Area M1 neurons are tuned with respect to locomotion speed
Female heterozygous Mecp2+/- female mice in 129-background were crossed to wild-type (WT) Camk2-Cre homozygous male mice in B6/C57 background. At 3 weeks of age (Fig. 1A), male Mecp2-null mice and WT littermates (F1 with C57&129 mixed genetic background) were implanted with a 3 mm cranial window over the anterior motor cortex (coordinates 1.6 mm lateral, 0.3 mm anterior to bregma) after AAV-flex-GCaMP6m was injected. Three weeks following the surgery (6 weeks of age), animals were acclimated to a head-fixed motorized foam wheel that forced the mice to locomote at five different speeds (rest, 15, 30, 45, and 60 mm/sec) (Fig. 1A). The training paradigm started with the rest condition and then the wheel speed increased stepwise, then another rest condition, followed by a trial starting at the top speed and decreasing stepwise (sequence: 0, 15, 30, 45, 60, 0, 60, 45, 30, 15, 0 mm/sec), as previously described [14]. Each condition lasted for 120 s. Initial training was performed daily for two days, followed by weekly chronic imaging over the same area (Fig. 1A, left). The mouse was placed under the two-photon microscope on an identical motorized foam wheel (Fig. 1A, right). The rotating dowel rod task was performed once per week to track the locomotion ability of imaged animals with disease progression. Data was acquired from 8 Mecp2-null and 9 WT littermate mice. The paw movement is automatically tracked by an algorithm [27]. Both front paws showed similar movements (Fig. 1B, C). Both WT and Mecp2-null mice performed successful locomotor paw movements on the motorized wheel. As observed by weekly testing on the wheel treadmill, Mecp2-null mice showed apparent motor deficit, reflected by shorter stride length and distance, as well as much lower stride numbers at each speed (Fig. 1D). Such motor deficit progressively worsened when the mice grow older (Fig. 1E). Data shown were analyzed from 6 weeks until the age of 10 weeks, as after that, half of the Mecp2-null mice were dead (and all were dead by 15 weeks of age).
Motor cortical neurons in Mecp2-null mice showed weakened response to locomotion speed
While the mice were performing the locomotion task on the treadmill, we monitored the neuronal activity of L2/3 pyramidal neurons in the M1 using a 2-photon microscope. A single plane of GCaMP-labeled L2/3 pyramidal cells in the anterior motor cortex (6 Hz, 512 × 512 resolution raster scan, 1 µm per pixel, 10–20 mW laser power) was selected by two-photon imaging (generally ~ 200 microns cortical depth). Mice were imaged once a week in the same area (but not the same group of neurons) until the window lost clarity, or the animal died (in the case of Mecp2-null mice). Diverse patterns of activity were observed in motor cortical neurons both during rest and locomotion (Fig. 2A). An automated cell detection, neuropil-demixing, and calcium deconvolution algorithm was used to generate cellular calcium traces [26]. Fifty-nine to 136 (mean 110, S.D. 17) neurons were analyzed in each animal at each time point. Interestingly, the firing rate of some neurons was related to the animal’s locomotion; one subpopulation of neurons demonstrated firing that correlated positively with speed (Fig. 2A, running-selective cells); another population was selective to the rest conditions (Fig. 2A, rest-selective cells).
To analyze the encoding of locomotion speed across the population of L2/3 motor cortical neurons, a custom fitting algorithm was produced that fit each cell’s activity to six different shapes (Fig. 2B): Speed-correlated (linear positive relationship to speed), running-specific (most active during locomotion conditions), fast-tuned (active selectively at high running speeds), speed anti-correlated (linear negative relationship to speed), rest-specific (most active during rest conditions), and slow-tuned (active selectively at intermediate speeds). Example event rate by locomotion tuning curves for each cell type from each genotype are shown in Fig. 2B with their respective fit values. Notably, cells with fit betas above 0.2 clearly resemble the respective fit type. Thus, this index was used as the minimum to classify cells (results were similar for other minimum fit betas). Neurons were allowed to be tuned to multiple fit types (i,e, we did not arbitrarily classify neurons to a single fit type). To be counted as a significant fit, the cell had to demonstrate a good fit (beta > 0.2) in both the first half (i.e. with the speed increasing stepwise) and the second half (with the speed decreasing stepwise) of the data.
The mean event rate by locomotion speed is plotted for each cell type, pooled into early (6–7 weeks of age, solid line) and late (9–10 weeks, dotted line) imaging sessions in Fig. 2C. Interestingly, there was a prominent decrease in the event rate of speed-correlated cells in Mecp2-null mice at later ages that was not observed in the WT mice (Fig. 2C, 1st panel). Running-specific (2nd panel), fast-tuned (3rd panel), and untuned cells (7th panel) were also less active in Mecp2-null mice compared to WT at the later ages. Rest-selective cells (5th panel) had similar or slightly increased firing in mutants compared to WT.
Loss of refinement in the population code for locomotion over time in Mecp2-null mice
The fraction of neurons with different locomotion tuning shapes was calculated per mouse per week, averaged within genotype, and plotted across weeks in Fig. 2E (blue, Mecp2-null, black, WT). The most prominent phenotype observable in the WT data was a gradual decrease in the fraction of running-selective cells (Fig. 2D, left 3 panels). This likely reflects the refinement of motor cortical ensemble activity known to occur with repeated training on a motor task [20, 28]. Rest-selective cells remained mostly stable over the same time frame. (Fig. 2D, fourth, fifth, and sixth panels), while the proportion of untuned cells slightly increased (Fig. 2D, right panel).
This iterative refinement of locomotion-related activity was not observed in Rett mice. The fraction of cells correlated to locomotion speed started off significantly lower in Rett mice and stayed low across training (Fig. 2D left 3 panels). The fraction of cells selectively active at rest, in contrast, was more common in the Rett mice (Fig. 2D, 4th and 5th panels), and this effect became more prominent with disease progression. Small fractions of neurons (~ 0.05–0.1) fired specifically at slow speeds (Fig. 2D, 6th panel), and this was not significantly different between genotypes. Taken together these results indicate that (1) neurons in the motor cortex encode different aspects of induced locomotion, (2) the proportion of neurons whose activity encodes locomotion speed decreases with task learning in WT mice, (3) this profile is lost in Mecp2-null mice, and (4) The proportion of neurons active during the rest condition increases with disease progression in Mecp2-null mice.
While there was considerable variability in the proportion of running-selective and rest-selective cells from mouse to mouse, the proportions were relatively consistent across time within individual mice, suggesting that the proportion depends on where precisely in the motor homunculus that was imaged [29]. We wondered if there also could be a fine-scale clustering of running-and rest-selective cells, similar to the previous observation that mouse M1 neurons that drive the same behavior are functionally clustered [17]. We therefore analyzed the relative physical locations of running-selective and rest-selective cells in our data set (Fig. 2E). Interestingly, we found that within-subtype intersomatic distances were significantly lower than across-subtype intersomatic distances (Fig. 2E, F). This was most prominent for rest-selective cells. Running-selective-to-Rest-selective intersomatic distances were slightly higher on average than the distance between all cells, suggesting that there is some repulsion between these cell types. While the within-subtype spatial clustering was mostly consistent across ages in WT (Fig. 2F top panel), this organization was lost after 6 weeks of age in Mecp2-null mice (Fig. 2F bottom panel).
Loss of motor cortical neuron ensemble synchronization with learning in Mecp2-null mice
We next looked for temporal patterns in the motor cortex ensemble activity during rest and locomotion. A clear feature observable in the data from WT mice was the emergence of synchronous co-activation of subpopulations of neurons starting on the 3rd–4th week of imaging (Fig. 3A left columns). Synchronous activity during locomotion, quantified using the Pearson correlation on deconvolved, thresholded calcium traces, increased from 7 to 10 weeks of age in WT (Fig. 3A left panels, Fig. 3B black lines) [30]. Sorting the correlations by distance (Fig. 3C) showed that the increase in synchrony with training was specific to cells that were greater than 25 microns apart; cells closer than this actually decreased their synchrony with training. This synchronization is similar to the synchronization of motor cortical neuronal ensemble activity known to occur with motor memory consolidation [19]. The events were too rare to be involved in triggering individual steps, and qualitative analysis of the animal’s behavior in relation to synchronous events (e.g. foot slips) did not show any clear correspondence, favoring the possibility that it reflects an internal phenomenon like memory consolidation rather than a motor command.
Interestingly, in Rett mice, synchronous events were not observed at any age (Fig. 3A, right panels), and synchrony did not increase over time (Fig. 3B blue lines). The lower synchrony compared to WT was observable across all intersomatic distances (Fig. 3C). This phenotype could possibly reflect a loss of circuit consolidation in the motor cortex. We next analyzed the synchrony of the different functional subtypes of neurons (Fig. 4). Correlation matrices were visualized sorted by running-selective (green) and rest-selective (magenta) neurons for WT (top) and Mecp2-null (bottom), for rest (left) and locomotion (right), across imaging weeks (Fig. 4A). Correlations were higher within cell type (Fig. 4A top-left quadrant of each matrix for running-selective, bottom-right quadrant for rest-selective) than between cell types (bottom-left and top-right quadrants). Synchrony increased over time in the WT but not the Mecp2-null (Fig. 4A top vs bottom panels, Fig. 4B–F), an effect that occurred across all cell types and was most clearly seen during locomotion. This loss of synchronization between behaviorally-relevant neurons with learning could contribute to the motor regression observed in these animals.