Working memory holds information for a few seconds while we read, calculate, or plan. Its capacity is limited, and over time the trace fades, but what exactly sets this time limit remained unclear. Classical models pointed to the prefrontal cortex, yet recent work has shown that the thalamus sustains cortical representations over long intervals. Research in this vein includes neighboring studies: miniature brains demonstrate habituation to stimuli, brain MRI learns not to forget, and memory depends not on the number of cells but on their order. Our work stands somewhat apart: it goes down to the molecular level and asks which lipid signals in the thalamus govern how long information is retained. Previously, we found a genetic locus in mice that influences working memory, and within it the receptor Gpr12, whose enhancement in the thalamus improved memory. What remained was to understand what activates this receptor under natural conditions.
First, we obtained a cryo-electron microscopy structure of Gpr12 in its active state, bound to a G protein. This revealed two regulatory sites: a classical orthosteric pocket and an unusual lateral pocket in the membrane. Mutations in the lateral pocket strongly altered receptor activity, while mutations in the orthosteric pocket did not, indicating ligand-dependent activation. To find the natural ligand, we tagged Gpr12 with an epitope tag, isolated it from mouse thalamus, and performed lipidomic analysis. Among the enriched lipids was anandamide, an endocannabinoid with a 20:4 chain and an ethanolamine headgroup. In cell assays, anandamide doubled Gpr12 activity, while related lipids (2-AG, arachidonic acid) had no effect. We then measured cyclic AMP in mouse thalamus in real time using the fluorescent sensor G-FLAMP2 while the animals performed a working memory task with delays of 15, 30, and 60 seconds. To test causality, we used the point mutation L171T, which abolished activation by anandamide but preserved the receptor's basal activity. Additionally, we applied proximity labeling with TurboID to see the protein environment of Gpr12 in thalamic neurons, and lipidomics after cognitive load. Finally, we administered anandamide to mice and an anandamide breakdown inhibitor to macaques, while simultaneously recording calcium activity in prefrontal cortex neurons.
The Gpr12 structure showed that the receptor does not self-activate, as previously thought, but switches from a basal state to an enhanced one under the action of a ligand. That ligand turned out to be anandamide: it bound directly to Gpr12 with a dissociation constant of about 172 µM and increased cAMP production in cells by 100%. In mouse thalamus, anandamide caused not an increase but a sustained decrease in cAMP that lasted exactly as long as the retention period in the task. In mice with elevated Gpr12 expression, this decrease was more pronounced and persisted at delays of 30 and 60 seconds, whereas in control animals it disappeared after just 15 seconds. The L171T mutation in the lateral pocket abolished both the cAMP suppression and the memory improvement, although the receptor's basal activity was preserved. Proximity labeling revealed anandamide synthesis and breakdown enzymes, NAPE-PLD and FAAH-1, next to Gpr12, indicating local on-demand production of the ligand. After a 30-minute task with long delays, anandamide levels in the thalamus rose, while levels of other lipids did not change. Administering anandamide to mice prolonged cAMP suppression and improved memory at delays of 30 and 60 seconds, but had no effect in mice lacking Gpr12. In macaques, blocking anandamide breakdown with the inhibitor URB-597 improved working memory in a task with delays up to a minute, and the effect was dose-dependent and followed an inverted U shape. Finally, anandamide delivered locally to the thalamus selectively enhanced prefrontal cortex neuron activity during the second half of the retention period, without affecting activity during other phases of the task.
The work links a molecular mechanism to behavior on timescales of up to a minute, something that had not been achieved before. Unlike neighboring studies that describe network or cellular correlates of memory, this one reveals a specific lipid signaling pathway that sets the duration of retention. It is the first identification of an endogenous ligand for Gpr12 and the first proof that anandamide controls working memory through this receptor. The results explain why the thalamus can sustain cortical representations longer than the cortex itself: local anandamide production triggers slow cAMP suppression that acts as a timer. The discovery also changes how we think about endocannabinoids: they are usually associated with memory impairment through CB1 receptors, whereas here anandamide improves memory through Gpr12, specifically in the thalamus.
The topic could develop in several directions. First, we need to determine exactly how cAMP suppression in the thalamus enhances prefrontal cortex activity: likely through modulation of HCN or GIRK ion channels and switching of thalamic neuron firing modes. Second, it is worth testing whether this mechanism operates in other working memory tasks and in other species, including humans. Third, the lateral pocket of Gpr12 is an attractive drug target: it bears little resemblance to analogous regions of the related receptors Gpr3 and Gpr6, promising high selectivity. Finally, one could search for other lipid signals that set temporal windows of cognitive processes through GPCRs.
The discovery could influence the neurobiology of memory, the pharmacology of cognitive disorders, and research on the endocannabinoid system. It offers a concrete molecular target for treating working memory deficits in schizophrenia, ADHD, Alzheimer's disease, and age-related decline.
The immediate steps are to test the role of Gpr12 and anandamide in animal models of cognitive impairment, and to develop selective modulators of the lateral pocket. In parallel, it is worth studying how this pathway interacts with other neuromodulators in thalamocortical loops.
The work directly addresses an unsolved problem in neuroscience: how molecular events at synapses are transformed into stable network states that support working memory. It also links lipid metabolism to cognitive function, which matters for understanding neurodegenerative diseases where lipid disturbances precede symptoms. Finally, it raises the question of why different endocannabinoids (anandamide and 2-AG) have opposite effects on memory even though they act through related receptors.
Anandamide got its name from the Sanskrit word "ananda," meaning bliss or joy. It is the same lipid that activates CB1 cannabinoid receptors, but in the thalamus it acts on a completely different receptor, Gpr12, and instead of impairing memory it improves it.
The idea of a molecular timer that counts out the seconds of information retention echoes the concept of the "mental chronometer" from Ted Chiang's "Story of Your Life," where the perception of time and memory plays a key role. More broadly, controlling the duration of working memory resembles the technologies in Philip K. Dick's "We Can Remember It for You Wholesale," except that instead of implants, it is an endogenous lipid.
Not published on arXiv. Here are its page, the PDF and the formalization: you can check it yourself, not only read our account.