This study assembled a cohort of 40 digestive NENs, including 26 NETs G3, 8 NETs G1/G2, and 6 NECs (5 of large and 1 of small cell type). Clinicopathological data are detailed in Table 1. NET G3 patients had a median age of 56 years (range 15–78), similar to NET G1/G2 (58 years, range 50–81), whereas NECs occurred in significantly older individuals (78 years, range 38–83; p = 0.042). Sex distribution did not significantly differ, although NET G3 showed a male predominance (69%). Median tumor size was identical in NET G3 and NET G1/G2 (2.5 cm), while NEC tended to be larger (7.1 cm). Within the NET G3 group, the pancreas was the predominant primary site (15/26, 58%), followed by the ileum (4/26), colon (3/26), stomach (2/26), and single cases arising in the duodenum and rectum. Median Ki-67 index was 25% (21–65) in NETs G3, intermediate between NETs G1/G2 (2%, 1–10) and NECs (80%, 75–90) (p < 0.001) with no overlap between NETs G3 and NECs. Median mitotic count per 2 mm² was likewise intermediate in NET G3 (5, range 0–33) compared with NET G1/G2 (1, 0–11) and NEC (35, 3–64) (p = 0.001). Strong SSTR2A expression (score 3+) was observed in 91% of NETs G3, 100% of NETs G1/G2 and in one NEC (p < 0.001).

Transcriptomic profiling (NanoString IO360) generated valid results in 34 quality-controlled samples, corresponding to 22 NETs G3, 7 NETs G1/G2, and 5 NECs. Targeted DNA/RNA sequencing yielded exploitable data in 20 samples after quality control, including 13 NETs G3, 4 NETs G1/G2, and 3 NECs.

Table 1 Clinico-pathological baselineMolecular Positioning of NET G3 as Compared to NET G1/G2 and NECGene Expression Profiling

Unsupervised hierarchical clustering analysis of IO360 genes separated neoplasms into two expression clusters (Fig. 1A). Cluster membership was not statistically associated with diagnostic categories, although most NETs G1/G2 and G3 grouped together (cluster II). Principal component analysis (PCA) confirmed separation of NECs from NETs, with no clear distinction between NETs G3 and NETs G1/G2 (Fig. 1B). The greatest transcriptomic divergence was observed between NETs G3 and NECs, where 226 genes were differentially expressed, as compared to the 138 differentially expressed genes between NETs G3 and NETs G1/G2 (Fig. 1C). The number of differentially expressed genes was also higher between NETs G3 and NECs than between NETs G3 and NETs G1/G2 in virtually all pathways (Fig. 1D). In addition, the pathway activity shift in NETs G3 compared to NECs presented a skewed profile, with several up- and downregulated pathways. In contrast, NETs G3 compared to NETs G1/G2 showed predominantly upregulated pathways, supporting a transcriptomic continuum between NET G1/G2 and G3 (Fig. 1E). In NET G3 vs NEC, the most pronounced negative shifts involved cell-cycle regulation and DNA-damage repair, together with metabolic-stress response, whereas positive shifts were observed in Hedgehog, JAK–STAT signaling, and autophagy. In NET G3 vs NET G1/G2, pathway modulation was predominantly positive, mainly involving extracellular-matrix remodeling, hypoxia/metabolic stress, and cytokine/JAK–STAT signaling, with only relevant negative shift in Hedgehog signaling. These trends are further illustrated in Fig. S1.

Fig. 1

Transcriptomic profile of NETs G3 as compared to NETs G1/G2 and NECs. Legend: A Unsupervised hierarchical clustering identified two main clusters without clear segregation of NETs G3 from NETs G1/G2 or NECs. B Principal component analysis demonstrated that the first two principal components (PC1 and PC2) accounted for 38% of the total variance and segregated NETs (G1/G2 and G3) from NECs. C Venn diagram showed that only a minority (n = 58) of the differentially expressed genes were shared between the NET G3 vs NEC and NET G3 vs NET G1/G2 comparisons. D The number of differentially expressed genes was higher in the NET G3 vs NEC comparison across most evaluated pathways. E Pathway activity analysis revealed a heterogeneous pattern in NET G3 vs NEC, with both up- and downregulated pathways, whereas NET G3 vs NET G1/G2 comparison was characterized predominantly by pathway upregulation

Targeted Sequencing

At the genomic level (next-generation sequencing analysis), the most frequent altered genes in both NETs G3 and NETs G1/G2 participated in epigenetic and chromatin control, with ZYMM3, BCOR, ATRX, KDM6A, and KDM5C among the most frequent 10 altered genes (Fig. 2A). In contrast, in NECs, tumor suppressor and cell cycle genes predominated. More broadly, NET G3 was overwhelmingly enriched in uniquely altered genes (n = 212), with limited overlap with both NET G1/G2 (23 genes) and NEC (54 genes) (Fig. 2B). Accordingly, a low J-index was observed in both comparisons (0.18 vs. NEC and 0.08 vs. G1/G2), underscoring low similarity. Indeed, NET G3 accounted for the majority of molecular alterations (798 events, including SNV, indels, and CNA affecting 286 genes), compared to NEC (79 events; 60 genes) and NET G1/G2 (34 events; 27 genes). Alteration burden per sample differed by category with borderline overall significance (Kruskal–Wallis p = 0.058), with a significant pairwise difference between NET G3 and NET G1/G2 (p = 0.033).

Fig. 2

NET G3 as compared to NETs G1/G2 and NECs. Legend. A Targeted deep sequencing revealed that NET G3 and NET G1/G2 most frequently harbored alterations in genes involved in epigenetic regulation, whereas cell cycle–related alterations predominated in NEC. B Only a limited number of molecular alterations were shared across the three diagnostic categories, with NET G3 exhibiting the highest number of uniquely altered genes. C NET G3 showed a high median number and the widest range of genomic alterations per sample. D Copy number losses represented the most frequent type of alteration in NET G3

NET G3 Heterogeneity and Core Characteristics

Across platforms, NETs G3 showed substantial heterogeneity. Transcriptomically, NETs G3 distributed across both unsupervised clusters (4/22 in Cluster I and 18/22 in Cluster II, Fig. 1A), indicating internal variability despite a common diagnostic label. The overlap of differentially expressed genes among NET G3 vs. NEC and G1/G2 was limited, as only 58 genes overlapped (hypergeometric p = 0.0398) (Fig. 1C). Moreover, only 15/58 genes were deregulated in the same direction.

NETs G3 showed a high number of genomic alterations (median 29, range 1-215) as compared to NETs G1/2 (median 7, range 3–17) and NECs (median 29, range 19–31) and were highly heterogeneous in terms of number of alterations per sample (Fig. 2C), dominated by copy number loss (Fig. 2D).

Despite the global molecular heterogeneity of NET G3, several reproducible group-level features were identified. Transcriptomically, in comparison with NECs, NETs G3 showed significant upregulation of genes involved in hormone signaling and morphologic control, including PRLR (log₂FC = + 5.85, p < 0.0001) and WNT4 (log₂FC = + 4.72, p < 0.0001), suggesting preserved well-differentiated signature (Fig. 3A). Against NETs G1/G2, NETs G3 showed positive pathway shifts in extracellular-matrix remodeling, metabolic reprogramming, and cytokine/JAK–STAT signaling (Fig. 1E). Indeed, among the most significantly upregulated transcripts were LAMB3 (log₂FC = + 4.50, p < 0.0001), a basal membrane laminin, and TNFAIP6 (log₂FC = + 3.27, p = 0.0001), a key regulator of extracellular matrix interactions; CASP3 (log₂FC = + 1.23, p < 0.0001), LIF (log₂FC = + 3.08, p < 0.0001), and IL1B (log₂FC = + 3.21, p = 0.0006) that are target and regulatory genes of the JAK–STAT–MAPK signaling; and the glycolytic enzyme HK2 (log₂FC = + 3.37, p = 0.0001). Similarly, EZH2, an epigenetic regulator associated with proliferative activity, was significantly upregulated in NET G3 compared with NET G1/G2 (log₂FC = + 1.49, p = 0.0002), while showing opposite regulation in the NET G3 versus NEC comparison (log₂FC = − 1.84, p = 0.0001) (Fig. 3B).

Fig. 3

Differential gene expression analysis of NETs G3 as compared to NETs G1/G2 and NECs. Legend: A, B Volcano plots depicting significantly upregulated (red) and downregulated (blue) genes in the NET G3 vs NEC (A) and NET G3 vs NET G1/G2 (B) comparisons. C Focused analysis of genes included in the Tumor Inflammation Signature (TIS) demonstrated low-magnitude and predominantly non-significant expression differences among NET G3, NET G1/G2, and NEC. D The composite TIS score was low in both comparisons. E Immune cell signature analysis demonstrated a predominantly low-to-intermediate tumor-infiltrating lymphocyte (TIL) profile in NET G3, with a small subset of cases exhibiting a TIL-enriched pattern. F TIL-enriched case 4 corresponded to a cecal pediatric NET G3 characterized by a low genomic alteration burden and prominent lymphocytic infiltration on conventional H&E histology

At the genomic level, NETs G3 showed alterations (SNVs, indels, CNAs) in epigenetic and chromatin control genes MEN1, DAXX, and ATRX in 5/13 cases, as compared to 2/4 of NETs G1/G2 and no NECs, underscoring a common neuroendocrine backbone profile of NETs independent from grade. Notably, ATRX alterations were not restricted to pancreatic primaries but also included a duodenal NET G3. Indeed, genes involved in cell replication network were significantly downregulated in NETs G3, as compared to NEC. These genes included RRM2 (log₂FC = − 2.68, p < 0.0001), TYMS (log₂FC = − 3.59, p < 0.0001), both involved in synthesis of DNA precursors, and DNMT1 (log₂FC = − 2.24, p < 0.0001), implicated in DNA methylation of subtelomeric regions (Fig. 3A).

In contrast, TP53 and RB1 alterations were a key feature of NECs (3/3 and 2/3 cases, respectively) as compared to NETs G3 (p = 0.002), where TP53 mutation alone was detected in a minority (2/13) of NETs G3 and RB1 was always intact. No TP53 and RB1 alterations were identified in G1/G2 NETs.

A key feature of NETs G3 in this dataset was the presence of frequent CN losses (average CN loss rate 64; range 6–207), occurring in specific chromosomal regions, recognized as recurrent large-segment deletions. Indeed, 8/13 (62%) NETs G3 presented recurrent deletions involving Xp22.33–p11.22 and Xq11.2–q28 loci, which contain several chromatin-regulators (KDM5C, KDM6A, BCOR, ZMYM3, DDX3X, STAG2, CUL4B), while 6/13 (46%) NETs G3 presented recurrent deletions involving the 10q11.21–q26.13 locus, which contains the histone demethylase ARID5B. This Xp/Xq deletion pattern was highly specific for NET G3 (specificity 97.6%; p = 0.0001 vs NET G1/G2 and NEC).

In addition, NETs G3 were also characterized by the presence of frequent PI3K/mTOR pathways mutations (PIK3CA, PTEN, TSC2, detected in 21% of cases). This observation was combined with the selective up-regulation of downstream metabolic and cytokine-signaling genes as compared to NETs G1/G2 at gene expression profiling. Also, one duodenal case demonstrated VHL mutation, and a global increased expression of LAMB3 and S100A9, involved in hypoxia-linked matrix remodeling, was observed in the NET G3 vs NET G1/G2 comparison.

Targetable Alteration and Immunotherapy Susceptibility in NET G3

According to the OncoKB database, targetable mutations were detected in 3 out of 13 (23%) NETs G3, namely ARID1A p.Gln524AlafsTer94, PIK3CA p.His1047Arg and PTEN p.Gly165Ar (Table 2).

Table 2 Pathogenic, likely pathogenic, and targetable mutations in NETs G3 according to the OncoKB database

Concerning the homologous recombination deficiency assessment, no case reached a formal genomic-instability threshold as defined by the GIM score, and recombination repair gene alteration counts did not differ across groups. Nevertheless, 61 losses and 2 SNVs of unknown significance were detected in HRR genes of NET G3 cases.

Regarding microsatellite instability, one NEC case was MSI, as was already known from the diagnostic workup, while one NET G3 carried an MSH2 p.Arg680Ter83 mutation, which was annotated as pathogenic for germline classification (Lynch Syndrome) in the ClinVar database.

For immune transcriptomics, the TIS score supported the notion of low immune activation. When individual TIS genes were analyzed in NET G3 vs NEC and NET G1/G2, none reached significance except for CMKLR1 (p = 0.011 for NET G3 vs. NEC), and the fold changes were low (Fig. 3C). The aggregate TIS scores were also low, namely 0.07 (NET G3 vs NEC) and 0.24 (NET G3 vs NET G1/G2) (Fig. 3D). When looking at immune cell signatures, NET G3 generally showed a low to intermediate TILs signature, with a notable subset of 2 cases with a TILs enriched profile (Fig. 3E). Complementary PCA and PLS-DA analyses of immune cell profiling scores suggested differences between NET G3 and NEC (Fig. S2). It is worth noting that one of these cases (case 4) occurred in a pediatric patient that had a cecal NET G3 with the lowest burden of genomic alterations and several TILs on conventional morphology (Fig. 3F).

Finally, no gene fusion was observed in the cohort studied.