Candidate serum metabolite biomarkers for differentiating gastroesophageal reflux disease, Barrett’s esophagus, and high-grade dysplasia/esophageal adenocarcinoma

Introduction

Incidence of esophageal adenocarcinoma (EA) has increased over 8-fold in recent decades, while median survival remains less than one year (1). Most EAs develop in a metaplastic epithelium termed Barrett’s esophagus (BE), which can arise in the setting of chronic gastroesophageal reflux (GERD). Besides recurrent GERD symptoms, which occur in 20–30% of US adults, the major known risk factors are white race, male sex, abdominal obesity, and smoking (2–4). The prevailing clinical paradigm is to identify BE patients, enroll them in long-term surveillance, and aggressively treat them before they develop invasive cancer. However, 95% of BE patients will never progress to cancer, while only ~5% of EA cases are diagnosed as a result of screening and surveillance. Upper endoscopy with biopsy, the standard method of diagnosing and monitoring BE, is both invasive and expensive, and its frequent yet variable implementation has not appreciably increased early detection or decreased EA mortality (1). Development of new and non-invasive methods for stratification and assessment of patients at risk of EA would address important unmet clinical needs (1, 4).

Metabolomics has emerged as a promising domain for identifying molecular signatures associated with cancer (5–10). In contrast to genomics, transcriptomics, or proteomics, metabolomics describes the study of concentrations and fluxes of low-molecular-weight metabolites present in biofluids or tissue (11, 12). As metabolic fluctuations lie downstream of alterations at the DNA, RNA, and protein level, metabolomics offers a sensitive and comprehensive functional read-out of biological systems. Metabolite levels and metabolic pathways are significantly altered in a broad spectrum of human cancers (13–16). Adaptive changes such as increased rates of aerobic glycolysis (the Warburg effect), amino acid metabolism, and lipid turnover are thought to be important features of the neoplastic process that facilitate sustained cellular proliferation and tumor expansion (15, 16).

Several past metabolomics studies, generally of modest sample size, have focused on esophageal adenocarcinoma (17–25). In one of the earlier reports, ¹H magic angle spinning-nuclear magnetic resonance (NMR) spectroscopy-based profiling of intact esophageal tissue samples identified a panel of metabolites that separated control tissues from tumor tissues and proximal histologically normal mucosa from individuals with EA (23). In a second study, Raftery and colleagues conducted a targeted mass-spectrometry (MS)-based analysis of eight nucleosides in serum samples from EA cases or healthy controls, and identified five as differentially abundant (19). Serum metabolites identified using NMR and LC-MS were combined to generate a signature capable of accurately discriminating between EA patients and healthy controls (24). Urinary metabolite panels identified by ¹H NMR spectroscopy have also facilitated separation of healthy controls from patients with BE or those with EA or esophageal squamous cell carcinoma (ESCC) (18). Most recently, serum levels of the amino acid L-proline, ketone body 3-hydroxybutyrate, and carbohydrate D-mannose were found to differ between EA cases versus healthy controls (21). Analyses in cell lines and isolated tissues specimens have described alterations in both energy metabolism and metabolically-relevant gene expression patterns across the disease continuum from BE to EA (26, 27). Collectively, while this overall body of work has suggested that metabolic perturbations are associated with BE/EA, very limited agreement between studies has been observed with respect to specific metabolite alterations, likely reflecting significant heterogeneity in study design, sample collection/processing, analytic platforms, and statistical approaches, as well as the confounding effects of modifying factors (age, race, smoking, etc).

In this discovery study, we used an optimized LC-MS targeted profiling platform to screen for differentially abundant serum metabolites among individuals with i) BE versus GERD, and ii) HGD/early-stage EA versus BE. We thus focused on clinically-relevant comparisons, rather than analyzing readily distinguishable groups such as cancer cases and healthy controls. A total of 322 serum samples (100 GERD, 122 BE, 100 HGD/EA) were selected from two prior epidemiologic studies of BE and EA and subjected to targeted metabolomic analysis of 57 metabolites, which collectively encompass several major metabolic pathways (glycolysis/TCA cycle, amino acid, nucleotide, and fatty acid metabolism). The potential utility of derived metabolite profiles to discriminate between these disease states was investigated through construction and evaluation of multivariate classification models.

Materials and Methods

Results

Discussion

The prevention and control of EA has become an increasingly important public health challenge, as incidence of this rare but lethal cancer has risen 8-fold in recent decades (1). Despite identification of an epithelial precursor lesion and several strong epidemiologic risk factors, current screening and surveillance efforts, based largely on upper endoscopy with biopsy of patients with GERD or BE, have not significantly improved early detection or lowered EA mortality (1, 4). An important unmet need remains the development of robust non-invasive methods of assessing and stratifying at-risk patients according to their level of current, and risk of future, disease progression (39). Here, we adopted metabolomics profiling methods to identify serum metabolites correlated with distinct disease intermediates along the pathway to EA. Targeted interrogation of 57 serum metabolites encompassing several major metabolic pathways revealed multiple metabolites that differed in abundance when comparing patients with BE versus GERD, and HGD/EA versus BE. Multivariate models based on candidate metabolite markers yielded moderate discrimination between the HGD/EA and BE case groups, with lesser separation observed for BE versus GERD.

Despite a large number of cancer metabolomics studies in recent years, relatively few have focused on EA (or BE). These mostly small-scale investigations employed various NMR- or MS-based methods to conduct either targeted or global metabolite profiling in tissue, serum, or urine (17–25). Considerable variability has been observed among reported findings, likely reflecting significant differences not only in analytic platforms and types of biospecimens, but also in study design, subject selection, and sample preparation (40). Differentially abundant metabolites identified by NMR in EA tissue relative to normal tissue have included phosphocholine, glutamate, myo-inositol, adenosine-containing compounds, uracil, and inosine (23). Studies comparing serum from EA patients versus healthy controls, using either MS or NMR, have described differences in multiple nucleosides, amino acids, and other metabolites (19, 21, 24, 25). NMR profiling of urine from healthy controls or EA/ESCC patients has identified differences in sucrose, pantothenate, urea, methanol, pyroglutamate, hypoxanthine, fucose, 2-aminobutyrate, and cis-aconitate (18). In total, minimal consensus has emerged regarding consistent metabolite-based signatures of EA, and our biological understanding of the inter-relationships between metabolite levels in tissues, blood, and urine remains limited. Furthermore, no previous studies have included at-risk GERD patients as a comparison group.

The individual metabolite marker candidates identified in our analysis belong to several classes of compounds. The nine metabolites differentiating BE from GERD (P<0.05) included seven amino acids or amino acid derivatives (creatine, homocysteine, 3-nitrotyrosine, hydroxyproline/aminolevulinate, arginine, tyrosine, ornithine), one fatty acid (linoleic acid), and one sugar alcohol (sorbitol). The four metabolites differentiating HGD/EA from BE (P<0.05) included one nucleotide derivative (urate), two amino acid derivatives (taurine and xanthurenate), and one sugar (erythrose). On a statistical level, the most significant signal observed was for urate (P=0.0002, q=0.01), in the HGD/EA versus BE comparison, and for creatine and homocysteine (q<0.15) in the BE versus GERD comparison. Urate also showed non-significant elevation in BE versus GERD (P=0.12), while homocysteine was similarly non-significantly elevated in HGD/EA versus BE (P=0.17).

It is of interest to note that several of the top identified metabolites (urate, homocysteine, and 3-nitrotyrosine) have been linked to inflammatory processes (41–43), as local and systemic inflammation are believed to play important roles in the pathogenesis of BE/EA (4). Urate is the monosodium form of uric acid, a derivative of purine nucleotides. High levels of circulating uric acid have been associated with a number of medical conditions, including gout, kidney stones, tumor lysis syndrome, metabolic syndrome, and diabetes (42), as well as with increased risk of several cancers (44, 45). Originally described as an antioxidant, uric acid has more recently been found to promote inflammation, both in its crystal and soluble forms, through activation of Toll receptors and stimulation of NFκb, respectively (42). The amino acid homocysteine, a derivative of methionine, functions in one-carbon metabolism reactions, which are important for nucleotide synthesis and DNA methylation. Elevated circulating homocysteine has been associated with increased risk of head and neck cancers, colorectal cancer, and colorectal adenoma recurrence (46–48). Homocysteine induces inflammatory cytokine production in cultured human cells, and has been mechanistically linked to inflammatory bowel disease (IBD) and atherosclerosis (41, 49). Nitrotyrosine (NT) is derived from nitration of the amino acid tyrosine, mediated by reactive oxygen species such as peroxynitrite anion. NT is considered a marker of oxidative stress, inflammation, and protein damage, and has been detected in a number of disease states such as gastritis and IBD (43). Past studies have also described elevated NT immunoreactivity in esophageal tissue specimens from patients with BE (43, 50) and EA (43, 51), and in the esophagus of an animal model of EA (52).

Local and systemic inflammation represent plausible drivers of altered metabolite levels, as do increased rates of local cell proliferation associated with cancer precursors and early lesions. Metabolite alterations observed across case types may or may not reflect underlying metabolite changes within the esophageal epithelium itself. Interrogation of paired tissue biopsies from the same individuals and time points in which serum specimens were isolated will be required to address this uncertainty and facilitate accurate biological interpretation of the reported metabolite differences.

Non-invasive clinical assessment tools for patients at risk of EA could be of use in at least two clinical scenarios: i) initial evaluation of patients with severe or chronic GERD symptoms, to identify those most likely to have BE (or advanced pathology); and ii) additional monitoring of BE patients between surveillance endoscopies, to identify those most likely to have HGD or early-stage EA. Preliminary risk models for BE or EA based on classic epidemiologic risk factors such as age, sex, BMI/WHR, smoking status, and GERD symptoms (53–56) have provided a promising starting point for improved risk stratification tools, but the modest classification accuracy of most models requires improvement to afford significant clinical utility. One emerging approach for enhancing non-invasive risk assessment is the Cytosponge, a cytological sampling device that yields esophageal epithelial cells for immunostaining or molecular analyses (57, 58). Our results suggest that the serum metabolome may represent another viable source of clinical information, and point to the potential utility of integrating novel candidate biomarkers with existing classical models to build more robust classifiers. Given that existing tissue-based variables such as dysplasia grade, DNA content flow cytometry, and somatic chromosome abnormalities provide strong discrimination among BE patients (59), serum-based assays will require substantial improvements to have direct clinical application. Future studies are also needed to probe whether serum metabolite signatures are specific to BE or HGD/EA, or may correlate with other conditions prevalent in a larger sample of at-risk individuals.

This discovery study had a number of key strengths. First, in contrast to past reports that primarily used healthy normal controls, our analysis focused on comparisons most relevant to clinical decision making and targeted risk assessment: BE versus GERD, and HGD/EA versus BE. Second, GERD, BE, and HGD/EA were all diagnosed according to standard definitions based on endoscopic and histopathologic evaluation, and serum specimens were processed according to uniform procedures and frozen at −80 °C within four hours for long-term storage. Third, biospecimens were linked with extensive epidemiologic and clinical data from high-quality studies of BE and EA. Statistical analyses were adjusted as needed for a number of potential confounders, including age, sex, smoking, and waist-to-hip ratio. Fourth, to avoid capturing metabolite alterations associated exclusively with advanced/incurable disease, EA cases were restricted to Stage 0/1. Finally, our LC-MS-based profiling platform has been optimized for sensitivity and accuracy, and used successfully to identify and/or validate serum metabolite biomarker candidates for a number of diseases (33, 60).

Our study also had several limitations. First, while a total sample size of 322 participants was larger than that used in most previous metabolomics studies of BE/EA, it is nonetheless still relatively small when considering the modest magnitude of disease-associated serum metabolite fluctuations typically observed, coupled with the anticipated degree of natural biologic (and technical) variability in such measurements (61). Given that participants in the present analysis were drawn from two distinct parent studies (SRD & SBEP), we chose to retain two separate BE case groups and conduct only within-study comparisons (BE versus GERD, and HGD/EA versus BE), to eliminate the potential for confounding by systematic differences between participants/specimens by parent study (eg. SBEP blood samples were all collected under fasting conditions prior to endoscopy, while SRD samples were collected in the field under non-fasting circumstances). This decision effectively reduced study power by decreasing the size of our BE comparison groups, and precluded a comparison of HGD/EA vs. GERD patients. Second, while our targeted LC-MS analysis enabled assessment of 57 metabolites in multiple metabolic pathways, these compounds represent only a very small fraction of the total serum metabolome. The combination of disparate detection technologies, such as MS and NMR, and the pairing of targeted and global profiling approaches, may facilitate identification of a richer set of differentiating metabolites (24). Exploratory global MS profiling in the present study, however, did not yield strong additional signals (data not shown). Third, while our analyses made use of extensive covariate data on study participants for statistical adjustments, often not conducted in metabolomics reports, the possibility remains for some degree of residual or unmeasured confounding, whereby observed metabolite differences between case types might reflect underlying subject/specimen characteristics. Medications and dietary patterns are two variables capable of influencing metabolite levels (62), which were not included in our analyses. However, given that we restricted EA cases to stage 0/1 (prior to surgical or chemoradiation treatment) and used refractory GERD cases as our primary “controls” (instead of healthy normals), substantial heterogeneity in medication usage between case groups appears relatively unlikely. It remains uncertain whether significant dietary differences exist between GERD, BE, and HGD/EA patients.

This study provides evidence that intermediate disease states preceding the development of EA correlate with detectable alterations in the serum metabolome. Our results expand upon the findings of a small number of past analyses by identifying metabolite differences across gradations of disease progression (BE versus GERD, and HGD/EA versus BE) that are most relevant to the clinical settings of screening and surveillance. Further studies in larger independent populations will be necessary to validate these results, and ultimately to evaluate the utility of serum metabolite profiles as a novel source of clinical information for improving non-invasive assessment and monitoring of patients at risk of EA.

Supplementary Material