Digital Image Analysis of Heterogeneous Tuberculosis Pulmonary Pathology in Non-Clinical Animal Models using Deep Convolutional Neural Networks

Spectinamide MBX-4888A exhibits favorable lesion and tissue distribution and promotes treatment shortening in advanced murine models of tuberculosis

The spectinamides are novel, narrow-spectrum semisynthetic analogs of spectinomycin, modified to avoid intrinsic efflux by Mycobacterium tuberculosis . Spectinamides, including lead MBX-4888A (Lee-1810), exhibit promising therapeutic profiles in mice, as single drugs and as partner agents with other anti-tuberculosis antibiotics including rifampin and/or pyrazinamide. To demonstrate that this translates to more effective cure, we first confirmed the role of rifampin, with or without pyrazinamide, as essential to achieve effective bactericidal responses and sterilizing cure in the current standard of care regimen in chronically infected C3HeB/FeJ mice compared to BALB/c mice. Thus, demonstrating added value in testing clinically relevant regimens in murine models of increasing pathologic complexity. Next we show that MBX-4888A, given by injection with the front-line standard of care regimen, is treatment shortening in multiple murine tuberculosis infection models. The positive treatment responses to MBX-4888A combination therapy in multiple mouse models including mice exhibiting advanced pulmonary disease can be attributed to favorable distribution in tissues and lesions, retention in caseum, along with favorable effects with rifampin and pyrazinamide under conditions achieved in necrotic lesions. This study also provides an additional data point regarding the safety and tolerability of spectinamide MBX-4888A in long-term murine efficacy studies.

Prediction of Tuberculosis From Lung Tissue Images of Diversity Outbred Mice Using Jump Knowledge Based Cell Graph Neural Network

Tuberculosis (TB), primarily affecting the lungs, is caused by the bacterium Mycobacterium tuberculosis and poses a significant health risk. Detecting acid-fast bacilli (AFB) in stained samples is critical for TB diagnosis. Whole Slide (WS) Imaging allows for digitally examining these stained samples. However, current deep-learning approaches to analyzing large-sized whole slide images (WSIs) often employ patch-wise analysis, potentially missing the complex spatial patterns observed in the granuloma essential for accurate TB classification. To address this limitation, we propose an approach that models cell characteristics and interactions as a graph, capturing both cell-level information and the overall tissue micro-architecture. This method differs from the strategies in related cell graph-based works that rely on edge thresholds based on sparsity/density in cell graph construction, emphasizing a biologically informed threshold determination instead. We introduce a cell graph-based jumping knowledge neural network (CG-JKNN) that operates on the cell graphs where the edge thresholds are selected based on the length of the mycobacteria's cords and the activated macrophage nucleus's size to reflect the actual biological interactions observed in the tissue. The primary process involves training a Convolutional Neural Network (CNN) to segment AFBs and macrophage nuclei, followed by converting large (42831*41159 pixels) lung histology images into cell graphs where an activated macrophage nucleus/AFB represents each node within the graph and their interactions are denoted as edges. To enhance the interpretability of our model, we employ Integrated Gradients and Shapely Additive Explanations (SHAP). Our analysis incorporated a combination of 33 graph metrics and 20 cell morphology features. In terms of traditional machine learning models, Extreme Gradient Boosting (XGBoost) was the best performer, achieving an F1 score of 0.9813 and an Area under the Precision-Recall Curve (AUPRC) of 0.9848 on the test set. Among graph-based models, our CG-JKNN was the top performer, attaining an F1 score of 0.9549 and an AUPRC of 0.9846 on the held-out test set. The integration of graph-based and morphological features proved highly effective, with CG-JKNN and XGBoost showing promising results in classifying instances into AFB and activated macrophage nucleus. The features identified as significant by our models closely align with the criteria used by pathologists in practice, highlighting the clinical applicability of our approach. Future work will explore knowledge distillation techniques and graph-level classification into distinct TB progression categories.

Drug distribution and efficacy of the DprE1 inhibitor BTZ-043 in the C3HeB/FeJ mouse tuberculosis model

BTZ-043, a suicide inhibitor of the Mycobacterium tuberculosis cell wall synthesis decaprenylphosphoryl-beta-D-ribose 2' epimerase, is under clinical development as a potential new anti-tuberculosis agent. BTZ-043 is potent and bactericidal in vitro but has limited activity against non-growing bacilli in rabbit caseum. To better understand its behavior in vivo, BTZ-043 was evaluated for efficacy and spatial drug distribution as a single agent in the C3HeB/FeJ mouse model presenting with caseous necrotic pulmonary lesions upon Mycobacterium tuberculosis infection. BTZ-043 promoted significant reductions in lung and spleen bacterial burdens in the C3HeB/FeJ mouse model after 2 months of therapy. BTZ-043 penetrates cellular and necrotic lesions and was retained at levels above the serum-shifted minimal inhibitory concentration in caseum. The calculated rate of kill was found to be highest and dose-dependent during the second month of treatment. BTZ-043 treatment was associated with improved histology scores of pulmonary lesions, especially compared to control mice, which experienced advanced fulminant neutrophilic alveolitis in the absence of treatment. These positive treatment responses to BTZ-043 monotherapy in a mouse model of advanced pulmonary disease can be attributed to favorable distribution in tissues and lesions, retention in the caseum, and its high potency and bactericidal nature at drug concentrations achieved in necrotic lesions.

Quantifying acute kidney injury in an Ischaemia-Reperfusion Injury mouse model using deep-learning-based semantic segmentation in histology

This study focuses on ischaemia-reperfusion injury (IRI) in kidneys, a cause of acute kidney injury (AKI) and end-stage kidney disease (ESKD). Traditional kidney damage assessment methods are semi-quantitative and subjective. This study aims to use a convolutional neural network (CNN) to segment murine kidney structures after IRI, quantify damage via CNN-generated pathological measurements, and compare this to conventional scoring. The CNN was able to accurately segment the different pathological classes, such as Intratubular casts and Tubular necrosis, with an F1 score of over 0.75. Some classes, such as Glomeruli and Proximal tubules, had even higher statistical values with F1 scores over 0.90. The scoring generated based on the segmentation approach statistically correlated with the semiquantitative assessment (Spearman's rank correlation coefficient=0.94). The heatmap approach localised the intratubular necrosis mainly in the outer stripe of the outer medulla, while the tubular casts were also present in more superficial or deeper portions of the cortex and medullary areas. This study presents a CNN model capable of segmenting multiple classes of interest, including acute IRI-specific pathological changes, in a whole mouse kidney section and can provide insights into the distribution of pathological classes within the whole mouse kidney section.

Application of convolutional neural network for analyzing hepatic fibrosis in mice

Recently, with the development of computer vision using artificial intelligence (AI), clinical research on diagnosis and prediction using medical image data has increased. In this study, we applied AI methods to analyze hepatic fibrosis in mice to determine whether an AI algorithm can be used to analyze lesions. Whole slide image (WSI) Sirius Red staining was used to examine hepatic fibrosis. The Xception network, an AI algorithm, was used to train normal and fibrotic lesion identification. We compared the results from two analyses, that is, pathologists' grades and researchers' annotations, to observe whether the automated algorithm can support toxicological pathologists efficiently as a new apparatus. The accuracies of the trained model computed from the training and validation datasets were greater than 99%, and that obtained by testing the model was 100%. In the comparison between analyses, all analyses showed significant differences in the results for each group. Furthermore, both normalized fibrosis grades inferred from the trained model annotated the fibrosis area, and the grades assigned by the pathologists showed significant correlations. Notably, the deep learning algorithm derived the highest correlation with the pathologists' average grade. Owing to the correlation outcomes, we conclude that the trained model might produce results comparable to those of the pathologists' grading of the Sirius Red-stained WSI fibrosis. This study illustrates that the deep learning algorithm can potentially be used for analyzing fibrotic lesions in combination with Sirius Red-stained WSIs as a second opinion tool in non-clinical research.

Reduced SARS-CoV-2 disease outcomes in Syrian hamsters receiving immune sera: Quantitative image analysis in pathologic assessments

There is a need to standardize pathologic endpoints in animal models of SARS-CoV-2 infection to help benchmark study quality, improve cross-institutional comparison of data, and assess therapeutic efficacy so that potential drugs and vaccines for SARS-CoV-2 can rapidly advance. The Syrian hamster model is a tractable small animal model for COVID-19 that models clinical disease in humans. Using the hamster model, the authors used traditional pathologic assessment with quantitative image analysis to assess disease outcomes in hamsters administered polyclonal immune sera from previously challenged rhesus macaques. The authors then used quantitative image analysis to assess pathologic endpoints across studies performed at different institutions using different tissue processing protocols. The authors detail pathological features of SARS-CoV-2 infection longitudinally and use immunohistochemistry to quantify myeloid cells and T lymphocyte infiltrates during SARS-CoV-2 infection. High-dose immune sera protected hamsters from weight loss and diminished viral replication in tissues and reduced lung lesions. Cumulative pathology scoring correlated with weight loss and was robust in distinguishing IgG efficacy. In formalin-infused lungs, quantitative measurement of percent area affected also correlated with weight loss but was less robust in non-formalin-infused lungs. Longitudinal immunohistochemical assessment of interstitial macrophage infiltrates showed that peak infiltration corresponded to weight loss, yet quantitative assessment of macrophage, neutrophil, and CD3+ T lymphocyte numbers did not distinguish IgG treatment effects. Here, the authors show that quantitative image analysis was a useful adjunct tool for assessing SARS-CoV-2 treatment outcomes in the hamster model.

Deep-Learning-Aided Detection of Mycobacteria in Pathology Specimens Increases the Sensitivity in Early Diagnosis of Pulmonary Tuberculosis Compared with Bacteriology Tests

The histopathological diagnosis of mycobacterial infection may be improved by a comprehensive analysis using artificial intelligence. Two autopsy cases of pulmonary tuberculosis, and forty biopsy cases of undetected acid-fast bacilli (AFB) were used to train AI (convolutional neural network), and construct an AI to support AFB detection. Forty-two patients underwent bronchoscopy, and were evaluated using AI-supported pathology to detect AFB. The AI-supported pathology diagnosis was compared with bacteriology diagnosis from bronchial lavage fluid and the final definitive diagnosis of mycobacteriosis. Among the 16 patients with mycobacteriosis, bacteriology was positive in 9 patients (56%). Two patients (13%) were positive for AFB without AI assistance, whereas AI-supported pathology identified eleven positive patients (69%). When limited to tuberculosis, AI-supported pathology had significantly higher sensitivity compared with bacteriology (86% vs. 29%, p = 0.046). Seven patients diagnosed with mycobacteriosis had no consolidation or cavitary shadows in computed tomography; the sensitivity of bacteriology and AI-supported pathology was 29% and 86%, respectively (p = 0.046). The specificity of AI-supported pathology was 100% in this study. AI-supported pathology may be more sensitive than bacteriological tests for detecting AFB in samples collected via bronchoscopy.

Artificial Intelligence-based Tumor Segmentation in Mouse Models of Lung Adenocarcinoma

BACKGROUND

Mouse models are highly effective for studying the pathophysiology of lung adenocarcinoma and evaluating new treatment strategies. Treatment efficacy is primarily determined by the total tumor burden measured on excised tumor specimens. The measurement process is time-consuming and prone to human errors. To address this issue, we developed a novel deep learning model to segment lung tumor foci on digitally scanned hematoxylin and eosin (H&E) histology slides.

METHODS

Digital slides of 239 mice from 9 experimental cohorts were split into training (n=137), validation (n=37), and testing cohorts (n=65). Image patches of 500×500 pixels were extracted from 5× and 10× magnifications, along with binary masks of expert annotations representing ground-truth tumor regions. Deep learning models utilizing DeepLabV3+ and UNet architectures were trained for binary segmentation of tumor foci under varying stain normalization conditions. The performance of algorithm segmentation was assessed by Dice Coefficient, and detection was evaluated by sensitivity and positive-predictive value (PPV).

RESULTS

The best model on patch-based validation was DeepLabV3+ using a Resnet-50 backbone, which achieved Dice 0.890 and 0.873 on validation and testing cohort, respectively. This result corresponded to 91.3 Sensitivity and 51.0 PPV in the validation cohort and 93.7 Sensitivity and 51.4 PPV in the testing cohort. False positives could be reduced 10-fold with thresholding artificial intelligence (AI) predicted output by area, without negative impact on Dice Coefficient. Evaluation at various stain normalization strategies did not demonstrate improvement from the baseline model.

CONCLUSIONS

A robust AI-based algorithm for detecting and segmenting lung tumor foci in the pre-clinical mouse models was developed. The output of this algorithm is compatible with open-source software that researchers commonly use.

Deep Learning Approaches and Applications in Toxicologic Histopathology: Current Status and Future Perspectives

Whole slide imaging enables the use of a wide array of digital image analysis tools that are revolutionizing pathology. Recent advances in digital pathology and deep convolutional neural networks have created an enormous opportunity to improve workflow efficiency, provide more quantitative, objective, and consistent assessments of pathology datasets, and develop decision support systems. Such innovations are already making their way into clinical practice. However, the progress of machine learning - in particular, deep learning (DL) - has been rather slower in nonclinical toxicology studies. Histopathology data from toxicology studies are critical during the drug development process that is required by regulatory bodies to assess drug-related toxicity in laboratory animals and its impact on human safety in clinical trials. Due to the high volume of slides routinely evaluated, low-throughput, or narrowly performing DL methods that may work well in small-scale diagnostic studies or for the identification of a single abnormality are tedious and impractical for toxicologic pathology. Furthermore, regulatory requirements around good laboratory practice are a major hurdle for the adoption of DL in toxicologic pathology. This paper reviews the major DL concepts, emerging applications, and examples of DL in toxicologic pathology image analysis. We end with a discussion of specific challenges and directions for future research.

Comparative Analysis of Pharmacodynamics in the C3HeB/FeJ Mouse Tuberculosis Model for DprE1 inhibitors TBA-7371, PBTZ169 and OPC-167832

Multiple drug discovery initiatives for tuberculosis are currently ongoing to identify and develop new potent drugs with novel targets in order to shorten treatment duration. One of the drug classes with a new mode of action are DprE1 inhibitors targeting an essential process in cell wall synthesis of Mycobacterium tuberculosis. In this investigation, three DprE1 inhibitors currently in clinical trials, TBA-7371, PBTZ169 and OPC-167832, were evaluated side-by-side as single agents in the C3HeB/FeJ mouse model presenting with caseous necrotic pulmonary lesions upon tuberculosis infection. The goal was to confirm the efficacy of the DprE1 inhibitors in a mouse tuberculosis model with advanced pulmonary pathology, and perform comprehensive analysis of plasma, lung and lesion-centric drug levels to establish pharmacokinetic-pharmacodynamic (PK-PD) parameters predicting efficacy at the site of infection. Results showed significant efficacy for all three DprE1 inhibitors in the C3HeB/FeJ mouse model after two months of treatment. Superior efficacy was observed for OPC-167832 even at low dose levels, which can be attributed to its low MIC, favorable distribution and sustained retention above the MIC throughout the dosing interval in caseous necrotic lesions where the majority of bacteria reside in C3HeB/FeJ mice. These results support further progression of the three drug candidates through clinical development for tuberculosis treatment.

Subunit vaccine protects against a clinical isolate of Mycobacterium avium in wild type and immunocompromised mouse models

The nontuberculous mycobacteria (NTM) Mycobacterium avium is a clinically significant pathogen that can cause a wide range of maladies, including tuberculosis-like pulmonary disease. An immunocompromised host status, either genetically or acutely acquired, presents a large risk for progressive NTM infections. Due to this quietly emerging health threat, we evaluated the ability of a recombinant fusion protein ID91 combined with GLA-SE [glucopyranosyl lipid adjuvant, a toll like receptor 4 agonist formulated in an oil-in-water stable nano-emulsion] to confer protection in both C57BL/6 (wild type) and Beige (immunocompromised) mouse models. We optimized an aerosol challenge model using a clinical NTM isolate: M. avium 2-151 smt, observed bacterial growth kinetics, colony morphology, drug sensitivity and histopathology, characterized the influx of pulmonary immune cells, and confirmed the immunogenicity of ID91 in both mouse models. To determine prophylactic vaccine efficacy against this M. avium isolate, mice were immunized with either ID91 + GLA-SE or bacillus Calmette-Guérin (BCG). Immunocompromised Beige mice displayed a delayed influx of innate and adaptive immune cells resulting in a sustained and increased bacterial burden in the lungs and spleen compared to C57BL/6 mice. Importantly, both ID91 + GLA-SE and BCG vaccines significantly reduced pulmonary bacterial burden in both mouse strains. This work is a proof-of-concept study of subunit vaccine-induced protection against NTM.