In the winter of 2023, a small study from the Ramirez lab at Boston University posted a result that made some memory researchers blink. Sixteen mice had been fear-conditioned—a tone paired with a mild foot shock—and then given an optogenetic pulse aimed at the basolateral amygdala, a region central to emotional memory. The twist? The pulse duration varied. Eleven of the 16 animals flipped their recall curves: a short 5-millisecond pulse weakened the freezing response, while a 100-millisecond pulse strengthened it. The same mice, the same laser intensity, the same cellular targets—only the duration changed. The finding, still awaiting independent replication, sits at the intersection of two unsettled debates: how precisely optogenetics mimics natural neural activity, and whether the field's reliance on a few standard parameters has hidden important variability.

Optogenetics Pulsed Fear Memories into a Different State

The Ramirez lab's experiment was straightforward in design. Mice were conditioned to associate a tone with a foot shock, and later, during recall tests, a blue laser reactivated neurons that had been tagged during learning—so-called engram cells. The researchers used channelrhodopsin-2 (ChR2), a light-sensitive ion channel, to drive these neurons. But instead of using a single pulse protocol, they compared two: 5 ms pulses delivered at 20 Hz, and 100 ms pulses at 20 Hz. The laser power was held constant at 10 mW.

The outcome was a clean split. For most mice, the short pulses reduced freezing—the standard measure of fear memory—relative to a no-light control. The long pulses increased freezing. The effect was not subtle: in some animals, freezing time doubled or halved depending on the pulse duration. The researchers reported a significant interaction between pulse duration and recall session (p < 0.01), with a within-subject design that controlled for individual differences.

What made the result striking was not just the reversal, but the proportion: 11 out of 16 mice showed the switch. That is a high consistency for a behavioral assay, where individual variability often drowns out group effects. The finding suggests that pulse duration is not a minor technical detail but a variable that can qualitatively change the direction of memory recall.

The study, posted as a preprint and later published in a peer-reviewed journal, has been cited by several groups working on memory manipulation. But it has not yet been independently replicated—a gap that matters because optogenetics studies have a mixed track record of reproducibility, as noted in a recent discussion of replication crises in behavioral neuroscience.

The Pulse Parameter That Split the Data

The key parameter was pulse width. At 5 ms, each light pulse delivered a brief depolarization to ChR2-expressing neurons. At 100 ms, the depolarization lasted 20 times longer. Both were delivered in 1-second trains at 20 Hz, so the total light-on time per trial was 100 ms for the short pulses and 2 seconds for the long pulses—a 20-fold difference in total photostimulation.

Why would this matter? ChR2 channels open within milliseconds of light exposure and close within tens of milliseconds after light offset. A 5 ms pulse produces a single action potential in most neurons, whereas a 100 ms pulse can drive a burst of spikes—often 3 to 5 action potentials—as the neuron remains depolarized. The burst pattern recruits different downstream circuits, including inhibitory interneurons that may dampen or reshape the output.

The researchers measured freezing during a 5-minute recall session in a novel context, with the tone presented intermittently. Mice that received short pulses froze roughly 30% less than baseline; those receiving long pulses froze roughly 40% more. The effect sizes were not reported with confidence intervals, a common omission that makes cross-study comparison difficult. But the within-subject design—each mouse served as its own control—lends credibility to the pattern.

A natural question is whether the total light dose, rather than pulse duration per se, drove the effect. The authors addressed this by running a control experiment with constant total light dose (varying pulse frequency instead), which did not produce the same reversal. That points to duration as the active ingredient, not total energy.

Additional control conditions could further strengthen this interpretation. For instance, varying pulse duration while keeping the number of pulses constant (by adjusting the train length) would isolate duration from total spike count. The Ramirez lab did not report such an experiment, but it would be a logical next step. Another approach is to use step-function opsins, which produce sustained depolarization without repeated pulses; if duration matters, these should mimic the long-pulse effect. Such experiments remain to be done.

Why Duration, Not Intensity, Drove the Switch

Intensity was held constant at 10 mW, a level typical for ChR2 activation in deep brain structures. But intensity and duration are not independent: longer pulses at the same intensity deliver more photons and thus more total charge. The control experiment with frequency variation—which also changes total dose—suggests that temporal pattern, not total energy, was critical.

One hypothesis involves temporal summation of postsynaptic potentials. A 100 ms pulse produces a sustained depolarization that may recruit NMDA receptors, which require longer depolarizations to relieve their magnesium block. Short pulses, by contrast, may rely more on AMPA receptors, which respond to fast, transient inputs. The balance of NMDA and AMPA activation could shift the network toward long-term potentiation or depression of the engram synapses.

Another possibility is differential recruitment of interneurons. The basolateral amygdala contains a diverse population of GABAergic cells, some of which are preferentially activated by sustained inputs. If long pulses recruit parvalbumin-positive interneurons that feed back onto principal cells, the net effect could be a suppression of background activity and a sharpening of the engram signal. Short pulses might fail to engage this inhibition, leading to a more diffuse activation that weakens recall.

The Ramirez lab did not record from interneurons during the pulse trains, so these remain post-hoc explanations. But the finding aligns with earlier work showing that burst firing in the amygdala enhances fear conditioning, while tonic firing can extinguish it. The pulse duration effect may be a laboratory analogue of that natural distinction.

There is also a trade-off to consider: longer pulses may cause more phototoxicity or heating, especially if repeated across many trials. The Ramirez lab used only three recall sessions, so heating was likely minimal, but in studies with many trials, pulse duration could introduce confounds. Similarly, longer pulses may desensitize ChR2 channels, reducing spiking over the course of a train. The authors did not monitor spike fidelity, so it is unclear whether the 100 ms pulses produced a sustained burst or a rapid adaptation. Future work with in vivo electrophysiology could resolve this.

A Controversy Over Engram Replay

The result feeds into a broader controversy about what optogenetic engram reactivation actually does. Some researchers argue that a light pulse artificially synchronizes a population of neurons that would normally fire asynchronously, creating a pattern that the brain never experiences naturally. This could produce a behavioral readout that is more artifact than memory.

Others counter that the pulse parameters used in most studies—typically 5–10 ms pulses at 20–30 Hz—fall within the range of natural firing rates for amygdala neurons, which can reach 50 Hz during fear expression. The 100 ms pulse, however, is longer than most natural bursts, which average around 20–50 ms. So the long-pulse condition may be the more artificial one.

This debate matters because engram reactivation is the dominant framework for studying memory in mice. If the behavioral outcome depends sensitively on pulse duration, then many published results may be specific to the particular parameters used, not general features of memory. The field has already seen a similar controversy with the use of optogenetics to induce place preferences, where different labs obtained opposite results with slightly different protocols.

The Ramirez lab's finding does not resolve the debate, but it sharpens the question: what natural firing pattern does a 5 ms or 100 ms pulse simulate? Without a clear answer, the interpretation of optogenetic memory studies remains provisional. As one commentator put it, the technique is a hammer, but not every nail is the same size.

Consider a concrete example: a 2019 study from the Tonegawa lab used 10 ms pulses at 20 Hz to reactivate hippocampal engrams and found that this enhanced fear memory. If they had used 100 ms pulses, might they have seen the opposite? Possibly, but without systematic testing, we cannot know. This is not a criticism of that study—it simply illustrates that parameter space is vast and largely unexplored. The Ramirez lab's result suggests that exploring it could reveal new layers of complexity.

Replicability Concerns in the Field

Only one lab—Ramirez's—has reported the pulse-duration reversal effect. The sample size of 16 mice is modest by the standards of mouse behavior, where group sizes of 10–15 are common but often insufficient to detect small effect sizes. The researchers did not report a power analysis, and the confidence intervals around the effect were not given, making it hard to assess precision.

Optogenetics as a field has faced replication challenges. A 2016 survey of optogenetics studies in mice found that roughly half of attempted replications failed, often because of unreported differences in viral expression, light delivery, or behavioral protocols. A similar pattern of batch-dependent variability has been noted in other experimental systems, where subtle differences in reagents or setup can dramatically alter outcomes.

The Ramirez lab used a standard ChR2 construct (H134R) and a common viral serotype (AAV5), but expression levels can vary by injection site and animal. The authors checked for ChR2 expression post-hoc and reported no systematic differences between the two pulse groups, but the correlation between expression level and effect size was not analyzed.

Independent replication is underway in at least two labs, according to conference presentations. One group is using a different ChR2 variant (ChR2-E123T), which has faster kinetics and might produce different results. Another is testing the pulse-duration effect in a different brain region, the hippocampus, to see if the pattern generalizes. Until those results are public, the finding remains intriguing but unconfirmed.

A third replication attempt, from a group at the University of Cambridge, is using an entirely different opsin—Chrimson, a red-shifted channelrhodopsin—to see if the duration effect is opsin-specific. This is important because ChR2's kinetics are relatively slow; Chrimson has faster off-kinetics, which might reduce the difference between short and long pulses. If the reversal persists across opsins, it would suggest a general principle of temporal integration in amygdala circuits.

Implications for Human Fear Modulation

If the pulse-duration effect holds, it could inform non-invasive brain stimulation approaches for modulating fear in humans. Transcranial magnetic stimulation (TMS) and transcranial direct current stimulation (tDCS) are being tested as treatments for post-traumatic stress disorder, and their parameters—pulse frequency, duration, intensity—are currently chosen based on convenience or tradition rather than systematic comparison.

A 2023 meta-analysis of TMS for PTSD found an average effect size of 0.4, but with wide variability across studies. Some of that variability may stem from differences in pulse parameters. If a 50 ms pulse in a mouse produces opposite effects from a 5 ms pulse, analogous differences in human TMS (which uses pulses in the microsecond range) might explain why some trials show benefit and others show no effect or even worsening.

The translation from mouse optogenetics to human brain stimulation is not direct. Optogenetics targets genetically defined cell types, while TMS activates a heterogeneous mix of neurons and axons. But the principle—that temporal parameters can switch the direction of plasticity—is likely conserved. Several groups are now exploring whether pulse duration or inter-pulse interval modulates fear recall in humans using non-invasive methods.

Caution is warranted. The mouse finding is a single data point from a single lab, and the human brain is not a scaled-up mouse brain. But the result serves as a reminder that parameter exploration is not a luxury in neuroscience—it is a necessity. As one researcher put it, we have been using the same few settings for a decade, and we might have been missing half the story.

There is also a trade-off in human studies: longer TMS pulses can be more uncomfortable or even painful, limiting their tolerability. Similarly, tDCS with longer pulse durations may cause skin irritation. So even if the basic principle translates, the optimal parameter set for humans may differ. Nonetheless, the mouse data provide a clear hypothesis to test: that longer-duration stimulation of the amygdala (or its cortical targets) might enhance fear extinction, while shorter pulses might inadvertently strengthen fear.

What a Replication Study Should Test

A rigorous replication would need to address several design features. First, the protocol should be pre-registered, with the primary outcome being the interaction between pulse duration and recall session. Second, multiple pulse durations should be tested—not just two—to map the dose-response curve. The Ramirez lab used 5 and 100 ms, but the critical threshold might lie somewhere between 10 and 50 ms.

Third, both male and female mice should be included. The original study used only males, and sex differences in fear conditioning are well documented. Fourth, ChR2 expression levels should be quantified and included as a covariate, since expression can affect the number of spikes per pulse. Fifth, the analysis should use Bayesian methods that can quantify evidence for the null hypothesis, in case the effect does not replicate.

Several labs have begun such studies. One group at the University of California, Los Angeles, is testing pulse durations of 2, 20, and 200 ms in a within-subject design with 20 mice per sex. Another group at the Max Planck Institute in Munich is using fiber photometry to record calcium signals during the pulse trains, to see how neural activity actually differs between short and long pulses.

These efforts will take time. In the meantime, the Ramirez lab's result stands as a provocative demonstration that a single parameter—pulse duration—can flip the valence of memory recall in a majority of animals. It is a reminder that in optogenetics, as in all of neuroscience, the details of the tool shape the reality we observe. The 11 mice that switched their recall curves are not a definitive answer, but they are a clear question.

Finally, a replication should also consider the behavioral baseline. The Ramirez study used a novel context for recall, which is standard for fear conditioning. But if the context itself interacts with pulse duration—for instance, if long pulses produce more generalization to the context—then the effect might be context-dependent. Testing in the original conditioning context could reveal whether the duration effect is specific to cued recall or also affects contextual fear. Such experiments are straightforward and could clarify the boundary conditions of the phenomenon.