It has become increasingly popular for people to take some neurostimulant drugs as nootropics1 that are expected to enhance cognition and learning2 beyond the initially approved therapeutic purposes such as curing attention deficits3. Likewise, brain stimulation techniques, including, e.g., transcranial magnetic stimulation4 and transcranial direct-current stimulation5, have also been extensively practiced in healthy subjects with similar expectations. However, while those drugs or stimulation techniques could help alleviating the relevant deficits, their resulting effects in the diseased conditions do not necessarily imply that healthy subjects could benefit from them in boosting the normal learning capabilities.
Learning is an intricate process that involves highly specific patterns of neuronal activation6–8, and the neocortex is known to be functionally relevant for the associative learning process9–11. In the past decade, specific manipulations of neuronal activities have been achieved by using the optogenetic technique12, advancing the understanding of learning and memory formation13. Although having become highly popular for animal experiments, the optogenetics technique shows little potential for applications in healthy humans due to the requirement of introducing exogenous genes in the brain. Here, we present a fundamentally different energy stimulation technique, mid-infrared modulation (MIM), which delivers mid-infrared light (MIR) energy through opened skull or even non-invasively through thinned intact skull to the brain and can significantly elevate neuronal firing rates in the targeted brain region. Notably, MIM induces neuronal firing in complete absence of any exogeneous gene. As a striking example, we demonstrate that MIM application in the auditory cortex of healthy adult mice during a sound-licking associative learning task boosts learning speed by ~ 50%.
We used a pulsed quantum cascade laser as the MIR light source for MIM in this study (see Supplemental Methods for details). A MIR fiber (core diameter 100 µm, numerical aperture 0.27) delivered the MIR to the target region of mouse brain (Fig. 1a), in a manner superficially similar to fiber-based optogenetics technique14. However, two major features fundamentally distinguish MIM from optogenetics. First, no exogeneous gene was introduced into the brain. Second, the wavelength of stimulation light was 5.6 µm, which is in the mid-infrared spectrum (3–50 µm, as defined by the ISO 20473 standard) and far beyond the visible (VIS, 0.38–0.78 µm) to near-infrared (NIR, 0.78–3 µm) spectrum used in optogenetics14.
We delivered the MIR with the following parameters: average irradiation power 9 ± 0.5 mW (5 measurements at the fiber tip), pulse width 300 ns, repetition rate 100 kHz, and irradiation duration 20 s (for detailed protocol, see Supplemental Methods). In order to yield repeatable and comparable results, we first placed the fiber tip closely above the cortical surface following craniotomy. After MIR delivery we performed immunohistochemistry for c-Fos (see Supplemental Methods), a widely-used molecular marker of neuronal activation15. C-Fos positive cells were found in a bullethead-shaped volume that was mostly within layer 2/3 of the cortex (Fig. 1b), where the axial and lateral extent of the distribution was both ~ 400 µm (Fig. 1c). Within this zone (e.g., outlined in Fig. 1b), the proportion of c-Fos positive cells among all cells was 27.8 / 24.0–32.1% (n = 21 slices from 7 animals, median / 1st – 3rd quartile, same notation for all subsequent data if not stated otherwise).
We next performed a dose-dependent test and found that the c-Fos cell count positively correlated with the irradiation time in our tested range between 5 s and 60 s (Fig. 1d, left 4 columns, c-Fos positive cell count per slice image, ‘5 s’: 16 / 11.75–23, n = 11; ’10 s’: 38 / 19.75–45, n = 9; ’20 s’: 46.5 / 39.5–63, n = 12; ’60 s’: 165 / 130–188, n = 14). Having shown that MIM application through opened skull could reliably induce neuronal activation, we applied MIM to the mice with thinned intact skull (thickness remaining: 44 / 37–48 µm, n = 12 measurements). MIM through thinned intact skull (irradiation time of 20 s) also induced neuronal activation (Fig. 1d, ‘Thinned 20 s’, c-Fos cell count: 24.5 / 19–42, n = 8). Comparing to the dose-dependent test, we estimate that the neuronal activation efficiency through thinned intact skull is about half (52%) of that through opened skull.
We conducted a new set of experiments wherein MIR and VIS light were delivered to each side of the cortex of individual mice respectively (both through opened skull, Fig. 1e), each via a fiber with the same geometric parameters and with the same irradiation power (9 mW) and time (20 s). Copious c-Fos positive cells were present in the MIR- but not the VIS-targeted regions (Fig. 1f). A few c-Fos positive cells were found at superficial locations within 70 µm from the cortical surface (i.e., layer 1); cells here are generally not considered to be pyramidal neurons16. Thus, we re-calculated the c-Fos positive cell count for this experiment by excluding these labeled superficial cells, and found that the count in the VIS region was near zero (Fig. 1g, MIR: 45.5 / 42–68, n = 8 slices from 4 animals; VIS: 1.5 / 1–3, n = 13 slices from 4 animals, p = 1e-4, rank sum test).
How does MIR light activate neurons without opsins? This activation was unlikely due to general heating, because previous studies have shown that heating (up to 2℃) by VIS light at a comparable power level (e.g., 15 mW in their studies versus the 9 mW of MIR in our study) either does not change or even suppresses neuronal activities through activation of an inwardly rectifying potassium conductance17,18. Furthermore, other studies have reported that pulsed NIR light induces neuronal activation not by general heating, but rather by a transient spatiotemporal gradient of light energy absorption in the tissue19,20. The higher water absorption coefficient for MIR relative to VIS or NIR light21 could be advantageous for efficiently establishing such gradients in a confined volume of brain tissue (see Fig. 1b, c for the spatial confinement). Such transient gradients may account for the efficiency of MIR energy to activate neurons without opsins. Moreover, the MIR light could induce conformational changes in particular proteins via nonlinear molecular resonance absorption22, further supporting the potential utility of MIR energies for modulating neuronal excitability and inducing firing activities.
To demonstrate that MIM could indeed induce neuronal firing, we performed in vivo single-cell loose-patch recording in cortical neurons throughout a course of MIM application (Fig. 1h). Figure 1i shows an example neuron that exhibited a greatly elevated firing rate in the MIM time window (across three consecutive trials). A summary of 5 recorded neurons (out of 9 loose-patched neurons) shows a significant elevation in the firing rate in the MIM window as compared to that in the pre- or post- MIM windows (Fig. 1j). These experiments above together establish a basic proof-of-principle for MIM in vivo, demonstrating that MIR energy delivery indeed induces neuronal firing in absence of any exogeneous gene.
We next employed two-photon Ca2+ imaging23–25 to visualize live neuronal population activities during MIM application (Fig. 2a, see also Supplemental Methods). Auditory cortical neurons were labeled with a Ca2+-sensitive fluorescent dye, Cal-520 (e.g., Fig. 2b). We performed real-time (40 Hz) two-photon imaging over the entire course of MIM application (40 s before, 20 s MIR irradiation, 60 s after). An example neuron (Fig. 2c) showed that MIM application reliably induced Ca2+ transients during the application time window over repeated trials. For each animal, we sequentially performed real-time imaging recordings (as the example above) at multiple cortical depths, which enabled mapping of the MIM-activated neurons (e.g., Fig. 2d). Note that a neuron was defined as MIM-activated if its trial-averaged Ca2+ activity level during the MIM window exceeded the activity level detected for the baseline window (before the irradiation window) by more than 2-fold (see Supplemental Methods).
Out of a total of 1532 neurons from 6 animals, we found 167 MIM-activated neurons (Fig. 2e, for the complete dataset see extended data Fig. 1). The activity level for the MIM-activated neurons recovered after the MIM window, returning to the same level as before irradiation (see also the average Ca2+ signal trace in Fig. 2e). Note that the proportion of MIM-activated neurons detected in these two-photon Ca2+ imaging experiments was 10.2 / 4.0–17.5% (n = 26 imaging focal planes from 6 animals), which was lower than that in the c-Fos imaging results (27.8 / 24.1–32.1%, n = 21 slices from 7 animals). We speculate that this difference could result from methodological differences, i.e., c-Fos imaging captures the cumulative neuronal activities from a prolonged time (the time required for c-Fos expression15 is much longer than the neuronal activation time), whereas two-photon Ca2+ imaging resolves specific neuronal activities in real time. Nevertheless, our results from c-Fos imaging, two-photon Ca2+ imaging, and single-cell loose-patch recording are highly consistent in demonstrating that MIM activates a substantial number of neurons in the targeted cortical region.
Previous studies have reported that the activities of auditory cortical neurons are involved in auditory associative learning for mice9,10,26. We therefore tested whether and how MIM application in the auditory cortex might affect the learning course for a sound-licking associative learning task24 (Fig. 3a). We recruited three cohort groups of naïve, healthy mice (8–9 weeks old at the beginning of training): the ‘control’ group that did not receive any MIM application, the ‘MIM-opened’ group that received MIM application through opened skull (a small craniotomy of ~ 200 µm diameter over the auditory cortex) and the ‘MIM-thinned’ group that received MIM application through thinned intact skull. All groups went through the same training program (consisting of 6 sessions). From session #2 on, animals in the two MIM groups received MIM application during the task engagement time windows (Fig. 3b, 25–35 s per engagement window with a 60 s pause between each). A typical example of a behavioral recording over consecutive trials during the learning process is shown in Fig. 3c, in which there were three successful response events (licking initiated within 0.5 s after sound stimulus onset) and one missed event (either no licking or licking initiated later than 0.5 s after sound).
The result showed that the licking action latency of successful response event became slightly shorter with increasing training sessions for all groups; however, no significant difference in the response latency among any pair of groups was found in any session (Fig. 3d). On the other hand, starting at the same level of response success rate in session #1 (Fig. 3e, success rate in session #1, ‘control’ group: 28.5 / 19.7–33.4%, n = 13; comparing to: ‘MIM-opened’ group: 25.9 / 17.9–38.2%, n = 14; p = 0.96; ‘MIM-thinned’ group: 18.5 / 8.2–35.4%, n = 8; p = 0.32; rank sum test), both the ‘MIM-opened’ and the ‘MIM-thinned’ groups achieved a significantly more increment in the response success rate over the first half of the learning course (Fig. 3e, increment of success rate from session #1 to session #3, ‘control’ group: 26.8 / 22.6–33.1%, n = 13; comparing to: ‘MIM-opened’ group: 40.5 / 30.4–57.8%, n = 14; p = 0.027; ‘MIM-thinned’ group: 50 / 37.8–57.1%, n = 8; p = 0.031; rank sum test). These results together suggest that MIM does not affect the behavioral action latency, but rather boosts the sensory-behavior associative learning speed.
We fitted a learning curve for each mouse (Fig. 3f shows the group average) using an exponential function27, and then defined the learning speed by the fitted exponent factor. Comparing to the ‘control’ group, MIM boosted learning speed by ~ 50% for either group with thinned intact skull or opened skull (Fig. 3g, fitted exponent factor, ‘control’ group: 0.29 / 0.25–0.33, n = 13; ‘MIM-opened’ group: 0.44 / 0.38–0.51, n = 14; p = 0.004; ‘MIM-thinned’ group: 0.45 / 0.32–0.54, n = 8; p = 0.03; rank sum test). Note that in this set of associative learning experiments we applied MIM repeatedly over multiple trials (Fig. 3b). Thus, the reduced neuronal activation efficiency through thinned intact skull obtained by the c-Fos imaging experiments with a single-shot MIM (Fig. 1d) is not incompatible with the result here that, MIM through thinned intact skull achieved nearly the same degree of learning acceleration as that of MIM through opened skull.
Associative learning is generally assumed to require synaptic plasticity, which in mammals is known to depend on the temporal coincidence of synaptic and somatic activities28–30. Thus, we speculate that the observed acceleration of sensory-behavior associative learning likely results from extra MIM-induced somatic activities (Fig. 2) that coincide temporally with the task-relevant synaptic input activities occurring during the engagement window. In this scenario, the increase in coincident synaptic and somatic activities would result in an overall increase in the amount of task-relevant synaptic plasticity during a given task engagement duration, thus accelerating the learning. From another viewpoint31, the MIM-induced activation of a subset of cortical neurons (Fig. 1, 2) during the learning task could promote the recruitment of these neurons into the engram cell population32, thus accelerating the learning.
In this study, we demonstrate MIM, a non-invasive and opsin-free neuronal stimulation technique that is fundamentally different from the well-known and widely-deployed optogenetics technique12,14. MIM reliably induces neuronal activations in a precisely targeted and spatially confined volume of brain tissue in vivo, in a manner similar to that of optogenetics but in complete absence of introducing exogeneous gene. Like that optogenetics application is shown to artificially alter memory, sensory perception or behavior control33,34, we show here that MIM application in the auditory cortex profoundly accelerates the sound-licking associative learning (Fig. 3e, g). Together, our results illustrate the utility of this promising technique for enhancing brain learning functions with mid-infrared light energy.