Visual perceptual learning in mice
A group of mice (nā=ā14) was subjected to a vPL protocol in which they were asked to discriminate two vertical gratings with equal contrast, but different spatial frequency (SF), made progressively more similar by making their SF closer to each other. Initially, the test grating had a SF of 0.522 c/deg, while the reference grating had a SF of 0.116 c/deg. All mice easily learned this discrimination task, with the percentage of correct choices increasing over the course of the training sessions (One-way RM ANOVA, Holm-Sidak method, Fā=ā24.276, DFā=ā1, pā<ā0.001) and finally achieving a performance level of at least 80% of accuracy in at least three subsequent sessions (Fig.Ā 1a). At this point, mice were divided in two groups: vPL mice (nā=ā7) underwent a perceptual learning task, practicing with a progressively more difficult discrimination as the SF of the test grating was made increasingly more similar to that of the reference grating; the other group, First-Step (FS) mice, (nā=ā7) continued to practice with the test grating maintained at the starting value of 0.522 c/deg. Notably, no differences were found in visual discrimination abilities between prospective FS and vPL mice during the initial phase (Two-way RM ANOVA, HolmāSidak method, Fā=ā1.334, DFā=ā1, pā=ā0.346, Fig.Ā 1b).
In vPL mice, we observed a robust improvement in visual discrimination abilities with practice, as shown by the progressive reduction in the minimum spatial frequency discrimination threshold (MDT) across sessions. In the first session, mean MDT was 0.282āĀ±ā0.016 c/deg, while this value reached 0.043āĀ±ā0.004 c/deg at the end of the vPL training procedure (One-way RM ANOVA on ranks, pā<ā0.001, Fig.Ā 1c). A clear vPL was also revealed by the increase in the percentage of correct choices for a given SF of the test grating (for example: reference grating of 0.116 c/deg vs test grating of 0.160 c/deg, nā=ā7, One-way RM ANOVA, HolmāSidak method, Fā=ā14.691, DFā=ā2, pā<ā0.001). In FS mice, the performance remained stable (Fig.Ā 1d).
A separate group of animals was used to probe the difference in the vPL task between male and female mice (nā=ā16, 8 males and 8 females). No difference could be found either in the achieved MDT (males: 0.044āĀ±ā0.003 c/deg, females: 0.039āĀ±ā0.004; t test, tā=ā0.794, DFā=ā14, pā=ā0.441, Supplementary Fig.Ā 1a) and the acquisition rate of the vPL (Two-way RM ANOVA, HolmāSidak method, Fā=ā0.521, DFā=ā1, pā=ā0.428, Supplementary Fig.Ā 1b)
To test the specificity of vPL for the stimulus orientation, we performed, in the same vPL mice previously trained with vertical stimuli (nā=ā7), an experiment of orientation shift in which the two vertical gratings were rotated by 90Ā°; afterwards, new trials were applied to assess the new MDT. We found that the visual grating discrimination abilities achieved by vPL mice were highly selective for the orientation of the stimulus; indeed, we found a marked impairment in their discrimination abilities occurring immediately after the rotation of the stimuli (nā=ā7, MDT before shift: 0.040āĀ±ā0.004 c/deg, DT after shift: 0.327āĀ±ā0.021 c/deg; paired t test, tā=āā13.084, DFā=ā6, pā<ā0.001, Fig.Ā 2a, b). The animals appeared unable to discriminate the newly oriented stimuli when the test grating was maintained at the same SF reached before the orientation shift, and the percentage of correct choices fell below the 70% criterion for two consecutive vPL sessions (gray bars, Fig.Ā 2a), until a higher SF difference between the two stimuli was achieved. With additional training, the animal performance reached a new perceptual plateau, and the MDT for horizontal grating was not significantly different from that achieved before the stimulus orientation change (nā=ā7, MDT before shift: 0.040āĀ±ā0.004 c/deg, MDT after shift and vPL training: 0.040āĀ±ā0.004 c/deg; paired t test on ranks, pā=ā1.000; Fig.Ā 2a, c). In contrast, FS mice were still able to perform the task when the two vertical gratings were rotated by 90Ā° (MDT before shift: 100% Ā± 0.0%; MDT after shift: 91% Ā± 0.035%; paired t test on ranks, pā=ā0.125; see Supplementary Fig.Ā 2).
Electrophysiological characterization of LM activity
To provide a general characterization of basic LM neuronal activity features, visual evoked potentials (VEPs) and single-units were recorded in a group of anesthetized naĆÆve mice (nā=ā7), by means of multichannel electrophysiological recordings. VEP recordings represent the elective method to study the function of visual areas and they have been extensively employed to assess visual capabilities, local cortical processing and the state of maturation of visual pathways56,57,58,59. Single-unit recordings were instead employed to locally characterize electrical responses of LM neurons.
To assess visual acuity, VEP responses were recorded from a silicon electrode inserted 3.6āmm lateral to lambda and advanced 200 Ī¼m within the cortex, in response to horizontal gratings of different SFs and maintained at 90% contrast. VEP acuity was obtained extrapolating VEP amplitude to 0āV. The average acuity was 0.408āĀ±ā0.039 c/deg (Supplementary Fig.Ā 3a). LM VEP acuity is just a measure of spatial resolution in LM visual neurons, but it does not necessarily reflect visual acuity in mice, as this requires recordings to be performed from V1. To measure contrast sensitivity, VEPs were recorded in response to gratings of a spatial frequency of 0.06 c/deg, at different contrasts. Contrast sensitivity was obtained extrapolating to 0āV the recorded VEP amplitudes. The average contrast threshold was 9.46% Ā± 2.051% (Supplementary Fig.Ā 3b).
To further characterize the electrical activity of LM neurons, single units in response to drifting sinusoidal gratings were recorded at multiple depths, spanning all cortical layers. Single units were clustered in seven ocular dominance (OD) classes on the basis of the ratio of contralateral to ipsilateral peak responses, according to the Hubel & Wieselās classification60. The great majority of all recorded cells (nā=ā145) fell into the intermediate OD classes, without a prevalence of either the contralateral (class 1) or the ipsilateral (class 7) eye (Fig.Ā 3a). LM electrical activity was evaluated measuring spontaneous (contra = 0.220āHz, ipsi = 0.274āHz, Fig.Ā 3b) and evoked discharge (contra = 3.008āHz, ipsi = 2.502āHz, Fig.Ā 3c), while orientation and direction selectivity were assessed calculating the orientation selectivity index (OSI, contra = 0.465, ipsi = 0.565, Fig.Ā 3d) and direction selectivity index (DSI, contra = 0.225, ipsi = 0.210, Fig.Ā 3e). No difference was found in the electrical activity recorded from the contralateral and ipsilateral eye in the spontaneous discharge or DSI (nā=ā6, paired t test, DFā=ā5, tā=āā1. 242 pā=ā0.269; tā=āā0.0298 pā=ā0.977, respectively) nor in the evoked discharge and OSI (nā=ā7, paired t test, DFā=ā6, tā=ā2.212 pā=ā0.069; tā=āā1.800 pā=ā0.122, respectively).
Chemogenetic inhibition of LM activity
In order to silence global LM neuronal activity, a constitutive viral vector was injected into the LM of naĆÆve mice, to induce the expression of hM4D(Gi)-mCherry, an inhibitory DREADD (designer receptors activated exclusively by designer drugs). Histological evaluations showed that the expression of hM4D(Gi)-mCherry was confined into LM, without entering the dorsal visual stream (Fig.Ā 4).
Two weeks after the injection of the viral vector, single units were recorded from LM in a group of mice (nā=ā9), to probe the effective chemogenetic inhibition of LM activity. In a group of mice (nā=ā5), electrical signals in response to alternate monocular stimulation of the two eyes were recorded before and 15āmin after an intraperitoneal (i.p.) administration of Clozapine N-oxide (CNO), a DREADD ligand. We found that CNO administration resulted in a marked suppression of the evoked activity in response to both the contralateral (pre-CNOā=ā2.473āĀ±ā0.690āHz, post-CNOā=ā0.789āĀ±ā0.112āHz; paired t test, tā=ā3.039, DFā=ā4, pā=ā0.038, Fig.Ā 5a) and ipsilateral (pre-CNOā=ā2.559āĀ±ā0.597āHz, post-CNOā=ā0.415āĀ±ā0.188āHz; paired t test, tā=ā4.180, DFā=ā4, pā=ā0.014, Fig.Ā 5a) eye stimulation. In contrast, no reduction in LM evoked activity was found in control mice (nā=ā4) recorded before (pre-SAL; contra = 1.708āĀ±ā0.325āHz, ipsi =ā1.681āĀ±ā0.294āHz) and 15āmin after an i.p. administration of saline (post-SAL; contra = 1.87āĀ±ā0.358āHz, ipsi = 1.274āĀ±ā0.288āHz) (paired t test, contra: on ranks, pā=ā0.375 and ipsi: tā=ā0.734, pā=ā0.516, Fig.Ā 5b). Notably, LM activity recorded from CNO and SAL mice was statistically different after the i.p. administration (post CNO vs post-SAL, t test, DFā=ā7, contra: tā=āā3.242, pā=ā0,014 and ipsi: tā=āā2.733, pā=ā0.029). In contrast, no difference could be found between the two groups of mice before the i.p. administration of CNO or saline, respectively (pre CNO vs pre SAL, t test, DFā=ā7, contra: tā=ā1.020, pā=ā0.342 and ipsi: tā=ā1.342, pā=ā0.221).
To evaluate the global impact of the hM4D injection, single-unit activity recorded from CNO and SAL mice was compared to that sampled from naĆÆve animals (nā=ā6). Only the evoked responses recorded from the contralateral and ipsilateral eyes of the animals treated with CNO (post-CNO) were significantly different compared to the responses recorded from naĆÆve mice (NaĆÆve, contra = 2.659āĀ±ā0.354āHz ipsi = 2.421āĀ±ā0.603āHz) (One-way ANOVA vs control, HolmāSidak method, DFā=ā4, contra: Fā=ā3.791, pā=ā0.02 and ipsi: Fā=ā3.919, pā=ā0.031, Fig.Ā 5c).
LM activity is required for vPL acquisition
Then, we used the same chemogenetic approach to test whether suppression of LM activity was able to significantly affect vPL. A group of mice (nā=ā16) was required to learn the vPL task two weeks after the hM4D injection. Once the 80% criterion was achieved, a subgroup of animals was subjected to the incremental phase of vPL with administration of CNO (nā=ā8, CNO mice), while a second subgroup of animals were tested on the same task with administration of saline (nā=ā8, SAL mice). Injections of either CNO or saline were performed 30āmin before each vPL session. The performance plateau reached by CNO mice (0.180āĀ±ā0.029 c/deg) was significantly different from that achieved by SAL mice (0.042āĀ±ā0.004 c/deg), with the former group of animals displaying a robust learning impairment both in the slope of the vPL curve and in the achieved perceptual plateau (Two-way RM ANOVA on ranks, HolmāSidak method, pā<ā0.001, Fig.Ā 6a).
The perceptual learning performance of CNO mice, but not that of SAL mice, was consistently different from the performance displayed by naĆÆve animals subjected to the same vPL task (Two-way RM ANOVA on ranks, Holm-Sidak method, vPL mice vs CNO mice: pā<ā0.001; vPL mice vs SAL mice pā=ā0.513; SAL mice vs CNO mice: pā<ā0.001; Fig.Ā 6a), with CNO mice reaching a significantly higher MDT compared to both SAL and vPL mice (One way ANOVA on ranks, Dunnās method, pā<ā0.001). Notably, no differences could be found between mice that were administered with CNO (prospective CNO) or with saline (prospective SAL) during the learning phase of the discrimination task (Two-way RM ANOVA, HolmāSidak method, Fā=ā0.0980, DFā=ā1, pā=ā0.757, Fig.Ā 6b). Moreover, the learning capabilities of the entire group of injected mice (nā=ā16) were not different from those of naĆÆve mice (nā=ā14, vPL and FS mice pooled together), ruling out the possibility that the observed impairment in vPL animals could be dependent to learning deficits due to the surgical procedures per se (Two-way RM ANOVA, HolmāSidak method, Fā=ā1.694, DFā=ā1, pā=ā0.200, Fig.Ā 6c).
To rule out the possibility that the deficits observed in CNO mice might be due to a visual impairment caused by the manipulation of LM activation, we also measured behavioral visual acuity (VA) after LM inactivation. VA was assessed through the Prusky water maze task61, testing the ability of injected mice to distinguish a visual grating from an homogeneous gray stimulus. Two weeks after a bilateral injection of hM4D, a separate group of mice (nā=ā8) was first trained with a low SF and then tested for the capability to distinguish higher SFs 30āmin after i.p. administration of saline. At the end of this first part of the procedure, the averaged VA was of 0.514āĀ±ā0.005 c/deg. Then, mice were retested in the same task 30āmin after i.p. administration of CNO. The VA measured in the same group of mice after CNO administration was 0.516āĀ±ā0.005 c/deg. Thus, no VA impairment could be found when LM activity was suppressed (paired t test, DFā=ā7, tā=āā0.678, pā=ā0.519, Fig.Ā 7).
LM activity is required for vPL retention
We then asked whether the vPL impairment caused by LM inactivation could be reversed by a treatment shift in which CNO was replaced by saline administration, and vice versa, whether the intact vPL abilities displayed by mice originally treated with saline might be impaired by a treatment shift in which saline was replaced by CNO administration. To this purpose, CNO and SAL mice were subjected to an experiment of administration shift (AS). When the animals reached their perceptual plateau, CNO mice were subjected to i.p. administration of saline, and then to additional vPL practice (CNO_SAL mice); vice versa, after reaching their perceptual plateau, SAL mice were subjected to administration of CNO, and then to additional vPL practice (SAL_CNO mice) (Fig.Ā 8). Both groups of animals were asked to perform additional vPL, starting 30āmin after the first administration of the new treatment. Injections were repeated 30āmin before each vPL session, as previously described for CNO and SAL mice.
CNO_SAL mice displayed a marked improvement in their vPL performance. While, immediately after AS, the MDT of CNO_SAL mice was not significantly different from that achieved before AS (paired t test, DFā=ā7, tā=ā2.004, pā=ā0.085, Fig.Ā 8a), they eventually reached a new MDT (0.048āĀ±ā0.003 c/deg) that was significantly lower than that reached before AS (paired t test, DFā=ā7, tā=ā4.379, pā=ā0.003, Fig.Ā 8a), and no longer different with respect to that achieved by both vPL and SAL mice (One-way ANOVA on ranks, DFā=ā2, pā=ā0.587). On the other hand, SAL_CNO mice were only able to discriminate the easiest set of stimuli of our vPL task immediately after AS (paired t test, test on ranks, pā=ā0.008, Fig.Ā 8b). When additional vPL was applied, SAL_CNO mice reached a new MDT (0.319āĀ±ā0.026 c/deg) that was significantly higher than that achieved before AS (paired t test on ranks, pā=ā0.008, Fig.Ā 8b) and then that achieved by vPL mice (t test, DFā=ā13, tā=āā8.702, pā<ā0.001).
When LM activity was suppressed, the vPL performance of CNO, but not that of SAL mice, was significantly different compared to the performance of vPL mice- naĆÆve animals subjected to the same vPL task (One-way ANOVA on ranks, Dunnās Method, pā<ā0.001, Supplementary Fig.Ā 4a). Instead, the MDT of SAL mice treated with CNO (SAL_CNO) but not that of CNO mice treated with SAL (CNO_SAL) was significantly different compared to the performance of vPL mice (One-way ANOVA on ranks, Dunnās Method, pā<ā0.001, Supplementary Fig.Ā 4b). Notably, we found that major perceptual impairments can be induced suppressing LM activity after vPL acquisition (SAL_CNO mice). The MDT achieved by SAL_CNO mice (MDTā=ā0.319āĀ±ā0.026 c/deg) was indeed significantly higher than the MDT achieved by CNO mice, i.e. those animals subjected to LM suppression from the very first vPL session (MDTā=ā0.180āĀ±ā0.029 c/deg) (t test on ranks, pā=ā0.021, Supplementary Fig.Ā 4c).
Taken together, these results show that interfering with the neuronal activity of LM leads to a marked impairment not only of vPL acquisition but also of its retention (Supplementary Fig.Ā 5a).
The role of top-down projections in vPL
To gain conclusive insights into the role played by top-down LM to V1 projections in vPL, we spatially confined our chemogenetic suppression to those secondary visual neurons directly projecting into V1 (LMā>āV1 projections) in a separate group of animals. To suppress LMā>āV1 projections, we induced the expression of a Cre-dependent hM4D delivered into LM by injecting retro-Cre within V1 borders. Through this double-injection strategy, we were able to selectively label and suppress only those LM neurons sending top-down projections to V1. As previously described for the chemogenetic suppression of LM, mice were subjected to the vPL task 30āmin after an i.p. administration of CNO (PRJ CNO mice, nā=ā7). A significant deficit in vPL was found when LMā>āV1 projections where selectively suppressed, with PRJ CNO mice reaching a mean MDT of 0.235āĀ±ā0.031 c/deg (Fig.Ā 9a). The final perceptual plateau reached by PRJ CNO mice was not different from that previously reported for CNO mice (Two-way RM ANOVA on ranks, HolmāSidak method, overall treatment, CNO mice vs PRJ CNO mice pā=ā0.333; SAL mice vs PRJ CNO pā<ā0.001; CNO mice vs SAL mice pā<ā0.001, Fig.Ā 9a).
We then tested whether the vPL deficit exhibited by PRJ mice could be rescued byĀ a shift to saline administration, as previously observed in CNO mice. When PRJ CNO mice achieved their perceptual plateau, they were subjected to additional vPL practice 30āmin after a saline i.p. administration (PRJ CNO_SAL mice). When LMā>āV1 projections were released from the chemogenetic suppression, PRJ CNO_SAL mice showed an improvement in their performance (0.235āĀ±ā0.031 c/deg vs 0.188āĀ±ā0.039 c/deg; paired t test, tā=ā2.576, DFā=ā6, pā=ā0.042, Fig.Ā 9b), continuing to improve until they achieved a new MDT (0.043āĀ±ā0.006 c/deg, paired t test, tā=ā6.606, DFā=ā6, pā<ā0.001, Fig.Ā 9b).
Similarly to what was observed when the global LM activity was suppressed, we found that PRJ CNO mice achieved a vPL plateau statistically different from SAL mice, but not from CNO mice (One-way ANOVA on ranks, pā<ā0.001, Fig.Ā 9c). However, when the activity of LMā>āV1 projections was rescued, PRJ CNO_SAL mice achieved a new perceptual plateau statistically different from SAL_CNO mice, but not from CNO_SAL mice (One-way ANOVA on ranks, pā<ā0.001, Fig.Ā 9d).
Overall, these results strongly show a key role for top-down projections that re-enter V1 from LM in the acquisition of vPL (Supplementary Fig.Ā 5b).