Polarization of low-grade inflammatory monocytes facilitates the pathogenesis of atherosclerosis. However, underlying mechanisms as well as approaches for resolving monocyte polarization conducive to the regression of atherosclerosis are not well established. In this report, we demonstrate that TRIF-related adaptor molecule (TRAM) mediated monocyte polarization in vivo and in vitro. TRAM controlled monocyte polarization through activating Src family kinase c-SRC, which not only induces STAT1/STAT5-regulated inflammatory mediators CCR2 and SIRP-α but also suppresses PPARγ-regulated resolving mediator CD200R. Enhanced PPARγ and Pex5 due to TRAM deficiency facilitated peroxisome homeostasis and reduction of cellular reactive oxygen species, further contributing to the establishment of a resolving monocyte phenotype. TRAM-deficient monocytes propagated the resolving phenotype to neighboring monocytes through CD200R-mediated intercellular communication. At the translational level, we show that TRAM-deficient mice were resistant to high-fat diet–induced pathogenesis of atherosclerosis. We further document that intravenous transfusion of TRAM-deficient resolving monocytes into atherosclerotic mice potently reduced the progression of atherosclerosis. Together, our data reveal that targeting TRAM may facilitate the effective generation of resolving monocytes conducive for the treatment of atherosclerosis.
BACKGROUND: Chronic inflammation initiated by inflammatory monocytes underlies the pathogenesis of atherosclerosis. However, approaches that can effectively resolve chronic low-grade inflammation targeting monocytes are not readily available. The small chemical compound 4-phenylbutyric acid (4-PBA) exhibits broad anti-inflammatory effects in reducing atherosclerosis. Selective delivery of 4-PBA reprogrammed monocytes may hold novel potential in providing targeted and precision therapeutics for the treatment of atherosclerosis. METHODS: Systems analyses integrating single-cell RNA sequencing and complementary immunologic approaches characterized key resolving characteristics as well as defining markers of reprogrammed monocytes trained by 4-PBA. Molecular mechanisms responsible for monocyte reprogramming were assessed by integrated biochemical and genetic approaches. The intercellular propagation of homeostasis resolution was evaluated by coculture assays with donor monocytes trained by 4-PBA and recipient naive monocytes. The in vivo effects of monocyte resolution and atherosclerosis prevention by 4-PBA were assessed with the high-fat diet-fed ApoE−/− mouse model with IP 4-PBA administration. Furthermore, the selective efficacy of 4-PBA-trained monocytes was examined by IV transfusion of ex vivo trained monocytes by 4-PBA into recipient high-fat diet-fed ApoE−/− mice. RESULTS: In this study, we found that monocytes can be potently reprogrammed by 4-PBA into an immune-resolving state characterized by reduced adhesion and enhanced expression of anti-inflammatory mediator CD24. Mechanistically, 4-PBA reduced the expression of ICAM-1 (intercellular adhesion molecule 1) via reducing peroxisome stress and attenuating SYK (spleen tyrosine kinase)-mTOR (mammalian target of rapamycin) signaling. Concurrently, 4-PBA enhanced the expression of resolving mediator CD24 through promoting PPARγ (peroxisome proliferator-activated receptor γ) neddylation mediated by TOLLIP (toll-interacting protein). 4-PBA-trained monocytes can effectively propagate anti-inflammation activity to neighboring monocytes through CD24. Our data further demonstrated that 4-PBA-trained monocytes effectively reduce atherosclerosis pathogenesis when administered in vivo. CONCLUSIONS: Our study describes a robust and effective approach to generate resolving monocytes, characterizes novel mechanisms for targeted monocyte reprogramming, and offers a precision therapeutics for atherosclerosis based on delivering reprogrammed resolving monocytes.
We present LazyNet, a compact one-step neural-ODE model for single-cell CRISPR activation/interference (A/I) that operates directly on two-snapshot (“pre → post”) measurements and yields parameters with clear mechanistic meaning. The core log–linear–exp residual block exactly represents multiplicative effects, so synergistic multi-locus responses appear as explicit components rather than opaque composites. On a 53k-cell × 18k-gene neuronal Perturb-seq matrix, a three-replica LazyNet ensemble trained under a matched 1 h budget achieved strong threshold-free ranking and competitive error (genome-wide r ≈ 0.67) while running on CPUs. For comparison, we instantiated transformer (scGPT-style) and state-space (RetNet/CellFM-style) architectures from random initialization and trained them from scratch on the same dataset and within the same 1 h cap on a GPU platform, without any large-scale pretraining or external data. Under these strictly controlled, low-data conditions, LazyNet matched or exceeded their predictive performance while using far fewer parameters and resources. A T-cell screen included only for generalization showed the same ranking advantage under the identical evaluation pipeline. Beyond prediction, LazyNet exposes directed, local elasticities; averaging Jacobians across replicas produces a consensus interaction matrix from which compact subgraphs are extracted and evaluated at the module level. The resulting networks show coherent enrichment against authoritative resources (large-scale co-expression and curated functional associations) and concordance with orthogonal GPX4-knockout proteomes, recovering known ferroptosis regulators and nominating testable links in a lysosomal–mitochondrial–immune module. These results position LazyNet as a practical option for from-scratch, low-data CRISPR A/I studies where large-scale pretraining of foundation models is not feasible.
Biological research traditionally relies on experimental methods, which can be inefficient and hinder knowledge transfer due to redundant trial-and-error processes and difficulties in standardizing results. The complexity of biological systems, combined with large volumes of data, necessitates precise mathematical models like ordinary differential equations (ODEs) to describe interactions within these systems. However, the practical use of ODE-based models is limited by the need for curated data, making them less accessible for routine research. To overcome these challenges, we introduce LazyNet, a novel machine learning model that integrates logarithmic and exponential functions within a Residual Network (ResNet) to approximate ODEs. LazyNet reduces the complexity of mathematical operations, enabling faster model training with fewer data and lower computational costs. We evaluate LazyNet across several biological applications, including HIV dynamics, gene regulatory networks, and mass spectrometry analysis of small molecules. Our findings show that LazyNet effectively predicts complex biological phenomena, accelerating model development while reducing the need for extensive experimental data. This approach offers a promising advancement in computational biology, enhancing the efficiency and accuracy of biological research.
We present LazyNet, a compact one-step neural-ODE model for single-cell CRISPR A/I that operates directly on two-snapshot (“pre → post”) measurements and yields parameters with clear mechanistic meaning. The core log–linear–exp residual block exactly represents multiplicative effects, so synergistic multi-locus responses appear as explicit components rather than opaque composites. On a 53k-cell × 18k-gene neuronal Perturb-seq matrix, a three-replica LazyNet ensemble trained under a matched 1-hour budget achieved strong threshold-free ranking and competitive error (genome-wide r≈0.67) while running on CPUs, compared with transformer and state-space baselines trained on a single V100 with the same time cap. A T-cell screen included only for generalization showed the same ranking advantage under the identical evaluation pipeline. Beyond prediction, LazyNet exposes directed, local elasticities; averaging Jacobians across replicas produces a consensus interaction matrix from which compact subgraphs are extracted and evaluated at the module level. The resulting networks show coherent enrichment against authoritative resources (e.g., large-scale co-expression and curated functional associations) and concordance with orthogonal GPX4-knockout proteomes, recovering known ferroptosis regulators and nominating testable links in a lysosomal–mitochondrial–immune module.
Social memory is essential for daily life, and its impairment is associated with psychiatric disorders. The dorsal CA2 (dCA2) region of the hippocampus is known to support social memory formation, yet how it processes social information remains unclear. Here we show that, in male mice, a specific group of dCA2 neurons-cholecystokinin-positive interneurons (CCK-INs)-acts as a hub that receives social cues and activates dCA2 pyramidal neurons (PyNs). Upon social encounters, CCK-INs are activated and suppress parvalbumin-positive interneurons (PV-INs) via GABAB receptors, thereby releasing PyNs from inhibition. The removal of Neuroligin 1, an autism-associated protein, from CCK-INs disrupts their function and leads to social memory deficits. Importantly, dCA2 CCK-INs receive direct glutamatergic inputs from the medial septum, and chemogenetic activation of these CCK-INs rescues memory deficits caused by septal inhibition. These findings delineate a important circuit mechanism for social memory and highlight potential therapeutic targets.