During early post-implantation development of the mouse embryo, the Anterior Visceral Endoderm (AVE) differs from surrounding visceral endoderm (VE) in its migratory behaviour and ability to restrict primitive streak formation to the opposite side of the egg cylinder. In order to characterise the molecular basis for the unique properties of the AVE, we combined single-cell RNA-sequencing of the VE prior to and during AVE migration, with high-resolution imaging, short-term lineage labelling, phosphoproteomics and pharmacological intervention. This revealed the transient nature of the AVE, the emergence of heterogeneities in AVE transcriptional states relative to position of cells, and its prominence in establishing gene expression asymmetries within the spatial constraints of the embryo. We identified a previously unknown requirement of Ephrin- and Semaphorin-signalling for AVE migration. These findings point to a tight coupling of transcriptional state and position in the AVE and reveal molecular heterogeneities underpinning its migratory behaviour and function.

Background: Automated cell-level characterization of the tumor microenvironment (TME) at scale is key to data-driven immuno-oncology. Artificial intelligence (AI)-powered analysis of hematoxylin and eosin (H&E) images scales and has recently been translated into diagnostics. However, robust TME analysis solely based on H&E data is bound by the stain's properties and by manual pathologist annotations, both in number and accuracy. In this study, we quantify the error introduced by pathologists' morphological assessment and mitigate this error by training AI-systems without manual pathologist annotations, using labels determined directly from IHC profiles. Methods: The work was carried out on 239 clinical NSCLC cases. CK-KL1, CD3+CD20, and Mum1 were used for defining carcinoma (CA), lymphocyte (LY), and plasma (PL) cells. For evaluation, representative regions were annotated by 3 trained pathologists. The workflow is based on co-registration of same-section H&E and IHC stained images with single cell precision. Cells were detected in H&E and labelled using rule-based algorithms that incorporated IHC information. This H&E data was used to train neural networks (NN). Results: (A) The inter-rater agreement of pathologists annotating on H&E is increased when information from registered IHC images is provided. (B) The concordance of pathologists on H&E-only compared to on H&E+IHC shows that pathologists miss or misclassify cells with a certain error. (C) NNs trained with IHC-based labels achieve similar performance for cell type classification on H&E as pathologists on H&E. Conclusion: This study demonstrates the value of combining histomorphological and IHC data for improved cell annotation. Our novel workflow provides a quantitative benchmark and facilitates training of accurate AI models for quantitative characterization of tumor and TME from H&E sections. A) Inter-rater agreement by metric, stain, and cell type By cell count, Pearson correlation By cell count, Pearson correlation By cell location, Krippendorff’s alpha By cell location, Krippendorff’s alpha Cell type H&E-only H&E+IHC H&E-only H&E+IHC CA 0.86 0.98 0.43 0.90 LY 0.88 0.99 0.21 0.76 PL 0.77 0.96 0.32 0.87 B) Performance of individual pathologists in H&E Against consensus in H&E+IHC Against own annotations in H&E+IHC Against own annotations in H&E+IHC Cell type By cell count, Pearson correlation By cell location, Precision By cell location, Recall CA 0.84 0.76 0.77 LY 0.78 0.70 0.60 PL 0.76 0.69 0.21 C) NN against annotator H&E+IHC consensus Cell Type By cell count, Pearson correlation CA 0.84 LY 0.92 PL 0.75 Citation Format: Thomas Mrowiec, Sharon Ruane, Simon Schallenberg, Gabriel Dernbach, Rumyana Todorova, Cornelius Böhm, Walter de Back, Blanca Pablos, Roman Schulte-Sasse, Ivana Trajanovska, Adelaida Creosteanu, Emil Barbuta, Marcus Otte, Christian Ihling, Hans Juergen Grote, Juergen Scheuenpflug, Viktor Matyas, Maximilian Alber, Frederick Klauschen. Immunohistochemistry-informed AI systems for improved characterization of tumor-microenvironment in clinical non-small cell lung cancer H&E samples [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 457.

Background: Mitotic rate is a readout routinely used for characterization of tumor samples. Standard methods to quantify cell division include manual pathologist counts based on hematoxylin and eosin (H&E) staining and either manual or automated assessment of immunohistochemistry (IHC) staining using phospho-histone H3 (pHH3). These suffer the drawbacks of high inter-pathologist variability in the case of H&E assessment and time inefficiency and false positive calls in case of pHH3 staining. Aiming to overcome these issues, we developed a mitosis detection model based on H&E-stained tissue sections alone. Methods: To develop and evaluate the model, we used 1032 H&E-stained tissue sections (156 used for model training) of pre-clinical pancreas cancer xenografts originating from mice that have undergone a series of experiments conducted to examine the pharmacodynamic effect of several anti-cancer protocols. We trained (i) a tissue segmentation model (segmenting tissue regions into ‘carcinoma’, ‘stroma’, ‘necrosis’, and ‘other’) and (ii) a segmentation model for pixel-level mitosis detection (segmenting ‘mitosis’ vs. ‘non-mitosis’). Regions predicted as mitosis were post-processed to represent individual dividing cells. The tissue segmentation model served as a filter to predict mitotic rate for carcinoma tissue areas only, which has not been accounted for in previous AI-based methods for mitotic rate prediction on H&E tissue. To evaluate the model on detecting mitotic events, the model was compared against a 5-pathologist consensus of mitotic count. The mitotic rates predicted by the model were used to infer differences between treatment groups (various treatments and dosages vs. control). Results: The mitosis detection model for quantifying rates of cell division in carcinoma regions of H&E-stained tissue sections showed a notable agreement with the 5-pathologist mitotic count consensus (Pearson correlation 0.92). Furthermore, the model correlated well with mitosis counts based on pHH3 IHC staining (Pearson correlation 0.78), which were available for 63 cases. Finally, when used to characterize the entire CDX cohort, the case level mitotic rate predicted by the model showed significant differences between treatment groups in line with or better than using a pHH3 IHC stain. Conclusion: This study demonstrates the efficacy and scalability of AI-based models for the quantification of mitotic rate based on H&E-stained tissue alone, presenting a time- and cost-efficient approach that also mitigates inter-annotator variability. Citation Format: Sharon Ruane, Lukas Ruff, Brian Reichholf, Christina Aigner, Emil Barbuta, Stephan Tietz, Olivér Atanaszov, Rosemarie Krupar, Simon Schallenberg, Maximilian Alber, Francesca Trapani, Frederick Klauschen. AI powered quantification of mitotic rate in H&E stained tissue detects significant differences between treatment groups of preclinical pancreas cancer xenografts. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5423.

Background: Cyclin-dependent kinase inhibitor p21 is a regulator of cell cycle progression. Due to its capacity to induce cell cycle arrest (CCA) when expressed in the nucleus, it is also considered a tumor suppressor and its presence can be used to evaluate the efficacy of anti-cancer treatment. Since pathologists cannot assess nuclear p21 status of cells using hematoxylin & eosin (H&E) stained tissue alone, the current state-of-the-art approach requires evaluation using immunohistochemistry (IHC). This process is time-consuming, adds additional cost and usually requires a separate section of the sample tissue. Further, manual evaluation of IHC stainings typically shows high inter-pathologist variability. In this study, we developed a deep learning model that predicts cell-level nuclear p21 status on H&E-stained tissue alone, aiming to bypass the IHC-staining step and all drawbacks associated with it. Methods: 99 tissue sections of pancreas cancer xenografts were stained by H&E, then restained for p21 (IHC). The samples originated from mice that had undergone experiments conducted to examine the pharmacodynamic effect of anti-cancer treatments. H&E and IHC image pairs were coregistered to micrometer level precision. A tissue segmentation model was trained to detect regions of ‘carcinoma’ in H&E. This model was used as a filter and only cells within the tumor region were considered for analysis. Individual cells were detected in the H&E image and these locations were transferred to the IHC image. A deep learning model was trained using IHC-informed labels to extract labels at scale from each IHC image. These labels were then transferred to the H&E image and used to train a second deep learning model which predicted nuclear p21 status from H&E alone. Results: IHC-informed labels were extracted with a balanced accuracy (BA) of 0.93. The resulting ‘H&E only’ nuclear p21 model achieved a cell-level BA of 0.83. A case level comparison of the share of predicted p21+ nuclei showed a Pearson correlation of 0.72 with the share of p21+ nuclei determined by the IHC-informed extracted labels. Further, when used to characterize all samples, the model detected significant differences between treatment groups. Conclusion: Nuclear p21 status can be detected at a cellular level in H&E images alone, using a deep learning model. This provides an opportunity to assess samples for cell cycle arrest status at scale in a standardized manner, without the need for IHC staining. Citation Format: Christina Aigner, Brian Reichholf, Maxime Emschwiller, Marija Pezer, Tobias Winterhoff, Simon Schallenberg, Rosemarie Krupar, Lukas Ruff, Sharon Ruane, Maximilian Alber, Frederick Klauschen, Francesca Trapani. Cell cycle arrest status predicted from H&E stained images using deep learning. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5441.