Morphology-based molecular classification of spinal cord ependymomas using deep neural networks.

Based on DNA-methylation, ependymomas growing in the spinal cord comprise two major molecular types termed spinal (SP-EPN) and myxopapillary ependymomas (MPE(-A/B)), which differ with respect to their clinical features and prognosis. Due to the existing discrepancy between histomorphogical diagnoses and classification using methylation data, we asked whether deep neural networks can predict the DNA methylation class of spinal cord ependymomas from hematoxylin and eosin stained whole-slide images. Using explainable AI, we further aimed to prospectively improve the consistency of histology-based diagnoses with DNA methylation profiling by identifying and quantifying distinct morphological patterns of these molecular ependymoma types. We assembled a case series of 139 molecularly characterized spinal cord ependymomas (nMPE = 84, nSP-EPN = 55). Self-supervised and weakly-supervised neural networks were used for classification. We employed attention analysis and supervised machine-learning methods for the discovery and quantification of morphological features and their correlation to the diagnoses of experienced neuropathologists. Our best performing model predicted the DNA methylation class with 98% test accuracy and used self-supervised learning to outperform pretrained encoder-networks (86% test accuracy). In contrast, the diagnoses of neuropathologists matched the DNA methylation class in only 83% of cases. Domain-adaptation techniques improved model generalization to an external validation cohort by up to 22%. Statistically significant morphological features were identified per molecular type and quantitatively correlated to human diagnoses. The approach was extended to recently defined subtypes of myxopapillary ependymomas (MPE-(A/B), 80% test accuracy). In summary, we demonstrated the accurate prediction of the DNA methylation class of spinal cord ependymomas (SP-EPN, MPE(-A/B)) using hematoxylin and eosin stained whole-slide images. Our approach may prospectively serve as a supplementary resource for integrated diagnostics and may even help to establish a standardized, high-quality level of histology-based diagnostics across institutions-in particular in low-income countries, where expensive DNA-methylation analyses may not be readily available.

Integrated proteomics spotlight the proteasome as a therapeutic vulnerability in embryonal tumors with multilayered rosettes.

BACKGROUND: Embryonal tumors with multilayered rosettes (ETMR) are rare malignant embryonal brain tumors. The prognosis of ETMR is poor and novel therapeutic approaches are desperately needed. Comprehension of ETMR tumor biology is currently based on only few previous molecular studies, which mainly focused on the analyses of nucleic acids. In this study, we explored integrated ETMR proteomics. METHODS: Using mass spectrometry, proteome data were acquired from 16 ETMR and the ETMR cell line BT183. Proteome data were integrated with case-matched global DNA methylation data, publicly available transcriptome data, and proteome data of further embryonal and pediatric brain tumors. RESULTS: Proteome-based cluster analyses grouped ETMR samples according to histomorphology, separating neuropil-rich tumors with neuronal signatures from primitive tumors with signatures relating to stemness and chromosome organization. Integrated proteomics showcased that ETMR and BT183 cells harbor proteasome regulatory proteins in abundance, implicating their strong dependency on the proteasome machinery to safeguard proteostasis. Indeed, in vitro assays using BT183 highlighted that ETMR tumor cells are highly vulnerable toward treatment with the CNS penetrant proteasome inhibitor Marizomib. CONCLUSIONS: In summary, histomorphology stipulates the proteome signatures of ETMR, and proteasome regulatory proteins are pervasively abundant in these tumors. As validated in vitro, proteasome inhibition poses a promising therapeutic option in ETMR.

Direct 3D Sampling of the Embryonic Mouse Head: Layer-wise Nanosecond Infrared Laser (NIRL) Ablation from Scalp to Cortex for Spatially Resolved Proteomics.

Common workflows in bottom-up proteomics require homogenization of tissue samples to gain access to the biomolecules within the cells. The homogenized tissue samples often contain many different cell types, thereby representing an average of the natural proteome composition, and rare cell types are not sufficiently represented. To overcome this problem, small-volume sampling and spatial resolution are needed to maintain a better representation of the sample composition and their proteome signatures. Using nanosecond infrared laser ablation, the region of interest can be targeted in a three-dimensional (3D) fashion, whereby the spatial information is maintained during the simultaneous process of sampling and homogenization. In this study, we ablated 40 µm thick consecutive layers directly from the scalp through the cortex of embryonic mouse heads and analyzed them by subsequent bottom-up proteomics. Extra- and intracranial ablated layers showed distinct proteome profiles comprising expected cell-specific proteins. Additionally, known cortex markers like SOX2, KI67, NESTIN, and MAP2 showed a layer-specific spatial protein abundance distribution. We propose potential new marker proteins for cortex layers, such as MTA1 and NMRAL1. The obtained data confirm that the new 3D tissue sampling and homogenization method is well suited for investigating the spatial proteome signature of tissue samples in a layerwise manner. Characterization of the proteome composition of embryonic skin and bone structures, meninges, and cortex lamination in situ enables a better understanding of molecular mechanisms of development during embryogenesis and disease pathogenesis.

Robust classification using average correlations as features (ACF).

MOTIVATION: In single-cell transcriptomics and other omics technologies, large fractions of missing values commonly occur. Researchers often either consider only those features that were measured for each instance of their dataset, thereby accepting severe loss of information, or use imputation which can lead to erroneous results. Pairwise metrics allow for imputation-free classification with minimal loss of data. RESULTS: Using pairwise correlations as metric, state-of-the-art approaches to classification would include the K-nearest-neighbor- (KNN) and distribution-based-classification-classifier. Our novel method, termed average correlations as features (ACF), significantly outperforms those approaches by training tunable machine learning models on inter-class and intra-class correlations. Our approach is characterized in simulation studies and its classification performance is demonstrated on real-world datasets from single-cell RNA sequencing and bottom-up proteomics. Furthermore, we demonstrate that variants of our method offer superior flexibility and performance over KNN classifiers and can be used in conjunction with other machine learning methods. In summary, ACF is a flexible method that enables missing value tolerant classification with minimal loss of data.

Spatial molecular profiling of a central nervous system low-grade diffusely infiltrative tumour with INI1 deficiency featuring a high-grade atypical teratoid/rhabdoid tumour component.

We performed spatial epigenetic and transcriptomic analyses of a highly unusual low-grade diffusely infiltrative tumour with INI1 deficiency (CNS LGDIT-INI1), which harboured a high-grade component corresponding to an atypical teratoid/rhabdoid tumour (AT/RT). Methylation profiles of both low-grade and high-grade components yielded high similarity with AT/RTs of the MYC subgroup, whereas RNA expression analyses revealed increased translational activity and MYC pathway activation in the high-grade component. Close follow-up of patients harbouring CNS LGDIT-INI1 is warranted.