Abstract
Neuroendocrine neoplasms (NENs) represent clinically and genetically heterogeneous malignancies, thus a comprehensive understanding of underlying molecular characteristics, prognostic signatures, and potential therapeutic targets is urgently needed.
Next-generation sequencing (NGS) and immunohistochemistry were applied to acquire genomic and immune profiles of NENs from 47 patients.
Difference was distinguished based on differentiation grade and primary localization. Poorly differentiated neuroendocrine carcinomas (NECs) and well-differentiated neuroendocrine tumors (NETs) harbored distinct molecular features; we observed that tumor mutational burden (TMB) and tumor neoantigen burden (TNB) were significantly higher in NECs versus NETs. Notably, we identified a 7-gene panel (MLH3, NACA, NOTCH1, NPAP1, RANBP17, TSC2, and ZFHX4) as a novel prognostic signature in NENs; patients who carried mutations in any of the 7 genes exhibited significantly poorer survival. Furthermore, loss of heterozygosity (LOH) and germline homogeneity in human leukocyte antigen (HLA) are common in NENs, accounting for 39% and 36%, respectively. Notably, HLA LOH was an important prognostic biomarker for a subgroup of NEN patients. Finally, we analyzed clinically actionable targets in NENs, revealing that TMB high (TMB-H) or gene mutations in TP53, KRAS, and HRAS were the most frequently observed therapeutic indicators, which granted eligibility to immune checkpoint blockade (ICB) and targeted therapy.
Our study revealed heterogeneity of NENs, and identified novel prognostic signatures and potential therapeutic targets, which directing improvements of clinical management for NEN patients in the foreseeable future.
Neuroendocrine neoplasms (NENs) have long been considered as a difficult malignancy to diagnose, monitor, and treat. The present study indicates that molecular profiling may complement histology to provide better diagnostic definition, therapy choice, and prognostic stratification of NENs. This study will be beneficial for an updated molecular-based classification of these neoplasms, novel targeted drug development, and improvement of prognosis.
Introduction
Neuroendocrine neoplasms (NENs) are a rare and heterogeneous group of tumors arising from diffuse neuroendocrine cells throughout the body, which most frequently occur in the gastrointestinal (GI) tract, pancreas, or pulmonary system.1 NENs encompass well-differentiated neuroendocrine tumors (NETs) and poorly differentiated neuroendocrine carcinomas (NECs), which are genetically and biologically distinct.2 Further classification is made based on grade, histology, and functionality.3 Nevertheless, current evaluation system is still insufficient to stratify the variable clinical behaviors and outcomes within each category. Molecular mechanisms which enable them to illustrate their clinical heterogeneity, from indolent to highly aggressive, and divergent responses to certain treatments, are lacking. Consequently, there is a high unmet need to characterize the molecular features of these heterogeneous NENs, ultimately advancing precise and patient-tailored therapeutic strategies.
Certain anti-tumor therapies, such as the mammalian target of rapamycin (mTOR) inhibitor everolimus and the multiple tyrosine kinase inhibitors (TKI) sunitinib, have been registered for management of progressive grade (G)1/G2 NETs.4,5 However, treatment options for high-grade G3 NETs or poorly differentiated NECs are limited. Treatment strategies for NECs are often extrapolated from the treatment paradigm for small-cell lung cancer (SCLC), which are generally managed with platinum-based chemotherapy as a first-line treatment,6 but no standard second-line therapy has been established. Immunotherapy with immune checkpoint blockade (ICB) has achieved outstanding breakthroughs in recent years, yielding pronounced clinical benefits in multiple tumor types.7 However, no conclusions regarding the efficacy of ICB in NENs have been drawn yet due to its rarity and heterogeneity. Therefore, it is essential and urgent to acquire comprehensive understanding of NENs at molecular level.
Next-generation sequencing (NGS) is able to perform thorough analyses of genetic alterations to understand the molecular landscapes of tumors, which enable us to gain more treatment choices and improves prognostication.8 Particularly, targeted NGS, in which clinically relevant genes are detected, is time-saving and cost-efficient.9 In recent years, NGS has been conducted in pancreatic NETs (pNETs), small intestinal NETs (SI-NETs), and NECs, providing new insights into novel potential carcinogenetic mechanisms of NENs.10-12 Integrated sequencing approach identified three subtypes of SI-NETs with significant impact on progression-free survival.13 Whole-genome landscape of pNETs has revealed the alteration of mTOR-related genes (TSC2, PTEN, and PIK3CA) and pinpointed activated mTOR signaling as a druggable mechanism in pNETs, enabling mTOR inhibitor everolimus to be approved by the Food and Drug Administration for the treatment of advanced pNETs.14 In addition, mutations in TP53, RB1, MYC, CCND1, KRAS, PIK3CA, PTEN, and BRAF have been reported in NECs.12,15 However, owing to the rarity of NENs, the number of cases examined in previous studies was limited. Thus, it is necessary to accumulate more cases worldwide to further understand the genetic landscape of NENs. Compelling evidence suggest that genomic and immune-related biomarkers, including microsatellite instability (MSI), tumor mutational burden (TMB), tumor neoantigen burden (TNB), germline human leukocyte antigen (HLA) homogeneity, HLA loss of heterozygosity (HLA-LOH), programmed death ligand-1 (PD-L1) expression, and CD8+ tumor-infiltrating lymphocytes (TILs), not only determine the clinical response to ICB but also have prognostic values in a variety of cancers.16-23 However, the landscape and clinical values of the abovementioned parameters have not been fully delineated in NENs yet.
Given the paucity of the aforementioned data, it urgently needs further investigation to comprehensively comprehend the molecular basis of NENs. In this study, targeted NGS was performed on tumor tissues from 47 patients with NENs. We delineated the landscape of genomic alterations, genomic biomarkers, and immunological signatures for the whole NENs cohort. Their differences were identified to distinguish NECs from NETs and comparisons were also made between different primary localization-derived NENs. To provide better prognostic stratification, novel survival-associated gene mutations, and immune signatures were identified. Furthermore, we dig into clinically actionable genetic alterations within NEN patients, which might render them eligible for off-label or experimental treatments to extend therapy options.
Materials and Methods
Patients and Sample Collection
A cohort of 47 NEN patients was identified from a retrospective database from January 2019 to May 2022. Clinicopathologic data were collected from electronic patient medical records. Tumor specimens were taken from archived formalin-fixed, paraffin-embedded (FFPE) tissue of local lesions or metastatic deposits. At the time of tissue collection, a matched peripheral blood sample was also collected to rule out germline mutations.
Targeted NGS and Mutation Analysis
Genomic profiling was conducted on tumor tissue containing a minimum of 20% tumor cells and matched peripheral blood samples. The GeneRead DNA FFPE Kit (Qiagen) and DNA blood mini kit (Qiagen) were used to isolate and extract genomic DNAs, respectively. The extracted DNAs were applied as templates to be amplified, then purified and analyzed by using YuceOne ICIs extensive targeted panel (1267 genes) in pathologists (CAP)-authenticated biotechnique laboratory (Yucebio, Shenzhen, China). SOAPnuke (V1.5.6) was used to filter sequencing reads with > 10% N rate and/or >10% bases with a quality score <20.24 The somatic single nucleotide variants (SNVs) and insertions and deletions (InDels) were detected using VarScan (V2.4).25 Next, an in-house method was applied to filter the possible false positive mutations. Last, SnpEff (V4.3) was used to perform functional annotation on the mutations detected in the tumor sample.26 Copy number variations (CNVs) were called using CNVkit (V0.8.1) to compare the exome-wide profile between tumors and matched peripheral blood.27 Allele-specific copy number and tumor purity were assessed using the ascatNgs (V3.1.0).
Protein-Protein Interaction (PPI) and Functional Enrichment Pathway Analyses
The mutated genes were selected to perform PPI-based and enrichment pathway analyses using Metascape database, which integrates GO, KEGG, STRING, and UniProt data resources. P < 0.05 was considered to be statistically significant.
Evaluation of Genomic Biomarkers
The MSI status was detected with MSIsensor (V0.2).28 An in-house tool was then used to recalculate and correct MSI values. The percentage of unstable sites was reported as the MSI sensor score. MSI scores ≥ 20 were defined as MSI-High (MSI-H). TMB was defined as the number of all nonsynonymous mutations and indels per megabase of the genome examined. TMB > 10 Muts/Mb was defined as TMB-High (TMB-H). Tumor neoantigen burden (TNB) was measured as the number of mutations that could generate neoantigens per megabase. TNB > 4.5 Neos/Mb was defined as TNB-High (TNB-H). HLA typing of the paired peripheral blood and tumor samples was performed from targeted sequencing data using POLYSOLVER (V1.0) and OptyType (V1.3.2). A scoring algorithm was then used to integrate the results that were used for further analysis.29 The bioinformatic tool LOHHLA with the default program settings was used to determine HLA maintenance or loss in the tumor.30
Immunohistochemistry (IHC) Assessment for PD-L1 and CD8
IHC staining assay for PD-L1 and CD8 was performed on FFPE tissue sections according to the manufacturer’s recommendations. PD-L1 expression was measured according to the Dako PD-L1 IHC 22C3 pharmDx assay. Tumor proportion score (TPS), defined as the percentage of tumor cells with complete or partial membrane staining (central or marginal tumor region), was adopted to calculate PD-L1 expression. Then, patients were divided as “negative” expression group (no signal), “low” expression group (TPS < 1%), “intermediate” expression group (1% ≤ TPS < 50%), and “high” expression group (TPS ≥ 50%) by following the standard recommendation as per prior publications.31 For CD8+ TILs analysis, primary antibody used for staining was anti-CD8 (SP57; Ventana, cat# 790-4460; 1:150). The BenchMark XT automated slide processing system (Ventana Medical Systems, Tucson, AZ) was used for IHC of CD8. Quantitative analysis of IHC staining was conducted using digital pathology software and confirmed by pathologists. The data are presented as density (total number of positive cells/mm2 area). According to the detection results of CD8+ TILs in large samples, the density of CD8+ TILs is divided into high and low. The density of CD8+ TILs in the top 40% is high, and the density of CD8+ TILs in the bottom 60% is low. This division method has been demonstrated to be scientific and reasonable.32
Inventory of Clinically Actionable Somatic Alterations and Putative Therapeutic Targets
Current clinical relevance and therapeutic implications of somatic alterations were determined based on the following publicly available databases: OncoKB, CiViC, CGI, My Cancer Genome, and ClinicalTrials.Gov. Intelligent variants Interpreter (IVI) developed by Yucebio lab was applied to aggregate and harmonize these databases. Clinical significance of potentially actionable alterations is defined according to OncoKB Therapeutic Level of Evidence V2. Subsequently, we curated the linked putative treatments for current NEN treatment options.
Statistical Analyses
Statistical analyses of the data from pairs of groups were performed by Fisher’s exact test for categorical variables and the unpaired t-test with Welch’s correction for continuous variables, respectively. The relationships between overall survival (OS), clinical characteristics, gene mutations, genomic, and immune-related biomarkers were determined by univariate/multivariate Cox proportional hazards analysis. Survival curves were generated by the Kaplan-Meier method, and the log-rank test was used to assess the statistical significance between the groups. Two-sided P < .05 was considered statistically significant. All of the statistical analyses were performed using the Python (V3.8.8) or GraphPad Prism (V8) (GraphPad, Inc., La Jolla, CA, USA).
Results
Patient characteristics
The clinicopathological characteristics of the 47 NEN patients are summarized in Table 1 and detailed in Supplementary Table S1. This cohort included 15 NETs and 32 NECs and was represented by 29 males and 18 females with a median age of 50 at the time of diagnosis. The majority of patients were at stage IV (39/47, 83%). The most common primary tumor location was GI tract (n = 13, 27.7%), followed by lung, thymus, and pancreas. Tumor tissues subjected to targeted NGS were taken from the primary sites (n = 21, 44.7%) and metastatic lesions, including liver metastases (n = 15, 31.9%) and metastatic lymph nodes (n = 11, 23.4%).
Clinicopathological features of 47 NEN cases.
| Characteristics | Number | Frequency (%) |
|---|---|---|
| Gender | ||
| Male | 29 | 61.7 |
| Female | 18 | 38.3 |
| Median age at diagnosis (range), years | 50 (8-82) | |
| Histology | ||
| NET | 15 | 46.9 |
| NEC | 32 | 68.1 |
| Ki-67 category | ||
| <3% | 1 | 2.1 |
| 3%-20% | 8 | 17.0 |
| >20% | 38 | 80.9 |
| AJCC staging | ||
| I | 1 | 2.1 |
| II | 1 | 2.1 |
| III | 6 | 12.8 |
| IV | 39 | 83.0 |
| Primary sites | ||
| GI tract | 13 | 27.7 |
| Lung | 9 | 19.1 |
| Thymus | 6 | 12.8 |
| Pancreas | 6 | 12.8 |
| Others (nose, throat, gall bladder, liver, breast, uterus, ovary, arm) | 10 | 21.3 |
| Unknown | 3 | 6.4 |
| Origins of tissue for targeted NGS | ||
| Primary tumor | 21 | 44.7 |
| Metastatic liver | 15 | 31.9 |
| Metastatic lymph node | 11 | 23.4 |
| Follow-up status | ||
| Alive | 18 | 38.3 |
| Dead | 28 | 59.6 |
| Lost | 1 | 2.1 |
| Characteristics | Number | Frequency (%) |
|---|---|---|
| Gender | ||
| Male | 29 | 61.7 |
| Female | 18 | 38.3 |
| Median age at diagnosis (range), years | 50 (8-82) | |
| Histology | ||
| NET | 15 | 46.9 |
| NEC | 32 | 68.1 |
| Ki-67 category | ||
| <3% | 1 | 2.1 |
| 3%-20% | 8 | 17.0 |
| >20% | 38 | 80.9 |
| AJCC staging | ||
| I | 1 | 2.1 |
| II | 1 | 2.1 |
| III | 6 | 12.8 |
| IV | 39 | 83.0 |
| Primary sites | ||
| GI tract | 13 | 27.7 |
| Lung | 9 | 19.1 |
| Thymus | 6 | 12.8 |
| Pancreas | 6 | 12.8 |
| Others (nose, throat, gall bladder, liver, breast, uterus, ovary, arm) | 10 | 21.3 |
| Unknown | 3 | 6.4 |
| Origins of tissue for targeted NGS | ||
| Primary tumor | 21 | 44.7 |
| Metastatic liver | 15 | 31.9 |
| Metastatic lymph node | 11 | 23.4 |
| Follow-up status | ||
| Alive | 18 | 38.3 |
| Dead | 28 | 59.6 |
| Lost | 1 | 2.1 |
Clinicopathological features of 47 NEN cases.
| Characteristics | Number | Frequency (%) |
|---|---|---|
| Gender | ||
| Male | 29 | 61.7 |
| Female | 18 | 38.3 |
| Median age at diagnosis (range), years | 50 (8-82) | |
| Histology | ||
| NET | 15 | 46.9 |
| NEC | 32 | 68.1 |
| Ki-67 category | ||
| <3% | 1 | 2.1 |
| 3%-20% | 8 | 17.0 |
| >20% | 38 | 80.9 |
| AJCC staging | ||
| I | 1 | 2.1 |
| II | 1 | 2.1 |
| III | 6 | 12.8 |
| IV | 39 | 83.0 |
| Primary sites | ||
| GI tract | 13 | 27.7 |
| Lung | 9 | 19.1 |
| Thymus | 6 | 12.8 |
| Pancreas | 6 | 12.8 |
| Others (nose, throat, gall bladder, liver, breast, uterus, ovary, arm) | 10 | 21.3 |
| Unknown | 3 | 6.4 |
| Origins of tissue for targeted NGS | ||
| Primary tumor | 21 | 44.7 |
| Metastatic liver | 15 | 31.9 |
| Metastatic lymph node | 11 | 23.4 |
| Follow-up status | ||
| Alive | 18 | 38.3 |
| Dead | 28 | 59.6 |
| Lost | 1 | 2.1 |
| Characteristics | Number | Frequency (%) |
|---|---|---|
| Gender | ||
| Male | 29 | 61.7 |
| Female | 18 | 38.3 |
| Median age at diagnosis (range), years | 50 (8-82) | |
| Histology | ||
| NET | 15 | 46.9 |
| NEC | 32 | 68.1 |
| Ki-67 category | ||
| <3% | 1 | 2.1 |
| 3%-20% | 8 | 17.0 |
| >20% | 38 | 80.9 |
| AJCC staging | ||
| I | 1 | 2.1 |
| II | 1 | 2.1 |
| III | 6 | 12.8 |
| IV | 39 | 83.0 |
| Primary sites | ||
| GI tract | 13 | 27.7 |
| Lung | 9 | 19.1 |
| Thymus | 6 | 12.8 |
| Pancreas | 6 | 12.8 |
| Others (nose, throat, gall bladder, liver, breast, uterus, ovary, arm) | 10 | 21.3 |
| Unknown | 3 | 6.4 |
| Origins of tissue for targeted NGS | ||
| Primary tumor | 21 | 44.7 |
| Metastatic liver | 15 | 31.9 |
| Metastatic lymph node | 11 | 23.4 |
| Follow-up status | ||
| Alive | 18 | 38.3 |
| Dead | 28 | 59.6 |
| Lost | 1 | 2.1 |
Genetic and Immune Landscapes of NENs
To depict the genetic landscape, the overall genetic variations were analyzed, which are summarized in Fig. 1 and detailed in Supplementary Table S1. The incidence of somatic variations, defined by SNVs, InDels, CNVs, and gene rearrangement/fusion, was 95.7% (45/47). A total of 511 somatic variants, including 393 SNVs, 55 InDels, 62 CNVs (including 43 copy number amplifications, 19 copy number deletions), and 1 fusion were identified (Fig. 1A). The different variant types of each gene with frequency above 4% were displayed in Fig. 1I. Among the detected mutated genes, TP53 (24/47, 51%) was the most frequently mutated gene, followed by RB1 (14/47, 30%), MYC (6/47, 13%), and FAT4 (6/47, 13%). In this cohort, 66 gene mutation was identified as oncogenicity based on ClinVar, OncoKB, and/or CIVIC databases, while most mutations remain unclear whether they were deleterious (Supplementary Table S1).
Landscape of genomic alterations, genomic, and immune-related biomarkers detected in NENs. A, Distribution of total somatic variations. B–H, Molecular analysis of MSI score (B), TMB (C), TNB (D), HLA-I homogeneity (E), HLA LOH (F), PD-L1 expression (G) and CD8+ TILs (H) in NENs cohort. I, Overview of genome-wide characteristics of the NENs cohort (n = 47) ordered by NET/NEC, primary localization, gender, age, and genomic biomarkers (TMB, TNB, HLA LOH, and HLA homogeneity).
As shown in Fig. 1B–H; Supplementary Table S1, the status of genomic and immune-related biomarkers was evaluated across our cohort. The results showed that MSI score ranged from 0.34 to 9.77 (mean = 3.582), indicating that all patients had low status of MSI (MSI score ≤ 20) (Fig. 1B). TMB was generally low with only 6 TMB-H cases (TMB ≥ 10 Muts/Mb), while TMB-L cases (TMB < 2.5 Muts/Mb) accounted for nearly 43% (n = 20) (Fig. 1C). The mean value of TMB was 5.921 Muts/Mb. As for TNB, its value was low in 14 cases (29.8%, TNB < 0.5 Neos/Mb), medium in 26 cases (55.3%), and high in 7 cases (14.9%, TNB > 4.5 Neos/Mb), ranging from 0.50 to 10.78 Neos/Mb (mean = 1.802 Neos/Mb) (Fig. 1D). Individuals whose HLA genotypes showed homozygous in at least one HLA-I gene were classified as having HLA-I germline homogeneity and HLA-I LOH is defined as the monoallelic loss of at least one HLA-I gene (HLA-A, HLA-B, and HLA-C). Patients with germline homogeneity or HLA-I LOH were considered to have HLA-I dysfunction and consequently a limited capacity to present tumor-derived neoantigens to cytotoxic T lymphocytes (CTLs).22,30 Overall, 36.2% (17/47) of patients had HLA-I homogeneity (Fig. 1E). Notably, HLA-A locus homogeneity is the most common (11/17, 65%) and HLA-A*11:01 genotype accounted for 64% (7/11) (Supplementary Table S1). Moreover, 39% (18/46) of patients had HLA-I LOH in our cohort, consisting of 6 strong (LOH of 3 genes), 5 moderate (LOH of 2 genes), and 7 weak (LOH of 1 gene) positive cases (Fig. 1F). Collectively, homogeneity and LOH in HLA-I are common in NENs.
According to the results of PD-L1 IHC staining, 47 patients were divided into 3 groups, low group with TPS < 1% (n = 43, 91.5%), intermediate group with TPS 1%-49% (n = 3, 6.4%), and high group with TPS ≥ 50% (n = 1, 2.1%) (Fig. 1G). Density of CD8+ TILs was evaluated in 32 cases, showing low in 28 cases (87.5%) and high in 4 cases (12.5%) (Fig. 1H). Taken together, both the level of PD-L1 expression and the density of CD8+ TILs were generally low in NENs.
Comparative Analyses of Mutational Landscape, Genomic, and Immune-Related Biomarkers Associated With Degree of Differentiation and Primary Localization
Given that NETs and NECs harbor distinct clinical and biological behaviors, comparative analyses were performed. The average number of somatic variations of NECs was significantly higher compared to that of NETs (14 vs. 5 variants/per case, P = .0058, Fig. 2A). In contrast with NETs, NECs carried a markedly higher frequency of TP53 mutation (68.8% vs. 6.7%, P < .0001) (Fig. 2B). RB1 mutation as well as TP53/RB1 co-mutation was exclusively detected in NEC patients in our cohort (Fig. 2B). In addition, we found that gene alterations of AR, FGFR3, GNAS, HRNR, KRAS, MYC, and NOTCH1 were also only detected in NECs but not in NETs (Fig. 1I; and Supplementary Table S1). The average score of MSI did not differ between the 2 groups (Fig. 2E). A significantly higher average TMB value was observed in NECs by contrast with NETs (7.515 vs. 2.522 Muts/Mb, P = .0082) (Figs. 2F, 1I). Similarly, the average value of TNB was also higher in NECs than NETs (2.308 vs. 0.7227 Neos/Mb, P = .0013) (Figs. 2G, 1I). Besides, the ratio of cases with HLA-I homogeneity seemed approximate between NETs and NECs, while proportion of cases with HLA LOH was relatively higher in NECs compared to NETs (48.4% vs. 20.0%, P = .1068) (Fig. 2C, 2D). The level of PD-L1 expression was generally low in both subtypes (Supplementary Table S1). Average density of CD8+ TILs in NECs was relatively lower than NETs (261.9 vs. 502.0 psc/mm2) (Fig. 2H). Collectively, it exists big differences, especially in the landscape of genomic biomarkers between NETs and NECs.
Comparison of mutational landscape, genomic, and immune-related biomarkers with regard to degree of differentiation and primary localization. A and E-H, Comparative analysis of average somatic variants (A), MSI score (E), TMB (F), TNB (G), CD8+ TILs (H) in patients with NETs (n = 15) and NECs (n = 32) using unpaired t-test with Welch’s correction. Data represent mean ± SEM. **P < .01; B-D, Comparative analysis of TP53, RB1 mutation, and TP53/RB1 co-mutation frequency (B) and the ratio of cases with HLA LOH (C) or HLA-I homogeneity (D) in patients with NETs and NECs using Fisher exact test. I-L, Comparative analysis of TP53 (I), RB1 (J) mutation frequency and the ratio of cases with HLA-I homogeneity (K) or HLA LOH (L) among different primary localization using Fisher exact test; M-P, Comparative analysis of average MSI score (M), TMB (N), TNB (O), CD8+ TILs (P) among different primary localization using unpaired t-test with Welch’s correction. Data represent mean ± SEM. *P <.05, **P < .01.
We also investigated their differences in regard to primary localization. NENs from GI tract carried a relatively higher frequency of TP53 mutation (9/13, 69.2%) than NENs from other locations, and RB1 displayed a comparatively higher mutation frequency in NENs from lung (4/9, 44.4%) in comparison with other locations, although no statistical significances were achieved (Fig. 2I, 2J). NENs from thymus carried a relatively lower average MSI score compared to NENs from other locations, which was significantly lower than NENs from lung (2.185 vs. 4.278, P = .0092, Fig. 2M). TMB and TNB did not differ significantly among groups (Fig. 2N, 2O). The ratio of cases with HLA-I homogeneity in GI tract (6/13, 46%) or unknown/others (6/13, 46%) was relatively higher than in other parts (Fig. 2K). The proportion of HLA LOH in NENs from thymus or pancreas was lower than NENs from other locations, but no statistical significance had been achieved among groups (Fig. 2L). NENs from lung displayed the lowest average density of CD8+ TILs, which was markedly lower than NENs from GI tract (115.6 vs. 301.8 psc/mm2,P = .0208) (Fig. 2P), suggesting that a more suppressive immune landscape exists in NENs from lung.
Protein-Protein Interaction (PPI) Network and Functional Pathway Analysis of Hub Mutant Signature
We performed PPI-based analysis using STRING database to determine whether the mutated genes functionally interacted with each other and were involved in tumorigenesis. The networks were visualized using Cytoscape and detailed in Supplementary Table S1. Mutated genes in NENs mainly clustered into 5 clusters and the top hub genes with the highest clustering included TP53, EP300, ATM, HRAS, and NRAS, illustrated by the size of circle (Supplementary Fig. S1A). When focusing on mutated genes in NECs, 5 clusters with the highest clustering of CREBBP are shown in Supplementary Fig. S1B. As for NETs, mutated genes just clustered into 2 clusters containing 18 nodes and 33 interacting pairs (Supplementary Fig. S1C). Next, we used the Metascape database to perform functional pathway analyses of the mutated genes. Supplementary Fig. S1D–S1F shows the most enriched GO term associated with regulation of pathways in cancer. In addition, enrichment of mutant genes in several signaling pathways involved in malignancy was demonstrated in the whole cohort. As shown in Supplementary Fig. S1G, the P53 pathway was the most affected (25/47, 53.2%), followed by the receptor tyrosine kinase (RTK)/RAS pathway (19/47, 40.4%) and PI3K/AKT/mTOR pathway (7/47, 14.9%). Other pathways including MYC, Notch, and Wnt were also involved.
Prognostic Impact of Somatic Mutations, Genomic and Immune-Related Biomarkers in NEN Patients
The median follow-up time was 19 months, and 28 (60.9%) subjects died of disease (NETs, 9; NECs, 19) (Table 1; Supplementary Table S1). We first investigated the relationship of clinicopathological characteristics including tumor histotype, age, and gender with patient survival in terms of OS by performing univariate Cox regression and Kaplan-Meier analyses. As shown in Supplementary Fig. S2, both age and gender had no clear relationship with OS in our cohort. The median OS of NEC patients was much worse than that of NET patients (23 vs. 44 months), although it did not reach significantly statistical difference (P = .09305).
Next, to explore the prognostic roles of gene mutations in NENs, univariate Cox regression analysis was conducted. Patients were categorized into wild-type and mutated type according to their gene-mutation status and the top 53 genes with mutation frequency more than 4% were analyzed. The results showed that there were seven OS-related gene (MLH3, NACA, NOTCH1, NPAP1, RANBP17, TSC2, and ZFHX4) mutations, which reached statistical significance in our cohorts (P < .05, Fig. 3A). Kaplan-Meier analysis further confirmed that above-mentioned 7 gene mutations were significantly associated with worse prognosis, respectively (Supplementary Fig. S3A–S3G). Besides, we also found patients with mutation in TPR had poorer prognosis (Supplementary Fig. S3H). Furthermore, to explore the combined predictability of the survival-associated genes, we investigated the effect of mutation status of any of the 7 genes on survival. That is, if a patient carries a mutation that belongs to any of the 7 genes (signature 1), then the patient is considered to be a mutation carrier. As shown in Fig. 3A, 3C, 12 patients who carried at least 1 mutation in at least 1 of the 7 genes exhibited significantly poorer survival, indicating that integrating mutations of multiple genes could better predict patient OS. To find the independent prognostic biomarkers, the univariate Cox statistical index (P < .05) was introduced into multivariate Cox. The results showed that MLH3, NOTCH1, and TSC2 mutations as well as signature 1 were independent factors affecting prognosis (Table 2). Next, we performed a stratified Kaplan-Meier analysis for mutations clustering to specific tumor histotypes. Among NECs, the mutations in DROSHA, NPAP1, and TSC2 were significantly correlated with poor OS (P < .05) (Supplemental Fig. S3I–SK). Interestingly, as shown in Fig. 3B, 3D, their combined predictability was further demonstrated by Kaplan-Meier and univariate Cox analyses (signature 2; P = 2.21e−06 and .00011, respectively). As for NETs, mutation in MLH3 had a strong relationship with worse OS (P = .00003) (Supplemental Fig. S3L).
Multivariate Cox analysis showed mutation in MLH3, NOTCH1, and TSC2 as well as signature 1 as independent prognostic markers of poor prognosis.
| Covariate | HR | 95% CI | P-value |
|---|---|---|---|
| MLH3 mutation | 29.03 | 5.90-142.79 | .00003 |
| NACA mutation | 0.10 | 0.00-14.30 | .36744 |
| NOTCH1 mutation | 6.64 | 1.76-24.99 | .00512 |
| NPAP1 mutation | 19.73 | 0.22-1.78e+03 | .19444 |
| RANBP17 mutation | 0.10 | 0.00-14.30 | .36744 |
| TSC2 mutation | 246.20 | 1.33-4.57e+04 | .03884 |
| ZFHX4 mutation | 4.84 | 0.57-41.44 | .14996 |
| Signature 1 | 13.87 | 1.48-130.04 | .0213 |
| Covariate | HR | 95% CI | P-value |
|---|---|---|---|
| MLH3 mutation | 29.03 | 5.90-142.79 | .00003 |
| NACA mutation | 0.10 | 0.00-14.30 | .36744 |
| NOTCH1 mutation | 6.64 | 1.76-24.99 | .00512 |
| NPAP1 mutation | 19.73 | 0.22-1.78e+03 | .19444 |
| RANBP17 mutation | 0.10 | 0.00-14.30 | .36744 |
| TSC2 mutation | 246.20 | 1.33-4.57e+04 | .03884 |
| ZFHX4 mutation | 4.84 | 0.57-41.44 | .14996 |
| Signature 1 | 13.87 | 1.48-130.04 | .0213 |
The significance of bold value is P value <.05.
Abbreviation: HR, hazard ratio.
Multivariate Cox analysis showed mutation in MLH3, NOTCH1, and TSC2 as well as signature 1 as independent prognostic markers of poor prognosis.
| Covariate | HR | 95% CI | P-value |
|---|---|---|---|
| MLH3 mutation | 29.03 | 5.90-142.79 | .00003 |
| NACA mutation | 0.10 | 0.00-14.30 | .36744 |
| NOTCH1 mutation | 6.64 | 1.76-24.99 | .00512 |
| NPAP1 mutation | 19.73 | 0.22-1.78e+03 | .19444 |
| RANBP17 mutation | 0.10 | 0.00-14.30 | .36744 |
| TSC2 mutation | 246.20 | 1.33-4.57e+04 | .03884 |
| ZFHX4 mutation | 4.84 | 0.57-41.44 | .14996 |
| Signature 1 | 13.87 | 1.48-130.04 | .0213 |
| Covariate | HR | 95% CI | P-value |
|---|---|---|---|
| MLH3 mutation | 29.03 | 5.90-142.79 | .00003 |
| NACA mutation | 0.10 | 0.00-14.30 | .36744 |
| NOTCH1 mutation | 6.64 | 1.76-24.99 | .00512 |
| NPAP1 mutation | 19.73 | 0.22-1.78e+03 | .19444 |
| RANBP17 mutation | 0.10 | 0.00-14.30 | .36744 |
| TSC2 mutation | 246.20 | 1.33-4.57e+04 | .03884 |
| ZFHX4 mutation | 4.84 | 0.57-41.44 | .14996 |
| Signature 1 | 13.87 | 1.48-130.04 | .0213 |
The significance of bold value is P value <.05.
Abbreviation: HR, hazard ratio.
Survival analysis identifying survival-related genes. A, The forest plots show the univariate Cox analysis results of 7 OS-related gene mutations and the combination of the 7 mutations (signature 1) in the NEN cohort. B, Univariate Cox analysis for the combination of all the mutations identified in 3 genes (DROSHA, NPAP1, and TSC2) (signature 2) in NECs. C, D, Kaplan-Meier survival analysis for the combination of signature 1 and signature 2. P-values were calculated by Cox and log-rank test, respectively.
Since genomic biomarkers have been reported to be associated with survival in various cancer types,19-21,23 we also tested their relationship with prognosis in our cohort. The result showed that TMB and TNB had no statistical significance on OS (Supplementary Fig. S4A, S4B). Besides, regardless of whether patients had germline HLA homogeneity, their prognoses were not significantly different (Supplementary Fig. S4C). We conducted further analysis to show the association between specific HLA-I locus homozygosity (HLA-A, HLA-B, and HLA-C) and survival. However, specific HLA-I locus homozygosity still showed limited prognostic value (Supplementary Fig. S4D, S4E). Next, we analyzed the predictive performance of HLA LOH with respect to OS. Notably, patients with LOH of HLA-I were associated with reduced survival in comparison with those with intact HLA-I (median OS: 15 vs. 44 months), although it did not reach significantly statistical difference (P = .06492, Supplementary Fig. S4F). Specifically, HLA LOH instead of HLA-I homozygosity significantly predicted poor survival in a subset of NEN patients who harboring low or medium TNB, low PD-L1 expression or low density of CD8+ TILs (Fig. 4F–4H vs. 4A–4D).
Kaplan-Meier survival analysis and Cox proportional risk regression analysis identified HLA LOH as a novel prognostic marker in a subset of NENs. The prognostic value of HLA-I homozygosity and HLA LOH in low or medium TMB (<10 Muts/Mb) (A and E), low or medium TNB (<4.5 Neos/Mb) (B and F), low PD-L1 expression (TPS < 1%) (C and G), and low density of CD8+ TILs (D and H) patients. P-values of Kaplan-Meier analysis were calculated by the log-rank test.
Identification of Clinically Actionable Mutations and Potential Therapeutic Targets
To determine treatment options for targeted therapy or immunotherapy in NENs, clinical relevance of somatic alterations is summarized in Fig. 5; Supplementary Table S2. When focused on somatic variants known as possible (and responsive) druggable targets against currently available (or under development) treatment agents, we observed 31 NEN patients (31/47, 66%) harboring one or more target-specific or general somatic aberrations could benefit from drugs related to immunotherapy or targeted therapy registered for another indication but not currently administered in NEN treatment. We found TP53 (n = 21; mutations resulting in TP53-deficiency), TMB-H (≥10; n = 6), KRAS (n = 5; mutations in exons 2 (G12 and T20M), and HRAS (n = 3; mutations in exons 2 (G12) and 3 (Q61)) to be the most frequently observed (target-specific or general) somatic aberrations, which granted eligibility to various possible treatment options, including ICB and novel small-molecular inhibitors, such as WEE1 inhibitor, FTase inhibitor, and CDK4/6 inhibitor. Pembrolizumab (PD-1 monoclonal antibody) is the only currently approved (level 1 evidence) immunotherapy indicated by TMB-H in NENs. In total, 5 of 6 cases with TMB-H (≥10 Muts/Mb) are NECs, indicating that they are potential responsive candidates toward ICB. Adavosertib, which is a specific WEE1 inhibitor, has been reported to be effective in TP53-mutant (including point mutation, truncation, etc.) cells in early phase clinical trial33,34 and TP53-deficient (caused by any type of mutation) RAS-mutant metastatic colorectal cancer.35 Thus, the 21 patients with TP53 mutation (resulting in TP53-deficiency), including 4 patients (P4, P16, P25, and P47) with both RAS- and TP53-mutations (both RAS and TP53-deficeiency) in our study may benefit from the application of Adavosertib. Besides, Abemaciclib (CDK4/6 inhibitor) and Tipifarnib (FTase inhibitor) have been reported to be precision therapy for KRAS-mutant (most often involved mutations in codon 12 or 13, Q61H and A146V) non-small cell lung cancer36 and HRAS-mutant (most often involved mutations in exons 2 (G12, G13), 3 (Q61), and 4 (K117 and A146)) head and neck squamous cell carcinomas,37 respectively. In our study, the 4 patients (P16, P24, P25, and P42) who had KRAS mutations in codon 12 may benefit from Abemaciclib, and the 3 patients (P5, P14, and P25) who had HRAS mutations in G12 and Q61 mutations may benefit from Tipifarnib. Thus, our study provides novel therapeutic targets which are potentially translated for NEN patients in the foreseeable future.
Clinically actionable somatic alterations related to targeted therapy or immunotherapy observed within NENs. NENs harboring current clinically actionable alterations are shown. Bottom tracks represent subclassification of NENs based on differentiation grade and primary localization. The right-hand side figure shows the number of samples harboring TMB-H or somatic alteration within the given gene.
Discussion
Currently, the diagnosis and management of NENs are largely dependent on tumor differentiation, grade, primary tumor localization, somatostatin expression, and clinical stage, while these parameters do not always accurately reflect the actual tumor biology. Although therapeutic strategies including somatostatin analog (SSA), TKIs, mTOR inhibitors, chemotherapy, and peptide receptor radionuclide therapy (PRRT) have greatly improved the management of NENs nowadays,38 significant heterogeneity in tumor leads to unpredictable or unsatisfied responses to certain therapies and distinct survival outcomes. Thus, there is still substantial well-acknowledged unmet clinical need for better molecular stratification within subtypes. Generally, it also lacks effective biomarkers to predict the clinical outcomes of NEN populations. Genomic characteristics and immune landscape will provide substantial information for better understanding of the biological behaviors and molecular abnormalities driving the development of NENs, and identify novel biomarkers to enable personalized therapeutic strategies. Therefore, it is of great significance to find predictive markers which can be used to improve prognostic stratification and extend treatment choices for NEN patients. In view of this, we carried out this pioneering study by performing targeted NGS and IHC assays to have an in-depth look into the genetic mutations, genomic and immune signatures of a cohort consisting of 47 NENs from various primary localizations and distinct differentiation statuses. Genetic mutation analysis showed that the most frequently altered gene is TP53 (51%), followed by RB1, MYC, and FAT4. Alterations in TP53 and RB1 have been reported in NECs of the GI (GIS-NECs).39 Functional enrichment pathway analysis indicated that the mutated genes were closely correlated with the P53, RTK/RAS, and PI3K/AKT/mTOR pathways, indicating their critical roles in tumorigenesis of NENs. Therefore, targeted therapy for these pathways is promising. The possible link between genetic alternations and patient prognosis was investigated. We found that the outcomes of patients with certain genetic mutations were worse than those of wild-type patients. Mutations in MLH3, NACA, NOTCH1, NPAP1, RANBP17, TSC2, and ZFHX4 and their combination (signature 1) were significantly correlated with poor OS, which are reported herein for the first time as predictors of poor prognosis in patients with NENs. According to previous studies in other cancer types, some of these 7 genes have been reported to regulate tumor progression and were identified as potential prognostic markers. For instance, correlation between gene alterations and poor prognosis has been observed for ZFHX4 in esophageal squamous cell carcinoma and TSC2 in hepatocellular carcinoma.40,41 Thus, further molecular studies are worthwhile to uncover the detailed mechanisms of above 7 mutant genes related tumorigenesis and aggressiveness in NENs.
NENs include 2 distinct entities: well-differentiated NETs and poorly differentiated NECs, which display dramatically distinct biological behaviors and prognosis.2,10,14 However, the intrinsic molecular differences underlying NETs and NECs have not been clearly illustrated. In our study, we verified that the genetic and immune landscapes of NECs are markedly dissimilar from those of NETs. Compared to NETs, NECs carried a significantly higher average number of mutations per case than NETs. Of note, frequency of TP53, RB1 mutation, and TP53/RB1 co-mutation was apparently higher in NECs than in NETs, which is in line with a previous study performed in pNENs, showing that intragenic mutations in the TP53 and RB1 genes were only identified in virtually all small cell and large cell NECs, but not in any NETs, indicating that inactivation of these 2 genes is a fundamental genetic feature of poorly differentiated NECs.42 Furthermore, the feature of TP53/RB1 co-mutation/loss has been applied to differentiate genomic profiles of large cell NECs (LCNECs) into 2 major subsets: SCLC-like and NSCLC-like.43 Thus, our finding is considered to be a major breakthrough regarding the classification of NENs utilizing genomic data and is of great importance for the treatment of NETs and NECs. Besides, gene alterations of AR, FGFR3, GNAS, HRNR, MYC, NOTCH1, and KRAS were also only detected in NECs but not in NETs in this cohort. Of course, this result is essential for further verification in a validation cohort.
In recent years, TMB and TNB, 2 novel predictive biomarkers, can be used to identify patients who are likely to benefit from ICB.44,45 TMB was generally low in this NEN cohort, which is in accordance with previous study.46 Significantly higher average values of TMB and TNB in NECs were observed in comparison with NETs, regardless of the site of tumor origin. Riet et al. also observed a relatively higher TMB in NECs than in NETs (average 7.515 vs. 2.522 Muts/Mb).46 This finding may explain why NEC patients benefit more from ICB than NET patients. According to a phase II basket trial in patients with non-pancreatic NENs treated with anti-CTLA-4 and anti-PD-1 blockade, patients with high-grade NECs had an objective response rate (ORR) of 44% vs. 0% in low/intermediate grade NETs.47 Considering that platinum-based chemotherapy regimens represent the only treatment usually proposed for poorly differentiated NECs, these results are encouraging and shed new light on immunotherapy as a rational approach for NEC patients. Concerning molecular differences related to primary localization of NENs, NENs from thymus carried a significantly lower level of MSI compared to NENs from other locations, suggesting a relatively stable genome of thymic NENs. Density of CD8+ TILs in NENs from lung was significantly lower than NENs from GI tract, possibly as consequence of a more suppressive immune landscape existing in NENs from lung. Those immunosuppressive cells including myeloid-derived suppressor cells (MDSCs), tumor-associated macrophages (TAMs), as well as regulatory T cells (Tregs) have been reported to infiltrate NENs, thus creating an immunosuppressed microenvironment to impairs the infiltration and activation of CD8+ TILs.48,49 With this regard, low-density infiltration of CD8+ TILs may reflect more immunosuppressive cell populations existing in NENs from lung. However, further exploration is required to validate the speculation.
In addition, NENs from GI tract carried a significantly higher frequency of TP53 mutation than NENs from thymus. These findings corroborate the fact that these tumors derived from different origins are actually different entities for which different therapeutic approaches may be explored.
HLA-I molecules are highly polymorphic, with variations located in the peptide-binding region and each variant binds a select repertoire of peptide ligands. The occurrence of HLA LOH, meaning the disruption of the ability to present neoantigens, is considered as one pattern of immune evasion.50 HLA LOH has been demonstrated to be clinically correlated with poor prognosis in various cancer,19,51,52 and mediates resistance to ICB,22 whereas there are no reports on LOH at HLA in NENs. Our study for the first time investigated the frequency and clinical significance of HLA-I function from the view of germline HLA-I status and somatic tumor HLA-I status. We found that HLA-I LOH and germline homogeneity are common in NENs, accounting for 39% and 36%, respectively. Our study has important implications for clinical translations. We found that HLA-I LOH played a more important role than germline homogeneity in clinical outcomes of patients with NENs, especially for the patients with low or medium TNB, low PD-L1 expression, and low density of CD8+ TILs. One potential explanation was the relatively lower prognostic impact of germline-level mutations on HLA-I function in NENs, which should be strictly proven in future studies. Indeed, we noticed that previous studies showed that the prognostic value of single germline HLA homogeneity was controversial.53
In the present study, we also sought to determine if genomic alterations in NENs could translate into clinically actionable targets. With various public databases, we observed a substantial group of NEN patients (66%) harboring clinically relevant and potentially targetable somatic aberrations which could possibly extend their treatment repertoire. Five NECs and a single high-grade NET (G3) with TMB-H (TMB ≥ 10 Mut/Mb) could be eligible for immune-based therapies such as ICB. When looking at TMB-H as a predictive factor for ICB, it was recently shown that TMB-H NECs can respond to pembrolizumab.54 In addition, we identified several novel small-molecular inhibitors such as WEE1 inhibitor, FTase inhibitor, and CDK4/6 inhibitor, which might be effective in NENs with corresponding mutant genes. It should be noted that most of the mutations defined as actionable gene alterations in our cohort are based on clinical evidence extracted from other cancer types. Although these genes have not been specifically studied in the NEN population, it paves the way for further exploration of their potential roles in driving tumorigenesis or mediating response to certain targeted therapies in NENs.
Compared to the previous NEN studies, our study has several distinctive findings. First, we provide a comprehensive view of the somatic mutation landscape by targeted NGS technology, which represents cost-effective compared to whole-genome sequencing or whole-exome sequencing. Second, due to lack of OS data, previous studies merely depicted the prevalence of certain mutated genes, but the prognostic value of each gene mutation had not been well elucidated.39,42,46 Based on the valuable OS data in our study, we were allowed to discover novel mutation biomarkers for predicting the prognosis of NEN patients. Third, this research is the first study to systemically discuss TNB, germline HLA-I mutation, somatic HLA-I LOH, and immune-related biomarkers in NENs. Fourth, our study has important implications for clinical translations. Our study stressed the prognostic significance of HLA-I LOH in NENs, which, to the best of our knowledge, had not been reported previously.
There are some limitations of this study. First, the true prevalence of prognosis-associated gene mutations or clinically targetable alterations in NENs might have been underestimated due to the limited genes in this targeted NGS panel. Besides, most patients in our study were given routine clinical care instead of biomarker or mutation-driven therapy, thus we do not yet know whether these identified associations between genomic alterations and specific drugs indeed translate into clinical responses in these patients. Therefore, precise biomarkers or mutation-guided prospective clinical trials are warranted to be designed in a larger cohort of NEN patients in the foreseeable future.
Conclusion
Our study delineated genomic landscape and immune signature in heterogenous and rare NENs, and identified potential therapeutic targets and novel prognostic biomarkers, pinpointing that molecular profiling could complement histology to provide better prognostic stratification and therapy choice for NEN patients.
Acknowledgments
We thank the patients and their families, the nurses, and the pathologists who participated in this study.
Ethics statement
This study was approved by the Institutional Ethics Committee (IEC) for Clinical Research of the First Affiliated Hospital of Sun Yat-sen University (approval number [2021]775).
Funding
This study was funded by the National Natural Science Foundation of China (grant number: 82002502), the National Natural Science Foundation of China (grant number: 82141104and 82070536), and the Natural Science Foundation of Guangdong Province (grant number: 2019A1515012027).
Conflict of Interest
The authors indicated no financial relationships.
Author Contributions
Conception/design: M.L., N.L., N.Z., J.C. Provision of study material or patients: N.Z., J.C. Collection and/or assembly of data: M.L., N.L., H.T., X.L., L.C., Y.W., Y.L., Y.L., S.W., W.W., M.C. Data analysis and interpretation: M.L., N.L., H.T. Technical support: X.L., L.C., Y.W., Y.L., Y.L., S.W., W.W., M.C., J.W. Manuscript writing: M.L., N.L. Manuscript revision: J.C., N.Z. Final approval of manuscript: All authors.
Data Availability
The datasets used and/or analyzed in this study are available from the corresponding author on reasonable request.
References
Author notes
Contributed equally.
