Spatial profiling reveals tissue-specific neuro-immune interactions in gastroenteropancreatic neuroendocrine tumors

Abstract

Gastroenteropancreatic neuroendocrine tumors (GEP-NETs) are heterogeneous malignancies that arise from complex cellular interactions within the tissue microenvironment. Here, we sought to decipher tumor-derived signals from the surrounding microenvironment by applying Digital Spatial Profiling (DSP) to hormone-secreting and non-functional GEP-NETs. By combining this approach with in vitro studies in human-derived organoids, we demonstrate the convergence of cell autonomous immune and pro-inflammatory proteins that suggests their role in neuroendocrine differentiation and tumorigenesis. DSP was used to evaluate the expression of 40 neural and immune-related proteins in surgically resected duodenal and pancreatic NETs (n=20) primarily comprised of gastrinomas (18/20). A total of 279 regions of interest were examined between tumors, adjacent normal and abnormal-appearing epithelium, and the surrounding stroma. The results were stratified by tissue type and Multiple Endocrine Neoplasia I (MEN1) status and protein expression was validated by immunohistochemical (IHC) staining. A tumor immune cell autonomous inflammatory signature was further evaluated by IHC and RNAscope, while functional pro-inflammatory signaling was confirmed using patient-derived duodenal organoids. Gastrin-secreting and non-functional pancreatic NETs showed a higher abundance of immune cell markers and immune infiltrate compared to duodenal gastrinomas. Compared to non-MEN1 tumors, MEN1 gastrinomas and preneoplastic lesions showed strong immune exclusion and upregulated expression of neuropathological proteins. Despite a paucity of immune cells, duodenal gastrinomas expressed the pro-inflammatory and pro-neural factor IL-17B. Treatment of human duodenal organoids with IL-17B activated NF-kB and STAT3 signaling and induced the expression of neuroendocrine markers. In conclusion, multiplexed spatial protein analysis identified tissue-specific neuro-immune signatures in GEP-NETs. Duodenal gastrinomas are characterized by an immunologically cold microenvironment that permits cellular reprogramming and neoplastic transformation of the preneoplastic epithelium. Moreover, duodenal gastrinomas cell autonomously express immune and pro-inflammatory factors, including tumor-derived IL-17B, that stimulate the neuroendocrine phenotype.

Potential mechanism of immune-autonomous IL-17B signaling in DGASTs.

Under normal physiological conditions, the submucosal Brunner’s glands are comprised primarily of morphologically similar epithelial glands interspersed with immune cells. Enteroendocrine cells marked by chromogranin A (CHGA) and synaptophysin (SYP) expression are relatively rare. Menin instability resulting from germline or somatic mutations may potentially derepress the expression of pro-inflammatory cytokines in stromal cells and the glandular epithelium. Upregulated IL-17B, a known pro-neuronal factor, acts on normal glandular tissue to induce neuroendocrine differentiation. The expression of IL-17B, TNF⍺, and IL-6 in DGASTs may be associated with the senescence associated secretory phenotype (SASP) that contributes to the classical features underlying well-differentiated G1/2 NETs, including slow proliferation, upregulated expression of CDKN2A/p16^INK4, strong stromal reactivity leading to desmoplasia and finally, an immune excluded tumor microenvironment permissive to disease progression. Figure rendered in Biorender.

Introduction

GEP-NETs comprise a complex group of cancers that display diverse molecular features, mixed differentiation status, and broad proliferative potential [1]. Gastrinomas are functional GEP-NETs that secrete gastrin peptide and most commonly arise in the pancreas, proximal duodenum, and gastric antrum comprising the “gastrinoma triangle” [2]. Among these, duodenal gastrinomas (DGASTs) associated with the Multiple Endocrine Neoplasia I (MEN1) Syndrome are the most malignant, presenting as small (70% <1.5cm), multicentric (90%) tumors that are more likely to metastasize, with up to 60% exhibiting lymph node metastases upon diagnosis [3]. Despite their association with clinically aggressive metrics, the signals that direct the emergence of MEN1 gastrinomas remain enigmatic.

Prior efforts to stratify GEP-NETs for treatment have largely centered on bulk sequencing of surgically resected tissues to identify genetic and epigenetic alterations [4–10]. Our previous transcriptomic analysis of a small cohort of human DGASTs identified expression of pro-inflammatory cytokines in the tumor and preneoplastic Brunner’s glands, indicating a role in neoplastic transformation [11]. These studies showed that activation of inflammatory signaling was immune cell-independent, as expression of these cytokines and downstream signaling was restricted to endocrine cells and the tumor stroma rather than the tumor immune compartment [11]. Consistent with these observations, prior immuno-typing of GEP-NETs showed that a majority of grade 1 and 2 (G1/2) tumors exhibit an immune excluded microenvironment compared to G3 NETs and neuroendocrine carcinomas (NECs) [8,12–15]. These studies suggest that GEP-NETs may be clinically stratified by their microenvironments, including the enrichment of immune infiltrates and cancer-associated fibroblasts [16]. Thus, further subtyping of GEP-NETs by immune and stromal status may elucidate approaches to therapeutically target the tumor microenvironment and unlock the potential for personalized medicine.

Until recently, the absence of reliable methods to unmix tumor- and stroma-derived signals has precluded a comprehensive understanding of how GEP-NETs arise in different tissues. Advances in spatial profiling technologies have addressed these limitations by illuminating cellular interactions in the tumor microenvironment with astonishing detail [17–20]. Here, we applied spatial profiling to clinical GEP-NETs comprised primarily of gastrinomas and elucidated a unique cell autonomous immune and pro-inflammatory signature associated with tumorigenesis.

Materials and Methods

Human Tissue Collection and Ethical Compliance

De-identified formalin-fixed paraffin-embedded (FFPE) sections of surgically resected GEP-NETs were collected under Institutional Review Board (IRB) approval #HUM00115310 from the Department of Pathology at the University of Michigan. Fresh normal-appearing duodenal mucosa was collected during Whipple procedures through the Tissue Acquisition and Mouse Resource Core (TACMASR) at the University of Arizona Comprehensive Cancer Center. Informed patient consent by the institutional parties was obtained prior to any tissue collection. All procedures involving human subjects conformed to the guidelines outlined by the IRB and Helsinki Declaration.

Nanostring GeoMx Digital Spatial Profiling

Five-micron FFPE tissue sections were baked at 65°C for 60 min prior to deparaffinization in xylene and rehydration in 100%, 90%, and 70% ethanol. Slides were washed in Tris Buffered Saline (TBS), then subjected to heat-mediated antigen retrieval using sodium citrate buffer (pH 6.0). Slides were incubated in a pressure cooker for 15 min then subjected to blocking and immunostaining steps as described in the Supplementary Methods. Expression analysis of 40 neural- and immune-related proteins was performed using the Nanostring Neural Cell Profiling Core supplemented with the Alzheimer’s Disease Module and Parkinson’s Disease Module (Nanostring Technologies, Seattle, WA). The Nanostring GeoMx Digital Spatial Profiler instrument was used to acquire a high resolution wide-field scan of each tissue section from which ROIs were selected for UV-activated oligo collection. Collected oligos were hybridized to unique probes and pooled for analysis using the Nanostring nCounter instrument, as described by the manufacturer.

The first DSP study was performed on a discovery cohort consisting of 4 surgically resected duodenal gastrinomas (DGAST), 3 pancreatic gastrinomas (PGAST), 2 non-functional pancreatic NETs (NF-PNET). Duodenal mucosa and pancreas from non-tumor bearing subjects were also included. Of these, the four duodenal gastrinoma-bearing patients were initially described in a previous study [11]. Eight ROIs were selected per slide for a total of 96 ROIs. For this study, ROIs were drawn across areas ranging 200 by 200 microns to 400 by 400 microns to yield similar nuclei counts across the different morphological features and tumors types. Selection area was normalized to housekeeping protein and background controls as described below. The second DSP study was performed on 12 different surgically resected duodenal gastrinomas. As the tumor specimens were more homogenous, all ROIs for this second study were selected at 200 by 200 microns. Of the 216 ROIs selected, 183 passed quality control parameters and were included for analysis. These included 47 normal BGs, 76 abnormal BGs, and 60 Tumor ROIs across the 12 patients. Quality control, normalization, and background subtraction steps were performed according to published Nanostring recommendations. Background subtraction was performed using the average of two IgG Isotype controls and protein expression values were normalized to the geomean of three housekeeping proteins.

DSP Data Visualization

Data analysis and visualization including heatmaps and PCA were completed with Python 3 using the “seaborn” and “pandas” analysis packages, as described in detail in the Supplementary Methods. Heatmaps were generated by plotting the z-scores for each marker for each of the tissue groups using the seaborn’s built-in heatmap function. Principal component analysis (PCA) was conducted on each dataset after z-score transformation, with no dimension reduction applied. PCA plots were generated by plotting the first and second principal components for each ROI on two orthogonal axes, and color coding according to the tissue type. PCA biplots were generated using a custom function to plot the projection of specific markers into PCA space by extracting the first and second loading coefficients. These coefficients were used to plot lines superimposed on the PCA plots, which point in the direction of variance corresponding to each marker.

Immunohistochemical and Immunofluorescence Staining

Five-micron FFPE sections were baked at 65°C for 60 min and then deparaffinized in xylene three times for 5 min. Tissues were rehydrated in a series of 100%, 90%, and 70% ethanol washes, and then rinsed with Tris Buffered Saline (TBS) for 5 min. Heat-mediated antigen retrieval was performed by heating slides in Tris-EDTA buffer (pH 9.0) (Sakura Finetek, Torrance, CA) for 30 min at 95°C. Slides were allowed to cool to 24°C for 15 min before washing and blocking with 5% bovine serum albumin (BSA) in TBS with 0.05% Tween-20 (TBST) for 1 h at 24°C. Next, slides were incubated in primary antibodies overnight (16 h) at 4°C. Slides were washed in TBST, and then incubated in anti-Mouse IgG or anti-Rabbit IgG secondary antibodies conjugated to Horse Radish Peroxidase (ImmPACT DAB, Vector Laboratories, Newark, CA) for 30 min at 24°C. Slides were washed, and then incubated in DAB substrate (ImmPACT DAB kit, Vector Laboratories) for 30–60 sec. Slides were immediately washed in running distilled water prior to counterstaining cell nuclei with diluted Gill’s Hematoxylin (Leica Biosystems, Wetzlar, Germany) and Bluing Reagent (Leica Biosystems). Slides were dehydrated in a series of 70%, 90%, and 100% ethanol, and then washed in xylene three times prior to mounting with Paramount mounting medium.

For immunofluorescence staining of FFPE slides, sections were processed as described previously and blocked in 10% donkey serum in 1% BSA TBST for 1 h at 24°C. Sections were then incubated in primary antibodies overnight at 4°C (Supplementary Methods). Slides were washed in TBST and then incubated in Alexa Fluor-conjugated secondary antibodies diluted 1:500 in TBST with 1% BSA (Thermo Fisher Scientific, Waltham, MA). Slides were mounted using Prolong Gold anti-fade mounting medium with 4’, 6-diamidino-2-phenylindole (DAPI) (Thermo Fisher Scientific). IHC and IF stained slides were imaged using the Olympus BX53F epifluorescence microscope (Center Valley, PA).

For immunostaining of duodenal organoids, the treated organoids were washed twice with pre-warmed Dulbecco’s PBS (DPBS). Organoids were fixed for 20 min at 24°C in 4% PFA that was pre-warmed to 37°C. Organoids were permeabilized in 0.5% Triton X-100 for 15 min at 24°C. Organoids were washed twice with DPBS and then blocked in 10% donkey serum and 1% BSA for 30 min at 24°C. Organoids were incubated in primary antibodies for 3 h at 24°C (Supplementary Methods).

Organoid Culture

Human duodenal organoid lines were generated from normal-appearing duodenal biopsies from two patients undergoing Whipple procedure [11]. Organoids lines were subcultured in complete organoid media as described previously (Supplementary Table 1). Media was supplemented with 10 μM SB202190, a selection p38 MAP kinase inhibitor and 10 nM nicotinamide (Tocris Bioscience, Bristol, UK) for the first day of passaging to prevent anoikis. Organoids were used for studies between passage numbers 5 and 10. Within 5–7 days post-seeding, duodenal organoids were treated with 100 ng/ml of recombinant human IL-17B (R&D Systems, Minneapolis, MN) for 48 h and processed for downstream RNA expression analysis. For whole cell protein analysis, organoids were treated with 50 ng/ml of IL-17B for 24 h, and a time-course treatment was conducted over 0, 0.5, 1, 2, 3, and 4 h to examine nuclear and cytoplasmic protein shuttling.

Cellular Fractionation and Western Blot

Following 4 h of IL-17B treatment, duodenal organoids were scraped from wells and collected in ice-cold PBS. Organoids were dissociated from the Matrigel (Corning, Corning, NY) by gently pipetting ten times and centrifuging for 5 min at 300 x g at 4°C. The organoid pellet was resuspended in 1 ml ice-cold PBS, and then centrifuged for 5 min at 300 x g at 4°C. To generate whole cell extracts, the pellet was resuspended in 200 μL of ice-cold RIPA buffer (Thermo Fisher Scientific) supplemented with 1X HALT protease and phosphatase inhibitor (Thermo Fisher Scientific). Organoids were lysed by passing through a 20G syringe ten times and vortexing. Following 40 min incubation on ice, organoid lysates were centrifuged for 15 min at 15,000 x g at 4°C and the supernatant was collected as the protein extract. Subcellular fractionation was performed using hypotonic and hypertonic lysis buffers to extract cytoplasmic and nuclear proteins, respectively. Detailed methods for fractionation and gel electrophoresis are described in the Supplementary Materials.

Quantitative Polymerase Chain Reaction (RT-qPCR)

RNA was extracted from duodenal organoids using the ReliaPrep miRNA Cell and Tissue Miniprep System following the manufacturer’s instructions (Promega, Madison, Wisconsin). Up to 1 microgram of cDNA was synthesized using SuperScript VILO after treatment with DNAse I to remove genomic DNA. Quantitative real-time PCR (RT-qPCR) was performed using 10 ng of cDNA under the following amplification conditions: an initial 2 min denaturation at 95°C, followed by 40 cycles of 15 s at 95°C and 1 min at 60°C. Target mRNA expression was normalized to Hypoxanthine phosphoribosyltransferase 1 (HPRT) as an endogenous control. Relative fold change was calculated using the 2^{−double deltaCt} method [22].

RNAScope

Fluorescent In Situ Hybridization was performed using the RNAscope Multiplex Fluorescence V2 Assay with pre-made RNAscope target probes C1–3 for human IL17B, TNF, and IL6 (Cat # 323100, Advanced Cell Diagnostics, Newark, CA). FFPE sections of human gastrinomas were baked and deparaffinized as previously described. RNAscope Multiplex fluorescence assay was performed according to the manufacturer’s kit instructions. Targets were visualized by developing the HRP signal with Akoya Opal 520, 570, 690 fluorophores according to the protocol using 1:750 dilution of each fluorophore (Cat #FP1488001, #FP1497001, #FP1501001, Akoya Biosciences, Marlborough, MA). Slides were then mounted using Prolong Gold anti-Fade mounting media with DAPI and imaged at 400X magnification using the Olympus BX53F epifluorescence microscope. Studies were independently repeated to confirm appropriate signal.

Statistical Analysis

Linear Mixed Modeling and Benjamini-Hochberg multiple test correction was applied to DSP analyses to test for statistically significant differences. For samples with unequal variance, the nonparametric Mann-Whitney test was applied. For cases with equal variance, unpaired Student’s T-test with Tukey post-test was applied for comparisons involving two groups, while one-way ANOVA with Sidak post-test was used for comparisons involved three or more groups.

Results

Gastrinomas and non-functional NETs exhibit distinct neuro-immune protein signatures.

Human duodenal gastrinomas express markers of neuroglial cells [23] and neuroglial reprogramming was recently implicated in the development of pancreatic and pituitary NETs [24]. To decipher whether GEP-NETs exhibit neuropathological features, we applied Nanostring Digital Spatial Profiling (DSP) technology to formalin-fixed paraffin-embedded (FFPE) tissue sections of three pancreatic gastrinomas (PGAST), two non-functioning pancreatic NETs (NF-PNET), and four duodenal gastrinomas (DGAST) (Table 1). We quantified the expression of a 40-plex human neuro-immune protein panel enriched with markers that are classically associated with Alzheimer’s and Parkinson’s diseases (Table 2). Consistent with previous reports [11], hematoxylin and eosin (H&E) stained images of DGASTs, PGASTs, and NF-PNETs showed the strongest stromal reaction in DGASTs (Figure 1A–C). Using the Nanostring tumor morphology markers, pan-cytokeratin (PanCK) and CD45 to epithelium and immune cells, respectively, we selectively analyzed the expression of the neuro-immune protein panel in eighty regions of interest (ROIs) spanning tumor lesions, stroma, and adjacent normal epithelium including the Brunner’s glands (Figure 1D–F).

Table 1.

Duodenal and pancreatic NETs included in the discovery cohort for digital spatial profiling.

ID	Clinical diagnosis	Age	MEN1 syndrome	MEN1 mutation	ZES	Grade	Ki-67	Mets	DSP
M2	Duodenal gastrinoma	57	-	c.1621G>A	-	1	2%	+	+
M4	Duodenal gastrinoma	52	+	c.784–9G>A	+				+
M5	Duodenal gastrinoma		UNK		UNK				+
M6, M6A	Duodenal gastrinoma		UNK		UNK				+
M8	Non-functional PNET	67	-	c.703G>A	-	2	7%	+	+
M9	Non-functional PNET	74	-		-	2	9%	+	+
M13	Pancreatic gastrinoma		UNK		UNK				+
M14	Pancreatic gastrinoma		UNK		UNK				+
M15	Pancreatic gastrinoma		UNK		UNK				+

Table 2.

Protein targets and antibody controls used for digital spatial profiling of human duodenal and pancreatic NETs.

Protein Target	Enrichment (Cell Type/Biological Process)	Nanostring Module
NeuN	Neuronal Differentiation	Neural Cell Profiling
Synaptophysin		Neural Cell Profiling
UCHL1		PD Pathology
Microtubule associated protein 2 (MAP2)		Neural Cell Profiling
Calbindin		PD Pathology
Alpha-synuclein (⍺-Syn)	Neuron and Endocrine	PD Pathology
Tyrosine Hydroxylase		PD Pathology
Neurofilament light	Neuron	Neural Cell Profiling
OLIG2	Oligodendrocyte	Neural Cell Profiling
Myelin basic protein	Oligodendrocyte/Schwann	Neural Cell Profiling
GFAP	Astrocyte/Inflammation	Neural Cell Profiling
S100B		Neural Cell Profiling
CD45	Microglia/Inflammation	Neural Cell Profiling
CD40		Neural Cell Profiling
CD39		Neural Cell Profiling
P2X7		AD Pathology
CD68	Microglia/Immune	Neural Cell Profiling
CD163		Neural Cell Profiling
CD11b		Neural Cell Profiling
Transmembrane protein 119 (TMEM119)		Neural Cell Profiling
AIF-1		Neural Cell Profiling
HLA-DR		Neural Cell Profiling
P2Y12		Neural Cell Profiling
Apo-E	Immune/Metabolism	AD Pathology
Apo-A1		PD Pathology
CD31	Endothelium	Neural Cell Profiling
Ki-67	Proliferation	Neural Cell Profiling
Tau	Ubiquitous	AD Pathology
Phospho-Tau (S404)		AD Pathology
TDP-43		AD Pathology
Phospho-TDP-43 (S409/S410)		AD Pathology
Amyloid-beta 1–42		AD Pathology
Ubiquitin		AD Pathology
Amyloid-beta precursor protein (APP)		AD Pathology
Amyloid-beta 1–40		AD Pathology
Park7		PD Pathology
Leucine rich repeat kinase 2		PD Pathology
FUS RNA-binding protein		PD Pathology
PTEN induced kinase 1 (PINK1)		PD Pathology
Phospho-Alpha-synuclein (S129)		PD Pathology
Histone H3	Ubiquitous/Housekeeping	Controls
GAPDH		Controls
S6		Controls
Ms IgG2a	Technical control	Controls
Ms IgG1		Controls
Rb IgG		Controls

Figure 1. Digital spatial profiling of duodenal and pancreatic neuroendocrine tumors.

(A) Hematoxylin and eosin-stained images of a duodenal gastrinoma (DGAST), (B) pancreatic gastrinoma (PGAST), and (C) non-functional pancreatic NET (NF-PNET). (D) Serial formalin-fixed paraffin-embedded sections of a DGAST stained with the GeoMx tumor morphology markers Pan cytokeratin (PanCK, green), CD45 (magenta), and SYTO13 (blue). Inset shows gastrin staining in red. White boxes indicate regions of interest (ROIs) from which oligo-conjugated antibodies were collected for multi-plexed protein analysis. Right panel shows higher magnification images of these boxed ROIs, including the transitioning Brunner’s glands (tBGs), stroma, and tumor. (E) A PGAST and (F) NF-PNET stained with the GeoMx tumor morphology markers, with right panels showing representative ROIs of adjacent tumor stroma and tumor regions.

Principal component analysis (PCA) was performed on tumor ROIs across DGASTs, PGASTs, and NF-PNETs to identify whether tumors exhibit distinct clustering based on their neuro-immune profiles (Figure 2A). Indeed, clustering of the three tumor types was primarily driven by immune marker enrichment and the expression of Ki-67 and neuroglial proteins (Figure 2A). Protein expression analysis of the tumor ROIs showed remarkably increased expression of immune markers, including CD45 and the leukocyte-specific receptor CD11b, in PGASTs and NF-PNETs compared to the DGASTs (Figure 2B and 2C). Further, the pancreatic tumors exhibited distinct immune expression profiles, with PGASTs showing elevated expression of CD163 and CD39, marking myeloid cells and Foxp3⁺ regulatory T cells, respectively [25,26]. In comparison, NF-PNETs showed elevated CD68 and CD40 expression, expressed by macrophages, dendritic cells, and B cells [26,27]. Consistent with the clinical pathology, NF-PNETs showed higher expression of Ki-67 compared to the other tumors (Table 1 and Figure 2C). Further immunohistochemical staining of the tumors confirmed these expression patterns while also revealing inter-tumor variations, with some NF-PNETs showing higher CD45 expression compared to gastrinomas (Figure 2D). Therefore, the application of DSP to this cohort of GEP-NETs unveiled distinct immune profiles across tumor types and suggests fundamental differences in the neuro-immune cell composition of the tumor microenvironment.

Figure 2. Gastrinomas and non-functional NETs exhibit distinct neuro-immune protein signatures.

(A) Principal Component Analysis (PCA) plot of neuro-immune protein expression in tumor ROIs of DGASTs (n=3 patients, n=30 ROIs), PGASTs (n=3 patients, n=23 ROIs), and NF-PNETs (n=2 tumors, n=12 ROIs). Each dot represents a unique tumor ROI while symbols indicate ROIs from a single patient. Symbol IDs match the IDs listed in Table 1. Superimposed lines point in the direction of variance corresponding to each protein marker and indicate the degree to which that protein defines the clustering on the x- and y-axes. A line pointing in any respective direction indicates that the protein is enriched in the population upon which it is superimposed, and more broadly, the axis upon which ROIs cluster. (B) Heatmap showing relative expression (Z-score) of the 40 proteins analyzed in the DSP neuro-immune profiling panel listed in order of their biological enrichment shown in Table 2. ROIs 0–29 indicate DGAST tumors, ROIs 30–52 indicate PGAST tumors, and ROIs 53–65 indicate NF-PNET tumors. (C) Tumor expression of select immune-related protein markers, in addition to the proliferative marker Ki-67, the endothelial marker CD31, and Ubiquitin. (D) Representative immunohistochemical (IHC) stained images showing the expression of CD45, CD39, Ki-67, and synaptophysin (SYP) in a DGAST, PGAST, and NF-PNET.

GEP-NETs show strong intra- and inter-tumoral variation in neuro-immune protein expression.

We next investigated intra-tumoral variation by evaluating neuro-immune protein expression across normal adjacent tissue, tumor stroma, and tumor lesions. We included a normal pancreas specimen (nPANC) and normal duodenum (nDUO) in our analysis for further comparison. Intra-tissue variation was driven by upregulated immune marker expression in nPANC and normal adjacent tissues, whereas neural markers were enriched in the tumor stroma and tumor lesions (Figure 3A and Supplementary Figure 1). PGASTs and NF-PNETs strongly expressed the neuroendocrine marker SYP as expected, however the tumors were marked by distinct neuroglial signatures (Figure 3B).

Figure 3. GEP-NETs show strong intra- and inter-tumoral variation in neuro-immune protein expression.

(A) PCA plot showing the variation in DSP protein expression across pancreas tissues. ROIs were taken from a normal pancreas specimen for comparison (nPANC, n=9 ROIs from one patient). The three patient PGASTs were further divided into PGAST stroma (S, n=6 ROIs), PGAST normal adjacent tissue (N, n=3 ROIs), and PGAST tumor (T, n=23 ROIs). The two patient NF-PNETs were divided into NF-PNET stroma (n=4), NF-PNET normal adjacent tissue (n=6), and NF-PNET tumor (n=12). Symbols indicate ROIs from a unique patient matching the IDs listed in Table 1. (B) Normalized protein expression of select markers across stroma, normal adjacent, and tumor ROIs of both PGASTs and NF-PNETs, respectively. *=p < 0.05, **=p < 0.01, ****=p < 0.0001 by one-way ANOVA with Tukey post-test. (C) PCA plot showing intra-tissue variation across the four patient DGASTs and a normal duodenum specimen (nDUO EPI, n=6 ROIs). Symbols indicate ROIs from a unique patient matching the IDs listed in Table 1. DGASTs were subdivided into adjacent epithelium (EP/EPI, n=16 ROIs), stroma (S, n=6 ROIs), normal-appearing Brunner’s glands (BG, n=8 ROIs), and tumor (T, n=30 ROIs). (D) Normalized protein expression of select markers across tissue categories. (E) Representative images of immunohistochemical staining for SYP and Ki-67 showing appropriate expression across the tissue categories. (F) Normalized protein expression of immune-related proteins across tissue categories in the four patient DGASTs. *=p < 0.05, **=p < 0.01, ***=p < 0.001, ****=p < 0.0001 by one-way ANOVA with Tukey post-test. (G) Representative images of CD45 expression in the adjacent epithelium, normal-appearing BGs, stroma, and DGAST tumor lesion.

We next examined the intra-tissue variation of these proteins in our cohort of DGASTs by comparing expression levels across the nDUO specimen, adjacent epithelium, tumor stroma, adjacent Brunner’s glands, and tumor lesions (Figure 3C and Supplementary Figure 1). DGAST tumor ROIs showed significant enrichment in SYP but showed reduced expression of Ki-67 and S100B compared to adjacent tissues (Figure 3D). IHC staining of DGASTs and adjacent tissues confirmed these expression patterns (Figure 3E). Most strikingly, DGAST tumor lesions showed strong exclusion of immune markers compared to the adjacent stroma and Brunner’s glands (Figure 3F and 3G). Taken together, GEP-NETs exhibit restricted immune profiles and upregulated SYP protein expression compared to the surrounding tumor stroma and adjacent epithelium. Furthermore, DGASTs and their associated Brunner’s glands show the highest extent of immune cell exclusion, consistent with a microenvironment that permits the reprogramming of these glands into preneoplastic lesions.

Reduced immune infiltration and increased neural protein abundance in DGASTs is consistent in an expanded patient cohort.

We next sought to extend these analyses to a larger cohort of GEP-NET-bearing patients with additional clinical annotation, including a diagnosis of MEN1 and Zollinger-Ellison Syndromes (ZES) (Table 3). We focused on DGASTs since these tumors are known to present with precursor lesions in the Brunner’s glands [28], thus marking them as an ideal model to study neoplastic development and tumor evolution. To investigate whether DGASTs upregulate their expression of neuroglial proteins during neoplastic development, we performed DSP on FFPE tumor sections from 12 patients, of which 11 were unique from the previous discovery cohort. We performed ROI selection on adjacent normal-appearing Brunner’s glands (nBG) and tumor lesions, in addition to transitioning Brunner’s glands (tBG) characterized by abnormal morphology, stromal thickening, and enrichment in PanCK expression. PCA and unbiased hierarchical clustering of 183 ROIs showed a progressive phenotype with tBGs clustering between nBGs and tumor ROIs (Figure 4A and Supplementary Figure 2). As anticipated, SYP expression and enrichment in neural protein abundance strongly defined the tumor cluster, while the nBG cluster was defined by elevated expression of immune-related markers (Figure 4A). Analysis of the three tissue classes showed a step wise increase in neural protein expression between tBGs and tumor lesions, and reduced expression of immune proteins, including markers of macrophage-like microglial cells [29] (Figure 4B–D). IHC staining of the DGASTs confirmed increased expression of neuroendocrine and neural-related proteins SYP and alpha-synuclein (⍺-Syn) in transitioning BGs and tumor lesions, coincident with immune cell exclusion (Figure 4E). In summary, DGASTs and preneoplastic lesions display alterations in their neuro-immune protein profiles, suggesting that the immunologically cold microenvironment favors epithelial reprogramming and neoplastic transformation.

Table 3.

Clinical demographics for duodenal gastrinomas and NETs included in the expanded digital spatial profiling analysis and validation studies.

ID	Clinical diagnosis	Age	Age at last f/u or death (years)	Survival after surgery (years)	Deceased (0) / Alive (1)	Without disease (0) / With (1)	MEN1 syndrome	MEN1 mutation	ZES	DSP
UOM-A001	Duodenal gastrinoma	33	49	16.4	1	1	+	not tested	+	+
UOM-A002	Duodenal gastrinoma	48	62	14.4	1	1	+	entire deletion	+	+
UOM-A003	Duodenal gastrinoma	47	62	16.4	1	0	+	c.1049+2delT	+	+
UOM-A004	Duodenal NET	75	76	1	0	0	+	c.512del	-	-
UOM-A005	Duodenal gastrinoma	52	68	16.6	1	1	+	c.784–9G>A	+	+
UOM-A006	Duodenal gastrinoma	35	47	12.6	1	1	+	c.784–9G>A	-	+
UOM-A007	Duodenal gastrinoma	50	44	5.4	1	1	+	not tested	+	-
UOM-A008	Duodenal gastrinoma	33	42	9.6	1	1	+	C.654+1, G>T	+	+
UOM-A009	Duodenal gastrinoma	51	64	13.7	1	0	-		+	+
UOM-A010	Duodenal gastrinoma	56	64	8.8	1	0	-		-	+
UOM-A011	Duodenal NET	31	39	9	1	0	-		-	-
UOM-A012	Duodenal NET	50	68	18.6	1	1	-		-	-
UOM-A013	Duodenal gastrinoma	62	63	1.7	0	1	-		+	-
UOM-A014	Duodenal NET	44	69	25.2	1	0	+	c.1548dupG	-
UOM-A015	Duodenal gastrinoma	38	44	6.7	0	0	+	UNK	+	+
UOM-A016	Duodenal gastrinoma	66	74	8.9	1	0	-		+	-
UOM-A017	Duodenal gastrinoma	66	69	3.6	0	1	-		+	+
UOM-A018	Duodenal NET	41	45	4.4	1	0	-		+	-
UOM-A019	Duodenal NET	64	68	3.9	0	0	-		-	-
UOM-A020	Duodenal NET	63	67	4.6	1	0	-		+	-
UOM-A021	Duodenal gastrinoma	75	82	7.8	1	0	-		+	+
UOM-A022	Duodenal gastrinoma	56	66	10.4	1	0	-		+	+
UOM-A023	Duodenal NET	63	72	9.1	1	0	-		-	-
UOM-A024	Duodenal NET	60	61	1.4	1	1	-		-	-
UOM-A025	Duodenal NET	58	68	10.1	1	1	-		-	-
UOM-A026	Duodenal NET	33	39	6.3	1	0	-		-	-
UOM-A027	Duodenal NET	65	66	1.6	1	0	-		-	-
UOM-A028	Duodenal NET	83	93	10.8	0	1	-		-	-
UOM-A029	Duodenal NET	58	65	7.5	0	1	-		+	-
UOM-A030	Duodenal NET	73	79	6.5	1	0	-		+	-

Figure 4. Reduced immune infiltration and increased neural protein abundance in DGASTs is consistent in an expanded patient cohort.

(A) PCA plot showing the variation in DSP protein abundance across different tissue categories in DGASTs from 12 tumor-bearing patients. Regions analyzed include normal-appearing adjacent Brunner’s glands (nBG, n=47 ROIs), transitioning Brunner’s glands (tBG, n=76 ROIs), and tumor (n=60 ROIs). (B) Quantitation of select neuroendocrine and neuroglial proteins. (C) Quantitation of immune cell markers and (D) microglial-related proteins across the tissue types. *=p < 0.05, **=p < 0.01, ***=p < 0.001, ****=p < 0.0001 by one-way ANOVA with Tukey post-test. (E) Representative IHC stained imaged of nBGs, tBGs, and DGASTs showing expression of the neuroendocrine proteins SYP and alpha synuclein (⍺-Syn), and the immune marker CD45.

MEN1 DGASTs show reduced immune infiltration and increased neuroglial protein expression compared to non-MEN1 tumors.

Menin, the protein product of the MEN1 gene, is known to regulate cell fate specification leading to NET development [23,24]. Thus, we next investigated whether the neuro-immune profiles of DGASTs differed among patients with a MEN1 diagnosis. In our expanded cohort, 7/12 patients were clinically diagnosed with MEN1 syndrome (Table 3). Upon follow up, 6 surviving MEN1 cases showed evidence of disease (5/6 or 83%), whereas the 4 surviving non-MEN1 patients showed no evidence of disease recurrence (0/4 or 0%). We stratified the DSP results by tissue class and MEN1 syndrome and identified striking differences in the immune profiles of MEN1 DGASTs compared to non-MEN1 patients (Figure 5A and Supplementary Figure 2). Specifically, the tBGs of MEN1 patients showed significantly reduced expression of immune markers compared to the tBGs of non-MEN1 DGASTs, and this was confirmed upon IHC staining of the tumors (Figure 5B). Further, MEN1 tumors showed significantly higher enrichment in neuroglial proteins compared to their non-MEN1 counterparts (Figure 5C). Lastly, non-MEN1 tBGs and tumor lesions exhibited significantly higher expression of Ki-67 and the endothelial marker CD31 (Figure 5D). These expression patterns were confirmed by IHC staining of DGAST lesions from both patient cohorts (Figure 5E). Thus, compared to non-MEN1 patients, the preneoplastic epithelium of MEN1 DGASTs exhibit a severely restricted immune microenvironment and MEN1 tumors show increased neuroglial differentiation and reduced proliferation.

Figure 5. MEN1 DGASTs show reduced immune infiltration and increased neuroglial protein expression compared to non-MEN1 tumors.

(A) Normalized expression of immune proteins across tissue categories in the 12 patient DGASTs stratified by MEN1 diagnosis. Protein expression was compared between normal-appearing adjacent Brunner’s glands (nBG MEN1 n=27, non-MEN1 n=20 ROIs), transitioning Brunner’s glands (tBG MEN1 n=51, non-MEN1 n=26 ROIs), and tumor (MEN1 n=40, non-MEN1 n=19 ROIs). *=p < 0.05, **=p < 0.01, ****=p < 0.0001 by two-way ANOVA with Benjamini-Hochberg multiple test correction. (D) Representative IHC stained images of CD45 and CD39 expression in nBG, tBG, and tumor lesions from MEN1 and non-MEN1 DGAST-bearing patients. (C) Normalized expression of neuroendocrine and neuroglial proteins and (D) Ki-67 and CD31 in the 12 patient DGASTs stratified by MEN1 status. (E) Representative IHC stained images of MEN1 and non-MEN1 DGASTs for Ki-67 and the neuroglial-associated proteins UCHL1 and S100B. Arrows indicate nuclear Ki-67 expression.

Tumor and stroma-derived IL-17B stimulate neuroendocrine differentiation downstream of NF-kB and STAT3 activation.

Given that DGASTs express pro-inflammatory cytokines [11] and exhibit features of senescence, we sought to define whether DGASTs exhibit features of the senescence-associated secretory phenotype (SASP). We first confirmed that DGASTs and preneoplastic lesions in the expanded patient cohort express pro-inflammatory cytokines, including IL-17B, TNF⍺, and IL-6 (Figure 6A and B). Additionally, DGASTs were shown to express phospho-STAT3 (Tyr705) known to be activated downstream of these cytokines (Figure 6B). Consistent with SASP and low Ki-67 expression, DGASTs showed robust expression of the cyclin dependent kinase inhibitors and senescence markers p16^INK4 (CDKN2A) and p21^Cip1 (CDKN1A) (Figure 6C). We next evaluated the functional role of cytokine signaling in NET development using two primary human organoid lines that were generated from normal-appearing adjacent duodenum obtained during Whipple procedure. We focused our investigation on IL-17B as others have shown a role for IL-17 in altering cell fate specification and pro-neuronal differentiation [30–32]. Moreover, of the six patient DGASTs examined by staining, four showed strong and diffuse IL-17B expression in SYP⁺ tumor cells, whereas the remaining 2/6 DGASTs and 3/3 PGASTs only showed sparse expression in the tumor stroma. Treatment of two human duodenal organoid lines with IL-17B (50 ng/ml) stimulated the expression of the neuroendocrine markers chromogranin A (CHGA), SYP, gastrin, and Nkx2.5. IL-17B treatment also increased the expression of the neural markers UCHL1 and ⍺-Syn, and the senescence markers p16^INK4 and p21^Cip1 (Figure 6D). Similarly, IL-17B increased the expression of neuroendocrine transcripts and CDKN2A mRNA levels in the two organoid lines (Figure 6E and F). We further showed the IL-17B activates the nuclear translocation of NF-kB/p65 and STAT3 proteins, suggesting that IL-17B acts through these pathways to alter the endocrine cell fate (Figure 6G and H).

Figure 6. Tumor and stroma-derived IL-17B stimulate neuroendocrine differentiation downstream of NF-kB and STAT3 activation.

(A) RNAscope of DGASTs and tBGs showing expression of TNF-⍺, IL-17B, and IL-6 in tumor lesions. (B) Representative immunofluorescence images of human DGAST tumors stained for IL-17B (white), phosho-STAT3 (Tyr705) (green), and SYP (red) protein expression. Four out of 6 patients showed strong, diffuse expression in the tumor. (C) Representative immunofluorescence images of DGASTs stained for the senescence markers p16^INK4 and p21 (green) merged with SYP (red). (D) Immunofluorescence staining for SYP, CHGA, gastrin, Nkx2.5, UCHL1, ⍺-Syn, and the senescence markers p16 and p21 in two human duodenal organoid lines treated with 50 ng/ml IL-17B for 48 h. DAPI shown in blue. (E) Relative mRNA expression of neuroendocrine markers and CDKN2A in a human duodenal organoid line treated with IL-17B for 48 h. (F) Expression of neuroendocrine transcripts and CDKN2A in a second human duodenal organoid line treated with IL-17B. n=5 experimental replicates for each line. *=p < 0.05, **=p < 0.01, ***=p < 0.001 by two-way ANOVA with Fisher’s LSD test. (G) Expression of NF-kB and STAT3 proteins in nuclear and cytoplasmic lysates of human duodenal organoids treated with IL-17B for 0–4 h. Histone H3 and β-tubulin shown as nuclear and cytoplasmic markers, respectively. (H) Expression of NF-kB and STAT3 in whole cell lysates of human duodenal organoids treated with increasing concentrations of IL-17B for 24 h.

Taken together, we show that DGASTs exhibit features of SASP, including upregulated expression of tumor-derived pro-inflammatory cytokines and cyclin-dependent kinase inhibitors. Moreover, we show that IL-17B exerts a functional effect on stimulating the neuroendocrine phenotype, consistent with prior reports showing that IL-17 alters cell fate commitment and neuronal differentiation. Thus, we conclude that immune-autonomous IL-17B signaling contributes to the pathogenesis and unique molecular features of DGASTs.

Discussion

Past investigations into the molecular features of GEP-NETs are limited in their ability to spatially resolve the complex cellular interactions that shape the tumor microenvironment. To address this gap, we applied Nanostring Digital Spatial Profiling (DSP) technology to human GEP-NETs to profile the expression of a 40-plex human neuro-immune protein panel developed for neurological diseases, including Alzheimer’s and Parkinson’s diseases. Investigating the expression of these proteins is significant as cellular reprogramming of neuroglial cells is implicated in the development of GEP-NETs [23,24]. Thus, deciphering whether GEP-NETs exhibit shared neuropathological features may elucidate novel therapeutic approaches for targeting these malignancies.

Our study identified robust expression of neuroglial proteins in gastrinomas arising in the duodenum and pancreas. Distinct immune profiles differentiated these tumors, with DGASTs and preneoplastic lesions exhibiting a microenvironment that strongly excludes immune cells. The immune-excluded environment shared across our cohort of DGASTs is consistent with other studies indicating low or absent T cell marker expression in GEP-NETs, with the lowest expression reported in SI-NETs [12,15,33]. Collectively, these studies help explain why immune checkpoint inhibition has seen limited success in GEP-NETs when administered as a monotherapy compared to other cancers with elevated tumor immune activity [34–36]. However, these studies are more nuanced as patients with poorly differentiated G3 NETs and NECs experience improved outcomes compared to G1/2 NETs, and higher response rates strongly correlated with the degree of immune infiltration in tumors [14]. The high-grade non-functioning PNETs in our patient cohort exhibited the highest expression of immune markers, and this is consistent with other reports showing that G3 NETs and poorly differentiated NECs are associated with higher levels of tumor-infiltrating lymphocytes and lower recurrence-free survival [8,13–15]. Surprisingly, MEN1-associated DGASTs in our cohort showed the lowest expression of immune infiltrate in the adjacent transitioning Brunner’s glands. This suggests that immune exclusion in preneoplastic tissues and precursor lesions creates a tumor permissive microenvironment that contributes to neuroendocrine differentiation and tumor progression. The decision to pursue adjuvant immunotherapy for GEP-NETs should be guided by additional factors, including the presence of MEN1 mutations, tumor differentiation status, and the site of tumor development. Importantly, our findings should be extended to larger multi-center studies to determine whether correlations exist between the molecular markers explored here and overall or progression free survival and treatment response.

Emerging studies have unveiled a role for the MEN1 gene and its protein product menin, in regulating cytokine expression and driving cellular senescence [37–39]. Given that MEN1 DGASTs exhibit features of the senescence-associated secretory phenotype, including cell cycle arrest (e.g. low Ki-67, high CDKN2A/p16^INK4 expression), increased production of cytokines (e.g. IL-17B and TNF⍺), immune evasion, strong desmoplastic reaction, and neuronal-differentiation, future work to understand these mechanisms is strongly warranted. A specific role for IL-17B in mediating these events in DGASTs is emphasized by our observation that SYP⁺ tumor cells and preneoplastic lesions strongly express IL-17B compared to PGASTs that only showed stromal IL-17B expression. IL-17B may facilitate hallmarks unique to DGASTs, including stromal activation leading to immune exclusion [40] and the reprogramming of preneoplastic tissues into neoplastic lesions with neuroendocrine features. Thus, therapeutic approaches to exploit the unique stromal and immune microenvironments of these tumors may prove beneficial.

Abbreviations

⍺-Syn

Alpha synuclein

AD

Alzheimer’s Disease

CHGA

Chromogranin A

DGAST

Duodenal Gastrinoma

DSP

Digital Spatial Profiling

FFPE

Formalin-fixed Paraffin-embedded

GEP-NET

Gastroenteropancreatic Neuroendocrine Tumor

IHC

Immunohistochemical Staining

MAP2

Microtubule-associated Protein

MEN1

Multiple Endocrine Neoplasia 1

nBG

Normal-appearing Brunner’s Glands

nDUO

Normal Human Duodenum

NEC

Neuroendocrine Carcinoma

NF-PNET

Non-functional PNET

nPANC

Normal Human Pancreas

NT

No treatment

PanCK

Pan Cyotkeratin

PCA

Principal Component Analysis

PD

Parkinson’s Disease

PGAST

Pancreatic Gastrinoma

PINK1

PTEN-induced Kinase

PNET

Pancreatic Neuroendocrine Tumor

ROI

Region of Interest

SASP

Senescence-associated Secretory Phenotype

SI-NET

Small intestinal NET

SYP

Synaptophysin

tBG

Transitioning Brunner’s Glands

TME

Tumor Microenvironment

TMEM119

Transmembrane protein 119

Data Availability Statement:

Data will be made available upon reasonable request to the corresponding author.