Recurrent Loss of Heterozygosity in Pancreatic Neuroendocrine Tumors

Abstract

Chromosomal aneuploidies are prognostic markers across a wide variety of tumor types, and recent literature suggests that pancreatic neuroendocrine tumors are no different. In this study 214 patients with grade 1, 2 or 3 pancreatic neuroendocrine tumors had their tissue examined for chromosomal copy number alterations using next generation sequencing. Univariate and multivariate statistical analyses were performed with all-cause mortality and disease-specific mortality as the end comparators. As such, the cohort stratified into three different clinically-relevant chromosomal subgroups: an indolent subgroup characterized by loss of chromosome 11 in relative isolation, an aggressive subgroup characterized by losses of chromosomes 1, 2, 3, 6, 10, 11, 16, and 22 and with no loss of chromosomes 4, 5, 7, 12, 14, 17, 19, and 20, and finally a heterogeneous third group with a subset of cases that behave even more aggressively than the aforementioned.

Introduction:

Pancreatic neuroendocrine tumors (PanNET) are a heterogeneous group of neoplasms with variable prognosis. They are generally slow-growing but progressive with a 5-year overall survival of only 33% [1]. This stands in contrast to pancreatic neuroendocrine carcinomas (PanNEC) which are more aggressive with less than 1-year expected survival; less than a quarter of patients with PanNECs are alive 2 years after diagnosis [1-2]. Currently PanNETs are pathologically stratified into three groups: grade 1, grade 2, and grade 3 by mitotic and Ki-67 proliferative indices [3]. Unfortunately, to date the standard classification system for PanNETs does not predict disease progression and survival with sufficient accuracy [1]. Medical uncertainty is due, in part, to a lack of understanding of the neogenesis of neuroendocrine tumors and the specific biologic mechanisms underlying their metastasis.

Pancreatic neuroendocrine tumors with DAXX and ATRX mutations or gene loss are often associated with a worse clinical outcome [1,4]. ATRX and DAXX mutations or gene loss are also associated with frequent losses of chromosomes 1,2,3,6,8,10,11,15,16,21 and 22 [4-7]. These losses are termed chromosomal instability, although these losses are recurring, involve entire chromosomes, and do not involve mid-chromosome or complex rearrangements, fragmentation, or other signs of true instability [7].

ATRX or DAXX genes are mutated in 40% of PanNETs, typically in association with MEN1 mutations, although DAXX and ATRX mutations can occur without MEN1 and MEN1 alterations can also be seen alone [2]. MEN1 encodes Menin, a tumor suppressor most well-known for its association with the syndrome termed multiple endocrine neoplasia type 1. This hereditary disease, caused by inherited mutations in MEN1, results in pituitary, parathyroid, and pancreatic neoplasms including PanNETs. MEN1-syndrome is rare, and most PanNETs are sporadic (80-90% non-syndromic) with somatic mutations in MEN1 found in 30-60% of sporadic cases [8].

The gene DAXX encodes a tumor suppressor that has a wide variety of functions in the cell and is known to interact with the tumor suppressor ATRX [9]. The DAXX/ATRX complex interacts with histone H3.3 at telomeres and other areas of genomic repeats. Previous studies have implicated a loss of the ATRX/DAXX complex in alternative lengthening of telomeres (ALT) and ultimately ascribe the chromatin instability seen in PanNETs to this mechanism [10].

ALT is a telomerase-independent way for linear chromosomes to extend telomeres after repeated rotations through the cell cycle. Shortened or disrupted telomeres, from repeated cell divisions, can halt further entry into the cell cycle and suppress neoplastic processes. Cells that repair telomeres either through telomerase or other mechanisms have a proliferative advantage [10]. The ALT mechanism results in end-to-end fusions of chromosomes, anaphase bridges, break-fusion events and ultimately lead to true chromosomal instability [11]. When DAXX or ATRX are mutated ALT is reportedly increased [12]. This simple biologic mechanism to explain recurrent chromosomal losses has come into question recently [13].

Quevedo et. al. found the eleven-chromosome loss of heterozygosity signature (losses of chromosomes 1,2,3,6,8,10,11,15,16,21 and 22), associated with mutations of MEN1, DAXX and/or ATRX, is likely the result of centromeric cohesion failure and mis-segregation, not necessarily ALT as previously described. Although they found indications that an ALT phenotype existed in these DAXX/ATRX/MEN1 mutant tumors, they postulate that the disruption of H3.3 deposition at genomic repeats, found both at telomeres and centromeres, could be the potential mechanism of genomic instability. H3.3 and an H3 variant, CENP-A, are responsible for assembling kinetochore proteins at the centromere during cell division. In their paper they found that CENP-A is mis-localized in DAXX/ATRX depleted cell lines. They also show that mutational events in DAXX, ATRX, and MEN1 precede the recurrent chromosome losses [13].

Lawrence et. al. postulated three different clinically-relevant subtypes of PanNETs [7]. Group 1: a poor prognostic subgroup characterized by recurrent multi-chromosome losses (specifically the loss of 10 chromosomes: 1, 2, 3, 6, 8, 10, 11, 16, 21, and 22) and mutations in, or loss of, the genes DAXX or ATRX; Group 2: an indolent group characterized by loss of heterozygosity of chromosome 11 only without loss of the other nine chromosomes and MEN1 mutations alone, without DAXX or ATRX alterations; and Group 3: a heterogenous group that did not fit the prior two subgroups. Herein we describe 240 PanNET cases from 214 different patients across two institutions. These 240 samples have been stratified according to a modified classification system devised by Lawrence et. al. with striking differences in clinical outcome suggesting possible new biomarkers which could improve prognostication. This is the largest single study looking at the correlation between chromosomal loss, as determined by next generation sequencing, and overall survival in patients with PanNETs.

Materials and Methods:

Case Selection:

After Institutional IRB approval a retrospective review of the archives of both the University of Chicago and the Memorial Sloan Kettering Cancer Center (MSKCC) was conducted. Cases with a histologic diagnosis of primary and/or metastatic well-differentiated (Grades 1, 2, or 3 by current WHO criteria) pancreatic neuroendocrine tumors were selected [1]. Neuroendocrine carcinomas, by current standards, were excluded. MEN1 syndromic PanNETs were also eliminated from the cohort. A total of 10 patients at the University of Chicago have a confirmed histologic diagnosis meeting both inclusion and exclusion criteria and 204 patients at MSKCC met the above criteria. From the University of Chicago slides were reviewed to verify the level of differentiation and exclude the aforementioned. A representative tumor slide from each case was chosen for analytical purposes and sequencing. At MSKCC slides were selected as part of clinical care and sequencing was performed when clinically indicated. All sequencing data at MSKCC is retrospectively reviewed. At both institutions demographic and histologic data including age at diagnosis, gender, tumor size, presence and site of metastases, ki-67 index, mitotic count, presence of lymphovascular and/or perineural invasion, and TNM stage, were tabulated from prior reports and tumor slides.

Sequencing at the University of Chicago:

A hybrid-capture panel targeting 1,213 cancer-associated genes (UCM-OncoPlus) was used to examine genomic alterations within the 12 tumor samples from 10 patients at the University of Chicago as part of a research trial. DNA extraction, DNA quantification, library preparation, and next-generation sequencing (NGS) mutational analysis and copy number calling were performed as described previously on a HIPAA-compliant high-performance computing system with an in-house developed bioinformatics pipeline [14]. Cases with low tumor purity were excluded (<10%).

Sequencing at MSKCC:

All cases were sequenced during the course of routine clinical care at MSKCC using MSK-IMPACT. Clinical indications for sequencing included advanced cancer stage, with treating physicians looking for possible therapeutic targets, or as a requirement for clinical research trial enrollment. MSK-IMPACT is a hybrid-capture panel targeting 341, 410, or 468 cancer-associated genes (MSK-IMPACTv3, MSK-IMPACTv5 and MSK-IMPACTv6 respectively). The 228 tumor samples from 204 patients at MSKCC underwent DNA extraction, DNA quantification, library preparation, and NGS mutational analysis with copy number calling per clinical protocol on a HIPAA-compliant high-performance computing system with an in-house developed bioinformatics pipeline [15]. FACETS was used to verify copy number calling when ambiguous by copy number log ratio alone [16]. Cases with low tumor purity were excluded (<10%).

Statistical Methods:

Survival analyses were performed in SAS 9.4 (SAS Institute, Cary, NC). Overall survival was defined as the interval between first diagnosis and death from any cause. Disease-specific survival was defined as the interval between first diagnosis and death from disease, with death from other causes censored. Univariate survival comparisons of discrete groups were performed with Kaplan–Meier method and survival curves were compared using the log rank test, with multiple comparisons corrected by Tukey's method. Cox proportional hazards regression was used for multivariate modeling, for univariate survival analyses of continuous variables, and to calculate univariate hazard ratios. The assumptions of Cox modeling were satisfied [17]. P values less than 0.05 were considered significant, and all p values were two-tailed.

Results:

Defining the Groups:

As previously detailed, recurrent loss of heterozygosity is common to DAXX/ATRX mutated PanNETs and has been described using a multitude of terms including, unfortunately, genomic instability and chromosomal instability - phrases that are both inaccurate and misleading. The index paper we reference (Lawrence et. al.) used the terms “recurrent 10 chromosome monosomy” and “10 chromosome loss of heterozygosity (LOH)” which are more accurate; however, other authors have referenced the loss of a slightly different number of chromosomes. For example, Quevedo et. al. referenced a “signature of loss of heterozygosity and copy-number alterations affecting 11 chromosomes” [13]. Importantly, mentioning only the lost chromosomes excludes which chromosomes must be preserved, thus a case with global LOH could inappropriately fall into this category. Additionally, many authors referenced the signature predominantly as a relative gain or absolute gain of the non-lost chromosomes. Nagano et. al. described a gain of chromosomes 5, 7, 12, 14, 17, and 20 in PanNETs and Zhao et. al. described both a loss of 1, 3, 6, 10, 11 and Y with a gain of 4, 5, 7, 9, 12, 14, and 17 [5,18].

All 240 cases in our group had their chromosome copy number plots visually inspected for features that match the above literature with the suggested chromosome losses and chromosome gains. This pattern was strikingly uniform, even though the language used to describe the fingerprint has been inconsistent (Figure 1A). 137/240 samples had a visual pattern representing recurrent losses and gains referenced hereafter as a “visual fingerprint.” To assist in defining this visual pattern the autosomal chromosomes were assessed for loss within the 137 cases: chromosome 1 was lost in 94% of fingerprint cases, chromosome 2 was lost in 100% of cases with a visual fingerprint, 3 was lost in 91% of fingerprint cases, 4 in 4% of cases, 5 in 5%, 6 in 98%, 7 in 1%, 8 in 77%, 9 in 8%, 10 in 93%, 11 in 97%, 12 in 3%, 13 in 14%, 14 in 2%, 15 in 71%, 16 in 98%, 17 in 4%, 18 in 21%, 19 in 3%, 20 in 1%, 21 in 81%, and 22 in 91%. Sex chromosomes were not included in the analysis. In sum, 8 chromosomes were recurrently lost at least 90% of the time and 8 chromosomes were recurrently non-lost at least 95% of the time in fingerprint cases.

Figure 1: Visual Fingerprints of Recurrent Chromosome Loss and more objective 8+8 Signature Criteria.

A) Copy number analysis is plotted per convention using the Log-ratio (LogR) of sample reads to expected reads [28]. The Y-axis for all graphs is LogR. The X axis for all graphs are the 23 chromosomes starting with autosomes in ascending order and terminating with X at right most position. Y is not plotted in males and X is normalized using sex-appropriate comparators. Mathematically, a region without gain or loss of genetic material will have a LogR of 0, which is indicated on the plot with a thick red line. Losses of chromosomes and sub-chromosomal regions will have a negative value, plotted below 0, and gains will have a positive value, plotted above 0. Although the axes are unlabeled, copy number analysis from eight different patients with the visual fingerprint of recurrent chromosome losses are pictured for a “low-power view” of the pattern of aneuploidies. A schematic representation of the visual fingerprint is presented on the bottom with all 22 autosomal chromosomes labeled and a red line at LogR=0. B) An example case with the visual fingerprint of chromosomal loss has been scored using 8+8 criteria. The LogR at 0 is indicated with a thick red line. Also, as above, the X-axis is the 23 chromosomes, autosomes first, in ascending order. Chromosome 23 is the X chromosome. And as above, the Y-axis is the log-transformed copy number measurements. For a case to meet criteria as having an “8+8 signature” it should have whole chromosome losses of chromosomes 1, 2, 3, 6, 10, 11, 16, and 22 while having no losses of chromosomes 4, 5, 7, 12, 14, 17, 19, and 20. To more objectively score, 1 point is assigned for each of the 16 listed chromosomes that matches expected criteria. All cases with a score of 14 or higher are considered positive for the signature. In this example case the patient had all 8 losses (diagramed below in red) and all 8 gains (diagramed in blue below) resulting in a score of 8+8=16, thus this case meets criteria for “8+8 signature.” If one of the chromosomes had not been lost, or had been lost where it should have been preserved, the patient would only receive a score of 7+8=15. Any chromosomes outside of the 16 listed, do not contribute to the assignment of an “8+8 signature”.

Using this information we have chosen to represent both the idea of recurrent chromosome losses and a relative or actual gain of other select chromosomes in the holistic term “8+ 8 signature.” For a case to have met “8+8 signature” criteria it must have had losses of chromosomes 1, 2, 3, 6, 10, 11, 16, and 22 and no losses of chromosomes 4, 5, 7, 12, 14, 17, 19, and 20. A simple mathematical equation was used to more objectively classify tumors as “8+8” or not (Figure 1B). Of the 137 cases with a visual fingerprint 96% (132/137) met the more objective “8+8” criteria. These 5 samples, from 5 patients, with the visual signature but without the objective “8+8” measurement were notably all without DAXX or ATRX mutations or focal gene loss. Additionally, 3 samples of the 103 without a visual signature met “8+8” criteria. The 3 samples all came from one patient with DAXX and MEN1 co-mutations. These 6 patients with either an 8+8 signature without a fingerprint or with a fingerprint without an 8+8 signature are detailed further in Supplemental Table 1.

The 8+8 signature group (n = 115 patients / n=135 samples) had a high frequency of MEN1, DAXX and/or ATRX mutations/ gene loss. Only four samples from four different patients with the 8+8 signature lacked MEN1, DAXX and ATRX DNA level alterations (4/115 patients; 4/135 samples). This 8+8 signature group roughly approximated the pNET group 1 in the Lawrence et. al. paper which had recurrent chromosome losses, defined subjectively, and mutations in MEN1, DAXX or ATRX. The remainder of this paper will refer to this “8+8 signature” group as the modified-pNET 1 group or mpNET1.

The pNET group 2 was characterized by loss of heterozygosity of chromosome 11 with few other changes in chromosomal copy number. This group also demonstrated mutations in MEN1 without DAXX or ATRX mutations. Criteria for this group were almost directly taken from Lawrence et. al. To qualify as a modified-pNET2 case (mpNET2) a MEN1 mutation was required as was chromosome 11 loss in relative isolation. Specifically, chromosome 11 loss in relative isolation was defined by two or less other whole-chromosome or partial chromosomal losses or gains that accompany the chromosome 11 loss. Fragmented chromosomes, defined as a chromosome with multiple non-uniform sub-chromosomal (regional) losses and gains, were obviously exclusionary for this group. DAXX and ATRX mutations did not co-occur in this group, with a single exception which will be discussed in the next section. There were no cases in our cohort of 240 that showed relatively isolated chromosome 11 loss without a MEN1 mutation.

Our modified-pNET group 3 encompassed all other PanNETs. This ‘wastebasket’ group was heterogenous. There are a few interesting sub-stratifications that can be seen in this group. Our modified-pNET3 group, unlike Lawrence et. al., included some cases with MEN1 mutations that did not meet the strict chromosomal requirements mentioned above.

Description of Cohorts:

As previously mentioned, 54% (115/214) patients fall into the mpNET1 category, with an “8+8” score of at least 14 (Figure 1B). A tiny fraction of cases, only 4%, are categorized as mpNET2 (9/214 patients). The remaining 42% of patients are best classified as mpNET3 group (90/214). These percentages are quite different from those of the index paper which reported 26% in the pNET1 group, 40% in the pNET2 group and 34% in the pNET3 group. This discrepancy will be addressed in the discussion. A representative image from each of the three chromosomal groups, matched for grade, can be found in supplemental materials.

The mpNET1 group consists of 115 patients and 135 samples, all with an 8+8 score of 14 or higher, by definition. Within this group 89% of samples (100 patients & 120 samples) have ATRX or DAXX mutations or focal homozygous gene loss with, or without, MEN1 mutations. 8% (11 samples from 11 patients) have MEN1 mutations or focal homozygous gene loss without ATRX or DAXX mutations. Finally, 3% (4 cases from 4 patients) are unusual without mutations or loss of MEN1, ATRX or DAXX seen by targeted NGS sequencing. The majority of ATRX and DAXX mutations identified in this cohort are frameshift, canonical splicing, or nonsense mutations predicted to produce either non-functional protein product or result in nonsense mediated decay (no protein). There are a lesser number of simple missense mutations. Homozygous focal gene loss is also a less common mechanism of gene loss for ATRX and DAXX, with 6p21.32 (DAXX) and Xq21.1 (ATRX) regional loss at least a log-fold beyond any whole chromosome losses of chromosome 6 or X. MEN1 mutations similarly are predominantly frameshift, canonical splicing or non-sense mutations resulting in a non-functional protein, with a lesser number of missense mutations and homozygous focal gene loss (11q13.1).

The mpNET2 group consists of 9 samples from 9 different patients. All 9 have MEN1 mutations and chromosome 11 monosomy, by definition. Six have chromosome 11 monosomy in complete isolation while three cases have up to two additional alterations including loss of one copy of chromosome 18, loss of one copy of chromosome 22 or gain of one copy of chromosome 14. There are no sub-chromosomal/regional gains or losses. Mutationally, among cases with matched normal samples to eliminate benign germline variations, six cases have MEN1 mutations in complete isolation, mostly frameshift alterations, with no other single gene somatic alterations and two cases have a co-occurring single gene alteration. One case has a co-occurring PTPRT missense mutation, and another strangely has a small sub-clonal ATRX missense alteration. The ATRX missense alteration is interesting, since ATRX and DAXX mutations are associated with the recurrent losses of heterozygosity seen in mpNET1. The MEN1 alteration in this case, with sub-clonal ATRX, has a variant allele fraction (VAF) of 76%, a reflection of both a relatively high tumor percentage and the loss of heterozygosity of the MEN1 locus. The ATRX mutation, on the other hand, has a VAF of only 7%, and is clearly subclonal. In addition the alteration in ATRX, p.Q1848K, is a simple missense alteration and not clearly detrimental to the function of ATRX. The particular ATRX p.Q1848K alteration has only been seen once before in Cosmic (https://cancer.sanger.ac.uk).

The mpNET3 group consists of 96 samples from 90 patients with a diverse array of copy number plots and genetic alterations. 11 patients have MEN1 mutations without loss of ATRX or DAXX, but often with other mutations. Copy number plots range from flat to completely fragmented with rare cases containing a visual fingerprint somewhat reminiscent of mpNET1, however they do not meet the strict criteria for an 8+8 signature call. This sub-group appears to have a poor clinical prognosis which will be discussed more in the next section. DAXX/ATRX mutations are rare in this group, occurring in only 5/90 patients, and most of these mutations are missense mutations, thus they may represent somatic passenger alterations, which could explain why there is little effect on chromosomal copy number. Within this group, mutations in the MTOR pathway, SETD2, VHL and BCOR were not uncommon. A large proportion had no mutations seen at all, despite adequate tumor purity, and two different patients had a BRAF V600E mutation. A summary of mpNET3 can be found in Supplemental Table 2.

Clinical Outcomes:

On univariate survival analyses, shorter disease-specific and overall survival are associated with stage IV disease, unresectable primary disease, increased mitotic count, increased Ki-67 index, tumor grade 3, and chromosomal group mpNET3 (see Table 1). ROC curves plotted for mitotic count and Ki-67 index do not identify any clearly optimal prognostic cut-point (data not shown). Supplemental Table 3 summarizes grade and stage by chromosomal subgroup.

	Pancreatic Neuroendocrine Tumor Outcomes (n=214)
	Disease-Specific Survival - Univariate Analyses				Overall Survival - Univariate Analyses
	n	Median DSS (months) (95% CI)	Hazard Ratio ‡	P value **	MedianOS (months) (95% CI)	Hazard Ratio ‡	P value **
Age at diagnosis			n.s.	0.07		n.s.	0.48
Sex				0.33			0.59
Male	113	151 (108-n.r.)	n.s.		118 (98-n.r.)	n.s.
Female	101	137 (95-n.r.)			137 (82-n.r.)
Stage				0.0003			<0.0001
I	9	n.r.	IV vs III: 3.1 (1.4-6.9)		n.r.	IV vs III: 3.1 (1.4-6.5)
II	38	174 (118-n.r.)	IV vs II: 4.7 (1.7-13.1)		174 (118-n.r.)	IV vs II: 5.4 (1.9-14.9)
III	49	n.r.	IV vs I: >10		n.r.	IV vs I: >10
IV	118	98 (65-174)	Other comparisons n.s.		89 (56-151)	Other comparisons n.s.
Tumor size			n.s.	0.14		n.s.	0.13
Mitoses			HR 1.151 (1.087-1.219) per additional mitotic figure	<0.0001		HR 1.151 (1.087-1.219) per additional mitotic figure	<0.0001
Ki-67 index			HR 1.089 (1.047-1.132) per percent Ki67	<0.0001		HR 1.089 (1.047-1.131) per percent Ki67	<0.0001
Lymphovascular invasion				0.052			0.051
Present	86	151 (118-n.r.)	n.s.		151 (118-n.r.)	n.s.
Not identified	27	n.r.			n.r.
Perineural invasion				0.21			0.14
Present	52	n.r.	n.s.		n.r.	n.s.
Not identified	51	n.r.			n.r.
Grade				<0.0001			<0.0001
1	58	151 (100-n.r.)	3 vs 1: 4.8 (2.1-10.6)		151 (98-n.r.)	3 vs 1: 5.0 (2.4-10.4)
2	116	137 (108-n.r.)	3 vs 2: 3.2 (1.7-6.0)		137 (108-174)	3 vs 2: 3.9 (2.2-7.0)
3	37	68 (20-n.r.)	2 vs 1: 1.5 (0.7-3.1) (n.s.)		40 (20-81)	2 vs 1: 1.3 (0.66-2.5) (n.s.)
Primary tumor resection				<0.0001			<0.0001
Resected	120	174 (137-n.r.)	0.22 (0.12-0.40)		174 (137-n.r.)	0.20 (0.12-0.36)
Not resected	94	82 (48-174)			63 (47-89)
Chromosome group				0.012			0.026
mpNET1	115	174 (118-n.r.)	mpNET3 vs mpNET2: ∮ (p=0.008)		151 (108-n.r.)	mpNET3 vs mpNET2: ∮ (p=0.02)
mpNET2	9	n.r.	mpNET3 vs mpNET1: 2.1 (1.2-3.5) (p=0.01)		n.r.	mpNET3 vs mpNET1: 1.8 (1.1-3.0) (p=0.03)
mpNET3	90	137 (87-n.r.)	mpNET2 vs mpNET1: ∮ (p=0.04)		95 (68-n.r.)	mpNET2 vs mpNET1: ∮ (p=0.09)
		Disease-SpecificSurvival - Multivariate Model (n=211)			Overall Survival - Multivariate Model (n=211)
		HR		P	HR		P
Stage		IV vs all other stages: 4.4 (2.2-8.7)		0.0004	IV vs all other stages: 4.3 (2.3-8.2)		0.0001
Tumor grade		3 vs 2: 2.3 (1.2-4.3) 3 vs 1: 3.7 (1.6-8.4) 2 vs 1: 1.7 (0.82-3.5) (n.s.)		<0.0001	3 vs 2: 2.9 (1.6-5.2) 3 vs 1: 4.1 (1.9-8.6) 2 vs 1: 1.4 (0.72-2.8) (n.s.)		<0.0001
Chromosomal group		mpNET3 vs mpNET2: ∮ mpNET3 vs mpNET1: 2.4 (1.4-4.2) mpNET1 vs mpNET2: ∮		0.0075	mpNET3 vs mpNET2: ∮ mpNET3 vs mpNET1: 2.0 (1.2-3.4) mpNET1 vs mpNET2: ∮		0.028

^**univariate p values for categorical and continuous variables were calculated by the log-rank and proportional hazards regression methods, respectively.

^‡Hazard ratios calculated by proportional hazards regression.

^∮Hazard ratios for comparisons with mpNET2 cannot be calculated, as no death events occurred in this group; provided p values were corrected by Tukey's method. DSS, disease-specific survival; n.r., not reached; n.s., not significant; OS, overall survival.

A Kaplan-Meyer curve for overall survival of the three mpNET groups is illustrated in Figure 2. Similar to Lawrence et. al. our mpNET2 is associated with an indolent clinical course. All nine mpNET2 patients are alive and eight of nine are without metastasis to date. A single patient has a metastasis of a neuroendocrine tumor to liver, however, the origin of this metastasis is of some debate. Although the patient is non-syndromic, they were negative for germline alterations in a panel of 88 genes including testing of MEN1, the patient had a synchronous gastric neuroendocrine tumor which was higher grade than the pancreatic neuroendocrine tumor. It is possible that the gastric primary could be the source of the neuroendocrine liver metastasis. All nine patients are alive and eight, excluding the aforementioned patient, are without disease recurrence.

Figure 2: Kaplan-Meier Survival Curve for All-Cause Mortality by PanNET Sub-groups.

The left Kaplan-Meier survival curve for all-cause mortality is stratified by pancreatic neuroendocrine tumor sub-group: mpNET1 (n=115), mpNET2 (n= 9), and mpNET3 (n=90). As in Lawrence et. al. mpNET2 is indolent with 100% survival in our small cohort. n.r. = not reached. mpNET1 tumors have an intermediate prognosis -- significantly better than the molecularly heterogeneous mpNET3 group (adjusted p=0.03), but trending toward significantly shorter survival than mpNET2 tumors (p=0.09). The right Kaplan-Meier survival curve for disease-specific mortality is again stratified by pancreatic neuroendocrine tumor sub-group. As discussed, mpNET2 has 100% survival. mpNET1 tumors again have an intermediate prognosis -- significantly better than the molecularly heterogeneous mpNET3 group (adjusted p=0.01), and significantly shorter survival than mpNET2 tumors (p=0.04).

mpNET1, with the “8+8 signature” is associated with a worse clinical outcome than mpNET2, however it appears to be a more standard, slowly progressive course than mpNET3. Only 7% (8/115) of cases are alive with no evidence of disease at the time of this study. 27% (31/115) had died of disease or with disease and 66% (76/115) are alive with persistent disease on therapy. The vast majority with disease have liver metastases that are either stable on therapy or slowly progressing. Some have bone or lymph node metastatic disease as well.

mpNET3 has subgroups with differing clinical outcomes, somewhat mirroring the heterogeneity of copy number changes and genetic mutations. Overall, 16% (14/90) are alive with no evidence of disease, twice the number as seen in mpNET1, but 36% (32/90) died of disease or with disease at the time of this study - an appreciably higher proportion than mpNET1. 59% (44/90) of mpNET3 patients are alive with persistent disease on therapy. Similar to mpNET1, the majority of patients with persistent disease have liver metastasis, although some have lymph node or bone disease as well. Interestingly, within this group, 8 of the 11 patients with MEN1 mutations in relative isolation, meaning without DAXX or ATRX, had died of disease and the remaining three all had liver metastasis.

A multivariate model was constructed including the three principal prognostic factors of interest: tumor stage, tumor grade, and chromosomal group. Mitotic count and Ki-67 index were excluded, as these individual parameters are factored into tumor grade, and tumor resectability was excluded due to high correlation with tumor stage. In this multivariate analysis, stage IV disease, tumor grade 3, and chromosomal group are all significantly associated with shorter disease-specific and overall survival, with survival differences observed between all three chromosomal groups.

Discussion

Chromosomal aneuploidies are important prognostic markers for numerous diseases. Within hematopoietic neoplasms, hyperdiploid B-lymphoblastic leukemia lymphoma (B-ALL) has a very favorable prognosis compared to other genetic alterations in B-ALL [19]. Similarly, multiple myeloma with trisomies of one or more of the odd-numbered chromosomes including 3, 7, 9, 11, 15, or 17 has a favorable prognosis [19]. Aneuploidies in solid tumors also have prognostic impacts. According to cIMPACT-NOW3, IDH-wild-type astrocytomas will soon be classified as glioblastomas, with a worse prognosis, when chromosome 7 is gained and chromosome 10 is lost [20-21]. And 20q has prognostic implications in colon cancer [22]. Thus, it is not surprising to see that PanNETs also demonstrates prognostic correlations with chromosomal aneuploidies.

Aneuploidy is common in PanNETs and is tightly associated with DAXX or ATRX gene loss or mutation. In our 8+8 signature aneuploidy cohort (mpNET1) ATRX and DAXX mutations are seen in 97% of samples (111/115 patients; 131/135 samples). In general, DAXX and ATRX mutations are not generally present in the mpNET2 or mpNET3 groups 6% of cases; 6/99 patients; 6/105 samples) and when these alterations are present they are more commonly missense mutations and classified as variants of uncertain significance. Additionally, the only ATRX or DAXX alteration in an mpNET2 case is present at a low VAF and is likely sub-clonal in addition to being a missense alteration.

Mutations in ATRX and DAXX, and the resultant loss of ATRX and DAXX protein expression can be demonstrated with immunohistochemical stains (IHC) [4,23]. Within pancreatic neuroendocrine tumors, the sensitivity and specificity of IHC to detect a DAXX mutation by DAXX IHC loss is 85% and 95%, respectively [24]. The sensitivity and specificity of ATRX IHC for ATRX mutations has not been established in pancreatic neuroendocrine tumors but ATRX IHC is commonly employed in the diagnosis of brain tumors which use the presence or absence of ATRX mutations in the diagnostic algorithm for diffuse gliomas (CNS WHO). Despite the apparent utility of DAXX and ATRX IHC to detect DAXX and ATRX mutations, routine IHC staining has not been implemented into most institutions in the diagnosis or prognostication of most panNETs.

Interestingly, DAXX and ATRX gene losses has been associated with better clinical outcome in some papers [25] and worse clinical survival in others [1,4,26-27]. Part of the reason for the discrepancy is that the studies use differing comparative end points. For example, it seems loss of either DAXX or ATRX predicts a decreased relapse-free survival, but not a shortened tumor-specific survival [4]. The other component contributing to the mixed data appears to be what the comparison group is, specifically asking the question, “DAXX/ATRX mutated tumors have a worse survival compared to whom?” When the comparison group is enriched in mpNET3 cases, or cases with liver metastasis, these DAXX/ATRX mutant cases may appear to do better. On the other hand if the comparison cases are enriched in mpNET1 cases, or low-grade, low-stage and indolent tumors, the DAXX/ATRX-mutant cases will appear to do worse.

The Lawrence et. al. index paper had 26% of cases classified in their pNET1 group, 40% in the indolent pNET2 group and 34% in the pNET3 group. Our cohort is markedly enriched in “bad actors” with 54% of patients in the mpNET1 category, only 4%, categorized as indolent mpNET2, and 42% of patients in mpNET3. This discrepancy is due to the method of data collection. The vast majority of cases in the cohort came from MSKCC, a referral cancer center, where only clinically relevant cases were sequenced during the course of clinical care. Clinical indications for sequencing are typically to assist in finding actionable mutations/biomarkers in patients with advanced cancer or to enroll patients on clinical trials, which also favors sequencing of patients with advanced disease. Lawrence et. al. sequenced cases experimentally with no selection bias for high-stage/high-grade cases, and likely reflects the general population of panNETs.

In our cohort of tumors, those with isolated chromosome 11 loss and co-occurring MEN1 mutations without DAXX or ATRX alterations (the mpNET2 group) have a strikingly good overall survival. The lower number of these mpNET2 tumors (n=9) somewhat limits power for survival comparisons. Nonetheless, mpNET2 tumors have unequivocally improved survival compared to mpNET3 tumors, and Kaplan-Meier curves show improved disease-specific survival over mpNET1 tumors as well. Furthermore, chromosomal group remained a significant prognostic factor in multivariate modeling correcting for tumor stage and grade.

Although mpNET2 cases, with MEN1 mutations but without DAXX or ATRX have the best survival, interestingly, MEN1 mutations without DAXX or ATRX alterations and without the “isolated” chromosome 11 loss have a much poorer prognosis (a sub-group of the mpNET3 group). This is remarkable, as it highlights the importance of taking chromosome 11 loss into account in conjunction with the mutational profile. This will be of some importance as NGS becomes more routine clinically. mpNET1 cases, the classic cases with recurrent losses of chromosomes and co-occurring DAXX or ATRX alterations (with or without MEN1 mutations), shows a slowly progressive, intermediate, expected disease course; mpNET1 does worse with respect to the indolent mpNET2 cases but overall survival is better than the mpNET3 group. Although, it is worth noting that there are twice the number of disease-free patients in mpNET3 compared to mpNET1.

As mentioned, the primary limitation of our data is in its statistical power due to the small number of mpNET2 cases. This is almost certainly the result of the skewed collection bias specifically favoring bad actors. Given the large proportion of panNETs with isolated chromosome 11 loss seen in the unbiased index study (40% in Lawrence et. al.), our low inclusion rate (4% mpNET2) may be a limitation but may also provide a useful data point. The low case numbers in this cohort provides additional evidence that mpNET2 are truly indolent tumors.

This is the largest single study looking at the correlation between chromosomal loss by NGS and overall survival in patients with PanNETs. As mentioned, the chromosomal groups do correlate with grade, thus assessing aneuploidies using molecular or cytogenetic techniques may be most useful in grade 2 cases without DAXX or ATRX loss as determined by molecular or immunohistochemical techniques. Figure 3 is adapted from Lawrence et. al. and it diagrams a possible workflow to help stratify pancreatic neuroendocrine tumors without the requirement for NGS. The clinical relevance of implementation of such a novel algorithm is an area for future study.

Figure 3: Genomic Features of mpNET1, mpNET2 and mpNET3 and Potential Non-NGS Workflow.

This figure is adapted from Lawrence et. al. Figure 6. At left is a schematic depiction of cohorts. Chromosomal aneuploidy was used to separate the three PanNET subgroups. These aneuploidies separated very well with single gene mutations or focal losses in MEN1, DAXX, and ATRX and with overall survival in agreement with the known literature and index paper. mpNET1 is characterized by recurrent 10 chromosome monosomy, defined as an "8+8" signature in this paper. This chromosomal group had ATRX and DAXX mutations in 97% of samples and were often with MEN1 mutations, but not always. Their clinical outcome was a predictable slowly progressive course. mpNET2 is characterized by loss of chromosome 11 in relative isolation in conjunction with MEN1 mutations. Their clinical course is extremely indolent. mpNET3 is our 'wastebasket group' with variable aneuploidies and variable mutations. A not insignificant subset had MEN1 mutations (12%) but this group rarely had DAXX and ATRX mutations. Statistically our mpNET3 cases had worse overall and disease-specific survival compared to mpNET1. However, given the collection bias for our samples, we are not confident that this is generalizable to the broad population. Additionally, twice the number of mpNET3 patients are disease free compared to mpNET1. Given these limitations we have labeled this group as having "variable" outcome, specifically noting that a subset behave very poorly, more poorly than mpNET1. At right, a possible workflow, that does not require NGS, is illustrated. Low-grade pancreatic neuroendocrine tumors (notably grade 2 – which were present in all three mpNET groups) may benefit most from additional immunohistochemical stains and FISH analysis, as chromosome group does stratify with grade. Sensitivity and specificity of ATRX IHC in detecting ATRX mutations in panNET is still an area of required investigation. Additionally, the exact FISH probes and number of FISH probes required to confidently call isolated loss of chromosome 11 is an area of necessary future study.

Supplementary Material

Supplemental Data File 1

Supplemental Table 1: Six patients with discrepant subjective copy number changes (visual fingerprint) and objective qualifications (8+8 score).

Supplemental Data File 2

Supplemental Table 2: Summary of mutational profiles for mpNET3.