Discovery of a structural class of antibiotics with explainable deep learning.

The discovery of novel structural classes of antibiotics is urgently needed to address the ongoing antibiotic resistance crisis1-9. Deep learning approaches have aided in exploring chemical spaces1,10-15; these typically use black box models and do not provide chemical insights. Here we reasoned that the chemical substructures associated with antibiotic activity learned by neural network models can be identified and used to predict structural classes of antibiotics. We tested this hypothesis by developing an explainable, substructure-based approach for the efficient, deep learning-guided exploration of chemical spaces. We determined the antibiotic activities and human cell cytotoxicity profiles of 39,312 compounds and applied ensembles of graph neural networks to predict antibiotic activity and cytotoxicity for 12,076,365 compounds. Using explainable graph algorithms, we identified substructure-based rationales for compounds with high predicted antibiotic activity and low predicted cytotoxicity. We empirically tested 283 compounds and found that compounds exhibiting antibiotic activity against Staphylococcus aureus were enriched in putative structural classes arising from rationales. Of these structural classes of compounds, one is selective against methicillin-resistant S. aureus (MRSA) and vancomycin-resistant enterococci, evades substantial resistance, and reduces bacterial titres in mouse models of MRSA skin and systemic thigh infection. Our approach enables the deep learning-guided discovery of structural classes of antibiotics and demonstrates that machine learning models in drug discovery can be explainable, providing insights into the chemical substructures that underlie selective antibiotic activity.

Discovering small-molecule senolytics with deep neural networks.

The accumulation of senescent cells is associated with aging, inflammation and cellular dysfunction. Senolytic drugs can alleviate age-related comorbidities by selectively killing senescent cells. Here we screened 2,352 compounds for senolytic activity in a model of etoposide-induced senescence and trained graph neural networks to predict the senolytic activities of >800,000 molecules. Our approach enriched for structurally diverse compounds with senolytic activity; of these, three drug-like compounds selectively target senescent cells across different senescence models, with more favorable medicinal chemistry properties than, and selectivity comparable to, those of a known senolytic, ABT-737. Molecular docking simulations of compound binding to several senolytic protein targets, combined with time-resolved fluorescence energy transfer experiments, indicate that these compounds act in part by inhibiting Bcl-2, a regulator of cellular apoptosis. We tested one compound, BRD-K56819078, in aged mice and found that it significantly decreased senescent cell burden and mRNA expression of senescence-associated genes in the kidneys. Our findings underscore the promise of leveraging deep learning to discover senotherapeutics.

Transcriptomic characterization of Lonrf1 at the single-cell level under pathophysiological conditions.

The LONRF family of proteins consists of three isozymes, LONRF1-3, which harbors RING (really interesting new gene) domain and Lon substrate binding domain. We have recently identified LONRF2 as a protein quality control ubiquitin ligase that acts predominantly in neurons. LONRF2 selectively ubiquitylates misfolded or damaged proteins for degradation. LONRF2-/- mice exhibit late-onset neurological deficits. However, the physiological implications of other LONRF isozymes remain unclear. Here, we analysed Lonrf1 expression and transcriptomics at the single-cell level under normal and pathological conditions. We found that Lonrf1 was ubiquitously expressed in different tissues. Its expression in LSEC and Kupffer cells increased with age in the liver. Lonrf1high Kupffer cells showed activation of regulatory pathways of peptidase activity. In normal and NASH (nonalcoholic steatohepatitis) liver, Lonrf1high LSECs showed activation of NF-kB and p53 pathways and suppression of IFNa, IFNg and proteasome signalling independent of p16 expression. During wound healing, Lonrf1high/p16low fibroblasts showed activation of cell growth and suppression of TGFb and BMP (bone morphogenetic protein) signalling, whereas Lonrf1high/p16high fibroblasts showed activation of WNT (wingless and Int-1) signalling. These results suggest that although Lonrf1 does not seem to be associated with senescence induction and phenotypes, LONRF1 may play a key role in linking oxidative damage responses and tissue remodelling during wound healing in different modes in senescent and nonsenescent cells.

Blocking PD-L1-PD-1 improves senescence surveillance and ageing phenotypes.

The accumulation of senescent cells is a major cause of age-related inflammation and predisposes to a variety of age-related diseases1. However, little is known about the molecular basis underlying this accumulation and its potential as a target to ameliorate the ageing process. Here we show that senescent cells heterogeneously express the immune checkpoint protein programmed death-ligand 1 (PD-L1) and that PD-L1+ senescent cells accumulate with age in vivo. PD-L1- cells are sensitive to T cell surveillance, whereas PD-L1+ cells are resistant, even in the presence of senescence-associated secretory phenotypes (SASP). Single-cell analysis of p16+ cells in vivo revealed that PD-L1 expression correlated with higher levels of SASP. Consistent with this, administration of programmed cell death protein 1 (PD-1) antibody to naturally ageing mice or a mouse model with normal livers or induced nonalcoholic steatohepatitis reduces the total number of p16+ cells in vivo as well as the PD-L1+ population in an activated CD8+ T cell-dependent manner, ameliorating various ageing-related phenotypes. These results suggest that the heterogeneous expression of PD-L1 has an important role in the accumulation of senescent cells and inflammation associated with ageing, and the elimination of PD-L1+ senescent cells by immune checkpoint blockade may be a promising strategy for anti-ageing therapy.

Senolysis by glutaminolysis inhibition ameliorates various age-associated disorders.

Removal of senescent cells (senolysis) has been proposed to be beneficial for improving age-associated pathologies, but the molecular pathways for such senolytic activity have not yet emerged. Here, we identified glutaminase 1 (GLS1) as an essential gene for the survival of human senescent cells. The intracellular pH in senescent cells was lowered by lysosomal membrane damage, and this lowered pH induced kidney-type glutaminase (KGA) expression. The resulting enhanced glutaminolysis induced ammonia production, which neutralized the lower pH and improved survival of the senescent cells. Inhibition of KGA-dependent glutaminolysis in aged mice eliminated senescent cells specifically and ameliorated age-associated organ dysfunction. Our results suggest that senescent cells rely on glutaminolysis, and its inhibition offers a promising strategy for inducing senolysis in vivo.

A 3D tissue model-on-a-chip for studying the effects of human senescent fibroblasts on blood vessels.

All human tissues experience aging that eventually causes organ dysfunction and disease. Cellular senescence was discovered in fibroblasts cultured in vitro. In adults, it is a primary defense mechanism against cancer, but also a major contributor to lifespan limits and disorders associated with aging. To assess how human blood vessels change in an aged environment, we developed an elementary tissue model-on-a-chip that comprises an in vitro three-dimensional model of a blood vessel embedded in a collagen gel with young or senescent skin fibroblasts. We found that senescent fibroblasts mechanically altered the surrounding extracellular matrix by exerting excessive traction stress. We then found that senescent fibroblasts induced sprouting angiogenesis of a microvessel via their senescence-associated secretory phenotype (SASP). Finally, we gathered evidence that the mechanical changes of the microenvironment play a role in sustaining SASP-induced angiogenesis. The model proved useful in monitoring morphological changes in blood vessels induced by senescent fibroblasts while controlling the proportion of senescent cells, and enabled the study of SASP inhibitors, a class of drugs useful in aging and cancer research.

Generation of a p16 Reporter Mouse and Its Use to Characterize and Target p16high Cells In Vivo.

Cell senescence plays a key role in age-associated organ dysfunction, but the in vivo pathogenesis is largely unclear. Here, we generated a p16-CreERT2-tdTomato mouse model to analyze the in vivo characteristics of p16high cells at a single-cell level. We found tdTomato-positive p16high cells detectable in all organs, which were enriched with age. We also found that these cells failed to proliferate and had half-lives ranging from 2.6 to 4.2 months, depending on the tissue examined. Single-cell transcriptomics in the liver and kidneys revealed that p16high cells were present in various cell types, though most dominant in hepatic endothelium and in renal proximal and distal tubule epithelia, and that these cells exhibited heterogeneous senescence-associated phenotypes. Further, elimination of p16high cells ameliorated nonalcoholic steatohepatitis-related hepatic lipidosis and immune cell infiltration. Our new mouse model and single-cell analysis provide a powerful resource to enable the discovery of previously unidentified senescence functions in vivo.