Although scientists have long studied how memories are formed in the brain, how certain memories persist over time for learning and cognitive function remains unclear.
Removing Ank2 function led to significantly impaired memory in mice after after two weeks. Under normal conditions, these mice showed standard locomotion, sociability, and recent memory immediately after learning.
Astrocytes lacking Ank2 formed significantly less physical contacts with nearby engram neurons, the specialized neurons for memory storage. Additionally, the maintenance of long-term potentiation (LTP) was impaired while normal synaptic transmission remained intact. The findings suggest that astrocytes stabilize the neural circuits required for preserving memories long after they are formed.
On the molecular level, researchers found that Ank2 is required for brain-derived neurotrophic factor (BDNF) signaling through the astrocytic TrkB.T1 receptor and IP3R2-mediated calcium signaling. In the absence of Ank2, calcium signaling weakened, astrocytes failed to undergo normal structural remodeling, and showed reduced ability to maintain contacts with memory-encoding neurons.
The researchers further demonstrated that hippocampal BDNF infusion normally strengthens long-term memory persistence, but this effect disappeared when astrocytic Ank2 was deleted, showing that Ank2 is essential for BDNF-dependent memory stabilization.
To determine whether astrocytic BDNF signaling alone is sufficient to enhance memory, the team developed an optogenetic tool called Opto-T1. Activation of this pathway promoted astrocyte remodeling, maintained long-term potentiation, and significantly enhanced remote memory without affecting recent memory.
“Our findings show that astrocytes are not passive support cells, but active regulators that determine how long memories last,” said Wuhyun Koh, PhD, senior research fellow at Institute for Basic Science (IBS) and corresponding author of the study. “By identifying Ank2 as a key regulator of astrocyte remodeling and BDNF signaling, we have uncovered a new mechanism that helps stabilize long-term memories and opens new avenues for understanding and potentially treating memory disorders.”
The researchers indicate the study provides a new framework for understanding how astrocytes contribute to neurological diseases.
BackgroundPropofol is widely used in pediatric anesthesia, but it has been implicated in adverse effects on brain development following repeated early-life exposure. Bag3, a co-chaperone protein involved in proteostasis and the neuronal stress response, may play a critical role in regulating synaptic function during early brain maturation.MethodsNeonatal mice were treated with propofol on postnatal days 5–7. Hippocampal neurogenesis was assessed via BrdU immunofluorescence. Synaptic proteins (PSD95, SNAP25) and Bag3 expression were measured by Western blotting. Behavioral performance in adolescence was evaluated using open-field, elevated plus-maze, Morris water maze, Y-maze, and T-maze tests.ResultsPropofol exposure significantly reduced proliferative activity (BrdU incorporation) in the dentate gyrus and decreased PSD95 and SNAP25 expression in both the cortex and hippocampus. Bag3 expression was markedly upregulated, accompanied by a mild increase in its phosphorylated form. Behaviorally, propofol-treated mice showed anxiety-like behavior and impairments in spatial learning and working memory.ConclusionThese findings suggest that early-life exposure to propofol impairs neurogenesis and synaptic plasticity. This process is temporally associated with the upregulation of stress-responsive co-chaperone BAG3, which precedes the of synaptic protein homeostasis. While causal relationship remains to be established, these findings identifies BAG3 as a candidate correlative marker of anesthetic-induced neurotoxicity and highlights it as a target for future mechanistic studies.
Treatment-resistant depression (TRD) remains a major clinical challenge. Intranasal esketamine, used adjunctively with an oral antidepressant, has been evaluated in randomized trials, but uncertainty persists regarding the magnitude and consistency of benefit, durability, and key harms. This systematic review and meta-analysis included randomized controlled trials comparing intranasal esketamine plus an oral antidepressant versus placebo nasal spray plus the same oral antidepressant in TRD. Acute induction (≈4 weeks) and maintenance randomized-withdrawal phases were analyzed separately. Depression outcomes were assessed primarily using the Montgomery–Åsberg Depression Rating Scale (MADRS), and functional outcomes using the Sheehan Disability Scale (SDS). Two reviewers independently screened studies, extracted data, and assessed risk of bias using RoB 2.0. Random-effects models pooled mean differences (MD) for continuous outcomes, risk ratios (RR) for binary outcomes, and hazard ratios (HR) for relapse prevention. Certainty of evidence was rated using GRADE. From 1,518 records, nine reports representing six unique RCTs (1,836 participants) were included. Four acute induction RCTs (n=937) showed greater symptom reduction at day 28 with esketamine (MADRS MD −2.99, 95% CI −5.10 to −0.89; I²=48.5%). Rapid improvement was evident by day 2 (MD −3.25, 95% CI −4.65 to −1.85). Esketamine increased day-28 response (RR 1.44, 95% CI 1.20–1.74) and remission (RR 1.52, 95% CI 1.20–1.92), corresponding to approximately +154 responders and +106 remitters per 1,000 patients, respectively, based on pooled control risks. Functioning improved (SDS MD −1.70, 95% CI −2.61 to −0.79). Two maintenance randomized-withdrawal RCTs (n=899) demonstrated reduced relapse risk with continued esketamine (HR 0.51, 95% CI 0.42–0.62; I²=0%). In acute induction, esketamine increased any treatment-emergent adverse event (TEAE) (RR 1.37, 95% CI 1.25–1.50) and discontinuation due to adverse events (RR 2.68, 95% CI 1.35–5.29), with notable increases in dissociation (RR 7.33, 95% CI 4.49–11.98) and blood pressure increased events (RR 3.96, 95% CI 2.24–7.01). Maintenance TEAE rates were similar between groups (RR 1.07, 95% CI 0.99–1.17). Intranasal esketamine plus an oral antidepressant provides rapid, modest acute improvement and reduces relapse risk during maintenance among stabilized responders/remitters, but increases acute adverse events, supporting use within supervised care and individualized benefit–risk assessment.
At the 74th American Society for Mass Spectrometry (ASMS) Conference in San Diego, the obvious story was hardware. Vendors showcased faster acquisition, higher sensitivity, alternative fragmentation, spatial workflows, and software ecosystems. New or highlighted platforms and workflows came from Waters, Thermo Fisher Scientific, Sciex, Bruker, Biognosys, and Evosep.
But after several days of talks, posters, hallway conversations, and interviews with senior figures in mass spectrometry (MS)-based proteomics, the deeper story was not simply that instruments are getting better. The field is beginning to look past the instrument. The mass spectrometer is still central, but the question is shifting: what has to happen around it for proteomics to become clinically useful, scalable, trusted, and routine?
Beyond the instrument
Jennifer Van Eyk, PhD, professor of cardiology, biomedical sciences, pathology, and laboratory medicine, and director of the Advanced Clinical Biosystems Research Institute at Cedars-Sinai Health Science University, put it most directly: “I think mass spec is no longer the limitation. We have the sensitivity, the throughput, and the accuracy at discovery and targeted levels.”
Jennifer Van Eyk, PhD [Gustav Ceder]
That is a remarkable statement in a field long defined by instrument performance. Van Eyk was not saying that MS innovation is finished. She pointed to continuing gains in quantitation, protein structure, conformational analysis, post-translational modifications (PTMs), top-down proteomics, and protein dynamics. But for clinical impact, she argued, the next bottlenecks are increasingly sample preparation, data analysis, standardization, harmonization, and quality control.
Joshua Coon, PhD, professor of biomolecular chemistry at the University of Wisconsin-Madison and the Pyle Chair at the Morgridge Institute for Research, saw instrument speed as the force opening new applications. Faster scanning mass analyzers are allowing deeper proteome coverage, more post-translational modification (PTM) measurements, and shorter runs. Ryan Kelly, PhD, professor of chemistry and biochemistry at Brigham Young University, framed the same shift as a throughput problem. “Now the mass spec is so fast that we need to figure out how to feed it faster,” he said. In plasma proteomics, Coon said, faster instruments, nanoparticle-based enrichment, and improved chromatography are moving the field from hundreds
Joshua Coon, PhD [Gustav Ceder]
toward thousands of detectable proteins in blood.
John R. Yates III, PhD, the John Lytton Young Endowed Chair in the department of integrative structural and computational biology at Scripps Research, highlighted electron activation dissociation methods and the possibility that high-throughput workflows could push MS deeper into plasma and population-level studies. He described targeted affinity platforms as powerful for “known knowns” because they measure targets defined in advance. “But with mass spectrometry,” he added, “you can look for unknown unknowns, which is where the gold lies.”
John R. Yates III, PhD [Gustav Ceder]
The point cuts to the heart of where the field now stands, and a recurring ASMS tension. The future of proteomics is not a choice between platforms. It is a division of labor. Targeted affinity technologies have become central to large-scale plasma proteomics and population studies. MS remains uniquely powerful for unbiased discovery, tissue proteomics, complex sample matrices, protein modifications, structural diversity, and biology that is not yet named.
From depth to trust
If the first era of modern proteomics was about seeing more, the next may be about measuring better. Devin Schweppe, PhD, assistant professor in the Department of Genome Sciences at the University of Washington, described the current moment as a “duality.” Instruments can now deliver deep coverage, and computational tools are making interpretation faster. Together, he said, they are creating “a comfort level with trusting the data.”
Devin Schweppe, PhD [Gustav Ceder]
Trust came up repeatedly. For discovery biology, a strong signal can be enough to generate a hypothesis. For clinical practice, it is not. Van Eyk said clinical-grade assays are “way harder than people think they are.” A research study can iterate. A clinical assay has to deliver the same measurement today, in five weeks, in six months, and years later. Once a test is locked, “you can’t go, ‘Oh no, we should have had this extra protein in there,’” she said. “It’s done.”
This distinction matters across assay types. Targeted MS methods such as multiple reaction monitoring (MRM) and parallel reaction monitoring (PRM) can provide absolute quantification, but only for preselected proteins. Data-independent acquisition (DIA), meanwhile, has moved discovery proteomics closer to translation by improving reproducibility and scalability. DIA is still often used for relative quantification, but its ability to capture patterns across tens or hundreds of proteins may become important as clinical decision-making moves beyond single biomarkers and reference intervals.
The field is responding to these demands. David Kotol, PhD, R&D manager at ProteomEdge, discussed an independently validated nine-protein plasma panel designed to improve emergency department triage and imaging decisions for patients with suspected venous thromboembolism, compared with D-dimer alone.
David Kotol, PhD [Gustac Ceder]
Kotol described a shift “from relative protein measurements toward robust, multiplexed absolute quantification.” He emphasized stable isotope-labeled protein standards added early in sample preparation to monitor digestion efficiency, downstream analytical variation, and multi-peptide quantification. These standards cannot remove variation introduced during sample collection, handling, or storage. But they can make the analytical workflow more transparent and transferable.
The clinical gap
Mathieu Lavallée-Adam, PhD, associate professor in the department of biochemistry, microbiology and immunology and director of the specialization in bioinformatics at the University of Ottawa, gave the least glamorous answer to what still blocks clinical translation. “My answer is going to be boring,” he said. “It’s going to be education.”
Mathieu Lavallée-Adam, PhD [Gustav Ceder]
Lavallée-Adam argued that many clinicians and biomedical researchers still do not fully understand what modern MS can do. Too often, the outside view is still: give me a list of differentially expressed proteins. But MS-based proteomics has moved beyond lists, into proteoforms, structural information, PTMs, protein dynamics, and flexible acquisition. “We’re past that now,” he said. “The main barrier is our inability to communicate the possibilities that we offer.”
Sasha Singh, PhD, assistant professor of medicine at Harvard Medical School, associate scientist at Brigham and Women’s Hospital, and director of proteomics research at the Center for Interdisciplinary Cardiovascular Sciences (CICS), described this translation role from inside a hospital environment. “That’s actually my role at the hospital,” Singh said. “I am a liaison between the technology and the application scientist.”
Sasha Singh, PhD [Gustav Ceder]
The translation is becoming harder because proteomics is diversifying. End users often need to distinguish among discovery MS, which can provide broad relative quantification; targeted MS, which can provide absolute concentrations for selected proteins; and targeted affinity proteomics, which can scale well for plasma cohorts but is limited by predefined assays and available binding reagents. Singh added that different technologies may produce profiles that do not fully overlap. Rather than treating that as a failure, she suggested it reveals something real: the circulation contains many subproteomes, and different technologies enrich different views.
AI with guardrails
No 2026 conference escapes artificial intelligence (AI), and ASMS was no exception. But the mood among the researchers was cautious rather than breathless.
Lavallée-Adam said agent-based AI was dominating conversations in his part of the field. The dream is seductive: put a sample on an instrument, ask an AI agent to maximize protein identifications or optimize a method, and let it select the best protocol. But he drew a clear line between potential and reality. “Are such agents really driving change? It’s unclear at this point,” he said. “I think it’s unproven.”
Still, AI-assisted acquisition strategies are entering workflows. Lavallée-Adam’s group works on real-time MS data acquisition, where software analyzes data as it is acquired and adapts the run to the biological question. Instead of measuring the same abundant proteins repeatedly, the system can decide it has seen enough and move on to new targets. In that sense, AI becomes less a magical oracle than an instrument assistant.
Faster instruments are generating more data, and faster analysis is needed to keep up. Schweppe also argued that open-source tools remain essential because they let laboratories build on one another’s work rather than rebuild it.
More than abundance
Much of the clinical proteomics effort is focused on plasma because it is minimally invasive and suitable for screening, longitudinal sampling, and routine monitoring. But even in blood, researchers are learning that plasma is only part of the story.
Roman Fischer, PhD, associate professor and head of the Discovery Proteomics Facility at the Target Discovery Institute, University of Oxford, pushed the conversation back toward biology. Plasma alone does not capture the full circulating system, he noted. Peripheral blood mononuclear cells, extracellular vesicles, microvesicles, and other compartments may contain disease-relevant information that conventional workflows miss. “We have to be more sophisticated in addressing the compartments of the blood,” Fischer said.
Roman Fischer, PhD [Gustav Ceder]
He also pointed to the proteoform problem. A single gene can give rise to many transcripts, isoforms, modified proteins, and glycosylated forms. These differences may affect activity, localization, disease pathways, and therapy response. Capturing that diversity is not possible with targeted affinity assays alone. It requires deeper characterization of the proteome, not only quantification.
Yates offered a clinical example. His group has been developing protein-footprinting approaches that can detect conformational changes in proteins in blood. In one transthyretin amyloid cardiomyopathy project, he said, abundance alone was not the answer. The important signal was how the protein folded or misfolded. That kind of assay moves proteomics beyond proteins going up or down, into structural disease biology.
Van Eyk’s work on remote sampling devices pointed to another future: patient-collected blood samples that make longitudinal cardiovascular studies easier, more inclusive, and better matched to real clinical questions.
In the background was a broader translational arc: discovery, verification, clinical validation, health economics, and access. Plasma proteomics highlights included Lekha Sleno, PhD, professor at Université du Québec à Montréal, who is combining nanoparticle enrichment with isotope-enabled targeted proteomics, and a CinderBio breakfast seminar featuring Fredrik Edfors, PhD, assistant professor at KTH Royal Institute of Technology and SciLifeLab, and Simion Kreimer, PhD, senior research project advisor in the Proteomics and Metabolomics Core at Cedars-Sinai Health Science University.
The seminar focused on accelerated plasma proteomics, rapid digestion workflows, stable isotope standards, Human Protein Atlas resources, and faster enzyme workflows that can reduce lead times. The common message was that sample preparation, quantification, and validation may become as decisive as instrument resolution.
The next bottleneck
ASMS 2026 was not short on technical spectacle. High-resolution instruments, electron-based fragmentation, narrow-window DIA, rapid acquisition, MS imaging, top-down workflows, and AI-enabled software all had their moment. But the most interesting conversations were less about spectacle than maturity.
Proteomics is no longer trying only to prove that it can see more. It is trying to prove that it can measure consistently, explain biology more deeply, support drug development, fit into clinical laboratories, and eventually improve patient decisions.
That means the next bottleneck is distributed across the ecosystem: sample preparation, standards, software, education, reimbursement, clinical menus, regulatory validation, open tools, and the ability to translate technical power into something a clinician can use.
Longer term, integrated proteomics, other omics, imaging, clinical data, and AI may support not only single biomarkers, but interpretable molecular patterns, longitudinal trajectories, and digital-twin-like models of patient biology.
The field spent decades making proteins visible. The next challenge is making proteomic measurements dependable enough to act on.
Jaak Panksepp spent nearly five decades mapping the primary-process affective systems of the mammalian brain across different species, producing a framework of considerable empirical power that is functionally invisible within psychiatry. Psychiatry has not built upon that literature in human contexts. Consequently, Affect has never occupied the foundational place in psychiatry that the evidence warrants. This paper attempts to close that gap and assembles twelve converging lines of evidence for subcortical primacy of Affect, including evidence not previously synthesized for this purpose: pseudobulbar affect, gelastic and dacrystic epilepsy, double dissociation of volitional and emotional facial expression, affective blindsight, and neonatal emotional behavior. The evidence shows that Affect is generated subcortically, the cortex modulates rather than creates it, and when cortical regulation is removed or impaired, affective states persist or intensify. From this evidence, personality is best understood as an individual’s position in a configuration space defined by subcortical Affect generation parameters and cortical regulatory capacity, with psychopathology occurring when signal intensity chronically exceeds or overwhelms regulatory capacity. This framework generates a specific clinical prediction: conditions that co-occur with personality pathology at rates incompatible with independent etiology are expressions of the same affective architecture through different conditioned channels, not independent diseases. Evidence confirms this directly: when personality pathology improves, its comorbid conditions decline substantially; when it does not, they persist. Further, the affective-regulatory framework transforms the clinical encounter by replacing a character verdict with a neurobiologically grounded account in which regulatory capacity is buildable and recovery is construction rather than correction.
Researchers in Japan have found that older adults with higher levels of vitamin C in their blood have a higher volume of gray matter in their brains and higher connectivity across brain regions involved in memory and attention. Published in PLOS One, their study raises the possibility of using dietary interventions to protect brain health as we age.
“What I found most fascinating about this research is that we were able to detect these subtle but significant associations between a single nutritional factor and large-scale brain networks by utilizing a robust, community-based cohort of over 2,000 older adults,” said Tomohiro Shintaku, MD, PhD, assistant professor of radiology at the Hirosaki University Graduate School of Medicine. “It truly highlights the potential impact of our everyday dietary habits on our brain structures.”
Previous research had linked diets high in vitamin C with lower risk of developing cognitive impairment in older adults. However, this study is the first to look directly at a potential link between vitamin C levels in blood and changes in brain structure and connectivity within a large participant cohort.
Shintaku and colleagues analyzed magnetic resonance imaging (MRI) scans and blood levels of vitamin C in 2,044 Japanese adults over the age of 64. They measured the volume of gray and white brain matter in each individual, as well as the connectivity between brain regions belonging to the default mode network, which are associated with important cognitive functions including autobiographical memory, future thinking, self-reference, and attention. Connectivity within the default mode network is also known to play a significant role in brain health, with lower connectivity being linked to cognitive impairment and Alzheimer’s disease, in addition to a number of psychiatric conditions including depression or schizophrenia.
After adjusting for other factors that affect brain structure and connectivity, such as age, physical activity, and education level, results showed that higher levels of vitamin C were associated with a higher volume of gray matter across several brain regions, and with higher connectivity within the default mode network.
“Our study demonstrates that higher plasma vitamin C levels are associated with better preserved structural connectivity of the default mode network, a key brain network involved in cognitive function,” said Shintaku. “This finding generates the exciting hypothesis that a diet rich in vitamin C might play a supportive role in maintaining brain health and mitigating age-related cognitive decline in older adults.”
More research will be needed to uncover the potential biological mechanisms driving this association and confirm whether there is a causal link between blood levels of vitamin C and changes in brain structure and connectivity. Going forward, the researchers plan to conduct studies looking at repeated measurements of vitamin C levels over time, accounting for additional lifestyle and nutritional factors that may influence these effects, and including cohorts with more diverse ethnicities and socioeconomic status.
A growing body of psychological and neuroscientific research suggests that moral injury (MI) involves maladaptive self-referential processing, including disruptions in moral identity, rigid negative self-appraisals, and impaired meaning-making following exposure to potentially morally injurious events (PMIEs). Building on Mindfulness-to-Meaning Theory (MMT), this paper proposes self-transcendence (ST)—a metacognitive state characterized by reduced self-focus, expanded awareness, transcendent affect, and prosocial meaning—as a potential mechanism for MI recovery. Within MMT, mindfulness practice is theorized to cultivate ST via decentering and meta-awareness, processes that broaden attentional scope, promote flexible cognitive reappraisal, and modulate habitual self-referential processing. Mindfulness and contemplative research further link ST to increased cognitive flexibility, reduced in shame-focused narrative self-processing, and adaptive integration of emotionally and morally disruptive experiences. Drawing on an integrative review of ST-consistent and MI-related mechanisms, this paper argues that fostering ST through mindfulness-based and contemplative practices may reduce rigid self-focus, expand interpretive frameworks of meaning, and support moral identity repair and meaning-making. Implications are discussed for designing interventions that intentionally cultivate ST as both standalone approaches and modular components, while acknowledging current limitations in measurement, readiness assessment, and the reliable induction of ST states.
IntroductionThe Attention Training Technique (ATT) is a brief metacognitive intervention recognised as a possibly efficacious standalone transdiagnostic treatment for emotional disorders. The cognitive and neuropsychological mechanisms underlying its clinical effects are of particular interest in understanding and developing the technique. The aim of the systematic review was to synthesise and evaluate the cognitive-attentional task performance and neurocognitive correlates of ATT in the context of theoretical mechanisms from which ATT is derived.MethodsFive electronic databases (PsycINFO, MEDLINE, PubMed, Web of Science and EMBASE) were searched from January 1990 to November 2025. Studies that used ATT as part of a metacognitive multi-component treatment package or combined with other therapy/technique(s) were excluded. Sample inclusion was diverse to capture effects on non-clinical and clinical individuals and across age groups for potential sub-group analyses.ResultsIn total, 20 studies with 1, 230 participants met the inclusion criteria. Four studies included clinical samples, four studies included non-clinical participants, two studies used experimental induction of pain or mind wandering, and 10 used healthy samples of which two used school children. Study quality varied from strong to weak with the majority receiving ‘moderate’ ratings. Across 14 cognitive-attentional tasks and three neural methodologies (EEG, fNIRS, fMRI), the review found small to large cognitive and neural effects associated with ATT. Nine cognitive tasks showed significant ATT-dependent effects in at least one study, with the most consistency shown on the emotional dot-probe. Neural findings across all methodologies converged, suggesting that ATT modulates cognitive control, frontoparietal, dorsal attention networks and reduces default mode network connectivity.DiscussionInterpretation and synthesis of findings based on the S-REF model are consistent with cognitive and neural effects involving reduced threat monitoring, improved executive control, and enhanced disengagement from self-referential processing; central theoretical mechanisms and design parameters of ATT. Where inconsistencies across study effects emerged, they may be due to heterogeneity in cognitive task and measurement factors and ATT protocol deviations. Future research on individual differences in neurocognitive effects associated with ATT across clinical and sub-clinical populations is needed. Studies must safeguard fidelity and adherence to the ATT protocol and improve reporting of these important factors.Systematic review registrationhttps://www.crd.york.ac.uk/prospero/, identifier CRD42024483053.
A cup of coffee can mean many things: a daily part of our routine, a moment of calm, a midday boost. But zoom in past the steam, past the roasted aromatics, down to the caffeine molecule itself, and a different story emerges. At the Texas A&M Health Institute of Biosciences and Technology, researchers have turned this everyday stimulant into something far more unexpected: a molecular “pause button” for engineered cells.
In a study published in the Journal of the American Chemical Society (JACS), the team unveiled CODS, a caffeine‑operated dissociation system built using AI‑guided de novo protein design. The paper, “AI‑Guided De Novo Design of a Caffeine‑Induced Protein Dissociation System,” describes how the group reprogrammed “an existing caffeine-responsive chemically induced proximity (CIP) module into a ligand-dependent dissociation system.”
“AI is changing how we design biology,” said senior author Yubin Zhou, MD, PhD. “Instead of relying only on protein parts that already exist in nature, we can now design new mini proteins with specific behaviors. Here, we used AI to help turn caffeine into a precise trigger for controlling engineered cells.”
A team of Texas A&M Health researchers led by Yubin Zhou, MD, PhD, is using caffeine to precisely control engineered cells, a step toward safer and more responsive therapies. [Texas A&M University]
The CODS system pairs a caffeine‑binding protein with a synthetic mini‑binder designed using the BindCraft platform. In the absence of caffeine, the two components stay locked together. Add caffeine, and the complex snaps apart, releasing the binder and shutting down the attached cellular function. As Tianlu Wang, PhD, a postdoctoral fellow in the Zhou lab, put it, “Many genetically-encoded molecular tools act like accelerators. CODS gives us something closer to a brake or pause button.”
The team demonstrated CODS across several biological contexts. In engineered gene circuits, caffeine addition sharply reduced transcriptional activity. In a rewired pyroptosis pathway, caffeine triggered inflammatory cell death by freeing the active domain of gasdermin D. And in perhaps the most translational example, CODS served as a conditional deactivator for CAR T cells,temporarily dampening their activity without destroying the therapeutic cells.
“Powerful therapies need powerful control,” Zhou said. “By combining AI‑designed proteins, high‑performance computing, and familiar small molecules, we are building a new language for communicating with engineered cells.”
The design process itself leaned heavily on computation. Graduate student Brendan McKee led the AI‑guided binder design and molecular modeling, while Tatsuki Nonomura spearheaded the molecular engineering and live‑cell validation. The Texas A&M High Performance Research Computing service provided the infrastructure needed to run large‑scale simulations. “High‑performance computing was essential for this project,” Zhou noted. “It helped us move from a conceptual idea to a functional molecular switch much faster.”
Although caffeine is not a therapeutic molecule, its safety and familiarity make it an appealing control signal. As Zhou emphasized, “Coffee will not replace medicine. But caffeine can help us imagine medicines that are more controllable, more responsive, and safer for patients.” The researchers’ next steps include further testing in therapeutic cells, animal models, and disease-relevant settings before moving toward clinical use.
CODS now joins a growing toolkit of AI‑designed molecular switches, offering a blueprint for future systems responsive to other safe, accessible molecules. As programmable cell therapies advance, the ability to modulate them with something as simple as caffeine may prove unexpectedly powerful.
The authors report that a patent application covering the CODS platform has been filed by Texas A&M University, with Y.Z., T.N., B.M., and T.W. listed as inventors (U.S. Provisional Patent Application No. 64/022,078).
