PBP-A, a cyanobacterial DD-peptidase with high specificity for amidated muropeptides, exhibits pH-dependent promiscuous activity harmful to Escherichia coli.

Penicillin binding proteins (PBPs) are involved in biosynthesis, remodeling and recycling of peptidoglycan (PG) in bacteria. PBP-A from Thermosynechococcus elongatus belongs to a cyanobacterial family of enzymes sharing close structural and phylogenetic proximity to class A ß-lactamases. With the long-term aim of converting PBP-A into a ß-lactamase by directed evolution, we simulated what may happen when an organism like Escherichia coli acquires such a new PBP and observed growth defect associated with the enzyme activity. To further explore the molecular origins of this harmful effect, we decided to characterize deeper the activity of PBP-A both in vitro and in vivo. We found that PBP-A is an enzyme endowed with DD-carboxypeptidase and DD-endopeptidase activities, featuring high specificity towards muropeptides amidated on the D-iso-glutamyl residue. We also show that a low promiscuous activity on non-amidated peptidoglycan deteriorates E. coli's envelope, which is much higher under acidic conditions where substrate discrimination is mitigated. Besides expanding our knowledge of the biochemical activity of PBP-A, this work also highlights that promiscuity may depend on environmental conditions and how it may hinder rather than promote enzyme evolution in nature or in the laboratory.

Cleavage of an engulfment peptidoglycan hydrolase by a sporulation signature protease in Clostridioides difficile.

In the model organism Bacillus subtilis, a signaling protease produced in the forespore, SpoIVB, is essential for the activation of the sigma factor σK, which is produced in the mother cell as an inactive pro-protein, pro-σK. SpoIVB has a second function essential to sporulation, most likely during cortex synthesis. The cortex is composed of peptidoglycan (PG) and is essential for the spore's heat resistance and dormancy. Surprisingly, the genome of the intestinal pathogen Clostridioides difficile, in which σK is produced without a pro-sequence, encodes two SpoIVB paralogs, SpoIVB1 and SpoIVB2. Here, we show that spoIVB1 is dispensable for sporulation, while a spoIVB2 in-frame deletion mutant fails to produce heat-resistant spores. The spoIVB2 mutant enters sporulation, undergoes asymmetric division, and completes engulfment of the forespore by the mother cell but fails to synthesize the spore cortex. We show that SpoIIP, a PG hydrolase and part of the engulfasome, the machinery essential for engulfment, is cleaved by SpoIVB2 into an inactive form. Within the engulfasome, the SpoIIP amidase activity generates the substrates for the SpoIID lytic transglycosylase. Thus, following engulfment completion, the cleavage and inactivation of SpoIIP by SpoIVB2 curtails the engulfasome hydrolytic activity, at a time when synthesis of the spore cortex peptidoglycan begins. SpoIVB2 is also required for normal late gene expression in the forespore by a currently unknown mechanism. Together, these observations suggest a role for SpoIVB2 in coordinating late morphological and gene expression events between the forespore and the mother cell.

Peptidoglycan-Targeted [18F]3,3,3-Trifluoro-d-alanine Tracer for Imaging Bacterial Infection.

Imaging is increasingly used to detect and monitor bacterial infection. Both anatomic (X-rays, computed tomography, ultrasound, and MRI) and nuclear medicine ([111In]-WBC SPECT, [18F]FDG PET) techniques are used in clinical practice but lack specificity for the causative microorganisms themselves. To meet this challenge, many groups have developed imaging methods that target pathogen-specific metabolism, including PET tracers integrated into the bacterial cell wall. We have previously reported the d-amino acid derived PET radiotracers d-methyl-[11C]-methionine, d-[3-11C]-alanine, and d-[3-11C]-alanine-d-alanine, which showed robust bacterial accumulation in vitro and in vivo. Given the clinical importance of radionuclide half-life, in the current study, we developed [18F]3,3,3-trifluoro-d-alanine (d-[18F]-CF3-ala), a fluorine-18 labeled tracer. We tested the hypothesis that d-[18F]-CF3-ala would be incorporated into bacterial peptidoglycan given its structural similarity to d-alanine itself. NMR analysis showed that the fluorine-19 parent amino acid d-[19F]-CF3-ala was stable in human and mouse serum. d-[19F]-CF3-ala was also a poor substrate for d-amino acid oxidase, the enzyme largely responsible for mammalian d-amino acid metabolism and a likely contributor to background signals using d-amino acid derived PET tracers. In addition, d-[19F]-CF3-ala showed robust incorporation into Escherichia coli peptidoglycan, as detected by HPLC/mass spectrometry. Based on these promising results, we developed a radiosynthesis of d-[18F]-CF3-ala via displacement of a bromo-precursor with [18F]fluoride followed by chiral stationary phase HPLC. Unexpectedly, the accumulation of d-[18F]-CF3-ala by bacteria in vitro was highest for Gram-negative pathogens in particular E. coli. In a murine model of acute bacterial infection, d-[18F]-CF3-ala could distinguish live from heat-killed E. coli, with low background signals. These results indicate the viability of [18F]-modified d-amino acids for infection imaging and indicate that improved specificity for bacterial metabolism can improve tracer performance.

Functional delineation of the luminal epithelial microenvironment in breast using cell-based screening in combinatorial microenvironments.

Microenvironment signals are potent determinants of cell fate and arbiters of tissue homeostasis, however understanding how different microenvironment factors coordinately regulate cellular phenotype has been experimentally challenging. Here we used a high-throughput microenvironment microarray comprised of 2640 unique pairwise signals to identify factors that support proliferation and maintenance of primary human mammary luminal epithelial cells. Multiple microenvironment factors that modulated luminal cell number were identified, including: HGF, NRG1, BMP2, CXCL1, TGFB1, FGF2, PDGFB, RANKL, WNT3A, SPP1, HA, VTN, and OMD. All of these factors were previously shown to modulate luminal cell numbers in painstaking mouse genetics experiments, or were shown to have a role in breast cancer, demonstrating the relevance and power of our high-dimensional approach to dissect key microenvironmental signals. RNA-sequencing of primary epithelial and stromal cell lineages identified the cell types that express these signals and the cognate receptors in vivo. Cell-based functional studies confirmed which effects from microenvironment factors were reproducible and robust to individual variation. Hepatocyte growth factor (HGF) was the factor most robust to individual variation and drove expansion of luminal cells via cKit+ progenitor cells, which expressed abundant MET receptor. Luminal cells from women who are genetically high risk for breast cancer had significantly more MET receptor and may explain the characteristic expansion of the luminal lineage in those women. In ensemble, our approach provides proof of principle that microenvironment signals that control specific cellular states can be dissected with high-dimensional cell-based approaches.

Towards Generalizability and Robustness in Biological Object Detection in Electron Microscopy Images.

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.

3D multiplexed tissue imaging reconstruction and optimized region of interest (ROI) selection through deep learning model of channels embedding.

Introduction: Tissue-based sampling and diagnosis are defined as the extraction of information from certain limited spaces and its diagnostic significance of a certain object. Pathologists deal with issues related to tumor heterogeneity since analyzing a single sample does not necessarily capture a representative depiction of cancer, and a tissue biopsy usually only presents a small fraction of the tumor. Many multiplex tissue imaging platforms (MTIs) make the assumption that tissue microarrays (TMAs) containing small core samples of 2-dimensional (2D) tissue sections are a good approximation of bulk tumors although tumors are not 2D. However, emerging whole slide imaging (WSI) or 3D tumor atlases that use MTIs like cyclic immunofluorescence (CyCIF) strongly challenge this assumption. In spite of the additional insight gathered by measuring the tumor microenvironment in WSI or 3D, it can be prohibitively expensive and time-consuming to process tens or hundreds of tissue sections with CyCIF. Even when resources are not limited, the criteria for region of interest (ROI) selection in tissues for downstream analysis remain largely qualitative and subjective as stratified sampling requires the knowledge of objects and evaluates their features. Despite the fact TMAs fail to adequately approximate whole tissue features, a theoretical subsampling of tissue exists that can best represent the tumor in the whole slide image. Methods: To address these challenges, we propose deep learning approaches to learn multi-modal image translation tasks from two aspects: 1) generative modeling approach to reconstruct 3D CyCIF representation and 2) co-embedding CyCIF image and Hematoxylin and Eosin (H&E) section to learn multi-modal mappings by a cross-domain translation for minimum representative ROI selection. Results and discussion: We demonstrate that generative modeling enables a 3D virtual CyCIF reconstruction of a colorectal cancer specimen given a small subset of the imaging data at training time. By co-embedding histology and MTI features, we propose a simple convex optimization for objective ROI selection. We demonstrate the potential application of ROI selection and the efficiency of its performance with respect to cellular heterogeneity.

Efficient semi-supervised semantic segmentation of electron microscopy cancer images with sparse annotations.

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.

MYC Deregulation and PTEN Loss Model Tumor and Stromal Heterogeneity of Aggressive Triple-Negative Breast Cancer.

Triple-negative breast cancer (TNBC) patients have a poor prognosis and few treatment options. Mouse models of TNBC are important for development of new therapies, however, few mouse models represent the complexity of TNBC. Here, we develop a female TNBC murine model by mimicking two common TNBC mutations with high co-occurrence: amplification of the oncogene MYC and deletion of the tumor suppressor PTEN. This Myc;Ptenfl model develops heterogeneous triple-negative mammary tumors that display histological and molecular features commonly found in human TNBC. Our research involves deep molecular and spatial analyses on Myc;Ptenfl tumors including bulk and single-cell RNA-sequencing, and multiplex tissue-imaging. Through comparison with human TNBC, we demonstrate that this genetic mouse model develops mammary tumors with differential survival and therapeutic responses that closely resemble the inter- and intra-tumoral and microenvironmental heterogeneity of human TNBC, providing a pre-clinical tool for assessing the spectrum of patient TNBC biology and drug response.

Digitize your Biology! Modeling multicellular systems through interpretable cell behavior.

Cells are fundamental units of life, constantly interacting and evolving as dynamical systems. While recent spatial multi-omics can quantitate individual cells' characteristics and regulatory programs, forecasting their evolution ultimately requires mathematical modeling. We develop a conceptual framework-a cell behavior hypothesis grammar-that uses natural language statements (cell rules) to create mathematical models. This allows us to systematically integrate biological knowledge and multi-omics data to make them computable. We can then perform virtual "thought experiments" that challenge and extend our understanding of multicellular systems, and ultimately generate new testable hypotheses. In this paper, we motivate and describe the grammar, provide a reference implementation, and demonstrate its potential through a series of examples in tumor biology and immunotherapy. Altogether, this approach provides a bridge between biological, clinical, and systems biology researchers for mathematical modeling of biological systems at scale, allowing the community to extrapolate from single-cell characterization to emergent multicellular behavior.

NlpC/P60 peptidoglycan hydrolases of Trichomonas vaginalis have complementary activities that empower the protozoan to control host-protective lactobacilli.

Trichomonas vaginalis is a human protozoan parasite that causes trichomoniasis, a prevalent sexually transmitted infection. Trichomoniasis is accompanied by a shift to a dysbiotic vaginal microbiome that is depleted of lactobacilli. Studies on co-cultures have shown that vaginal bacteria in eubiosis (e.g. Lactobacillus gasseri) have antagonistic effects on T. vaginalis pathogenesis, suggesting that the parasite might benefit from shaping the microbiome to dysbiosis (e.g. Gardnerella vaginalis among other anaerobes). We have recently shown that T. vaginalis has acquired NlpC/P60 genes from bacteria, expanding them to a repertoire of nine TvNlpC genes in two distinct clans, and that TvNlpCs of clan A are active against bacterial peptidoglycan. Here, we expand this characterization to TvNlpCs of clan B. In this study, we show that the clan organisation of NlpC/P60 genes is a feature of other species of Trichomonas, and that Histomonas meleagridis has sequences related to one clan. We characterized the 3D structure of TvNlpC_B3 alone and with the inhibitor E64 bound, probing the active site of these enzymes for the first time. Lastly, we demonstrated that TvNlpC_B3 and TvNlpC_B5 have complementary activities with the previously described TvNlpCs of clan A and that exogenous expression of these enzymes empower this mucosal parasite to take over populations of vaginal lactobacilli in mixed cultures. TvNlpC_B3 helps control populations of L. gasseri, but not of G. vaginalis, which action is partially inhibited by E64. This study is one of the first to show how enzymes produced by a mucosal protozoan parasite may contribute to a shift on the status of a microbiome, helping explain the link between trichomoniasis and vaginal dysbiosis. Further understanding of this process might have significant implications for treatments in the future.

Development and Multicenter Case-Control Validation of Urinary Comprehensive Genomic Profiling for Urothelial Carcinoma Diagnosis, Surveillance, and Risk-Prediction.

PURPOSE: Urinary comprehensive genomic profiling (uCGP) uses next-generation sequencing to identify mutations associated with urothelial carcinoma and has the potential to improve patient outcomes by noninvasively diagnosing disease, predicting grade and stage, and estimating recurrence risk. EXPERIMENTAL DESIGN: This is a multicenter case-control study using banked urine specimens collected from patients undergoing initial diagnosis/hematuria workup or urothelial carcinoma surveillance. A total of 581 samples were analyzed by uCGP: 333 for disease classification and grading algorithm development, and 248 for blinded validation. uCGP testing was done using the UroAmp platform, which identifies five classes of mutation: single-nucleotide variants, copy-number variants, small insertion-deletions, copy-neutral loss of heterozygosity, and aneuploidy. UroAmp algorithms predicting urothelial carcinoma tumor presence, grade, and recurrence risk were compared with cytology, cystoscopy, and pathology. RESULTS: uCGP algorithms had a validation sensitivity/specificity of 95%/90% for initial cancer diagnosis in patients with hematuria and demonstrated a negative predictive value (NPV) of 99%. A positive diagnostic likelihood ratio (DLR) of 9.2 and a negative DLR of 0.05 demonstrate the ability to risk-stratify patients presenting with hematuria. In surveillance patients, binary urothelial carcinoma classification demonstrated an NPV of 91%. uCGP recurrence-risk prediction significantly prognosticated future recurrence (hazard ratio, 6.2), whereas clinical risk factors did not. uCGP demonstrated positive predictive value (PPV) comparable with cytology (45% vs. 42%) with much higher sensitivity (79% vs. 25%). Finally, molecular grade predictions had a PPV of 88% and a specificity of 95%. CONCLUSIONS: uCGP enables noninvasive, accurate urothelial carcinoma diagnosis and risk stratification in both hematuria and urothelial carcinoma surveillance patients.

Efficient semi-supervised semantic segmentation of electron microscopy cancer images with sparse annotations.

Electron microscopy (EM) enables imaging at a resolution of nanometers 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 a 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, and semi-supervised learning as well as the next steps for the mitigation of the manual segmentation bottleneck.

Segmentation of cellular ultrastructures on sparsely labeled 3D electron microscopy images using deep learning.

Focused ion beam-scanning electron microscopy (FIB-SEM) images can provide a detailed view of the cellular ultrastructure of tumor cells. A deeper understanding of their organization and interactions can shed light on cancer mechanisms and progression. However, the bottleneck in the analysis is the delineation of the cellular structures to enable quantitative measurements and analysis. We mitigated this limitation using deep learning to segment cells and subcellular ultrastructure in 3D FIB-SEM images of tumor biopsies obtained from patients with metastatic breast and pancreatic cancers. The ultrastructures, such as nuclei, nucleoli, mitochondria, endosomes, and lysosomes, are relatively better defined than their surroundings and can be segmented with high accuracy using a neural network trained with sparse manual labels. Cell segmentation, on the other hand, is much more challenging due to the lack of clear boundaries separating cells in the tissue. We adopted a multi-pronged approach combining detection, boundary propagation, and tracking for cell segmentation. Specifically, a neural network was employed to detect the intracellular space; optical flow was used to propagate cell boundaries across the z-stack from the nearest ground truth image in order to facilitate the separation of individual cells; finally, the filopodium-like protrusions were tracked to the main cells by calculating the intersection over union measure for all regions detected in consecutive images along z-stack and connecting regions with maximum overlap. The proposed cell segmentation methodology resulted in an average Dice score of 0.93. For nuclei, nucleoli, and mitochondria, the segmentation achieved Dice scores of 0.99, 0.98, and 0.86, respectively. The segmentation of FIB-SEM images will enable interpretative rendering and provide quantitative image features to be associated with relevant clinical variables.

Urinary Comprehensive Genomic Profiling Correlates Urothelial Carcinoma Mutations with Clinical Risk and Efficacy of Intervention.

The clinical standard of care for urothelial carcinoma (UC) relies on invasive procedures with suboptimal performance. To enhance UC treatment, we developed a urinary comprehensive genomic profiling (uCGP) test, UroAmplitude, that measures mutations from tumor DNA present in urine. In this study, we performed a blinded, prospective validation of technical sensitivity and positive predictive value (PPV) using reference standards, and found at 1% allele frequency, mutation detection performs at 97.4% sensitivity and 80.4% PPV. We then prospectively compared the mutation profiles of urine-extracted DNA to those of matched tumor tissue to validate clinical performance. Here, we found tumor single-nucleotide variants were observed in the urine with a median concordance of 91.7% and uCGP revealed distinct patterns of genomic lesions enriched in low- and high-grade disease. Finally, we retrospectively explored longitudinal case studies to quantify residual disease following bladder-sparing treatments, and found uCGP detected residual disease in patients receiving bladder-sparing treatment and predicted recurrence and disease progression. These findings demonstrate the potential of the UroAmplitude platform to reliably identify and track mutations associated with UC at each stage of disease: diagnosis, treatment, and surveillance. Multiple case studies demonstrate utility for patient risk classification to guide both surgical and therapeutic interventions.

A multi-omic analysis of MCF10A cells provides a resource for integrative assessment of ligand-mediated molecular and phenotypic responses.

The phenotype of a cell and its underlying molecular state is strongly influenced by extracellular signals, including growth factors, hormones, and extracellular matrix proteins. While these signals are normally tightly controlled, their dysregulation leads to phenotypic and molecular states associated with diverse diseases. To develop a detailed understanding of the linkage between molecular and phenotypic changes, we generated a comprehensive dataset that catalogs the transcriptional, proteomic, epigenomic and phenotypic responses of MCF10A mammary epithelial cells after exposure to the ligands EGF, HGF, OSM, IFNG, TGFB and BMP2. Systematic assessment of the molecular and cellular phenotypes induced by these ligands comprise the LINCS Microenvironment (ME) perturbation dataset, which has been curated and made publicly available for community-wide analysis and development of novel computational methods ( synapse.org/LINCS_MCF10A ). In illustrative analyses, we demonstrate how this dataset can be used to discover functionally related molecular features linked to specific cellular phenotypes. Beyond these analyses, this dataset will serve as a resource for the broader scientific community to mine for biological insights, to compare signals carried across distinct molecular modalities, and to develop new computational methods for integrative data analysis.

Computational multiplex panel reduction to maximize information retention in breast cancer tissue microarrays.

Recent state-of-the-art multiplex imaging techniques have expanded the depth of information that can be captured within a single tissue sample by allowing for panels with dozens of markers. Despite this increase in capacity, space on the panel is still limited due to technical artifacts, tissue loss, and long imaging acquisition time. As such, selecting which markers to include on a panel is important, since removing important markers will result in a loss of biologically relevant information, but identifying redundant markers will provide a room for other markers. To address this, we propose computational approaches to determine the amount of shared information between markers and select an optimally reduced panel that captures maximum amount of information with the fewest markers. Here we examine several panel selection approaches and evaluate them based on their ability to reconstruct the full panel images and information within breast cancer tissue microarray datasets using cyclic immunofluorescence as a proof of concept. We show that all methods perform adequately and can re-capture cell types using only 18 of 25 markers (72% of the original panel size). The correlation-based selection methods achieved the best single-cell marker mean intensity predictions with a Spearman correlation of 0.90 with the reduced panel. Using the proposed methods shown here, it is possible for researchers to design more efficient multiplex imaging panels that maximize the amount of information retained with the limited number of markers with respect to certain evaluation metrics and architecture biases.

Discovery, isolation, heterologous expression and mode-of-action studies of the antibiotic polyketide tatiomicin from Amycolatopsis sp. DEM30355.

A genomic and bioactivity informed analysis of the metabolome of the extremophile Amycolatopsis sp. DEM30355 has allowed for the discovery and isolation of the polyketide antibiotic tatiomicin. Identification of the biosynthetic gene cluster was confirmed by heterologous expression in Streptomyces coelicolor M1152. Structural elucidation, including absolute stereochemical assignment, was performed using complementary crystallographic, spectroscopic and computational methods. Tatiomicin shows antibiotic activity against Gram-positive bacteria, including methicillin-resistant Staphylococcus aureus (MRSA). Cytological profiling experiments suggest a putative antibiotic mode-of-action, involving membrane depolarisation and chromosomal decondensation of the target bacteria.

Metabolic labeling of the bacterial peptidoglycan by functionalized glucosamine.

N-Acetylglucosamine (GlcNAc) is an essential monosaccharide required in almost all organisms. Fluorescent labeling of the peptidoglycan (PG) on N-acetylglucosamine has been poorly explored. Here, we report on the labeling of the PG with a bioorthogonal handle on the GlcNAc. We developed a facile one-step synthesis of uridine diphosphate N-azidoacetylglucosamine (UDP-GlcNAz) using the glycosyltransferase OleD, followed by in vitro incorporation of GlcNAz into the peptidoglycan precursor Lipid II and fluorescent labeling of the azido group via click chemistry. In a PG synthesis assay, fluorescent GlcNAz-labeled Lipid II was incorporated into peptidoglycan by the DD-transpeptidase activity of bifunctional class A penicillin-binding proteins. We further demonstrate the incorporation of GlcNAz into the PG layer of OleD-expressed bacteria by feeding with 2-chloro-4-nitrophenyl GlcNAz (GlcNAz-CNP). Hence, our labeling method using the heterologous expression of OleD is useful to study PG synthesis and possibly other biological processes involving GlcNAc metabolism in vivo.

A multiplex implantable microdevice assay identifies synergistic combinations of cancer immunotherapies and conventional drugs.

Systematically identifying synergistic combinations of targeted agents and immunotherapies for cancer treatments remains difficult. In this study, we integrated high-throughput and high-content techniques-an implantable microdevice to administer multiple drugs into different sites in tumors at nanodoses and multiplexed imaging of tumor microenvironmental states-to investigate the tumor cell and immunological response signatures to different treatment regimens. Using a mouse model of breast cancer, we identified effective combinations from among numerous agents within days. In vivo studies in three immunocompetent mammary carcinoma models demonstrated that the predicted combinations synergistically increased therapeutic efficacy. We identified at least five promising treatment strategies, of which the panobinostat, venetoclax and anti-CD40 triple therapy was the most effective in inducing complete tumor remission across models. Successful drug combinations increased spatial association of cancer stem cells with dendritic cells during immunogenic cell death, suggesting this as an important mechanism of action in long-term breast cancer control.