Alternative Lengthening of Telomeres and Differential Expression of Endocrine Transcription Factors Distinguish Metastatic and Non-metastatic Insulinomas

Insulinomas are the most common functional pancreatic neuroendocrine tumor (PanNET) type and are diagnosed by the triad of hypoglycemic symptoms, low blood glucose concentrations, and relief of symptoms after glucose administration. Most insulinomas are indolent tumors: there are few mitoses (low grade) and metastases are very rare [1, 2]. In contrast, 40–50% of non-functional PanNETs present with liver metastases at time of initial diagnosis [3, 4]. Surgery for insulinomas is primarily indicated to alleviate symptoms of hypoglycemia. Survival after surgery is not different from the general population [5, 6]. There are no insulinoma-specific international recommendations for follow-up after surgery [5, 7]. Nevertheless, about 10% of insulinoma patients develop metastases that are mostly present at time of initial diagnosis, but may sometimes develop years after resection of the primary tumor [8]. Median survival is less than 2 years in patients with metastatic insulinoma, similar to metastatic non-functional PanNET [9‐11].

Because of the rarity of metastatic insulinomas, little is known about the mechanisms of tumorigenesis. Recent sequencing studies identified recurrent YY1 gene mutations in insulinomas; however, relapses or metastases were rarely reported in the studied cohorts [12‐15]. There is a need to increase our understanding of insulinoma development, in particular of metastatic insulinomas. This will improve identification of patients at risk for recurrence, who may benefit from follow-up after surgery and thereby earlier detection and treatment of metastases.

Metastatic non-functional PanNETs are more common and have been characterized more extensively. Next-generation sequencing studies have demonstrated that sporadic PanNETs harbor relatively few gene mutations. The genes that are most frequently mutated are MEN1, ATRX, and DAXX [12, 16, 17]. Mutually exclusive mutations in ATRX or DAXX—coding for two chromatin-modifying proteins which form a histone chaperone complex—are associated with the alternative lengthening of telomere phenotype (ALT) [18]. Immunohistochemical loss of ATRX and DAXX can be used as a surrogate marker of inactivating mutations of ATRX or DAXX, respectively [18, 19]. The presence of either of these alterations is associated with recurrence and liver metastases [20‐25]. In addition to ATRX or DAXX mutations and ALT, loss of ARID1A, loss of H3K36 trimethylation (H3K36me3) by SETD2 dysfunction, and CDKN2A deletions were also reported to be of prognostic value for non-functional PanNET [26]. A recent finding in non-functional PanNETs is that alpha- or beta-cell types-of-origin may predict clinical behavior [20, 27]. Beta-cell like non-functional PanNETs, marked by expression of the endocrine transcription factor PDX1, were generally indolent, while almost all relapses were observed in the group of alpha-cell like non-functional PanNETs marked by ARX expression. Furthermore, somatic mutations in ATRX, DAXX, and MEN1 and acquisition of ALT were observed more often in the alpha-type non-functional PanNETs [20, 27].

The cohort consisted of 35 primary insulinomas. The clinical characteristics of the patients are listed in Table 2. All cases were confirmed to be clinically functioning insulinomas by the treating hospital clinician. Three patients developed liver metastases during follow-up, one of whom also had local recurrence and tumor depositions around the uterine appendages. While all metastastic primary cases initially presented with symptomatic hypoglycemia, only two out of three cases had hypoglycemic events at relapse by liver metastasis (Table 3). Overall survival could not be analyzed as there were no deaths during follow-up. The metastatic primary insulinomas were larger than the indolent primary insulinomas, but this did not reach statistical significance (mean 32 indolent insulinomas 1.48 cm, SD 0.45, mean 3 metastatic insulinomas 7.20 cm, SD 3.12, t test p = 0.086; Fig. 1). In addition, biopsies of liver metastases of two patients with inoperable primary insulinoma were available (Table 3), as well as a biopsy of a corresponding liver metastasis of one of the primary insulinomas in the cohort (patient 2, Table 3). When the sizes of the primary tumors of the inoperable insulinoma patients were included, the five metastatic insulinomas were significantly larger than the indolent cases (mean 1.48 cm, SD 0.45, metastatic mean 6.16 cm, SD 2.63, t test P = 0.016). A size cut-off of 3 cm perfectly separated all metastatic from indolent insulinomas.

Positive or scattered insulin expression was observed in all insulinomas. Positive or scattered glucagon expression was seen in about 60% of insulinomas. PDX1 expression was observed in 34/35 primary insulinomas (97%), and 2/3 liver metastases; in contrast, only 3/35 (9%) of primary insulinomas were ARX positive, while all liver metastases were ARX positive (Fig. 1, Tables 2 and 3). All ARX positive primary insulinomas developed metastases during follow-up. The ARX-positive percentage of cells varied between 10 and 90% of cells (Fig. 1, Table 3), and PDX1 expression was intermediate to strong in 70–100% of cells for all positive cases. Two (of 3) metastasizing primary insulinomas were multifocal (two tumors per patient). To exclude the possibility of the smaller tumor being the insulinoma and the larger tumor (tested in the TMA) possibly being an ARX-positive non-functional PanNET, whole slides of all multifocal tumors were stained for ARX, PDX1, and insulin. ARX, PDX1, and insulin expression was identical between tumors of the same patient. Thus, strong ARX expression in more than 10% of cells identified metastatic insulinomas with a 100% sensitivity and specificity. Of note, one of the metastatic insulinomas showed areas with < 10% ARX positive cells with scattered positive cells when assessing the whole slide (Patient 1, Fig. 1c). Although no obvious heterogeneity was seen between different tumor cores when scoring for ARX and PDX1, it is conceivable that the number of ARX-positive cells was underestimated using TMA cores for other cases. We therefore sought to further confirm our results by reviewing all ARX-negative insulinomas by digital image analyses on the TMA similar to the Ki67 count [30]. In all these cases, expression was observed in far less than 10% of cells, most often absent or in less than 1% of cells.

Of 33 insulinomas interpretable by telomere FISH, two cases (6%) had ALT which both developed liver metastases during follow up (Fig. 1; Table 3). All tested liver metastasis biopsies had ALT. All insulinomas (n = 34) with tumor tissue present in the TMAs had retained ATRX expression, only one case had heterogenic DAXX loss (also ALT positive). Insulin, glucagon, ARID1A, H3K36me3 IHC, and CDKN2A FISH were only tested on 31 cases, as for the last four insulinoma cases, no unstained TMA slides were available. All cases (n = 30) with tumor tissue present in the TMA had retained H3K36me3 and ARID1A expression. One case with homozygous CDKN2A deletion and one case with monosomal CDKN2A were observed (of 21 interpretable cases; Table 3). The case with homozygous CDKN2A loss developed liver metastases. The metastatic primary insulinomas were grade 1 (2/3) or grade 2 (1/3), and both metastasis biopsies were grade 2 (Ki67 labeling index, Table 3).

All metastatic insulinomas in this cohort were larger than 3 cm. Strikingly, all metastatic insulinoma lesions showed ARX expression, which was not observed in any of the indolent primary insulinomas. Four out of five (80%) metastatic insulinomas had ALT—not reported before in insulinomas—while none of the indolent insulinomas showed ALT. Interestingly, the two recent studies that identified ARX as marker for PanNET relapse after surgery, also included non-metastatic insulinomas: Cejas et al. found no ARX immunohistochemical expression in 17 primary insulinomas [27] and Chan et al. reported ARX mRNA expression (alpha signature) in one out of three insulinomas. None of the insulinomas in these studies metastasized, but remarkably, the ARX-expressing case in the study of Chan et al. was also ATRX mutated and very large (8 cm), while all the negative cases were small (< 2cm) and ATRX, DAXX, MEN1 wild type [20]. These results are in line with our observation that there is an association between size, ALT, and ARX expression in insulinomas. To our knowledge, ALT has not been described before in any insulinoma. Retention of ATRX/DAXX protein expression in one of the insulinomas with ALT may be explained by non-truncating mutations, translocations, or other underlying mutations causing the ALT phenotype [34]. In the literature, presence of somatic ATRX and DAXX mutations or protein loss detected by immunohistochemistry is uncommon in sporadic insulinoma—in contrast to non-functional PanNET [12‐15, 26]. A recent large whole-genome sequencing study definitively established that insulinomas and non-functional PanNETs have distinct genetic underpinnings, and recurrent copy number variations together with ATRX and DAXX mutations are a characteristic feature of non-functional PanNETs [12]. However, such alterations might be more prevalent in metastatic insulinomas, of which only few are present in the cohorts previously reported in the literature. Interestingly, in a cohort with multiple rare malignant/metastatic insulinomas, a high number of chromosomal aberrations was strongly associated with metastases [35]. As ATRX/DAXX mutations and ALT correlate with copy number variations and chromosomal instability [21, 36], this may be a reflection of ATRX/DAXX mutations that were not tested at that time. Loss of the tumor suppressor CDKN2A has been reported once before in a malignant insulinoma [37] and was recently described as a marker of malignant behavior in non-functional PanNET [26].

Several studies have reported large insulinomas to be malignant more often, and the late symptomatology suggests relatively low or acquired insulin production (Fig. 2) [1, 38, 39]. The ARX transcription factor is not expected to be expressed in insulinomas [40], as it is not expressed in pancreatic beta-cells [41, 42]. In contrast, around 50–60% of non-functional sporadic PanNETs express ARX [27]. Although focal nesidioblastosis was considered as explanation for ARX expression, it was deemed highly unlikely due to the characteristic tumor morphology, random peptide hormone expression, and the presence of metastases [43].

Acquired insulin production could be the result of transdifferentiation of ARX positive non-functional PanNETs or (subclinical) glucagonomas. The latter phenomenon has been shown in mice [44, 45], and a recent report described a human non-functional PanNET that progressed into a metastatic insulinoma with liver metastases over the course of 2 years [46]. Importantly, all malignant cases in this study had hypoglycemic symptoms at presentation, which resolved after surgery of the primary tumor. All but one metachronous liver metastases (patient 2) gave new episodes of symptomatic hypoglycemia when diagnosed. Interestingly, in patient 2, both the primary and the corresponding liver metastasis only had few PDX1-positive cells (potential insulin producing cells), so perhaps the solitary liver metastasis (3 cm) did not grow to a size in which a low percentage of cells could have caused symptoms.

There are several strengths of this study. The assay to assess expression of the nuclear transcription factor ARX is very robust. Whole slides that were stained show no variable expression due to fixation, and pancreatic islets, enteroendocrine duodenal/antrum cells serve as positive internal controls. Although several other markers have been proposed based on molecular analyses, protein, or mRNA expression, this is the first study to show a clear association of one single immunohistochemical marker with malignant behavior in insulinomas [35, 47, 48].

The small cohort size and number of metastatic cases is a limitation of this study, caused by the fact that metastatic insulinomas are very rare. Validation of our findings in other cohorts would be of important value. For the limitations inherent to cut-off scoring systems and TMAs, we have tried to minimize their effect by scoring blinded by two independent observers, confirming key percentages by digital analysis, and staining whole slides to confirm immunohistochemical expression when necessary.

To conclude, large tumor size (> 3 cm) was confirmed to be a strong marker of metastatic behavior in insulinomas. In addition, we found that large metastatic insulinomas are driven by tumorigenic mechanisms often seen in non-functional PanNETs, but not in small indolent insulinomas. Furthermore, we demonstrate that ARX—which is normally not present in beta-cells—is expressed in a subset of insulinomas and is associated with large tumor size, ALT, and metastatic disease. In contrast, ARX expression was absent in any of the small indolent insulinomas in this study or the previous literature. We hypothesize that cellular differentiation and tumorigenic mechanisms more closely related to non-functional PanNETs are important for the development of malignant insulinoma. Our findings with regard to rare metastatic insulinomas may be of value to personalize follow-up and treatment strategies in the future.