Faculty Bibliography

BACKGROUND: Tumor heterogeneity presents a major challenge to cancer diagnosis and treatment. In addition to interpatient tumor variability, intratumoral heterogeneity characterized by distinct molecular and phenotypic profiles is increasingly recognized as a major cause of therapy resistance and cancer recurrence. Because DNA methylation patterns are largely responsible for determining cell-type-specific functioning, we hypothesized that distinct DNA methylation and proteomic alterations could be identified in histologically-defined invasive and proliferative tumor cell populations in human isocitrate dehydrogenase 1 (IDH1)- mutated and wild-type glioblastoma (GBM).
METHOD(S): Formalin-fixed paraffin-embedded tissue sections of human adult IDH1-mutated and wild-type GBM were laser-microdissected (LM) into perinecrotic pseudopalisading tumor cells (PPCs), non-pseudopalisading tumor core cells (NPPCs), invasive subpial spread (SPS) and perivascular satellitosis tumor cells and brain adjacent to tumor cells prior to analysis and compared to non-microdissected tumor (NMT) and/or germline DNA. Genomewide DNA methylation and chromosomal copy numbers were determined with Infinium MethylationEPIC 850K BeadChip and intratumoral DNA methylation patterns compared by unsupervised hierarchical clustering. Label-free quantitative liquid chromatography-mass spectrometry of proteins was performed and proteins differentially expressed across LM areas subjected to pathway enrichment analysis.
RESULT(S): Unsupervised hierarchical classification of DNA methylation patterns for each LM area and NMT demonstrated remarkable clustering for all patients, based on methylation probe and methylated gene patterns. Proteomics analysis showed upregulation of hypoxia-inducible factor-1 inducible proteins in hypoxic PPCs. Out of 1819 proteins quantified, 5 were overexpressed and 9 underexpressed more than 10-fold in SPS compared with NPPCs and associated with alterations in metabolism, transport, extracellular matrix and apoptosis. Correlation of protein expression and DNA methylation patterns was noted.
CONCLUSION(S): Compared to NPPCs, SPS cells migrating toward the invasive edge share a relatively consistent epigenetic and proteomic signature, suggesting potentially targetable common mechanism(s) of invasion shared among GBM

In the version of this Letter originally published, the competing interests statement was missing. The authors declare no competing interests; this statement has now been added in all online versions of the Letter.

BACKGROUND AND PURPOSE/OBJECTIVE:Few studies have shown MR imaging features and ADC correlating with molecular markers and survival in patients with glioma. Our purpose was to correlate MR imaging features and ADC with molecular subtyping and survival in adult diffuse gliomas. MATERIALS AND METHODS/METHODS:promoter methylation, and overall survival. RESULTS:wild-type glioma. Other MR imaging features were not statistically significant predictors of survival. CONCLUSIONS:wild-type gliomas.

Visual inspection of histopathology slides is one of the main methods used by pathologists to assess the stage, type and subtype of lung tumors. Adenocarcinoma (LUAD) and squamous cell carcinoma (LUSC) are the most prevalent subtypes of lung cancer, and their distinction requires visual inspection by an experienced pathologist. In this study, we trained a deep convolutional neural network (inception v3) on whole-slide images obtained from The Cancer Genome Atlas to accurately and automatically classify them into LUAD, LUSC or normal lung tissue. The performance of our method is comparable to that of pathologists, with an average area under the curve (AUC) of 0.97. Our model was validated on independent datasets of frozen tissues, formalin-fixed paraffin-embedded tissues and biopsies. Furthermore, we trained the network to predict the ten most commonly mutated genes in LUAD. We found that six of them-STK11, EGFR, FAT1, SETBP1, KRAS and TP53-can be predicted from pathology images, with AUCs from 0.733 to 0.856 as measured on a held-out population. These findings suggest that deep-learning models can assist pathologists in the detection of cancer subtype or gene mutations. Our approach can be applied to any cancer type, and the code is available at https://github.com/ncoudray/DeepPATH .

A methionine substitution at lysine-27 on histone H3 variants (H3K27M) characterizes ~80% of diffuse intrinsic pontine gliomas (DIPG) and inhibits polycomb repressive complex 2 (PRC2) in a dominant-negative fashion. Yet, the mechanisms for this inhibition and abnormal epigenomic landscape have not been resolved. Using quantitative proteomics, we discovered that robust PRC2 inhibition requires levels of H3K27M greatly exceeding those of PRC2, seen in DIPG. While PRC2 inhibition requires interaction with H3K27M, we found that this interaction on chromatin is transient, with PRC2 largely being released from H3K27M. Unexpectedly, inhibition persisted even after PRC2 dissociated from H3K27M-containing chromatin, suggesting a lasting impact on PRC2. Furthermore, allosterically activated PRC2 is particularly sensitive to H3K27M, leading to the failure to spread H3K27me from PRC2 recruitment sites and consequently abrogating PRC2's ability to establish H3K27me2-3 repressive chromatin domains. In turn, levels of polycomb antagonists such as H3K36me2 are elevated, suggesting a more global, downstream effect on the epigenome. Together, these findings reveal the conditions required for H3K27M-mediated PRC2 inhibition and reconcile seemingly paradoxical effects of H3K27M on PRC2 recruitment and activity.

Visual inspection of histopathology slides is one of the main methods used by pathologists to assess the stage, type and subtype of lung tumors. Adenocarcinoma (LUAD) and squamous cell carcinoma (LUSC) are the most prevalent subtypes of lung cancer, and their distinction requires visual inspection by an experienced pathologist. In this study, we trained a deep convolutional neural network (inception v3) on whole-slide images obtained from The Cancer Genome Atlas to accurately and automatically classify them into LUAD, LUSC or normal lung tissue. The performance of our method is comparable to that of pathologists, with an average area under the curve (AUC) of 0.97. Our model was validated on independent datasets of frozen tissues, formalin-fixed paraffin-embedded tissues and biopsies. Furthermore, we trained the network to predict the ten most commonly mutated genes in LUAD. We found that six of them-STK11, EGFR, FAT1, SETBP1, KRAS and TP53-can be predicted from pathology images, with AUCs from 0.733 to 0.856 as measured on a held-out population. These findings suggest that deep-learning models can assist pathologists in the detection of cancer subtype or gene mutations. Our approach can be applied to any cancer type, and the code is available at https://github.com/ncoudray/DeepPATH .

Patients with DICER1 predisposition syndrome have an increased risk to develop pleuropulmonary blastoma, cystic nephroma, embryonal rhabdomyosarcoma, and several other rare tumor entities. In this study, we identified 22 primary intracranial sarcomas, including 18 in pediatric patients, with a distinct methylation signature detected by array-based DNA-methylation profiling. In addition, two uterine rhabdomyosarcomas sharing identical features were identified. Gene panel sequencing of the 22 intracranial sarcomas revealed the almost unifying feature of DICER1 hotspot mutations (21/22; 95%) and a high frequency of co-occurring TP53 mutations (12/22; 55%). In addition, 17/22 (77%) sarcomas exhibited alterations in the mitogen-activated protein kinase pathway, most frequently affecting the mutational hotspots of KRAS (8/22; 36%) and mutations or deletions of NF1 (7/22; 32%), followed by mutations of FGFR4 (2/22; 9%), NRAS (2/22; 9%), and amplification of EGFR (1/22; 5%). A germline DICER1 mutation was detected in two of five cases with constitutional DNA available. Notably, none of the patients showed evidence of a cancer-related syndrome at the time of diagnosis. In contrast to the genetic findings, the morphological features of these tumors were less distinctive, although rhabdomyoblasts or rhabdomyoblast-like cells could retrospectively be detected in all cases. The identified combination of genetic events indicates a relationship between the intracranial tumors analyzed and DICER1 predisposition syndrome-associated sarcomas such as embryonal rhabdomyosarcoma or the recently described group of anaplastic sarcomas of the kidney. However, the intracranial tumors in our series were initially interpreted to represent various tumor types, but rhabdomyosarcoma was not among the typical differential diagnoses considered. Given the rarity of intracranial sarcomas, this molecularly clearly defined group comprises a considerable fraction thereof. We therefore propose the designation "spindle cell sarcoma with rhabdomyosarcoma-like features, DICER1 mutant" for this intriguing group.

Pineoblastoma is a rare and highly aggressive brain cancer of childhood, histologically belonging to the spectrum of primitive neuroectodermal tumors. Patients with germline mutations in DICER1, a ribonuclease involved in microRNA processing, have increased risk of pineoblastoma, but genetic drivers of sporadic pineoblastoma remain unknown. Here, we analyzed pediatric and adult pineoblastoma samples (n = 23) using a combination of genome-wide DNA methylation profiling and whole-exome sequencing or whole-genome sequencing. Pediatric and adult pineoblastomas showed distinct methylation profiles, the latter clustering with lower-grade pineal tumors and normal pineal gland. Recurrent variants were found in genes involved in PKA- and NF-κB signaling, as well as in chromatin remodeling genes. We identified recurrent homozygous deletions of DROSHA, acting upstream of DICER1 in microRNA processing, and a novel microduplication involving chromosomal region 1q21 containing PDE4DIP (myomegalin), comprising the ancient DUF1220 protein domain. Expresion of PDE4DIP and DUF1220 proteins was present exclusively in pineoblastoma with PDE4DIP gain.

IDH-mutant astrocytomas are significantly less aggressive than their IDH-wildtype counterparts. We analyzed The Cancer Genome Atlas dataset (TCGA) and identified a small group of IDH-mutant, WHO grade II-III astrocytomas (n = 14) with an unexpectedly poor prognosis characterized by a rapid progression to glioblastoma and death within 3 years of the initial diagnosis. Compared with IDH-mutant tumors with the typical, extended progression-free survival in a control group of age-similar patients, the tumors in the rapidly progressing group were characterized by a markedly increased level of overall copy number alterations ([CNA]; p = 0.006). In contrast, the mutation load was similar, as was the methylation pattern, being consistent with IDH-mutant astrocytoma. Two of the gliomas (14%) in the rapidly progressing, IDH-mutant group but none of the other grade II-III gliomas in the TCGA (n = 283) had pathogenic mutations in genes (FANCB and APC) associated with maintaining chromosomal stability. These results suggest that chromosomal instability can negate the beneficial effect of IDH mutations in WHO II-III astrocytomas. The mechanism of the increased CNA is unknown but in some cases appears to be due to mutations in genes with a role in chromosomal stability. Increased CNA could serve as a biomarker for tumors at risk for rapid progression.