ObjectiveComplex phenotypes captured on histological slides represent the biological processes at play in individual cancers, but the link to underlying molecular classification has not been clarified or systematised. In colorectal cancer (CRC), histological grading is a poor predictor of disease progression, and consensus molecular subtypes (CMSs) cannot be distinguished without gene expression profiling. We hypothesise that image analysis is a cost-effective tool to associate complex features of tissue organisation with molecular and outcome data and to resolve unclassifiable or heterogeneous cases. In this study, we present an image-based approach to predict CRC CMS from standard H&E sections using deep learning.DesignTraining and evaluation of a neural network were performed using a total of n=1206 tissue sections with comprehensive multi-omic data from three independent datasets (training on FOCUS trial, n=278 patients; test on rectal cancer biopsies, GRAMPIAN cohort, n=144 patients; and The Cancer Genome Atlas (TCGA), n=430 patients). Ground truth CMS calls were ascertained by matching random forest and single sample predictions from CMS classifier.ResultsImage-based CMS (imCMS) accurately classified slides in unseen datasets from TCGA (n=431 slides, AUC)=0.84) and rectal cancer biopsies (n=265 slides, AUC=0.85). imCMS spatially resolved intratumoural heterogeneity and provided secondary calls correlating with bioinformatic prediction from molecular data. imCMS classified samples previously unclassifiable by RNA expression profiling, reproduced the expected correlations with genomic and epigenetic alterations and showed similar prognostic associations as transcriptomic CMS.ConclusionThis study shows that a prediction of RNA expression classifiers can be made from H&E images, opening the door to simple, cheap and reliable biological stratification within routine workflows.

BackgroundThe regulation of specific target genes by transcription factors is central to our understanding of gene network control in developmental and physiological processes yet how target specificity is achieved is still poorly understood. This is well illustrated by the Hox family of transcription factors as their limited in vitro DNA-binding specificity contrasts with their clear in vivo functional specificity.ResultsWe generated genome-wide binding profiles for three Hox proteins, Ubx, Abd-A and Abd-B, following transient expression in Drosophila Kc167 cells, revealing clear target specificity and a striking influence of chromatin accessibility. In the absence of the TALE class homeodomain cofactors Exd and Hth, Ubx and Abd-A bind at a very similar set of target sites in accessible chromatin, whereas Abd-B binds at an additional specific set of targets. Provision of Hox cofactors Exd and Hth considerably modifies the Ubx genome-wide binding profile enabling Ubx to bind at an additional novel set of targets. Both the Abd-B specific targets and the cofactor-dependent Ubx targets are in chromatin that is relatively DNase1 inaccessible prior to the expression of Hox proteins/Hox cofactors.ConclusionsOur experiments demonstrate a strong role for chromatin accessibility in Hox protein binding and suggest that Hox protein competition with nucleosomes has a major role in Hox protein target specificity in vivo.Electronic supplementary materialThe online version of this article (doi:10.1186/s13072-015-0049-x) contains supplementary material, which is available to authorized users.

Chromothripsis and kataegis are frequently observed in cancer and may arise from telomere crisis, a period of genome instability during tumorigenesis when depletion of the telomere reserve generates unstable dicentric chromosomes 1 – 5 . Here we examine the mechanism underlying chromothripsis and kataegis using an in vitro telomere crisis model. We show that the cytoplasmic exonuclease TREX1, which promotes the resolution of dicentric chromosomes 4 , plays a prominent role in chromothriptic fragmentation. In absence of TREX1, the genome alterations induced by telomere crisis primarily involve Breakage-Fusion-Bridge cycles and simple genome rearrangements rather than chromothripsis. Furthermore, we show that the kataegis observed at chromothriptic breakpoints is the consequence of cytosine deamination by APOBEC3B. These data reveal that chromothripsis and kataegis arise from a combination of nucleolytic processing by TREX1 and cytosine editing by APOBEC3B.

PURPOSE Precision medicine trials in glioblastoma (GBM) are often conducted at tumor recurrence. However, second surgeries for recurrent GBM are not routinely performed, and therefore, molecular data for trial inclusion are predominantly derived from the primary sample. This study aims to establish whether molecular targets change during tumor progression and, if so, whether this affects precision medicine trial design. MATERIALS AND METHODS We collected 186 pairs of primary-recurrent GBM samples from patients receiving chemoradiotherapy with temozolomide and sequenced approximately 300 cancer genes. MGMT, TERT, and EGFRvIII status was individually determined. RESULTS The molecular profile of our cohort was identical to that of other GBM cohorts ( IDH wild-type [WT], 95%; EGFR amplified, approximately 50%), indicating that patients amenable to second surgery do not represent a specific molecular subtype. Molecular events in IDH WT GBMs were stable in approximately 80% of events, but changes in mutation status were observed for all examined genes (range, approximately 90% and 60% for TERT and EGFR mutations, respectively), and such changes strongly affected targeted trial size and design. A similar pattern of GBM driver instability was observed within MGMT promoter–methylated tumors. MGMT promoter methylation status remained prognostic at tumor recurrence. The observation that hypermutation at GBM recurrence was rare (8%) and not correlated with outcome was relevant for immunotherapy-based treatments. CONCLUSION This large cohort of matched primary and recurrent IDH WT tumors establishes the frequency of GBM driver instability after chemoradiotherapy with temozolomide. This allows per gene or pathway calculation of trial size at tumor recurrence, using molecular data of the primary tumor only. We also identify genes for which repeat surgery is necessary because of low mutation retention rate.

Patient-derived tumor xenograft (PDX) mouse models are a versatile oncology research platform for studying tumor biology and for testing chemotherapeutic approaches tailored to genomic characteristics of individual patients’ tumors. PDX models are generated and distributed by a diverse group of academic labs, multi-institution consortia and contract research organizations. The distributed nature of PDX repositories and the use of different metadata standards for describing model characteristics presents a significant challenge to identifying PDX models relevant to specific cancer research questions. The Jackson Laboratory and EMBL-EBI are addressing these challenges by co-developing PDX Finder, a comprehensive open global catalog of PDX models and their associated datasets. Within PDX Finder, model attributes are harmonized and integrated using a previously developed community minimal information standard to support consistent searching across the originating resources. Links to repositories are provided from the PDX Finder search results to facilitate model acquisition and/or collaboration. The PDX Finder resource currently contains information for 1985 PDX models of diverse cancers including those from large resources such as the Patient-Derived Models Repository, PDXNet and EurOPDX. Individuals or organizations that generate and distribute PDXs are invited to increase the ‘findability’ of their models by participating in the PDX Finder initiative at www.pdxfinder.org.