Metabolomics Proling of EUS-FNA Sample Predicts Advanced Pancreatic Adenocarcinoma Prognosis

Background Effective anti-tumor medicine for individual cases with different prognosis risks is urgent for pancreatic adenocarcinoma (PAC) therapy. Metabolic therapy is a promising strategy for PAC. Endoscopic ultrasound-guided ne needle aspiration (EUS-FNA) is a suitable technique for the biopsy of pancreatic mass. Here, we rst perform metabolome analysis with EUS-FNA samples to identify prognostic biomarkers and potential therapeutic targets in advanced PAC.


Introduction
Pancreatic adenocarcinoma (PAC) is one of the top ten leading causes of death from cancer in the world.
(1) Advanced PAC is highly lethal with median survival time less than 1 years. The traditional chemoagents were always powerless, restricted by many factors, such as the hypo-vascularly anaerobic microenvironment. Clinically, the urgent need for PAC therapy is an effective anti-tumor medicine for individual cases with different prognosis risks.
Metabolic process is highly active in the pancreas due to its complex exocrine and endocrine functions.
Numerous studies have proved the distinctive metabolic features in the PAC anaerobic microenvironment, such as the alteration in carbon source to fuel anabolic processes, which might provide great clues in the development of metabolic therapy strategy. (2,3) Metabolomics is an omics technology that enables the large-scale quanti cation of small molecules termed metabolites. Several studies have investigated tumor metabolism in the discrimination between benign or malignant behavior and prognosis prediction in PAC. Urine, serum, or resected tissue samples with surgery were mostly applied. Among them, surgical samples were the most representative but rarest, or even unavailable, for advanced PAC. Endoscopic ultrasound-guided ne needle aspiration (EUS-FNA) is a suitable technique for the biopsy of pancreatic mass to obtain local cytological samples. Clinically, samples from EUS-FNA might be delivered to genetic sequencing to guide individual therapy.(4) In the current study, we performed metabolome analysis with EUS-FNA samples to identify prognostic biomarkers and potential therapeutic targets in advanced PAC.

Design and Patients
Previously, we conducted a retrospective study to investigate metabolomic differences between pancreatic head and body/tail cancer. Thirty PACs located in the head and 30 PACs located in the body/tail were enrolled; these samples were from patients who underwent EUS-FNA at the Department of Endoscopy, Fudan University Shanghai Cancer Center (FUSCC) between October 2018 and March 2019.
With EUS-FNA, all patients were pathologically diagnosed by positive or suspicious cytology for malignancy. Survival data was collected according to FUSCC cancer patients' survival follow-up protocol.
The follow-up period was measured as the date of clinical therapy to the date of the last hospital or telephone visit. All cases were allowed for a follow-up period of more than 1 year. All hospitalized patient were given informed consent about the potential sample collection when received biopsy, surgery, or et el., approved by the institutional review board of the FUSCC. Hence, need for another speci c approval was waived. All cases were given informed consent about the procedure of EUS-FNA.

Sample collection and preparation
The detail of EUS-FNA procedure was described as before.(5) Two experienced endoscopists performed all the procedures with linear longitudinal ultrasound endoscopes (Pentax EG3270UK and EG-3870UTK). After inserting the needle into targeted area, a 10 ml syringe was set to activate the suction. The cytological material was rapid on-site evaluated for malignancy. When the positive diagnosis was obtained, the remain cytological samples were ushed into tubes and stored at -80 °C until use. All the materials were nally delivered to the cytological and histological examination.
The cytological sample was mixed with methanol, then vortexed for 30 s, and ultrasound for 20 min. The sample was centrifuged at 12000 rpm for 15 min. The supernatant (200 μL) was transferred to vial for Liquid chromatography coupled with mass spectrometry (LC-MS) analysis. Equal volumes (20 μL) of each sample were mixed and pooled into the quality control sample.

LC-MS analysis
All experiments were performed on an Ultimate 3000LC platform (Thermo) and equipped with Hyper gold C18 (100x2.1mm 1.9 μm) column. Chromatographic separation conditions including column temperature at 40 °C, ow rate at 0.35 mL/min, automatic injector temperature at 4 °C, injection volume at 10 μL. The mobile phase A was water plus 5 % acetonitrile and 0.1% formic acid. The mobile phase B was acetonitrile plus 0.1% formic acid. Metabolomics feature extraction and preprocessed were performed with Compound Discoverer software (Thermo). The raw data was normalized and edited into twodimensional data matrix by excel 2010 software, including retention time, compound molecular weight, samples and peak intensity.

Survival model development
The model development and statistical analyses were performed using the R version 4.0.3. All analyses were set at two-sided p value <0.05 as the threshold for statistical significance. All metabolites were delivered into univariate Cox analysis for overall survival (OS). The candidate metabolites with P-value < 0.05 were selected for multivariate Cox analysis. The risk-score was calculated based on the identi ed prognostic metabolites and each coefficient. Based on mean value of risk-score, patients were strati ed into high-risk and low-risk groups. Forest plots were draw to demonstrate the Hazard ratio (HR) of prognostic metabolites. The risk-score was further combined with clinical parameters to construct a integrated predictive model. Kaplan-Meier plot was depicted for different risk groups. To cluster the metabolites, the Weighted Gene Co-expression Network Analysis (WGCNA) was performed with the R package. In the sample clustering, one outlier was identi ed and excluded. A soft threshold of 3 was set, based on the scale free topology model. In the module clustering, the minimal metabolite number was set as 50. The enrichment analysis was performed in online website MetaboAnalyst. (6)

Results
The clinical features of all 60 PACs were described in Table 1. The mean age was 64.6 years, and 38 cases were male. The lesions were located in the pancreatic head in 30 cases and body/tail in 30 cases.
The mean size was 36.0 mm (range, 21.0-60.5 mm). Clinically, 48 (80.0%) cases received gemcitabinebased chemotherapy, 2 cases received mFOLFIRINOX, and the remaining 10 cases did not receive any chemotherapy. In LC-MS analysis, 2,550 features and 2,307 features were obtained at the electrospray ionization negative and positive (ESI − and ESI + ) ionic modes. A total of 587 metabolites were identi ed and annotated based on compound molecular weight and peak intensity. For metabolites with multiple items, the average level was calculated, and 402 unique metabolites were nally included in the analysis.
Collectively, all patients with PAC were followed up for at least 1 year. The median follow-up time was 352 days, ranging from 0 day to 762 days. Forty-ve patients died at the time of the last follow-up (January 15, 2021), which included all the ten cases without any chemotherapy. The median survival time was 390 days, ranging from 324 days to 492 days. Initially, 16 metabolites were identi ed by univariate analysis (Fig. 1). Further multivariate analysis indicated that cholesterol glucuronide (HR = 1.181, 95%CI = 1.092-1.276, p < 0.001) and taurocholic acid 3-sulfate (HR = 1.132, 95%CI = 1.058-1.210, p < 0.001) were the independent predictors for overall survival (OS). As shown in the human metabolome database, both cholesterol glucuronide and taurocholic acid 3-sulfate were associated with bile acid metabolism. In PAC, the head tumor always induced the obstructive dilation of the bile duct and cholestasis instead of the body/tail tumor. As shown in Supplemental Fig. 1, the level of taurocholic acid 3-sulfate was slightly higher in cancers with head location and bile duct dilation, than body/tail location and bile duct nondilation. As for cholesterol glucuronide, no any signi cant difference in such two conditions.
In accordance with the coe cient of prognostic metabolites, the risk score was calculated and strati ed into the high-risk (n = 15) and low-risk (n = 45) groups. The C-index of the risk score was 0.64 (95% CI: 0.70-0.59). The areas under the curve at 1 and 2 years were 0.685 and 0.588, respectively ( Fig. 2A). As shown in Fig. 2B, the levels of the two metabolites were higher in the high-risk group than in the low-risk group (p < 0.001 for cholesterol glucuronide, p < 0.05 for taurocholic acid 3-sulfate). The Kaplan-Meier plot indicated that the high-risk group had signi cantly worse prognosis than the low-risk group (p < 0.001), with median survival times of 195 and 480 days, respectively (Fig. 2C).
Further, clinical parameters were incorporated into the Cox model with risk score to construct the predictive model. With univariate and multivariate analyses, therapy (HR = 0.270, 95% CI = 0.097-0.747, p = 0.012; Fig. 3) and risk score (HR = 1.983, 95% CI = 1.362-2.887, p < 0.001; Fig. 3) were found to be the independent predictors for prognosis. The predictive model was depicted in the nomogram (Fig. 4), with a C-index of 0.67 (95% CI: 0.73-0.62). A credible predictive model based on the metabolites was nally constructed for PAC prognosis.
Subsequently, we attempted to investigate the metabolic pro le in different risk groups. As shown in the volcano plot (Fig. 5A), 29 metabolites showed higher levels in the high-risk group than in the low-risk group, and the difference was statistically signi cant (p < 0.05). The detailed level of signi cant metabolites is shown in the heatmap (Fig. 5B).
Finally, we performed WGCNA to cluster the metabolites and identi ed three distinct modules, namely, turquoise, blue, and grey (Figs. 6A and 6B). All the three showed no signi cant association with clinical parameters (Supplemental Fig. 2). A total of 243 metabolites were clustered into the turquoise module and majorly enriched in the glucose-alanine cycle. A total of 150 metabolites were clustered into the blue module and majorly enriched in α-linolenic acid metabolism. Only nine metabolites were clustered into the grey module, and enriched in no metabolic pathway. Details are shown in Figs. 6C and 6D.

Discussion
The current study validated the ability of metabolomics pro le by using EUS-FNA-acquired cytological samples in the prediction of PAC prognosis. A two-metabolite risk score (cholesterol glucuronide and taurocholic acid 3-sulfate) predicted the OS. We further depicted a nomogram by combining the risk score and clinical parameters. Finally, WCGNA clustered all metabolites into three modules, which majorly enriched into the glucose-alanine cycle and α-linolenic acid metabolism.
Several studies investigated tumor metabolism in the prognosis prediction of PAC through the use of urine, serum, or surgical samples. Battini et al. indicated that surgical sample-based metabolic biomarkers can predict the long-term survival of resectable PAC. (7) Thereafter, serum samples were also proved to be useful in the prediction of recurrence-free survival.(8) However, the reliability of urine and serum samples may be questioned because they are susceptible to several factors, including gender, age, body mass, and metabolic diseases.(9-11) Surgical samples are suitable and representative but infeasible for most advanced pancreatic cancer. EUS-FNA is a suitable technique for the biopsy of pancreatic mass to obtain local cytological samples. Clinically, samples from EUS-FNA might also be delivered to genetic sequencing to guide individual therapy.(4) However, the application of EUS-FNA in metabolomics has never been reported. In Shanghai Cancer Center, we established a metabolomics analysis system with EUS-FNA-acquired cytological samples, which proved to be reliable. Here, we initially validate the predictive value for advanced PAC prognosis with this system.
In the 402 identi ed unique metabolites, only cholesterol glucuronide and taurocholic acid 3-sulfate were proved to be independent predictors for OS. Glucuronidation is a pathway for the detoxi cation and elimination of toxic substances. Cholesterol glucuronide is generated in the liver by UDP glucuronyltransferase, and it is reported to prevent cholesterol gallstone initiation via bilirubin biotransformation.(12) Bile acid sulfation is a minor pathway in normal circumstances but increases in the presence of cholestasis to accelerate renal clearance and decrease intestinal reabsorption. Physically, both the two metabolites are involved in the feedback control of bile acid accumulation. Experiments already revealed the toxicity of bile acid on pancreatic acinar cell and ductal epithelium. (13,14) In our data, both metabolites signi cantly increased in the high-risk group and independently predicted PAC prognosis. We speculated that elevation of the two metabolites was a protective consequence of cholestasis induced by PAC progression. The predictive value of bile duct dilation was also evidence of such a conjecture. However, one point should be noted. PAC in the head results in bile duct dilation and cholestasis, but PAC in body/tail presents negative bile duct dilation and suffers no direct stimulation from bile acid. Our data did not show quite signi cant difference of both two metabolites as for tumor location and bile duct dilation. We speculated that bile acid might suffer from stacked excretion in all locations and is exerted on PAC cells via systemic circulation.
The WGCNA found that all the metabolites could be clustered into three distinct modules, and they were enriched in glucose-alanine cycle or α-linolenic acid metabolism. Cancer shows different metabolic features, compared with normal counterparts, such as the alteration in carbon source to fuel anabolic processes. (15) Alanine, a non-essential amino acid, was reported to be secreted by pancreatic stellate cell and outcompeted glucose-derived carbon in PAC to fuel the tricarboxylic acid cycle. (16) The current data veri ed that the glucose-alanine cycle was highly active in PAC, which indicate its may be potential target for future anti-cancer therapy. Studies already proved the protective effect of α-linolenic acid in several in ammatory conditions, such as ulcerative colitis and hepatic steatosis. (17,18) Meanwhile, epidemiologic evidence on the risk of PAC remains controversial. (19) As a member of n-3 polyunsaturated fatty acids, α-linolenic acid plays a distinctive role of carbon source substitution in athletes' aerobic performance when the carbohydrate reserve is consumed. (20) In advanced PAC, whether α-linolenic acid contributes to the carbon source into anaerobic metabolism still needs further investigation.
Our work had several limitations. First, the primary target of the cohort was metabolic pro le in different PAC locations. Thus, head and body/tail PAC with even numbers were enrolled. Such a condition might generate selection bias to the current analysis, to some extent. Second, the current predictive model should be validated in further external cohorts.
In conclusion, our data validated the ability of metabolic pro les by using EUS-FNA-acquired cytological samples in the prediction of PAC prognosis. A risk score formulated by cholesterol glucuronide and taurocholic acid 3-sulfate, which both were associated with bile acid accumulation, was constructed. Furthermore, a nomogram was depicted to predict the survival probability for advanced PAC. Enrichment analysis indicated that glucose-alanine cycle and α-linolenic acid metabolism demonstrated high activity in advanced PAC, which might be a potential target in future therapy.

Declarations
Ethics approval and consent to participate All hospitalized patient were given informed consent about the potential sample collection when received biopsy, surgery, or et el., approved by the institutional review board of the Fudan University Shanghai Cancer Center. Hence, need for another speci c approval was waived. All cases were given informed consent about the procedure of EUS-FNA.

Consent for publication
Not applicable.

Availability of data and materials
The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. Identi cation of the prognosis-associated metabolites for PAC. In the forest plot, 16 metabolites were identi ed by univariate analysis. Multivariate analysis indicated that cholesterol glucuronide and taurocholic acid 3-sulfate were the independent predictors for OS.

Figure 2
Prognostic analysis of the two metabolites' risk score. The AUC at 1 and 2 years were 0.685 and 0.588, respectively (A). Level of the two metabolites in the high-risk and low-risk group were shown (B). Kaplan-Meier plots showed PAC OS between the high-risk group and low-risk group (C).

Figure 3
Identi cation of the metabolites and clinical associated predictors for PAC survival. With the univariate and multivariate analyses, therapy and risk score were the independent predictors for prognosis.

Figure 4
Nomogram of the predictive model for PAC prognosis. The survival probability is shown in A, and median survival time is shown in B.

Figure 5
Level of metabolites in different risk groups. As shown in the volcano plot (A), 29 metabolites showed higher levels in the high-risk group than in the low-risk group, with the difference being statistically