Explore the vast range of titles available.

Deep neural networks display impressive performance but suffer from limited interpretability. Biology-inspired deep learning, where the architecture of the computational graph is based on biological knowledge, enables unique interpretability where real-world concepts are encoded in hidden nodes, which can be ranked by importance and thereby interpreted. In such models trained on single-cell transcriptomes, we previously demonstrated that node-level interpretations lack robustness upon repeated training and are influenced by biases in biological knowledge. Similar studies are missing for related models. Here, we test and extend our methodology for reliable interpretability in P-NET, a biology-inspired model trained on patient mutation data. We observe variability of interpretations and susceptibility to knowledge biases, and identify the network properties that drive interpretation biases. We further present an approach to control the robustness and biases of interpretations, which leads to more specific interpretations. In summary, our study reveals the broad importance of methods to ensure robust and bias-aware interpretability in biology-inspired deep learning.

Metastasis is the major cause of cancer-related deaths. Neuroblastoma (NB), a childhood tumor has been molecularly defined at the primary cancer site, however, the bone marrow (BM) as the metastatic niche of NB is poorly characterized. Here we perform single-cell transcriptomic and epigenomic profiling of BM aspirates from 11 subjects spanning three major NB subtypes and compare these to five age-matched and metastasis-free BM, followed by in-depth single cell analyses of tissue diversity and cell-cell interactions, as well as functional validation. We show that cellular plasticity of NB tumor cells is conserved upon metastasis and tumor cell type composition is NB subtype-dependent. NB cells signal to the BM microenvironment, rewiring via macrophage mgration inhibitory factor and midkine signaling specifically monocytes, which exhibit M1 and M2 features, are marked by activation of pro- and anti-inflammatory programs, and express tumor-promoting factors, reminiscent of tumor-associated macrophages. The interactions and pathways characterized in our study provide the basis for therapeutic approaches that target tumor-to-microenvironment interactions. The bone marrow is a common site of metastasis for neuroblastoma patients. Here, the authors perform single cell RNA-seq and ATAC-seq of bone marrow aspirates from 16 subjects and show conservation of tumor cell plasticity in metastases and identify tumor-to-bone marrow cell signals that trigger tumor promoting monocytes.

Different manufacturing processes and storage conditions of biotherapeutics can lead to a significant variability in drug products arising from chemical and enzymatic post-translational modifications (PTMs), resulting in the co-existence of a plethora of proteoforms with different physicochemical properties. To unravel the heterogeneity of these proteoforms, novel approaches employing strong cation-exchange (SCX) high-performance liquid chromatography (HPLC) hyphenated to mass spectrometry (MS) using a pH gradient of volatile salts have been developed in recent years. Here, we apply an established SCX-HPLC-MS method to characterize and compare two rituximab-based biotherapeutics, the originator MabThera® and its Indian copy product Reditux™. The study assessed molecular differences between the two drug products in terms of C-terminal lysine variants, glycosylation patterns, and other basic and acidic variants. Overall, MabThera® and Reditux™ displayed differences at the molecular level. MabThera® showed a higher degree of galactosylated and sialylated glycoforms, while Reditux™ showed increased levels of oligomannose and afucosylated glycoforms. Moreover, the two drug products showed differences in terms of basic variants such as C-terminal lysine and N-terminal truncation, present in Reditux™ but not in MabThera®. This study demonstrates the capability of this fast SCX-HPLC-MS approach to compare different drug products and simultaneously assess some of their quality attributes.

Cardiac fibrosis is mediated by the persistent activity of myofibroblasts, which differentiates from resident cardiac fibroblasts in response to tissue damage and stress signals. The signaling pathways and transcription factors regulating fibrotic transformation have been thoroughly studied. In contrast, the roles of chromastin factors in myofibroblast differentiation and their contribution to pathogenic cardiac fibrosis remain poorly understood. Here, we combined bulk and single-cell CRISPR screens to characterize the roles of chromatin factors in the fibrotic transformation of primary cardiac fibroblasts. We uncover strong regulators of fibrotic states including Srcap and Kat5 chromatin remodelers. We confirm that these factors are required for functional processes underlying fibrosis including collagen synthesis and cell contractility. Using chromatin profiling in perturbed cardiac fibroblasts, we demonstrate that pro-fibrotic chromatin complexes facilitate the activity of well-characterized pro-fibrotic transcription factors. Finally, we show that KAT5 inhibition alleviates fibrotic responses in patient-derived human fibroblasts. Cardiac fibrosis arises from persistent myofibroblast activity. This study reveals how chromatin factors control scar-forming cells in the heart and shows that inhibiting KAT5 can reduce harmful cardiac fibrosis.

Monoclonal antibody N -glycosylation is a critical quality attribute influencing therapeutic safety and efficacy, and is strongly influenced by bioprocess design. NISTCHO, a publicly available Chinese hamster ovary producer cell line, is increasingly encouraged for use as a reference system. However, the impact of feeding strategies on cellular performance and N -glycosylation has not been assessed. Here, we applied multivariate analysis of compositional N -glycan data to assess how feeding strategies influence N -glycan composition of cNISTmAb. We varied feeding strategies in frequency, glucose supply, and galactose/manganese supplementation. Feeding frequency had minimal impact on quality attributes but strongly affected culture performance, with every-other-day feeding improving titers and cell-specific productivity. High glucose availability supported growth and productivity. Low glucose strategies reduced titers and shifted N -glycosylation towards non-galactosylated and fucosylated species, despite lactate accumulation remaining within favorable ranges. Galactose and manganese consistently increased antibody galactosylation, with galactose additionally serving as an auxiliary carbon source, extending cell viability. Importantly, mAb glycation remained stable across all feeding strategies at harvest. Overall, these results demonstrate that feed composition and timing can be used to tune both cellular performance and mAb glycosylation, establishing NISTCHO as a robust benchmark for standardized process-quality studies.

Immune cells need to sustain a state of constant alertness over a lifetime. Yet, little is known about the regulatory processes that control the fluent and fragile balance that is called homeostasis. Here we demonstrate that JAK-STAT signaling, beyond its role in immune responses, is a major regulator of immune cell homeostasis. We investigated JAK-STAT-mediated transcription and chromatin accessibility across 12 mouse models, including knockouts of all STAT transcription factors and of the TYK2 kinase. Baseline JAK-STAT signaling was detected in CD8 + T cells and macrophages of unperturbed mice—but abrogated in the knockouts and in unstimulated immune cells deprived of their normal tissue context. We observed diverse gene-regulatory programs, including effects of STAT2 and IRF9 that were independent of STAT1. In summary, our large-scale dataset and integrative analysis of JAK-STAT mutant and wild-type mice uncovered a crucial role of JAK-STAT signaling in unstimulated immune cells, where it contributes to a poised epigenetic and transcriptional state and helps prepare these cells for rapid response to immune stimuli. Bock and colleagues perform integrative analysis of JAK-STAT mutant mice and find JAK-STAT signaling regulates CD8 + T cell and macrophage homeostasis by contributing to a poised epigenetic and transcription-regulatory state, preparing cells to rapidly respond to stimuli.

Manufacturing of biopharmaceuticals involves recombinant protein expression in host cells followed by extensive purification of the target protein. Yet, host cell proteins (HCPs) may persist in the final drug product, potentially reducing its quality with respect to safety and efficacy. Consequently, residual HCPs are closely monitored during downstream processing by techniques such as enzyme-linked immunosorbent assay (ELISA) or high-performance liquid chromatography combined with tandem mass spectrometry (HPLC-MS/MS). The latter is especially attractive as it provides information with respect to protein identities. Although the applied HPLC-MS/MS methodologies are frequently optimized with respect to HCP identification, acquired data is typically analyzed using standard settings. Here, we describe an improved strategy for evaluating HPLC-MS/MS data of HCP-derived peptides, involving probabilistic protein inference and peptide detection in the absence of fragment ion spectra. This data analysis workflow was applied to data obtained for drug products of various biotherapeutics upon protein A affinity depletion. The presented data evaluation strategy enabled in-depth comparative analysis of the HCP repertoires identified in drug products of the monoclonal antibodies rituximab and bevacizumab, as well as the fusion protein etanercept. In contrast to commonly applied ELISA strategies, the here presented workflow is process-independent and may be implemented into existing HPLC-MS/MS setups for drug product characterization and process development.

Deep neural networks display impressive performance but suffer from limited interpretability. Biology-inspired deep learning, where the architecture of the computational graph is based on biological knowledge, enables unique interpretability where real-world concepts are encoded in hidden nodes, which can be ranked by importance and thereby interpreted. In such models trained on single-cell transcriptomes, we previously demonstrated that node-level interpretations lack robustness upon repeated training and are influenced by biases in biological knowledge. Similar studies are missing for related models. Here, we test and extend our methodology for reliable interpretability in P-NET, a biology-inspired model trained on patient mutation data. We observe variability of interpretations and susceptibility to knowledge biases, and identify the network properties that drive interpretation biases. We further present an approach to control the robustness and biases of interpretations, which leads to more specific interpretations. In summary, our study reveals the broad importance of methods to ensure robust and bias-aware interpretability in biology-inspired deep learning.

Ex vivo cell cultures are reductionist models that enable cost effective, precisely controlled experiments with fewer ethical concerns than in vivo conditions. This results in their extensive use in diverse studies, especially large-scale genetic and drug perturbation screens. Although it is commonly accepted that ex vivo models do not fully recapitulate in vivo conditions, the molecular effects of model systems on cells in homeostasis and on cells undergoing drug or genetic perturbations are poorly understood. Using a CRISPR knockout (KO) screen with transcriptome read-out (Perturb-seq) of hematopoietic progenitor cells cultured ex vivo and grown in vivo, we analyzed the effects of ex vivo culture on unperturbed and perturbed cells. Unperturbed cells cultured ex vivo generally showed reduced basal interferon signatures and increased growth and metabolism signatures. These differences in unperturbed cells translated to differences between KO effects observed in vivo and ex vivo. We validated this impact of the model system on KO effects in an additional dataset of genetic KOs in bulk-sorted splenic immune cells, which confirmed our results. Interestingly, genes and molecular pathways with different KO effects were partly predicted by differences between unperturbed cells cultured in vivo and ex vivo. We therefore evaluated the performance of state-of-the-art models in predicting in vivo KO effects from ex vivo KO effects. This proved challenging, demonstrating the need for further developments. In summary, our study reveals differences in culture models, suggests approaches to improve culture models, and provides a test case for computational predictions of perturbation effects.

Cardiac fibrosis is mediated by the persistent activity of myofibroblasts, which differentiate from resident cardiac fibroblasts in response to tissue damage and stress signals. The signaling pathways and transcription factors regulating fibrotic transformation have been thoroughly studied. In contrast, the roles of chromatin factors in myofibroblast differentiation and their contribution to pathogenic cardiac fibrosis remain poorly understood. Here, we combined bulk and single-cell CRISPR screens to characterize the roles of chromatin factors in the fibrotic transformation of primary cardiac fibroblasts. We uncover strong regulators of fibrotic states including Srcap and Kat5 chromatin remodelers. We confirm that these factors are required for functional processes underlying fibrosis including collagen synthesis and cell contractility. Using chromatin profiling in perturbed cardiac fibroblasts, we demonstrate that pro-fibrotic chromatin complexes facilitate the activity of well-characterized pro-fibrotic transcription factors. Finally, we show that KAT5 inhibition alleviates fibrotic responses in patient-derived human fibroblasts.