Genome-wide loss of heterozygosity predicts aggressive, treatment-refractory behavior in pituitary neuroendocrine tumors

Abstract

Pituitary neuroendocrine tumors (PitNETs) exhibiting aggressive, treatment-refractory behavior are the rare subset that progress after surgery, conventional medical therapies, and an initial course of radiation and are characterized by unrelenting growth and/or metastatic dissemination. Two groups of patients with PitNETs were sequenced: a prospective group of patients (n = 66) who consented to sequencing prior to surgery and a retrospective group (n = 26) comprised of aggressive/higher risk PitNETs. A higher mutational burden and fraction of loss of heterozygosity (LOH) was found in the aggressive, treatment-refractory PitNETs compared to the benign tumors (p = 1.3 × 10^-10 and p = 8.5 × 10^-9, respectively). Within the corticotroph lineage, a characteristic pattern of recurrent chromosomal LOH in 12 specific chromosomes was associated with treatment-refractoriness (occurring in 11 of 14 treatment-refractory versus 1 of 14 benign corticotroph PitNETs, p = 1.7 × 10^-4). Across the cohort, a higher fraction of LOH was identified in tumors with TP53 mutations (p = 3.3 × 10^-8). A machine learning approach identified loss of heterozygosity as the most predictive variable for aggressive, treatment-refractory behavior, outperforming the most common gene-level alteration, TP53, with an accuracy of 0.88 (95% CI: 0.70-0.96). Aggressive, treatment-refractory PitNETs are characterized by significant aneuploidy due to widespread chromosomal LOH, most prominently in the corticotroph tumors. This LOH predicts treatment-refractoriness with high accuracy and represents a novel biomarker for this poorly defined PitNET category.

Keywords: Aggressive pituitary tumor; Pituitary carcinoma; Pituitary neuroendocrine tumor; Treatment-refractory pituitary tumor.

Fig. 1

Genomic alterations and clinical characteristics of patients with aggressive, treatment-refractory and benign PitNETs. (a) Schematic diagram showing relationship between the retrospective and prospective groups and clinical behavior. (b) Summary of the tumor lineages in the treatment-refractory and benign subsets and metastatic disease status at the time of the data freeze (treatment-refractory subset). (c) Oncoprint summarizing recurrently altered driver genes in treatment-refractory (left: n = 23) and benign (right: n = 69) PitNETs. Patients who had multiple tumor samples are represented by the union of alterations among all samples. Patient demographics and clinicopathologic features are on the top, followed by common genetic alterations (frequencies are shown on the right in percent per clinical category). USP8 status is reported if whole-exome recapture was performed

Fig. 2

Genomic instability due to loss of heterozygosity (LOH) is frequent in aggressive, treatment-refractory PitNETs. (a) The heatmap shows LOH status for individual chromosomes with each blue box demonstrating LOH covering at least 75% of the given chromosome and reflects the first sequenced resection. Top tracks report clinicopathologic features such as lineage and treatment, again for the first sequenced resection. Bottom tracks report features of genomic instability, including whole-genome duplication status (WGD), fraction of genome altered (FGA), and fraction of LOH based on median values (median FGA = 0.22, median fraction of LOH = 0.03). Copy-number alteration (CNA) data are unavailable for two patients as annotated. (b) Cohort segregated into treatment-refractory and benign, displaying the fraction of LOH by lineage. Two-sided Wilcoxon rank sum tests are shown. (c) The percentage of samples with classifications of LOH: LOH-H (fraction LOH higher than the median = 0.03) with or without rcLOH (involving chromosomes 1, 2, 3, 4, 6, 10, 11, 15, 17, 18, 21, and 22), or LOH-L (fraction of LOH < 0.03)

Fig. 3

Hypodiploidy is identified by sequencing and fluorescence in situ hybridization. (a) Integer total copy number of patient TR-9 estimated by FACETS, displaying extensive LOH including recurrently lost chromosomes across the treatment-refractory cohort. Chromosomes 1, 9, and 19 highlighted to compare to FISH. (b) Fluorescence in situ hybridization (FISH) analysis on patient TR-9. Probes against 1p, 1q, 19p, 19q, and CEP9 demonstrate that loss of one copy of chromosome 1, 9, and a segment of 19 including 19p validating the FACETS integer copy-number estimates. (c) and (d) provides schematic timelines, which outline the clinical course, in a patient with a treatment-refractory lactotroph (TR-15) and a treatment-refractory corticotroph (TR-2) PitNET, respectively. Clone trees from the two patients highlight mutations, LOH and signatures that are either shared or unique to the primary tumor and a metastasis

Fig. 4

Performance of random forest classifier. (a) On the left, the number of times a feature is the root of a regression tree is plotted against the average depth of the first node for that feature (mean minimal depth) with the size of each data point representing total number of nodes that utilize the feature for splitting; the fraction of LOH is the top feature for partitioning aggressive, treatment-refractory behavior. Fraction of LOH is also the top feature for predicting aggressive, treatment-refractory behavior when ranking each feature by the mean decrease in accuracy (middle) and Gini index (right). (b) Plot showing the accuracy and precision of our model on the test dataset with prediction on the x-axis and the reference truth on the y-axis