Introduction
Neuroendocrine neoplasms (NENs) are a class of rare and heterogeneous tumors whose molecular pathogenesis is an open issue. NENs are characterized by a body-wide distribution because they develop from neuroendocrine system cells, which are spread throughout the whole body [
1‐
3]; they mainly arise in gastrointestinal and pulmonary tract, but can also arise from thyroid, pituitary gland, lung, breast or larynx or other organs and tissues [
4]. These neoplasms can occur both in sporadic form and in hereditary syndromes such as multiple endocrine neoplasia type 1 and 2 (MEN1 and MEN2), von Hippel–Lindau disease (VHL), neurofibromatosis type 1 (NF1) and the tuberous sclerosis complex (TSC) [
5,
6].
NENs originating from pancreas and the gastro-intestinal tract, the gastroenteropancreatic NENs (GEP-NENs) are among the most common forms. The World Health Organization (WHO) and the International Agency for Research on Cancer (IARC) classified GEP-NENs based on tumor primary sites and on the morphological differentiation features by which these neoplasms can be divided into the well-differentiated tumors (NETs) and poorly-differentiated carcinomas (NECs) [
4]. According to proliferation index and mitotic count, GEP-NENs have been categorized into low (G1, Ki67 < 3%), intermediate (G2, Ki67 3–20%) and high (G3, Ki67 > 20%) grades [
7]. Particularly, NET comprised well differentiated tumors with G1, G2 and G3 grade, while NEC comprised poorly differentiated carcinomas with G3 grade [
7].
The thyroid NENs are tumors of parafollicular C-cells that are conventionally known as medullary thyroid carcinomas (MTCs) [
8]; they represent 3–5% of all thyroid carcinomas and can develop, in ~ 30% of the cases, in the context of MEN2 syndromes [
9]. In MTCs the Ki-67 index, conventionally used for GEP-NENs classification, is difficult to assess, being often lower than 1%, so a classification based on this parameter is not currently used [
4].
Specific genomic profiles and genetic signatures have been previously observed among GEP-NENs with different primary sites and degrees of differentiation and in MTCs, with pancreatic NENs being the best described in the literature. In pancreatic NETs, somatic mutations in MEN1, DAXX, ATRX, PTEN, TSC2 and members of the mTOR signaling pathway were observed [
10‐
12]. Moreover, sporadic pancreatic NETs also present germline mutations in the DNA repair genes MUTYH, CHEK2 and BRCA2 [
11]. On the other hand, gastrointestinal NETs (GI-NETs) frequently show mutations in CDNK1B gene [
13,
14]. In contrast, both pancreatic and intestinal NECs commonly show mutations in TP53 and RB1 and may share mutations in KRAS and SMAD4 [
13,
15,
16]. In MTCs mutations in RET gene were described, affecting tumor microenvironment and angiogenesis, and this has been linked to poor prognosis compared to MTCs that are RAS mutated [
4,
17]. Overall, from the genomic point of view, the loss of chromosome 18 has been reported in small bowel NETs, even if the biological significance of this alteration is still unknown [
18]. However, in pancreatic NETs, the loss of genetic material has been described more often than chromosomal gains [
11].
Generally, NENs are characterized by a relatively indolent rate of growth and by the ability to secrete peptide hormones and biogenic amines that are used as biomarkers [
18]. However, over the latest 40 years the incidence and the prevalence of these tumors have increased more than sixfold in the United States [
19] and, due to non-specific symptomatology and lack of early markers, many NENs show metastatic profile at diagnosis, making it sometimes impossible to identify the primary site of tumor lesion [
3,
20,
21].
A further problem, in addition to a more accurate classification, is the lack of specific markers for NEN diagnosis; Chromogranin A (CgA), synaptophysin (Syn), 5-Hydroxyindoleacetic Acid (5-HIAA), neuron-specific enolase (NSE) and cluster of differentiation 56 (CD56) (neural cell adhesion molecule) are currently used for GEP-NENs diagnosis [
18,
22] and Calcitonin for MTCs [
23]. In GEP-NENs, both Syn and CgA are highly expressed in well-differentiated neoplasms, whereas poorly differentiated carcinomas often maintain synaptophysin positivity while losing CgA expression and acquiring NSE expression [
18]. CgA is characterized by low sensitivity and specificity and the tests can give lots of false-positive elevations [
24,
25]. Equally, the prognostic role of 5-HIAA remains controversial [
26]. For MTC diagnosis, calcitonin is a sensitive tumor marker because it correlates with C-cell mass and burden of the neoplasms [
23], but this has also some limitations, such as inter-assay variability, concentration-dependent half-life and rapid degradation [
23].
To meet this clinical need, in our study we aimed to identify novel prognostic factors and biomarkers for the improvement of the histologic and pathologic evaluation of NENs which is a key component of clinical management [
27,
28]. Particularly, our attention was focused on genome, transcriptome and miRNome profiles of tumor biopsies through a multi-omics approach. The case cohort studied included 66 specimens from GEP-NENs at different grades and MTCs. In few cases only metastatic tissues were available, mainly among neoplasms with gastroenteropancreatic primary location, and these were analyzed as a separate group. Moreover, in order to link our results to clinical management, serum samples of NENs patients were analyzed to determine presence and concentration in the serum of NEN patients of the miRNAs highly expressed within the corresponding tumor tissue. This study design allowed the identification of a subset of molecules able to discriminate healthy from sick subjects, as well as to find some miRNAs significantly correlating with clinical-pathologic features of the neoplasms. These might have a strong impact for diagnostic and prognostic purposes respectively, as therapeutic sequence in patients with NENs is still debated [
29,
30].
Moreover, altogether the obtained results revealed the ataxia telangiectasia mutated (ATM) signaling among the most significantly impacted at different levels, considering gene variants as mutations and amplifications and miRNA expression deregulation. Indeed, this might represent a putative targetable pathway in the treatment of NENs.
Subjects and methods
Patients characteristics and pathological assessment
Tumor biopsies from 46 NEN patients (Thyroid n = 17, Pancreas n = 14, Intestine n = 12 and Lung n = 3) were collected by the biobank of the “Istituto Nazionale Tumori-IRCCS-Fondazione G. Pascale” (Naples, Italy) and by the Department of Clinical Medicine and Surgery, Endocrinology Unit of Federico II University (Naples, Italy). Out of 46 tumor tissues, 19 were FFPE sections (Formalin fixed paraffin embedded) while 27 were frozen sections. In addition, a validation set of 20 previously isolated RNAs from MTCs were obtained from Endocrinology Unit, Department of Medicine (DIMED), University of Padua. Patient characteristics and samples pathology were summarized in Table
1.
Table 1
Patients’ clinical data
Gender |
Male | 33 (50%) |
Female | 33 (50%) |
Tumor location |
Thyroid | 37 (56.1%) |
Pancreas | 14 (21.2%) |
Intestine | 12 (18.2%) |
Lung | 3 (4.5%) |
Metastases | 19 (28.8%) |
Grading classification (WHO 2019) for p-NET, I-NET and lung-NET |
NET G1 (Ki-67% ≤ 2) | 11 (38%) |
NET G2 (Ki-67% 3–20) | 11 (38%) |
NEC G3 (Ki-67% > 20) | 7 (24%) |
Genetic syndrome |
Men1 | 1 (2%) |
Men2 | 3 (7%) |
ND | 1 (2%) |
Status |
Live | 53 (80.3%) |
Dead | 7 (10.6%) |
nd/progression | 6 (9.1%) |
Cases have been reviewed by an expert pathologist (FT) and graded and staged according to WHO 2017 and 2019 classification criteria (NET-G1, NET-G2, NET-G3, NEC-G3) on tissue sections. The 4 main categories are distinguished on the basis of the proliferative activity, measured through the mitotic count and the Ki67 expression. In our cohort, high-grade (G3) specimen were all classified as poorly differentiated according to Hematoxylin/Eosin staining, thus they all fall within NECs. Medical records have been reviewed for clinical information, including histologic parameters, assessed on standard H&E-stained slides combined with immunohistochemical staining with neuroendocrine markers, and tumor location. Immunohistochemical staining for Ki67 (clone MM1, Leica), Chromogranin A (clone 5H7, Leica, Wetzlar, Germany, ready to use), Synaptophysin (clone 27G12, Leica, ready to use) and Calcitonin (clone CL1948, Leica, ready to use) has been performed using the En Vision method (DAKO, Denmark) following the manufacturer’s instruction.
For miRNA validation in liquid biopsies, serum from 42 NEN patients (6 MTCs and 36 GEP-NENs) and 34 healthy subjects were obtained.
DNA and RNA isolations were performed from both FFPE and frozen sections using FFPE DNA Purification Kit (Cat. 47400, Norgen Biotek Corp, Thorold, Canada) and RNA/DNA Purification Kit (Cat. 48700, Norgen Biotek Corp.) respectively, according to the manufacturers protocol. Nucleic acids were quantified with Qubit 2.0 fluorometer using Qubit RNA HS assay kit and Qubit DNA HS assay kit (Thermo Fisher Scientific, Waltham, Massachusetts, USA). The assessment of nucleic acids integrity (DIN and RIN) was performed with Agilent 4150 TapeStation System (Agilent Technologies, Santa Clara, CA, USA).
Mutational profiling
Libraries for mutational profiling were prepared, starting from 40 ng of DNA as input material, with TruSight™ Oncology500 kit (Cat. 20028213, Illumina, San Diego, California, USA) according to manufacturer’s protocol. These consist in a targeted-capture of 523 cancer-relevant genes. The libraries were sequenced on NextSeq 500 (Illumina) in 2 × 150 bp or 2 × 75 bp in paired end mode. Sequencing data were analyzed using the TruSight Oncology 500 Local App Version 2.2 (Illumina) to identify variants, gene amplifications, TMB and MSI. The genomic coordinates of all the identified variants were subsequently converted to hg38 using LiftOver [
31].
Variants were annotated with Annovar [
32]. Those with coverage depth lower than 100 and variants occurring with a frequency higher than 5% in 1000G or GnomAD were discarded. The sequence variants annotated as “benign” or “likely benign” in ClinVar database were also filtered out. Furthermore, variants classified as synonymous or in intergenic positions were discarded. Oncoplots were generated using Maftools [
33] on R (v4.0.2). Functional analysis was performed using Ingenuity Pathway Analysis (IPA, Qiagen, Hilden, Germany) and only the pathways with more than 1.3 in –log of the adjusted p-value were considered.
Array-CGH
Array-CGH analysis was performed using Agilent Oligonucleotide Array-Based CGH for Genomic DNA Analyis-Enzimatic Labeling (Agilent Technologies, Santa Clara, CA, USA), starting from 500 ng of DNA, following the manufacturer’s instructions. The microarray includes 60.000 oligonucleotide probes. Genomic DNA samples and reference samples were labeled with Cy5 and Cy3, respectively, using the SureTag DNA Labeling Kit and following Agilent Enzymatic Labeling protocol. Labeled genomic DNA was purified using the reaction Purification Column provided with the kit. After the hybridization protocol, slides were scanned using Agilent SureScan Dx Microarray Scanner G5761A. Image files were analyzed using Agilent Cytogenomics 5.0.0.38 Software, and genomic coordinates were evaluated according to GRCh38/hg38. The measure of success of profiling for each sample was based on array data sample quality indices (derivative log ratio scores). Circos plot was generated using circos (v0.69.9) [
34].
Trancriptome profiling
Libraries preparation for transcriptome analysis was performed employing the TruSeq RNA Exome kit (Cat. 20020189, Illumina) for FFPE samples and TruSeq Stranded Total RNA kit (Cat. 20020598, Illumina) for frozen samples, starting from 200 and 400 ng of RNA as input materials, respectively, according to manufacturers’ guidelines. 46 libraries were sequenced on NextSeq 500 (Illumina) using 2 × 75pb paired end and 14 libraries on HiSeq 1500 (Illumina) using 1 × 50 single read.
Raw reads were pre-processed using FastQC software [
35] to evaluate raw sequences quality and adapter sequences were removed using cutadapt (v3.3) [
36]. Alignment was performed on human genome version hg38 (release 34) with STAR (v2.7.8a) [
37] and expressed transcripts were identified using featureCounts on Rsubread (v2.0.1) [
38]. Differential expression analysis was performed using DESeq2 package (v1.28.1) in R [
39], with default parameters. Transcripts were considered differentially expressed if they showed |FC| ≥ 1.5 and adjusted p-value ≤ 0.05. Fusion transcripts detection was performed with STAR-Fusion tool [
40], setting default parameters and only fusion transcripts with more than 10 junction reads were considered.
Small-RNA profiling
Small-RNA libraries were prepared with NEXTFLEX Small RNA-Seq Kit v3 (Cat NOVA-5132-06, Perkin Elmer, Massachusetts, USA) starting from 50 ng of RNA as input, according to manufacturer’s guidelines. The libraries were sequenced on NextSeq 500 (Illumina) using 1 × 75 bp single read. miRNA-Seq data analysis was performed using the automated pipeline iSmaRT [
41]. Target prediction was performed using miRWalk [
42]. Only targets validated and present on TargetScan or miRDB with more than 10 reads in at least 60% of the samples were considered. Gene ontology plot was generated using the library GOplot on R [
43]. Network was generated using Cytoskape v 3.9.0 [
44].
Serum was obtained from whole blood samples by centrifugation at 1900×g for 10 min at 4 °C. The supernatant was further centrifuged at 16,000×g for 10 min at 4 °C and stored in aliquots of 0.5 ml at − 80 °C until analysis. The extraction of total RNA from 200 µl of serum was performed within 1 year of storage at − 80 °C using miRNeasy Serum/Plasma Kit (Qiagen, Hilden, Germany) according to the manufacturer’s instructions. Briefly, 1 µl of UniSp2 spike-in (Qiagen, Hilden, Germany), a control for the quality of RNA extraction, was combined with the lysis buffer before mixing with the serum. Total RNA (including miRNAs) was eluted in 14 µl of RNase-free water. Reverse transcription was performed using miRCURY LNA™ RT Kit (Qiagen, Hilden, Germany) according to the manufacturer’s instructions. Briefly, 1 µl of UniSp6 spike-in (Qiagen, Hilden, Germany), a control for the quality of RT reaction, was added to the reaction mix including 2 µl of total RNA, nucleic acid mix buffer and reverse transcriptase in a final volume of 10 µl. RT mix was incubated for 60 min at 42 °C and for 5 min at 95 °C. cDNA was stored at − 20 °C until analysis.
Expression value of hsa-mir-106b-3p, hsa-mir-143-3p, hsa-mir-144-3p, hsa-mir-150-5p, hsa-mir-18a-5p, hsa-mir-21-5p, hsa-mir-222-3p, hsa-mir-26a-5p, hsa-mir-335-5p, hsa-mir-361-3p, hsa-mir-375, hsa-mir-7-5p, and hsa-mir-942-5p was determined by real-time PCR using miRCURY LNA miRNA primers (Qiagen, Hilden, Germany) and miRCURY LNA™ SYBR Green PCR Kit (Qiagen, Hilden, Germany), with the instrument CFX384 (Biorad, USA). PCR cycling conditions were 95 °C for 2 min, 40 cycles of 95 °C for 10 s, 56 °C for 60 s and melting curve analysis 60–95 °C. The maximum cycle threshold (Ct) value was set at 280. UniSp2 and UniSp6 were used as control genes. Experiments were carried out in triplicates for each data point, and data analysis was done by using CFX Maestro Software (Biorad, USA). Data were expressed as relative expression using the 2-ΔΔCt method (compared to healthy patients).
Discussion
NENs represent a heterogeneous group of neoplastic diseases originating from neuroendocrine cells distributed throughout the body. Conventionally, this cancer types are considered very difficult to diagnose due to the lack of molecular and prognostic markers and the difficulties in identifying the primary site of origin. Being mostly indolent, they are often already metastatic at diagnosis [
3] and, indeed, selected NEN specimens are being included in clinical studies aiming at tumor origin identification [
58]. Thus, the identification of early and specific diagnostic and prognostic markers as well as novel therapeutic targets is crucial. The current landscape of NENs genetic knowledge is heterogeneous, with well-defined traits by high-throughput studies for some anatomic sites such as pancreatic NENs, but relatively low information for other sites. Moreover, the increasing interest in circulating biomarkers offers a new perspective for earlier NENs diagnosis [
59]. Furthermore, some studies have been conducted to emphasize the potential of miRNAs as biomarkers and for easier grade stratification and tissue discrimination in GEP-NENs [
54,
60‐
62] and as promising diagnostic and prognostic factors in MTC [
63‐
65].
In the present paper, a multi-omics approach has been applied to analyze a diversified NENs cohort composed of GEP-NENs and MTC primary tumors and few GEP-NEN derived metastasis. Given the heterogeneity of the cohort, with a small sample size for each group making it difficult to identify a histotype specific molecular signature, and considering the recently emerged concept that NENs from different organ systems inter-relate clinically and genetically [
4], we focused on highlighting common molecular and functional features that might represent useful and effective NEN biomarkers and therapeutic targets.
The gene mutational landscape revealed, over the already known key drivers specifically associated with pancreatic NETs, pancreatic NECs [
11,
66] and MTCs [
6], a common signature represented by a set of genes most frequently mutated, among which those involved in genome stability maintenance and DNA damage recombination emerged, including MDC1, BRCA1 and ATM. In particular, MDC1 was never specifically associated to NENs before, except for one paper observing the presence of either MDC1 or ATM somatic mutations in RET and RAS negative MTCs [
67]. In our cohort, instead, MDC1 resulted the top mutated gene in absolute among all NEN specimens analyzed. It is a key component of the DNA damage response, binding to γ-H2AX at DNA double-strand breaks, and participating in the recruitment of key factors including ATM, BRCA1, and TP53 [
68]. MDC1 loss of function could negatively affect both homologous recombination and non-homologous end joint repair pathways and co-mutation with some of its key co-factors has been proposed as potential marker for radiosensitivity [
69]. Interestingly, within our cohort MDC1 resulted to be co-mutated with ATM in 2 out of 7 ATM-mutated samples and with BRCA1 in 4 out of 7 BRCA1-mutated ones. Moreover, we also found one ATM-BRCA co-mutation case (Fig.
2C). With these premises, not surprisingly, ATM signaling resulted among the most significantly impacted at gene level, together with other pathways involving BRCA and DNA repair, or enrolling ATM downstream targets such as G2/M and G1/S checkpoints and other DNA damage-induced ones (Fig.
2D) [
70]. On the other hand, complementary analyzes carried out on DNA and RNA samples, revealed a different behavior of GEP-NENs and MTCs regarding the tendency to form large chromosome rearrangements or gene amplifications and fusion transcripts. Indeed, as also observed before, GEP-NENs were more prone to this kind of events. Moreover, we observed a good concordance between the amplification/deletion patterns observed at chromosome level and gene expression data (Fig.
3). Given the absence of matched normal tissues, we did not sought to investigate differential expressed genes in tumor tissues except to have an indication of a possible deregulation among GEP-NENs of different grades even if this was not the main focus of the present work. As expected, several transcripts were found differentially expressed between grades, functionally targeting most of the pathways observed to be affected at the gene level, and, thereafter, post-transcriptionally through miRNA targeting (data not shown). Indeed, in the attempt to mainly focus on molecular features in common to the analyzed neoplasms, we proceeded by analyzing tissue miRNAs to identify those over-expressed in the various kind of investigated NENs. This led to the identification of a set of 623 commonly expressed miRNAs, with the top expressed being also mostly shared among different samples. Anyway, even in this case, a set of miRNAs resulted to be expressed at different level in low with respect to high-grade GEP-NENs, thus corroborating previous findings pointing to miRNAs as useful biomarkers for grade stratification in this class of NENs [
54,
61]. In this context, considering our and others’ evidence of high TMB in high-grade GEP-NENs and the proposed role for PDL1 expression in GEP-NENs grade stratification [
71], a positive response of NETs-G3 and NECs to immunotherapy may be desirable and the impact of miRNA-mediated deregulation on PDL1 and other genes involved in the immunological synapse should be deeply investigated. In our case, although we observed PDL1 coding transcript over-expression in G3 vs G2 tumors, this appears not to be dependent on miRNA-mediated post-transcriptional regulation, although the limited number of high-grade tumors may have undermined the significance of the results (data not shown). Moreover, through a pilot sequencing by NGS of serum miRNAs from 3 patients, multiple miRNA molecules could be detected, with more than 95% of them being previously identified in the corresponding tissues. This finding reinforced the hypothesis that overexpressed miRNAs may be specifically released and evaluated as circulating NEN biomarkers.
Based on these results, we selected a set of 13 miRNAs to be evaluated in serum samples as possible circulating NEN biomarkers. We found that these miRNAs were overall significantly overexpressed in NEN patients compared to healthy subjects. This result represents a remarkable one, because for the first time a set of circulating miRNAs overexpressed in NENs patients could potentially represent a pathological signature for diagnostic purposes. A larger cohort with higher sample number for each histotype would be needed to confirm the data and select the most suitable molecule combination to be assessed for specific and reliable results.
Very interestingly, the mRNA targets of the selected miRNA panel are linked to ATM signaling (Fig.
4D) that we found emerging among the most significantly impacted at multiple levels (Fig.
5D). Taken together, we can speculate that ATM may represent a novel druggable pathway, in addition to the widely used inhibitors of mTOR [
72‐
74], whose regulation by ATM has been also demonstrated [
75,
76].
Indeed, following the success of PARP inhibitors, ATM inhibitors have been proposed in the therapeutic exploitation of DNA Damage Response (DDR) in cancer [
77]. Several synthetic molecules have been already developed and demonstrated to induce significant sensitization to radiation and DNA-damaging chemotherapeutic agents [
78], and some of them are undergoing clinical trials in combination with radiation therapy [
79] or with PARP inhibitors [
80].
In fact, according to an experimentally proven hypothesis, the use of an ATM inhibitor, shutting down the MDC1-mediated DDR pathway, together with PARP inhibitors, which rescue endogenous BIN1 expression (that increases cell death due to DNA damage), is able to generate a new ‘BRCAness-independent’ synthetic lethal effect in cancerous cells [
81].
Taken together, our results reinforce this hypothesis, but further experimental and preclinical evidences are needed by establishing in vitro and in vivo models demonstrating their effectiveness and potential clinical application in NENs.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.