Guillaume Thibault

SUMMARY: Neurosignaling is increasingly recognized as a critical factor in cancer progression, where neuronal innervation of primary tumors contributes to the disease’s advancement. This study focuses on segmenting individual axons within the prostate tumor microenvironment, which have been challenging to detect and analyze due to their irregular morphologies. We present a novel deep learning-based approach for the automated segmentation of axons, AxonFinder, leveraging a U-Net model with a ResNet-101 encoder, based on a multiplexed imaging approach. Utilizing a dataset of whole-slide images from low-, intermediate-, and high-risk prostate cancer patients, we manually annotated axons to train our model, achieving significant accuracy in detecting axonal structures that were previously hard to segment. Our analysis includes a comprehensive assessment of axon density and morphological features across different CAPRA-S prostate cancer risk categories, providing insights into the correlation between tumor innervation and cancer progression. Our paper suggests the potential utility of neuronal markers in the prognostic assessment of prostate cancer in aiding the pathologist’s assessment of tumor sections and advancing our understanding of neurosignaling in the tumor microenvironment.

Machine learning approaches have the potential for meaningful impact in the biomedical field. However, there are often challenges unique to biomedical data that prohibits the adoption of these innovations. For example, limited data, data volatility, and data shifts all compromise model robustness and generalizability. Without proper tuning and data management, deploying machine learning models in the presence of unaccounted for corruptions leads to reduced or misleading performance. This study explores techniques to enhance model generalizability through iterative adjustments. Specifically, we investigate a detection tasks using electron microscopy images and compare models trained with different normalization and augmentation techniques. We found that models trained with Group Normalization or texture data augmentation outperform other normalization techniques and classical data augmentation, enabling them to learn more generalized features. These improvements persist even when models are trained and tested on disjoint datasets acquired through diverse data acquisition protocols. Results hold true for transformerand convolution-based detection architectures. The experiments show an impressive 29% boost in average precision, indicating significant enhancements in the model’s generalizibality. This underscores the models’ capacity to effectively adapt to diverse datasets and demonstrates their increased resilience in real-world applications.

Electron microscopy (EM) enables imaging at nanometer resolution and can shed light on how cancer evolves to develop resistance to therapy. Acquiring these images has become a routine task; however, analyzing them is now the bottleneck, as manual structure identification is very time-consuming and can take up to several months for a single sample. Deep learning approaches offer a suitable solution to speed up the analysis. In this work, we present a study of several state-of-the-art deep learning models for the task of segmenting nuclei and nucleoli in volumes from tumor biopsies. We compared previous results obtained with the ResUNet architecture to the more recent UNet++, FracTALResNet, SenFormer, and CEECNet models. In addition, we explored the utilization of unlabeled images through semi-supervised learning with Cross Pseudo Supervision. We have trained and evaluated all of the models on sparse manual labels from three fully annotated in-house datasets that we have made available on demand, demonstrating improvements in terms of 3D Dice score. From the analysis of these results, we drew conclusions on the relative gains of using more complex models, semi-supervised learning as well as next steps for the mitigation of the manual segmentation bottleneck.

Mapping spatial interactions of cancer, immune and stromal cells present novel opportunities for patient stratification and for advancing immunotherapy. While single-cell studies revealed significant molecular heterogeneity in prostate tumors, there is currently no understanding of how immune cell heterogeneity impacts spatial coordination between tumor and stromal cells in localized tumors. Here, we used cyclic immunofluorescent imaging on whole-tissue sections to uncover novel spatial associations between cancer and stromal cells in low- and high-grade prostate tumors and tumor-adjacent normal tissues. Our results provide a spatial map of 699,461 single-cells that show epigenetic and molecular differences in distinct clinical grades. We report unique populations of mast cells that differentially express CD44, CD90 and Granzyme B (GZMB) and demonstrate GZMB+ mast cells are spatially associated with M2 macrophages in prostate tumors. Finally, we uncover recurrent neighborhoods that are primarily driven by androgen receptor positive (AR+) stromal cells and identify transcriptional networks active in AR+ prostate stroma.

The drivers of ductal carcinoma in situ (DCIS) to invasive ductal carcinoma (IDC) transition are poorly understood. Here, we conducted an integrated genomic, transcriptomic, and whole-slide image analysis to evaluate changes in copy-number profiles, mutational profiles, expression, neoantigen load, and topology in 6 cases of matched pure DCIS and recurrent IDC. We demonstrate through combined copy-number and mutational analysis that recurrent IDC can be genetically related to its pure DCIS despite long latency periods and therapeutic interventions. Immune “hot” and “cold” tumors can arise as early as DCIS and are subtype-specific. Topologic analysis showed a similar degree of pan-leukocyte-tumor mixing in both DCIS and IDC but differ when assessing specific immune subpopulations such as CD4 T cells and CD68 macrophages. Tumor-specific copy-number aberrations in MHC-I presentation machinery and losses in 3p, 4q, and 5p are associated with differences in immune signaling in estrogen receptor (ER)-negative IDC. Common oncogenic hotspot mutations in genes including TP53 and PIK3CA are predicted to be neoantigens yet are paradoxically conserved during the DCIS-to-IDC transition, and are associated with differences in immune signaling. We highlight both tumor and immune-specific changes in the transition of pure DCIS to IDC, including genetic changes in tumor cells that may have a role in modulating immune function and assist in immune escape, driving the transition to IDC.Implications:We demonstrate that the in situ to IDC evolutionary bottleneck is shaped by both tumor and immune cells.

Summary of Supplementary Tables Supplementary Table S1. Clinicopathological characteristics of patient cohort Supplementary Table S2. Antibody panel for cyclic IF Supplementary Table S3. Summary of mutations and neoantigen expression in the recurrence cohort (Related to 1C, 6A) Supplementary Table S4. Summary of altered pathways in DCIS/IDC (Fig 1F) Supplementary Table S5. Enriched gene sets between DCIS and IDC (Fig 3A) Supplementary Table S6. Identified chromosomal regions associated with immune changes (Related to Fig. 5) Supplementary Table S7. Mutational, neoantigen and rsSNP frequency of commonly mutated genes in TCGA and Abba cohorts (related to Fig 6C,E)

Summary of Supplementary Tables Supplementary Table S1. Clinicopathological characteristics of patient cohort Supplementary Table S2. Antibody panel for cyclic IF Supplementary Table S3. Summary of mutations and neoantigen expression in the recurrence cohort (Related to 1C, 6A) Supplementary Table S4. Summary of altered pathways in DCIS/IDC (Fig 1F) Supplementary Table S5. Enriched gene sets between DCIS and IDC (Fig 3A) Supplementary Table S6. Identified chromosomal regions associated with immune changes (Related to Fig. 5) Supplementary Table S7. Mutational, neoantigen and rsSNP frequency of commonly mutated genes in TCGA and Abba cohorts (related to Fig 6C,E)

The drivers of ductal carcinoma in situ (DCIS) to invasive ductal carcinoma (IDC) transition are poorly understood. Here, we conducted an integrated genomic, transcriptomic, and whole-slide image analysis to evaluate changes in copy-number profiles, mutational profiles, expression, neoantigen load, and topology in 6 cases of matched pure DCIS and recurrent IDC. We demonstrate through combined copy-number and mutational analysis that recurrent IDC can be genetically related to its pure DCIS despite long latency periods and therapeutic interventions. Immune “hot” and “cold” tumors can arise as early as DCIS and are subtype-specific. Topologic analysis showed a similar degree of pan-leukocyte-tumor mixing in both DCIS and IDC but differ when assessing specific immune subpopulations such as CD4 T cells and CD68 macrophages. Tumor-specific copy-number aberrations in MHC-I presentation machinery and losses in 3p, 4q, and 5p are associated with differences in immune signaling in estrogen receptor (ER)-negative IDC. Common oncogenic hotspot mutations in genes including TP53 and PIK3CA are predicted to be neoantigens yet are paradoxically conserved during the DCIS-to-IDC transition, and are associated with differences in immune signaling. We highlight both tumor and immune-specific changes in the transition of pure DCIS to IDC, including genetic changes in tumor cells that may have a role in modulating immune function and assist in immune escape, driving the transition to IDC.Implications:We demonstrate that the in situ to IDC evolutionary bottleneck is shaped by both tumor and immune cells.

S1. Overview of patient cohort. S2. Whole exome sequencing of matched DCISand IDC transition. S3. Mutationalprofiling of the DCIS-to-IDC transition. S4.Phenotypic immune properties of patient cohort. S5.Immune-related genomicchanges and neoantigen load.