subjective bias that hinders accurate diagnosis. Efforts to avoid image interpretation errors have focused on using machine learning and artificial intelligence methods (ML-AI) to process all available images and more accurately determine the classification of a pancreatic cyst. This work is mainly in the nascent stage, and it shows greater promise in organs like the lung in which the tumor signal-to-background ratio is high. In contrast, the solid components of pancreatic tumors are frequently camouflaged, making them difficult to detect (20). For patients with indeterminate cystic lesions of sufficient size, the next logical step is often to sample the cyst fluid and conduct (bio)chemical analyses.

Use of the EUS-FNA procedure has become more widespread, as the information contained within pancreatic cyst fluid has proven valuable (21–23). This section, and the following sections, will focus on EUS-FNA fluid analysis. In the case of pancreatic cyst fluid, microscopic examination of fluid contents (cytology) has the potential to accurately diagnose a pancreatic cyst with high-grade dysplasia or adenocarcinoma. However, the examination is often non-diagnostic because of the paucity of tumor cells in the fluid, with an estimated 31% success rate for obtaining cells for cytological analysis (24). Furthermore, the sample volume required for slide preparation may exceed 500 µL, which is often unavailable. Although this is a common clinical practice for diagnosing pancreatic cysts, the limited cellularity of pancreatic cyst fluid yields relatively low clinical utility; thus, we will omit any further discussion of cytology.

Biomarkers indicative of cancer have been studied, published, and established in the clinic as screening and diagnostic tools. Every cancer is different, but several biomarkers consistently present themselves during clinical investigations (25, 26). Their ubiquity makes them generally poor predictors of cancer type, but they unquestionably retain utility as indicators of cancer or progression toward cancer when the possibility of cancer arises. In the case of pancreatic cancer, carcinoembryonic antigen (CEA) and amylase have been used extensively in the clinic to support rule-out diagnoses. However, neither marker is individually indicative of either non-mucinous or mucinous pancreatic cyst fluid, with low sensitivity and/or specificity (27, 28). Other markers that suffer similar fates in clinical settings include CA19-9 and glucose; however, glucose has maintained clinical relevance for determining mucinous pancreatic cysts, with a sensitivity and specificity over 80% (29–32). For clinical labs, however, any of these markers represent a seemingly sensible method, as they are all amenable to automated processing on clinical analyzers, with little training or upgrades needed. Importantly, this reliance on convenience is still rife, causing clinical problems for diagnosis.

Clinical analyzers and their associated assays have been developed to work with highly proteinaceous fluids such as blood, serum, or plasma, and some are compatible with alternative biofluids such as urine or saliva. None of these fluids compare with pancreatic cyst fluid’s relatively high complexity and inter-sample variability. A basic review of published investigations into proteomic analyses of pancreatic cyst fluid reveals thousands of individual biomaterials present in any given sample (18, 33, 34). Still more problematic is that high concentrations of biomaterials have been established to cause inaccuracy and/or decreased reproducibility/repeatability in standard assay formats (35–37). For example, viscous materials like mucins present challenges in basic liquid handling, leading to erroneous sample volume transfers and assay results. Furthermore, the sample volume requirements for clinical analyzers often preclude the replicate analysis of multiple markers in single pancreatic cyst fluid samples. The balance at hospital clinics between high-throughput analysis and sample preservation is certainly skewed toward high throughput. Furthermore, the standards set by the analysis of readily available blood samples are often inappropriate for pancreatic cyst fluid samples. Although assays for markers of interest could be run at lower volumes, the analyzers employed in the clinic often require 200–500 µL of sample per run. When these factors are taken together, it is unsurprising that assays of biomarkers that lack sensitivity and specificity, and inhibit multiple-marker analyses due to sample consumption, inevitably lack the capacity for accurate diagnosis.

However, there are alternative protein markers still undergoing translational research that can provide useful diagnostic information. For instance, vascular endothelial growth factor (VEGF) has been shown to be a reliable marker of notoriously difficult to diagnose serous cystic neoplasms (SCNs) (38–40). SCNs are non-mucinous pancreatic cysts that are difficult to distinguish from potentially malignant intraductal papillary mucinous neoplasms (IPMNs) and mucinous cystic neoplasms (MCNs) using imaging or the previously described protein markers (41). Although they constitute an estimated 1%–2% of all pancreatic cysts, avoiding unnecessary surgical intervention for SCNs will improve the quality of life for patients (42). In addition to VEGF, the monoclonal antibody Das-1, originally used as a marker of colonic epithelial phenotype, has shown utility in predicting malignancy in conditions of the upper GI tract (43, 44). Importantly, Das-1 has displayed the unique ability to stratify low- and high-grade dysplasia and invasive cancer in mucinous pancreatic cyst fluid.

Research into the genetic drivers of pancreatic cancer has found long lists of markers in DNA, RNA, miRNA, and mRNA, and has recently made the logical step toward combination marker panels investigating several genetic components at once. Again, most of the tools have not translated into clinical utility, except for DNA markers that can be amplified and quantified in high-throughput platforms (45, 46). Next-generation sequencing (NGS) is a technique that emerged on the market in the mid-2000s, and has provided a wealth of information for researchers in biology and biochemistry. Every clinical sample prepared for extraction is subject to volume requirements to meet the demand of the extraction method. Although sample preparation for NGS can be conducted with lower volumes, to counter low-genomic-content samples, general estimates for required sample volume are around 500 µL (47). This means that NGS is expected to consume significant resources from a limited pancreatic cyst fluid sample, further constraining the utility of other diagnostic tools.

Regardless, although the fluid is typically acellular, NGS has produced rich genetic information from pancreatic cyst fluid samples. Studies investigating genetic markers or defects have found that NGS can accurately diagnose specific cyst types. A prime recent example of this is the study by Paniccia et al., wherein they investigated 1,933 pancreatic cyst fluid samples for genomic content and were able to diagnose those cysts with negative predictive values (NPVs) and positive predictive values (PPVs) greater than 90% in most cases (48). They also made use of imaging where available to supplement NGS data and further improve the NPV and PPV in select cases. With the cyst type identified, a more practical treatment course can be determined, and this information indeed leads to better diagnostic outcomes for patients. But it is again important to note that cyst type alone does not confer sufficient information about the degree of risk of progression to cancer. Recent efforts have focused on improving analysis algorithms to identify cyst types better and eventually stratify dysplasia.

Recent years have seen a tremendous increase in interest in biological exosomes (49). Broadly speaking, exosomes are vesicles ranging in size from 10—200 nm that are released from cells and contain unique chemical and genetic information. Exosomes are found in many distinct human biofluids, with the majority of research focused on blood-based exosomes, which contain information from any tissue source secreting into the bloodstream. Blood-based exosomes are ideal targets for the analysis of most disease states, including, if not especially, cancer (50). It has been hypothesized that cells use exosomes to inform their local community of different signals and activation patterns (51). Until recently, however, efficient isolation of such material from biological sources was complicated at best, and near impossible at worst.

Multiple methods, including ultracentrifugation, size exclusion chromatography, and immunoaffinity have been used to isolate exosomes with varying degrees of success and recovery. However, a recent study by Hinestrosa et al. demonstrated a clever method for isolating exosomes from plasma, wherein they used an electrode array to attract the charged membrane of exosomes to a solid support from which they can be washed and resuspended for analysis (52). This study investigated exosomes to identify pancreatic, ovarian, and bladder cancers, and analyzed the resulting samples with commercially available protein quantification kits. The compiled data were used to train an algorithm for multiple markers that could accurately detect early-stage cancer, and displayed a high sensitivity for pancreatic cancer. These results suggest that exosomes should continue to be a source of diagnostic information, and it is likely that with some further marker optimization and refined assay development this isolation technology has the potential for widespread adoption in cancer diagnostics.

An underutilized approach in the search for clinically translatable diagnostic assays in pancreatic cancer has been the analysis of enzyme activity, as opposed to enzyme mass (53–56). Immunoassays dominate the landscape of biomarker analysis and have been readily adapted into automated systems that improve throughput, reproducibility, and sample handling. ELISA-based methods have provided an incomparable wealth of data about protein expression and are now a mainstay in research and clinical laboratories. However, immunoassays quantify the amount of protein present, which can be ambiguous when investigating enzyme markers as active (or activatable) enzymes are responsible for signaling cascades, metabolism, and phenotypical responses (57–59). There have been publications over the years examining protease cleavage, enzymatic redox reactions, and other activity measurements. Yet still, these assays have struggled to translate meaningfully to clinical settings in the same way that immunoassays have over the decades. One possible reason is that, although protease cleavages are an amplification process, proteolytic activity on substrates is not necessarily as efficient or rapid a process as the commonly used secondary reactions in ELISA formats. Due to the lower amplification of optically active products, it is difficult for traditional optical methods to detect and quantify low activity. It is even more complicated when sample volumes are limited.

Clinical assays rely on optical detection methods including luminescence, fluorescence, and absorbance. New protein biomarkers are routinely identified using mass spectrometry (MS) methods. However, for many of these candidate markers, the expression level is below the limits of detection for traditional optical methods. A path forward is the use of alternative detection technologies such as Raman spectroscopy. Raman spectroscopy is based on an inelastic light scattering event that imbues the scattered photons with chemical and structural information owing to the excitation of molecular vibrational modes. Although only approximately 1 in 100,000,000 photons are predicted to produce Raman scattering, the introduction of a coinage metal (Au, Ag, or Cu) surface has been shown to enhance the Raman scattering intensity by multiple orders of magnitude, leading to ultrasensitive detection limits (60, 61). First discovered in 1974, surface-enhanced Raman spectroscopy (SERS) has become a more practical tool as the cost and size of Raman equipment have dramatically decreased (62, 63). Recent investigators have used SERS to detect single-molecule Raman signals (SMSERS), which provides additional support for the utility of SERS in clinical assays for low-concentration and/or low-volume samples (64–66). Furthermore, the chemical fingerprint information provided by Raman scattering enables discrimination of multiple dye classes (i.e., fluoresceins, rhodamines, cyanines), or dyes with variable isotopic labeling, enabling intrasample multiplexing at incredibly low concentrations (67–70).

The authors of this review have translated assays using Raman-active dyes that have shown detection limits below those of traditional fluorescence measurements (71). Furthermore, Suresh et al. have published the first example of translating a SERS assay for protease activity of gastricsin to a clinically ready platform with a small cohort of human pancreatic cyst fluid samples (72). Pancreatic cyst fluid is a notoriously tricky medium with which to run assays or optically quantify results, and the simplest solution is significant sample dilution to diminish interfering substances. The ultrasensitivity of SERS and a protease activity assay that displayed reduced interference from the sample milieu provided an ideal assay platform for high-dilution quantification. Importantly, the published results were from assays of 1 µL of pancreatic cyst fluid, supporting the analysis of multiple markers indicative of pancreatic cancer from single pancreatic cyst fluid samples without high volume requirements. Currently the group is engaged in a statistically powered study with a larger cohort and a panel of three biomarker assays that is expected to produce an NPV of >95% for identifying non-mucinous pancreatic cyst fluid. Again, all three assays are run using only 12 μL of clinical sample, meaning that these efforts do not disrupt current standards of clinical care, and analyze a sample volume that would be incompatible with contemporary clinical assays.