Circulating miRNA Increases the Diagnostic Accuracy of Chromogranin A in Metastatic Pancreatic Neuroendocrine Tumors

Simple Summary Despite its varying sensitivity and decreased specificity, chromogranin A (CgA) is the most widely used biomarker for neuroendocrine tumors. The most common factor affecting its diagnostic accuracy is the use of proton pump inhibitors (PPIs). Our aim was to investigate circulating miRNA expression profiles in pancreatic neuroendocrine tumors (pNET) and pheochromocytomas/paragangliomas (PPGL) to find miRNAs which could be used as biomarkers along with CgA in these patients. MiRNA expression profiles were determined with next generation sequencing and validated by quantitative real time PCR in 74 samples obtained from patients and healthy volunteers treated with PPI. We observed a global downregulation of miRNAs in NET compared to controls. A set of miRNAs in combination with CgA resulted in the best discrimination of pNET irrespective of PPI treatment and a combination of miRNAs increased the diagnostic utility of CgA even in pNET patients with low CgA. Abstract Chromogranin A (CgA) is the most widely accepted biomarker for neuroendocrine tumors (NET) but its diagnostic accuracy is dependent on tumor type and the use of proton-pump inhibitors (PPI). We investigated the diagnostic value of circulating miRNAs along with CgA in pancreatic neuroendocrine tumors (pNET). 74 serum samples from patients with pNET (n = 25, nonfunctioning), pheochromocytoma/paraganglioma (PPGL, n = 20), healthy individuals with normal CgA (n = 29) including 10 samples from 5 healthy individuals with and without current PPI treatment were collected. MiRNA expression profiles were determined using next-generation sequencing, followed by validation with individual TaqMan assays. A global downregulation of miRNAs was observed in patients with NET compared to controls. MiRNA expression of 33 miRNAs was able to discriminate tumor samples from controls. No miRNA alone could be considered as an applicable biomarker for pNET or PPGL. However, using a logistic model, the combination of a set of miRNAs increased the discriminatory role of CgA irrespective of PPI treatment. In pNET patients with normal CgA level our regression model yielded high (89.4%) diagnostic accuracy (AUC: 0.904, sensitivity: 66.6%, specificity: 96.5%). A set of miRNAs increased the diagnostic utility of CgA in pNET even in patients with low CgA.


Introduction
Chromogranins are polypeptide prohormones being the major constituents of dense-core secretory granules in neuroendocrine cells and they are co-secreted with peptide hormones and amines [1].
Neuroendocrine tumors (NETs) originate from neuroendocrine cells distributed widely throughout the body [2]. They are rare neoplasms but their incidence and prevalence are steadily rising [3]. Pancreatic NET (pNET) is the most common neuroendocrine tumor that can either be functioning or nonfunctioning according to hormone production [2]. Nonfunctioning (NF) pNETs comprising the largest group of pNETs being asymptomatic result in delayed diagnosis and often present as metastatic at diagnosis that significantly affect prognosis and survival [2,4].
In addition, many oncological and non-oncological conditions, including renal failure, non-endocrine tumors (e.g., prostate, breast, thymus, uterus, colon), chronic gastritis, and current proton pump inhibitor therapy (PPIs) decrease the specificity CgA [6,9], but CgA is the only routinely measured circulating marker in pNET. Previous studies, including one from our research group, showed that both short-and long-term applications of PPIs and other acid-suppressive medications result in a significant increase of CgA in blood, thus decreasing its clinical utility in the management of NETs [10,11].
Neuron-specific enolase (NSE) is a cytosolic enzyme of neurons and neuroendocrine cells, that is predominantly expressed in small-cell lung cancer and poorly differentiated NETs [12,13]. Its diagnostic value, particularly its specificity for neuroendocrine tumors is rather poor [8,14,15]. 5-Hydroxyindoleacitic acid (5-HIAA), as the main metabolite of serotonin, is a useful biomarker of NETs in patients with carcinoid syndrome [12]. Other biomarkers, such as insulin, glucagon, or gastrin are specific for subtypes of neuroendocrine cells [12].
Pheochromocytomas/paragangliomas (PPGLs) are rare neuroendocrine tumors of chromaffin cell origin. These are found in the adrenal medulla (pheochromocytomas, PCC) and the paravertebral ganglia of the nervous system (paragangliomas, PGL) [16]. They can produce catecholamines, but PGLs located above the diaphragm are often non-secretory [16]. In PPGL, as CgA is co-secreted with catecholamines, besides the determination of catecholamines and their metabolites, elevated CgA may indicate tumor mass and malignancy; therefore, it may be used to monitor clinical response and tumor relapse especially when the measurement of plasma metanephrines is not accessible [16].
Similarly to gene expression, miRNA expression is also cell type-specific and due to their presence in circulation and their stability, they are considered as promising biomarker candidates in several tumor types [17,18]. Their previously described expression alterations in pNET and PPGL raise their potential usefulness as circulating biomarkers [19,20]. A multianalyte biomarker (NETest) encompassing 51 different NET-related transcripts showed a significantly more sensitive and efficient method in NET diagnosis compared to single biomarkers [21], however, its availability and costs may limit its usefulness [12].
We aimed to investigate the serum miRNA expression profile in patients with pNET to identify potentially useful miRNA biomarkers. PPGL and healthy individuals with and without PPI treatment were also studied to challenge the diagnostic utility of identified miRNAs along with CgA.

Evaluation of Cell Contamination Based on miRNA Signature
As intracellular miRNA concentration is much higher compared to those measured in a cell-free environment, cellular contamination is an important factor to avoid. Therefore, first, we assessed whether cellular contamination in our serum samples was present. In the collected samples visually, no sign of hemolysis could be seen. The ratio of red blood cell-enriched miR-451a to the reference miR-23a-3p was revealed as the most sensitive method for the detection of hemolysis compared to hemoglobin measurement techniques (Coulter ® AcT diff Analyzer and spectrophotometric method) [22]. Assessing miR-451a to miR-23a-3p ratio no red blood cell (RBC) contamination was detected in any of our samples.MiRNAs which are considered indicating red-blood-cell or platelet cellular contamination [23][24][25] also showed low variance (0.4 for miR-142-3p and 0.3 for miR451a/miR-23a-3p, respectively). Only samples without cellular contamination were included.

Analysis of Circulating miRNAs by Next-Generation Sequencing
Our exploratory study included 24 samples: 8 pNET (4 patients with low-and 4 with high-CgA), 8 PPGL patients and 8 control samples from 6 healthy individuals (2 of them were also treated with PPI, samples before and after PPI treatment were both included). Altogether, 1525 miRNA were detected in NGS profiling of whole serum samples. While nine miRNAs were the most abundant with an >500 UMI reads/sample representing 0.05% of all detected miRNAs, the great majority of miRNAs (1461 of 1525, 95%) were in the low-expression range (below 50 UMI reads/sample) ( Figure 1A,B). Total (global) miRNA expression in patients with pNET or PPGL was significantly decreased compared to normal controls ( Figure 1C).
Low abundance (<50 UMI read number) miRNAs were excluded from further analysis because these miRNAs could not be validated in our experimental set up [26]. 33 miRNAs differentiated pNET/PPGL patients from normal controls irrespective of PPI treatment (Table 1, Figure 1D). In pNET 19 and 7 under-and over-expressed miRNA, while in PPGL 25 and 6 under-and over-expressed miRNAs compared to healthy controls were found (Table 1). However, PNET and PPGL patients could not be differentiated by serum miRNA expression ( Figure 1D). Additionally, we did not find significant differences in healthy controls before and after PPI treatment.

miRNA Validation
Of the 33 significantly differently expressed miRNAs between controls and tumor sera samples we selected 6 underexpressed miRNAs (let-7b-5p, let-7i-5p; miR-143-3p; miR-30d-5p; miR-451a; miR-486-5p) in pNET/PPGL samples compared to controls) based on miRNA abundance for further investigation. Additionally, we also validated the most overexpressed miR-203a-3p despite its relatively lower abundance (avg. UMI read number in pNET/PPGL samples: 68 vs. avg. UMI read number in controls: 19). The selected miRNAs were measured on an extended sample cohort consisting of 25 pNET and 20 PPGL patients (together with those included in the exploratory cohort), and 29 control samples (5 with PPI treatment, 24 without PPI treatment; including samples from our exploratory cohort).
All the 6 underexpressed miRNAs were confirmed by RT-qPCR as well ( Figure 2A). After dissecting tumor groups, the underexpression of miR-30d-5p, miR-451a, and miR-486-5p was significant in pNET patients with high CgA levels and let-7b-5p in PPGL cases compared to healthy controls without PPI treatment ( Figure 2B).
The overexpression of miR-203a-3p was not confirmed by RT-qPCR in any of the samples but one pNET patient with extremely high CgA (2490.8 ng/mL).

Association of miRNA Expression with Clinicopathological Parameters
We could not find a difference in miRNA expression profiles of pNET/PPGL samples compared to healthy controls treated with PPI. However, let-7b-5p, let-7i-5p; miR-143-3p; miR-30d-5p; miR-451a and miR-486-5p showed lower expression in samples with high (>100 ng/mL) CgA levels compared to samples with normal CgA level (≤100 ng/mL) ( Figure 3A). The expression of all 6 miRNAs showed a negative correlation with serum CgA level when all samples were included in the analysis. While investigating correlation in different patient groups the negative correlation between miRNAs and CgA remained significant only in the healthy control group but not in pNET or PPGL group (Table 2).

Association of miRNA Expression with Clinicopathological Parameters
We could not find a difference in miRNA expression profiles of pNET/PPGL samples compared to healthy controls treated with PPI. However, let-7b-5p, let-7i-5p; miR-143-3p; miR-30d-5p; miR-451a and miR-486-5p showed lower expression in samples with high (>100 ng/mL) CgA levels compared to samples with normal CgA level (≤100 ng/mL) ( Figure 3A). The expression of all 6 miRNAs showed a negative correlation with serum CgA level when all samples were included in the analysis. While investigating correlation in different patient groups the negative correlation between miRNAs and CgA remained significant only in the healthy control group but not in pNET or PPGL group (Table 2).

Healthy Controls (with & without PPI)
Investigating miRNAs in relation to the tumor grade, in pNET patients a tendency of inverse correlation between miRNAs: let-7i-5p, miR-30d-5p, and miR-451a and grade was observed ( Figure 3B).
In samples obtained from PPGL patients only miR-486-5p showed an inverse correlation with CgA ( Table 2). As different miRNA signatures have been described in malignant vs. benign [27] and sporadic vs. hereditary PPGL [28] we aimed to investigate serum miRNAs in correlation to these clinicopathological parameters. We found that in serum samples obtained from PPGL patients with germline mutations (SDHB, RET, VHL, or NF1) miR-486-5p and miR-30d-5p were downregulated compared to sporadic patients ( Figure 3C). In patients with SDHB germline mutation beside the underexpression of miR-30d-5p and miR-486-5p, let-7b-5p and let-7i-5p also showed downregulation compared to serum samples of patients with SDHB wild type ( Figure 3D).

Diagnostic Value of the Circulating miRNAs in PPGL/pNET
In the discrimination of healthy controls from pNET and PPGL patients, we performed receiveroperating characteristic (ROC) analysis for CgA and individual miRNAs. Binary logistic regression models were generated for combinations of miRNAs and combinations of CgA with miRNAs. Altogether 6 miRNAs (let-7b-5p, let-7i-5p, miR-143-3p, miR-30d-5p, miR-451a, and miR-486-5p) were assessed individually and in combinations (Table 3). In the pNET group compared to CgA the use of individual let-7b-5p, miR-30d-5p, miR-451a, and miR-486-5p in ROC analysis resulted in better  Clinically, the most challenging pNET patients are those with normal serum CgA levels (<100 ng/mL). Therefore, miRNA expression profiles are of particular interest in these patients. The use of the individual expression of let-7b-5p or miR-143-3p yielded better discrimination from healthy controls compared to CgA alone (Table 3). However, the combination of CgA value with an expression of 4 miRNAs in a logistic regression model showed the best discrimination between pNET-low CgA group compared to healthy controls including normal or elevated CgA with the accuracy of 89.4% (AUC: 0.904, sensitivity: 66.6%, specificity: 96.5%) ( Table 3).

Analysis of Deregulated pNET/PPGL Tissue miRNAs in Serum Samples
To investigate the potential origin of miRNAs in serum, we collected the data of 12 studies reporting tissue miRNA expression profiles in pNET [29][30][31][32][33][34][35] and PPGL [27,28,[36][37][38]. By cross-referencing differentially expressed miRNAs from those studies with our current findings 9 miRNAs in pNET and 3 miRNAs in PPGL were found commonly dysregulated in both tissue and serum. (Table 4). Among these, only miR-203a-3p was overexpressed both in pNET tissue and serum. All other miRNAs were upregulated in tumor tissues and downregulated in serum samples. In PPGL, while miR-101-3p showed higher expression in malignant vs. benign PPGLs, in serum, it was downregulated compared to controls. However, miR-16-5p and miR-451a as putative tumor suppressor miRNAs were downregulated in both tissue and serum (Table 4).

Discussion
Tumors of the neuroendocrine system are rare entities but represent serious diagnostic and therapeutical challenges. Especially, when they are hormonally inactive, the diagnosis is usually delayed due to a lack of specific circulating biomarkers. The delayed diagnosis influences prognosis and survival [2]. Therefore, novel circulating biomarkers are needed. To our best knowledge, no study on the diagnostic accuracy of circulating miRNAs in neuroendocrine tumors in relation to serum CgA has been published.
Our data revealed a low global miRNA expression (abundance) in sera of patients with pNET/PPGL compared to normal controls. The same phenomenon was also observed in blood samples obtained from patients with pituitary adenoma [26]. Interestingly, in tumor tissues, a general downregulation of miRNAs compared to normal tissues was also detected suggesting their general tumor-suppressive role instead of being oncogenes [39]. Based on the individual miRNA measurements, Mitchell et al., 2008 [17] reported that the presence of tumors does not lead to a generalized increase in circulating miRNAs. However, specific individual tumor-derived miRNAs can show increased expression in circulation. With the technical advancement of NGS, total miRNA abundance can be determined easier and more accurately than earlier used platforms (i.e., microarray) allowed. Still, both the causes and the role of the decreased global miRNA expression in circulation related to endocrine tumors are waiting to be further explored.
The origin of circulating miRNAs is still not clearly defined [40]. The high-abundance cell-free circulating miRNAs derive primarily from blood and endothelial cells, and tumor-derived miRNAs are possibly among the moderate or low abundance range which highlights the importance of the accurate detection of moderate-low abundance miRNAs [17,24,40]. The overexpressed miRNAs can originate directly from tumor tissues [17]. By cross-referencing these with the tissue miRNA expression we identified 3 miRNAs commonly dysregulated in both tissue and circulation. MiR-203a-3p was overexpressed both in pNET tissue and serum. Similarly, to our findings, in breast cancer, the same miR-203a-3p was overexpressed both on a tissue level and in exosomes secreted by tumor cells [41,42]. Functionally, it exhibits oncogenic effects in colon and breast cancer, and through MYC and it influences taxane sensitivity [41,43,44]. These data raise the possibility that miR-203a-3p may potentially be originated from tumor cells and can be considered as an oncomiR. However, its low abundance in the sera of patients with NET limits its wide clinical applicability as a tumor marker. Among downregulated miRNAs, miR-16-5p and miR-451a were both downregulated in PPGL tissues and sera indicating their roles as potential tumor suppressors that are well-established in other cancers [45,46].
Based on the expression pattern of 33 miRNAs we were able to separate pNET/PPGL patients from normal controls even on PPI treatment. Based on serum miRNA expression we could not distinguish samples of pNET from those of PPGL patients. This may reflect the common enterochromaffin cell type origin of these tumors [47].
Based on NGS data, several miRNAs showed (mostly inverse) correlation with serum CgA level. However, after validation on an extended sample set, this correlation remained significant only in healthy individuals but not in pNET or PGGL patients (except for miR-486-5p in PPGL patients). This may be due to the loss of the great majority of miRNAs showing correlation with CgA during tumorigenesis (as the global expression of miRNAs decreased in patients compared to healthy individuals) or due to technical issues. NGS allows us to detect any miRNA with low or very low copies but RT-qPCR has a detection limit, in our experimental settings miRNAs with reads <50 UMI could not be validated with RT-qPCR.
Increased CgA level upon PPI treatment decreases the clinical utility of CgA in neuroendocrine tumors. Performing ROC analysis, we found that in pNET samples CgA alone was less effective in discrimination of pNET sera from healthy controls. The addition of miRNAs to CgA increased the diagnostic value of CgA. Importantly, in patients with pNET with normal CgA level where CgA alone was not informative, the combination of CgA with a miRNA panel could reach high diagnostic value (AUC: 0.904).
In pNET patients, the grade, determined by the Ki-67 proliferation index, has an important impact on prognosis [48]. Although the low sample number represents a limitation of our study, the expression of miR-451a, let-7i-5p, and miR-30d-5p showed a tendency of an inverse correlation with a grade in pNET patients. Circulating miR-451a was reported underexpressed in several tumor types and it was described as an immune-regulatory miRNA [45,[49][50][51] that may explain its dysregulation according to tumor grade in pNET. Let-7i-5p is a well-known tumor suppressor miRNA which expression is negatively correlated with prognosis [52]. In several studies, it was underexpressed in sera of cancer patients vs. healthy controls [52] but the explanation behind its decreasing expression and its role in pNET patients with higher-grade remains to be clarified. MiR-30d-5p was also reported downregulated in tumor (hepatocellular carcinoma) cells and secreted exosomes and its high expression in hepatocellular carcinoma exhibited improved overall survival [53]. These data suggest that miR-30d-5p, measured both in tissue and in liquid biopsy samples, may indeed have a role in relation to tumor grade, progression, and survival.
Previously, in pNET, only one study investigated circulating miRNA levels compared to healthy controls [30]. The authors found a very similar miRNA expression profile compared to healthy volunteers [30]. Only miR-193b was described to be more abundant in the serum of patients and it was also more abundant in pNET tissue compared to islet cells. However, by analyzing both tissue and serum samples of 6 patients, there was a much lower correlation between expression levels in serum and tissue compared to the mean correlation coefficient within the same tissue [30]. In another study, 5 miRNAs were identified differentially expressed in the sera of pancreatic ductal adenocarcinoma patients compared to pNET but no comparison between pNET and healthy controls was performed [54]. None of these 5 miRNAs showed overlap with our results comparing pNET vs. controls. In a very recent publication, 6 upregulated and 11 downregulated miRNAs were identified in plasma exosome fraction of pNET patients (no functionality reported) compared to chronic pancreatitis [55]. Again, none of the identified miRNAs was identical to those in our study although they had different starting material, different measurement methods, and most importantly, they compared their data to chronic pancreatitis.
The decrease of miR-486-5p and miR-30d-5p was detected in patients with genetically determined PGGLs (mutation in either SDHB, RET, VHL, or NF1 genes) compared to sporadic cases. Let-7b-5p, let-7i-5p, miR-30d-5p, and miR-486-5p were downregulated in SDHB mutant cases compared to SDHB wild type cases indicating that the etiology of PPGL may influence the circulating miRNA pool. This observation needs further studies because no data about serum and tissue miRNA differences between sporadic and genetic PPGL cases were found. Four studies investigated circulating miRNAs in PPGLs [27,36,56,57]. No significant difference was demonstrated between benign and malignant cases but elevated levels of miR-96-5p; miR-182-5p and miR-21-3p in PPGL patients compared to healthy controls were reported [27,36,56]. We could not confirm these results probably due to differences in experimental design (mirVana™ PARIS™ Kit vs. miRNeasy Serum Kit for isolation; TaqMan assays vs. QIAseq™ miRNA Library Kit and droplet digital PCR vs. next-generation sequencing).
As a summary, we assessed for the first time the diagnostic accuracy of circulating miRNAs in relation to CgA in patients with pNET. Using the miRNA profile, we were able to discriminate pNET/PPGL samples from healthy controls, and a global downregulation of circulating miRNA pool was observed in patients with neuroendocrine tumors compared to healthy individuals. Similarly, to CgA, miRNA signature alone was not applicable as a biomarker for PPGL/pNET if PPI treatment is administered. However, the combination of a set of miRNAs together with CgA, even in the most challenging group (pNET patients with low CgA level or receiving PPI treatment) represents a promising diagnostic tool.

Patients and Controls
Altogether, 74 consecutive serum samples sent for CgA measurement to the Department of Laboratory Medicine, Semmelweis University were collected (Table 5) between 1 November 2017 and 24 July 2019. Samples were obtained from patients with pancreas nonfunctioning neuroendocrine tumor (pNET, n = 25), with pheochromocytoma-paraganglioma (PPGL, n = 20) and from healthy individuals (n = 29 samples from 24 unrelated cases). PPGL and healthy individuals treated with PPI were included as controls because in clinical practice elevated CgA levels most often can be found in these cases.  All patients with pNET (12 females, 13 males; avg age: 63.4 years) had metastatic disease. Low-CgA (<100 ng/mL) pNET group consisted of 9, high-CgA (>100 ng/mL) pNET group consisted of 16 patients. Grade of pNET was determined according to WHO 2017 classification of tumors of endocrine organs based on the Ki-67 index (Table 5) [58].
PPGL group consisted of 12 females and 8 males with an average age of 41.8 years. Of them, 8 were benign, 11 were malignant (1 bilateral, 1 locally receive, and 9 metastatic) and there was no information available in one case. 14 of the 20 PPGL cases were sporadic, while 6 cases were associated with germline gene mutations in one of the PPGL susceptibility genes ( Table 5). All PPGL patients had elevated CgA levels (avg. CgA: 1751 ng/mL).
Altogether 24 healthy controls were included in the study. Among 29 control samples, 10 sera from 5 healthy volunteers were collected with and without proton pump inhibitor (PPI) treatment. Samples from these individuals were included if high-dose PPI at least for 2 weeks were administered [11]. In our discovery analysis, we used 8 control samples from 6 healthy individuals, 2 samples were taken from two cases after PPI treatment. None of the controls had the endocrine disease, tumor, or renal insufficiency. Eight samples obtained from patients with pNET and 8 samples obtained from PPGL patients were evaluated. In our validation set, we used all samples used in the discovery analysis along with 17 additional samples obtained from pNET, 12 samples obtained from PPGL patients, and 21 samples obtained from healthy controls ( Table 5)

Sample Handling and RNA Extraction
Upon arriving samples were immediately processed for sera separation by centrifugation with 3000× g for 10 min at 4 • C. Then, samples were aliquoted and refrigerated at −20 • C until further processing. Samples for RNA isolation were further processed by centrifugation with 16,000× g for 15 min at 4 • C to deplete platelets.
For total RNA extraction containing small RNA fraction miRNeasy Serum/Plasma Kit (Qiagen, Cat #217184, Hilden, Germany) was applied. Custom cel-miR-39 spike-in control was added to samples during RNA extraction following the manufacturer's instruction (Qiagen, Cat # 219610) [26]. Cellular contamination of serum samples was studied by evaluation of expression of miRNAs related to red blood cells and platelets.

miRNA Expression Profiling by Next-Generation Sequencing
MiRNA expression profiling and NGS data analysis were done as we earlier reported [26,59]. For miRNA sequencing, libraries were prepared using 5 ul RNA and the QIAseq™ miRNA Library Kit (Qiagen, Cat #331505) following the manufacturer's instructions. Next-generation sequencing was run on Illumina MiSeq instrument using MiSeq Reagent Kit v3 150-cycle (MS-102-3001, Illumina, San Diego, CA, USA. For count normalization TMM was applied as it is one of the best performing algorithm for comprehensive miRNA profiling studies [60].

Statistical Analysis
Shapiro-Wilks test was applied to reveal data distribution. A comparison of differential expression was done by unpaired T-test or Mann-Whitney U test and one-way ANOVA followed by Tukey post-hoc test depending on data distribution and grouping. P-value adjustments were applied by Benjamini-Hochberg. For hierarchical cluster analysis, complete linkage clustering with Kendall's Tau distance measuring method was used. The receiver-operator characteristic (ROC) analysis of CgA and miRNAs' were used for their discriminatory role between different groups. Binary logistic regression analyses were applied to evaluate the relationships between miRNAs and the incidence of PPGL/pNET and to find the best logistic model. p-Value < 0.05 was considered significant. Statistical analyses were done using the R statistical programming language.

Conclusions
As a summary, we assessed for the first time the diagnostic accuracy of circulating miRNAs in relation to CgA in patients with pNET. Using the miRNA profile, we were able to discriminate pNET/PPGL samples from healthy controls, and a global downregulation of circulating miRNA pool was observed in patients with neuroendocrine tumors compared to healthy individuals.
In a comparison of pNET or PPGL groups with controls, levels of CgA, and miRNAs showed significant differences. If samples from controls treated with PPI were included either CgA or miRNA signature alone was not applicable as biomarker for PPGL/pNET. In addition, neither CgA nor any miRNA alone was able to discriminate low CgA pNET patients from controls treated with PPI. However, the combination of CgA concentration and expression level of miRNAs in a regression model yielded higher AUC compared to those obtained using individual markers, demonstrating that our model was effective in distinguishing different groups in every comparison. Our results are in line with previous findings showing that in other cancer types, independently of individually being significantly different or not, combining more factors in a regression model increases the discriminatory value of biomarkers [61,62]. Therefore, the combination of a set of miRNAs together with CgA, even in the most challenging group (pNET patients with low CgA level or receiving PPI treatment) represents a promising diagnostic tool.  Henriett Butz is a recipient of Bolyai Research Fellowship of Hungarian Academy of Sciences and her work is supported by the ÚNKP-19-4 New National Excellence Program of The Ministry for Innovation and Technology.

Acknowledgments:
We are grateful to Barna Vásárhelyi, the director of the Department of Laboratory Medicine of Semmelweis University for his support of this work and the critical review of the manuscript.

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