Taufik A. Valiante

Controlling whether an experience is encoded into long-term memory and thus later remembered is a crucial aspect of memory. Here we examined single-neuron activity in the human brain while participants formed memories for novel visual stimuli and later retrieved them with confidence ratings. Category-selective amygdala neurons exhibited elevated firing during encoding of later remembered items relative to later forgotten items. This effect was predictive of the episodic aspect of the memory, was restricted to the preferred category of each cell, emerged earlier and was stronger in the amygdala compared to the hippocampus, and did not depend on stimulus valence and arousal. Successful memory formation was accompanied by enhanced spike-field coherence between the activity of amygdala category cells and hippocampal field potentials. These findings demonstrate that stimulus-specific recruitment of the amygdala predicts memory formation and suggest category cells in the right amygdala as a target for interventions to treat disorders affecting episodic memory in humans.

Theoretical models propose that pattern separation, a computation thought to be implemented by the hippocampus, allows us to differentiate between familiar and novel items. However, whether pattern separation operates in the human hippocampus remains contested, with no established single-cell correlate. We recorded the single neuron activity of 3506 neurons in the human brain while 97 patients performed a recognition memory task in which novel images similar to previously seen images led to false memories and associated behavioral errors. We identified two kinds of memory selective neurons distributed across the brain: those responding differently to falsely recognized novel and correctly recognized familiar images in a manner compatible with pattern separation, and the other signaling the subject’s choice. At the population level, these cells predicted mnemonic ground truth in the hippocampus and the decision in the pre-supplementary motor area, illustrating the progression from mnemonic signals to decisions. Removing pattern separation-signaling cells abolished the continuous memory strength gradient present in the hippocampus, suggesting a role of these cells in separating memories of different strength. These results establish a single cell correlate for mnemonic pattern separation in the human hippocampus and show its behavioral relevance in episodic memory.

The decision criterion is foundational to theories of decision-making, yet little is known about its neural underpinnings. In memory-based decision making, the criterion sets the minimal memory strength for something to be familiar, but whether or how it is distinctly represented from memory strength is unknown. We recorded single neurons in the medial frontal cortex (MFC) and medial temporal lobe (MTL), both implicated in memory-based decisions, while participants made decisions under different decision criteria. We identified criterion-selective (CS) neurons in the MFC that tracked the criterion regardless of memory strength, and memory-selective (MS) neurons in both regions that tracked memory strength regardless of the criterion. CS neurons signaled the criterion before MS neurons signaled memory strength, and a race model incorporating both neuron types outperformed one using MS neurons alone. These findings reveal two independent cellular substrates, one for the decision criterion and one for memory strength, whose joint activity underlies memory-based decisions.

Pursuing longer-term goals relies on representations of task context that can be maintained over long periods of time and switched flexibly when goals change. Little is known about the neural substrate for maintaining context representations on the minute-timescale on which human behaviour frequently persists. Utilizing recordings in neurosurgical patients (total n = 25), we examined how neurons in the human medial frontal cortex and hippocampus persistently represent context. When cued explicitly, context was encoded in both brain areas and changed rapidly at task boundaries. The hippocampus exhibited a temporally dynamic code with fast decorrelation over time, preventing cross-temporal generalization. Medial frontal cortex exhibited a static code that decorrelated slowly, allowing generalization across time. When context needed to be inferred as a latent variable, the hippocampus encoded context with a static code. These findings reveal two formats for encoding minute-timescale context representations that were differentially engaged based on task demands.

Studying the role of individual neurons in behavior and disease requires tools for controlling neural activity with high spatiotemporal resolution. Implantable nanophotonic neural probes are capable of delivering targeted photostimulation to enable genetically distinct neurons to be selectively controlled, yet face barriers to achieving scalable emitter densities and lack sensors for monitoring feedback signals relevant to device operation. To address this, we developed laser-integrated nanophotonic neural probes, which feature hybrid-integrated laser diodes (LD) and thermo-optic photonics switches for scalable emitter addressing and on-chip sensors for monitoring optical power and temperature during photostimulation. Devices were fabricated at a commercial silicon photonics foundry on 200-mm diameter silicon-on-insulator (SOI) wafers in an active visible-light platform and were controlled using a custom-developed electronic circuit board. Each device features a hybrid-integrated InGaN LD which couples 450-nm light into a reconfigurable photonic switching tree for delivering spatially resolved photostimulation through 16 emitters along a 3-mm implantable shank. Using the on-chip photodetectors, we demonstrate how devices can enable switching tree calibration as well as output power monitoring during photostimulation. Furthermore, using the on-chip temperature sensors, we show how device temperature perturbations resulting from LD and thermo-optic switch activation can be directly monitored during photostimulation to ensure temperature fluctuations remain below 1 °C. We validate our design by delivering high spatiotemporal photostimulation during an in vivo optogenetic experiment with simultaneous Neuropixels recording to monitor evoked responses. Overall, these scalable integrated devices offer a pathway for neuroscientists to conduct fiberless optogenetic experiments with greater control and precision.

Deep brain stimulation (DBS) is an established therapy for advanced medication-resistant Parkinson's disease (PD), yet its ability to alter the course of the disease remains uncertain. Although preclinical research using toxin-induced PD models demonstrate neuroprotective effects, clinical studies in PD patients undergoing DBS have not substantiated these findings. This disconnect may be attributed to factors such as the initiation of DBS late in disease, stimulation protocols targeting symptoms rather than pathology, and the limited translational relevance of animal models lacking hallmark alpha-synuclein (α-Syn) aggregation. Incorporating α-Syn-based models may bridge this gap by facilitating the discovery of early electrophysiological biomarkers of pathological progression, refining stimulation parameters to enhance α-Syn clearance, and assessing if early DBS intervention can mitigate neurodegeneration. Yet, only a limited number of DBS studies have employed α-Syn models to date. This review examines the translational gap between preclinical neuroprotection claims and clinical outcomes, focusing on how α-Syn-based models could resolve current limitations in DBS research. Prioritizing these models could clarify whether DBS has the potential to extend beyond symptomatic relief and directly engage PD's underlying neurodegenerative mechanisms. Achieving this goal requires systematic investigation of DBS influences on α-Syn accumulation and its electrophysiological correlates in disease-relevant models.

Why do some experiences endure in memory better than others? Here we explore the possibility that learning fluctuates rhythmically several times per second, with fortuitously timed experiences being more memorable. Although such fleeting opportunities for encoding would evade our awareness, they are predicted by a prominent model describing how theta rhythms in the brain coordinate memory—the Separate Phases for Encoding and Retrieval (SPEAR) model. In a preregistered study, we adapted a dense sampling approach to reconstruct the millisecond time course of memory encoding in n = 125 participants. We found that memory encoding fluctuated at a theta rhythm (3–10 Hz), that these rhythms were not a by-product of rhythmic attention and that—like theta rhythms in the brain—memory rhythms were modulated by putative markers of acetylcholine. Our findings provide behavioural evidence consistent with the SPEAR model of episodic memory.

Mesial ( a.k.a., medial) temporal lobe epilepsy (MTLE) is the most common focal epilepsy ^1,2 and, in drug-resistant cases, is treated by surgical removal of the anterior temporal lobe, which often shows neuronal loss and gliosis consistent with hippocampal sclerosis (HS) ² . MTLE with HS has minimal contribution from germline genetic variation ³ , and is associated with prior precipitating insults such as prolonged childhood seizures and head trauma ^4–6 . Somatic variants in Ras-MAPK pathway genes were recently reported in a few MTLE surgical specimens ^7,8 , but their prevalence, clinical relevance, and underlying biological mechanisms remain unknown. Targeted duplex sequencing of hippocampal DNA from 462 surgical resections revealed significant enrichment of deleterious somatic variants in MTLE versus controls, with >40% of MTLE specimens harboring activating Ras-MAPK variants in PTPN11 , NF1 , BRAF , KRAS , and twelve genes not previously associated with focal epilepsy. Eight Ras-MAPK genes showed positive clonal selection in MTLE. Increased somatic variant burden predicted worse surgical outcome. Somatic Ras-MAPK variants at ultra-low (9,10 ; in contrast, Alzheimer disease hippocampi exhibited microglial enrichment of Ras-MAPK variants, consistent with prior reports ¹¹ . Single-nucleus RNA sequencing showed increased expression of Ras-MAPK genes in neurons and upregulation of pathways mediating neurogenesis and neural development in MTLE. Functional validation of novel, recurrent PTPN11 variants confirmed gain-of-function, while cellular modeling in induced pluripotent stem cells demonstrated proliferative/survival advantages for mutant cells in mosaic culture. Overall, our data suggest that somatic Ras-MAPK variants and acquired risk factors may converge on clonal competition in the hippocampus to modulate epilepsy risk.

Why do some experiences endure in memory better than others? Here, we explore the possibility that learning fluctuates rhythmically several times per second, with fortuitously timed experiences being more memorable. Although such fleeting opportunities for encoding would evade our awareness, they are predicted by a prominent model describing how theta rhythms in the brain coordinate memory – the Separate Phases for Encoding and Retrieval (SPEAR) model. Here, in a pre-registered study, we adapted a dense sampling approach to reconstruct the millisecond time-course of memory encoding in n=125 participants. We found that memory encoding fluctuated at a theta rhythm (3-10 Hz), that these rhythms were not a byproduct of rhythmic attention, and that—like theta rhythms in the brain—memory rhythms were modulated by putative markers of acetylcholine. Our findings provide behavioral evidence consistent with the SPEAR model of episodic memory.

Working memory (WM), the active retention of information over short periods, is a fundamental cognitive function, yet its underlying neural mechanisms remain unclear. In rodents, cue-selective “time cells” fire at specific timepoints after a WM memory cue, collectively forming sequences that encode cue-memory and elapsed time, providing a temporal code for maintaining information across time intervals. Whether similar dynamics support WM in the human brain is unknown. Here, we analyzed intracranial single-neuron recordings from medial frontal and medial temporal regions of patients performing a WM task. We found time cells with temporally tuned activation during WM maintenance, collectively forming robust sequences that tiled a delay period. Time-cell coordination in the hippocampus predicted successful WM maintenance, whereas in the pre-supplementary motor area it reflected memory load. Furthermore, we identified distributed cue-selective time cells that encoded both the identity of a memorandum and elapsed maintenance time, providing a temporally structured mnemonic code that complements persistent firing of concept cells. Together, these findings establish time-cell sequences as a conserved neural mechanism supporting human working memory.