Unsupervised representation learning of chromatin images identifies changes in cell state and tissue organization in DCIS.

Ductal carcinoma in situ (DCIS) is a pre-invasive tumor that can progress to invasive breast cancer, a leading cause of cancer death. We generate a large-scale tissue microarray dataset of chromatin images, from 560 samples from 122 female patients in 3 disease stages and 11 phenotypic categories. Using representation learning on chromatin images alone, without multiplexed staining or high-throughput sequencing, we identify eight morphological cell states and tissue features marking DCIS. All cell states are observed in all disease stages with different proportions, indicating that cell states enriched in invasive cancer exist in small fractions in normal breast tissue. Tissue-level analysis reveals significant changes in the spatial organization of cell states across disease stages, which is predictive of disease stage and phenotypic category. Taken together, we show that chromatin imaging represents a powerful measure of cell state and disease stage of DCIS, providing a simple and effective tumor biomarker.

Germinal Center Dark Zone harbors ATR-dependent determinants of T-cell exclusion that are also identified in aggressive lymphoma.

The germinal center (GC) dark zone (DZ) and light zone (LZ) regions spatially separate expansion and diversification from selection of antigen-specific B-cells to ensure antibody affinity maturation and B cell memory. The DZ and LZ differ significantly in their immune composition despite the lack of a physical barrier, yet the determinants of this polarization are poorly understood. This study provides novel insights into signals controlling asymmetric T-cell distribution between DZ and LZ regions. We identify spatially-resolved DNA damage response and chromatin compaction molecular features that underlie DZ T-cell exclusion. The DZ spatial transcriptional signature linked to T-cell immune evasion clustered aggressive Diffuse Large B-cell Lymphomas (DLBCL) for differential T cell infiltration. We reveal the dependence of the DZ transcriptional core signature on the ATR kinase and dissect its role in restraining inflammatory responses contributing to establishing an immune-repulsive imprint in DLBCL. These insights may guide ATR-focused treatment strategies bolstering immunotherapy in tumors marked by DZ transcriptional and chromatin-associated features.

Imaging and AI based chromatin biomarkers for diagnosis and therapy evaluation from liquid biopsies.

Multiple genomic and proteomic studies have suggested that peripheral blood mononuclear cells (PBMCs) respond to tumor secretomes and thus could provide possible avenues for tumor prognosis and treatment evaluation. We hypothesized that the chromatin organization of PBMCs obtained from liquid biopsies, which integrates secretome signals with gene expression programs, provides efficient biomarkers to characterize tumor signals and the efficacy of proton therapy in tumor patients. Here, we show that chromatin imaging of PBMCs combined with machine learning methods provides such robust and predictive chromatin biomarkers. We show that such chromatin biomarkers enable the classification of 10 healthy and 10 pan-tumor patients. Furthermore, we extended our pipeline to assess the tumor types and states of 30 tumor patients undergoing (proton) radiation therapy. We show that our pipeline can thereby accurately distinguish between three tumor groups with up to 89% accuracy and enables the monitoring of the treatment effects. Collectively, we show the potential of chromatin biomarkers for cancer diagnostics and therapy evaluation.

Detecting radio- and chemoresistant cells in 3D cancer co-cultures using chromatin biomarkers.

The heterogenous treatment response of tumor cells limits the effectiveness of cancer therapy. While this heterogeneity has been linked to cell-to-cell variability within the complex tumor microenvironment, a quantitative biomarker that identifies and characterizes treatment-resistant cell populations is still missing. Herein, we use chromatin organization as a cost-efficient readout of the cells' states to identify subpopulations that exhibit distinct responses to radiotherapy. To this end, we developed a 3D co-culture model of cancer spheroids and patient-derived fibroblasts treated with radiotherapy. Using the model we identified treatment-resistant cells that bypassed DNA damage checkpoints and exhibited an aggressive growth phenotype. Importantly, these cells featured more condensed chromatin which primed them for treatment evasion, as inhibiting chromatin condensation and DNA damage repair mechanisms improved the efficacy of not only radio- but also chemotherapy. Collectively, our work shows the potential of using chromatin organization to cost-effectively study the heterogeneous treatment susceptibility of cells and guide therapeutic design.

Self-supervised representation learning for surgical activity recognition.

PURPOSE: Virtual reality-based simulators have the potential to become an essential part of surgical education. To make full use of this potential, they must be able to automatically recognize activities performed by users and assess those. Since annotations of trajectories by human experts are expensive, there is a need for methods that can learn to recognize surgical activities in a data-efficient way. METHODS: We use self-supervised training of deep encoder-decoder architectures to learn representations of surgical trajectories from video data. These representations allow for semi-automatic extraction of features that capture information about semantically important events in the trajectories. Such features are processed as inputs of an unsupervised surgical activity recognition pipeline. RESULTS: Our experiments document that the performance of hidden semi-Markov models used for recognizing activities in a simulated myomectomy scenario benefits from using features extracted from representations learned while training a deep encoder-decoder network on the task of predicting the remaining surgery progress. CONCLUSION: Our work is an important first step in the direction of making efficient use of features obtained from deep representation learning for surgical activity recognition in settings where only a small fraction of the existing data is annotated by human domain experts and where those annotations are potentially incomplete.