Study Cohort
The study was approved by the University Medical Center (UMC) Utrecht Biobank Research Ethics Committee. Tissue-microarrays (TMAs) were constructed of primary sporadic insulinomas resected between 1991–2017 and 1997–2017 for the UMC Utrecht and Amsterdam UMC, respectively. Inherited cases (e.g., MEN1 syndrome) were excluded. Neuroendocrine tumor diagnosis was confirmed by an experienced gastrointestinal pathologist (LAAB). Three 0.6-mm cores per tumor were randomly taken from annotated tumor areas in formalin-fixed paraffin-embedded (FFPE) blocks. In case of multiple tumors, the largest tumor was used. Biopsies of insulinoma liver metastases were identified by a search in the UMC Utrecht pathology archive.
Information on age, sex, multifocality, surgery type, surgery date, tumor size, location, grade, resection margin, and lymph nodes was collected from pathology reports. If possible, macroscopic tumor size was used. Free margins were interpreted as R0, also if the distance was less than 1 mm from the resection margin. Medical records were reviewed to collect information on the functional status of the tumor, presence of genetic syndromes, and follow-up. Events of tumor relapse (local recurrence, liver metastases, or other metastases) were either histologically proven or diagnosed by the treating clinician. The first radiological evidence of proven relapse was used as event time point. Follow-up time is counted from date of surgery until described events, death, or was censored at the last visit to a relevant hospital clinician (surgery, internal medicine, endocrinology, gastroenterology, or oncology) or most recent clinic visit. For overall survival, any cause of death and the most recent clinic visit were used. Relapse was defined as any distant metastasis (liver or other location) or local recurrence. Relapse-free, distant-free, and liver metastases-free survival were censored at last visit to a relevant hospital clinician.
Immunohistochemistry
Four-micrometer sections of FFPE tissue were cleared at 60 °C and deparaffinized in xylene. Endogenous peroxidase was blocked by immersion in 0.6% H
2O
2 (7210, Merck, Kenilworth, USA) in methanol for 15 min. Antigen retrieval was performed by cooking slides in a 10 mM citrate (pH 6) or 10/1 mM Tris/EDTA (pH 9) solution for 20 min. Nonspecific binding was reduced by with Protein Block Serum Free (X0909, Dako, Santa Clara, United States of America). Antibodies were diluted in normal antibody diluent (Immunologic, Duiven, The Netherlands) and applied on the slides (Table
1). After incubation of post antibody blocking solution for 15 min (Immunologic), the secondary antibody Poly-HRP-goat anti Mouse/Rabbit IgG (cat. no. VWRKDPVB110HRP, Immunologic) was incubated for 30 min. Peroxidase activity was detected by DAB (D5637, Sigma, St. Louis, USA) or Bright-DAB (cat. no. VWRKBS04–110, Immunologic) as chromogen for 8 min. After all incubation steps, except the protein block, slides were washed with PBS-Tween-20 0.1% four times. Slides were counterstained with hematoxylin and mounted with Pertex (Histolab, Askim, Sweden).
Table 1
Antibodies and protocol variations
DAXX | Atlas antibodies, Bromma, Sweden | HPA008736 | Rabbit PAB | ARS/pH6 20 min | 1:100 1 h RT | Bright-DAB | Negative if positive nuclear staining < 5% of tumor cells |
ATRX | Sigma, St. Louis, MO | HPA0001906 | Rabbit PAB | ARS/pH9 20 min | 1:400 overnight 4 °C | DAB | Negative if positive nuclear staining < 5% of tumor cells |
ARX | Millipore, Burlington, MA | MABN102 | Mouse MAB (11F6.2) | ARS/pH6 20 min | 1:2000 1 h RT | DAB | Positive if weak nuclear staining > 50% or intermediate/strong nuclear staining > 10% of tumor cells |
PDX1 | Abcam, Cambridge, UK | ab134150 | Rabbit MAB (EPR3358(2)) | ARS/pH6 20 min | 1:2000 1 h RT | DAB | Positive if weak nuclear staining > 50% or intermediate/strong nuclear staining > 10% of tumor cells |
Ki67 | Immunologic, Duiven, The Netherlands | VWRKILM9252-C05 | Mouse MAB (MIB1) | ARS/pH6 20 min | 1:200 1 h RT | DAB | Digital image analysis of nuclear expression in at least 2000 tumor cells |
Glucagon | Cell Marque, Rocklin, CA | 259A-15 | Rabbit PAB | ARS/pH6 20 min | 1:100 1 h RT | Bright-DAB | Positive if cytoplasmic staining > 10% of tumor cells, scattered if < 10% of tumor cells |
Insulin | Dako, Santa Clara, CA | A564 | Rabbit PAB | ARS/pH6 20 min | 1:100 1 h RT | DAB | Positive if cytoplasmic staining > 10% of tumor cells, scattered if < 10% of tumor cells |
H3K36me3 | Abcam, Cambridge, UK | ab9050 | Rabbit PAB | ARS/pH6 20 min | 1:2000 1 h RT | DAB | Negative if positive nuclear staining < 30% of tumor cells |
ARID1A | Abcam, Cambridge, UK | ab182560 | Rabbit MAB (EPR13501) | ARS/pH6 20 min | 1:1000 1 h RT | DAB | Negative if positive nuclear staining < 5% of tumor cells |
Scoring was performed by at least two independent researchers (WMH, WS, LAAB), blinded for each other’s results and clinical information. Disagreements were resolved by consensus.
For ARX and PDX1, negative protein expression in tumor tissue was defined as weak nuclear staining in < 50% of cells or strong nuclear staining in < 10% of cells. Positive expression was defined as weak nuclear staining in > 50% of cells or intermediate/strong nuclear staining > 10% of cells [
27]. For insulin and glucagon, cytoplasmic staining of > 10% of cells was considered positive expression for the respective peptide hormone. Normal islets, containing a mix of cells expressing or not expressing the respective peptide hormone, were used as positive and negative controls, respectively. If < 10% of cells had expression, cases were called scattered. DAXX, ATRX, and ARID1A were considered negative if < 5% of cells had positive nuclear staining and if there was non-tumoral tissue present with positive nuclear staining serving as internal control, e.g., islets of Langerhans, stromal cells, endothelial cells, or lymphocytes [
26,
28]. All negative cases in the TMA were also stained on whole sections to confirm the results. For H3K36me3 loss, a cut-off of 30% of cells was used [
26,
29]. Negative cases without a positive internal control were non-informative. All cytoplasmic staining was ignored.
Ki67 labeling index (LI) was counted in at least 2000 cells by digital image analysis with Sectra (PACS, Sectra AB, Linköping, Sweden), as previously described [
30]. Digital counts were confirmed by visual assessment. PanNETs were graded by the 2017 WHO criteria (Ki67 G1 < 3%, G2 3 to 20%, G3 > 20%) [
31]. If the pathology report also mentioned tumor grade based on Ki67 or mitoses per 10 HPF, the highest grade was used for further analysis as the location of tumor cores not always represents the most proliferative region.
Fluorescence In Situ Hybridization
After deparaffinization in xylene, 4 μm FFPE sections for CDKN2A/CEN9 FISH were pre-treated in 0.2 N HCL for 20 min, cooked in a 10 mM citrate buffer (pH 6) for 20 min and washed in PBS. Slides were then digested in proteinase K buffer for 10 min at 37 °C (5 μM Tris–HCL, 1μM EDTA, 1 μM NaCl, 10 mg/L Proteinase K), washed with PBS and dried. Ten microliters of CDKN2A/CEN9 probe mix (CDKN2A/CEN 9 Dual Color probe, Zytolight, Bremerhaven, Germany) was applied per slide. Slides were denatured at 78 °C for 5 min and cooled on ice for 5 min. Hybridization was performed in a ThermoBrite (Abbott Laboratories, Chicago, IL) at 37 °C overnight. After removing coverslips, slides were washed in washing buffer (WB) 1 (0.4× SCC, 0.5% NP-40, 73 °C), WB 2 (2× SCC, 0.1% NP-40, room temperature), WB 3 (2× SCC, room temperature), and PBS, for 2, 1, 5 min, and 20 s respectively. Nuclei were counterstained and mounted with Vectashield with DAPI (H-1200, Vector laboratories, Amsterdam, The Netherlands).
Slides for telomere/centromere FISH were cooked in 10 mM citrate buffer (pH 6) for 20 min, washed in dH2O and dried. Probes (TelC-Cy3 F1002 PNA 180723PL-01, Cent-FITC, F3013 172865, Panagene, Daejeon, Republic of Korea) were diluted in hybridization mix (50% deionized Formamide, 50% SCC 4×, 5% Dextran sulphate, Tween-20 0.5%) at a 400 nM concentration and applied on the slides. After 5 min denaturation at 84 °C, slides were cooled on ice for 5 min before hybridizing at 37 °C overnight. After removing coverslips, slides were washed in two cycles of 1× WB (70% Formamide, 30% dH2O, 10 mM Tris, 15 min) and 3× PBS (2 min each time). Nuclei were counterstained with DAPI in PBS 2 μg/ml (Sigma-Aldrich, D9542), and coverslips were mounted with Vectashield (H-1000, Vector laboratories).
Slides were stored at 4 °C before imaging, and viewed with a Leica DM5500 B using appropriate excitation and emission filters. Images were made at 100× magnification with a Z stack of 14 steps in Leica application Suite X (Leica Microsystems, Rijswijk, The Netherlands).
The number of
CDKN2A gene probe and centromere probe signals were counted in at least 50 tumor cells for each case (WMH, WS). At least nine photographs of tumor tissue (confirmed on H&E) were made for counting of cells. Only intact non-overlapping nuclei with at least one centromere probe were counted. Multiple signals separated ≤ 1 signal distance were counted as one. If no signals were observed (gene and centromere) in the tumor and surrounding stromal cells or if there was too much background, cases were called non-informative. Homozygous loss was defined as at least 20% of counted cells lacking
CDKN2A probe signals with at least one CEP9 probe. Hemizygous deletion of
CDKN2A was concluded if 45% of counted cells had one
CDKN2A probe and two CEP9 probes; monosomy of chromosome 9 (which can be considered hemizygous loss) if 15% of cells had one
CDKN2A probe and one CEP9 probe. Cut-off values were based on previous literature [
26,
32]. If results were discordant, additional photos were made and at least 100 cells were counted (WMH).
ALT positivity was defined as ultra-bright, intra-nuclear telomere FISH signals, 10× the signal intensity of cumulative single telomere sum intensities in normal stromal/endothelial cells, which are present in more than 1% of cells [
18,
33]. The percentage of ALT cells > 1% was determined on × 20 magnification by visual assessment (WMH) in areas of tumor tissue (annotated in H&E slide), with a cut-off of more than 20 ultra-bright foci per tumor core (max 2000 cells). If less than 1% by visual assessment, all ultra-bright foci were counted. All ultra-bright foci were confirmed at × 100 magnification. Using the same laser/microscope settings, representative ultra-bright foci near stromal cells were photographed and digital grayscale TIFF images exported for each fluorophore channel. Telomere signals were quantified using Telometer (a free custom software ImageJ plug-in, downloaded from
demarzolab.pathology.jhmi.edu/telometer). Rolling ball radius was set at the maximal telomeric signal diameter, which was measured per photograph. Signals were separated by the draw function if necessary. Centromere signals were used as positive hybridization efficiency control and to confirm specific probe binding.