Article Figures & Data
Figures
- Figure 1.
Tactile environment enrichment sharpens spatial acuity in adjacent-whisker discrimination. A, Awake, head-fixed mice are trained on a whisker spatial discrimination go/no-go task while being simultaneously imaged. The animal licks for water reward on the same go whisker stimulation, but withholds licking on any other whisker stimulation. In the final form of the task, the no-go whisker is always adjacent to the go whisker. B, An example lick raster of a discrimination task session. Go trials (left) and no-go trials (middle) are randomly interleaved. Each trial is 12 s, and the stimulus onset is at 3.5 s (first vertical line). The animal has a response window of 0.2–2 s after stimulus onset (second vertical line signifies the end of response window). Trials where animals lick prematurely (0.5 s before and 0.19 s after stimulus onset) are excluded. A well-trained animal typically has <5% premature trials. Tactile task performance is measured by the ratio of hit rate to false alarm rate. In this example session, the animal’s discrimination performance is 4.67. C, Discrimination task performance is significantly better in enriched animals (control bootstrap mean = 2.64, error bar: 2.5th percentile = 1.99, and 97.5th percentile = 3.43, N = 7885 trials, 6 mice; enriched bootstrap mean = 5.98, error bar: 2.5th percentile = 3.68, and 97.5th percentile = 9.40, N = 13,123 trials, 7 mice; p < 0.05). D, The hit rates were the same between control and enriched animals (control bootstrap mean = 84.07%, error bar: 2.5th percentile = 76.92%, and 97.5th percentile = 90.17%, N = 7885 trials, 6 mice; enriched bootstrap mean = 84.36% error bar: 2.5th percentile = 78.09%, and 97.5th percentile = 89.52%, N = 13,123 trials, 7 mice; p > 0.05). E, Lowered false alarm rate is the main behavior improvement in enriched animals (control bootstrap mean = 37.82%, error bar: 2.5th percentile = 30.49%, and 97.5th percentile = 45.17%, N = 7885 trials, 6 mice; enriched bootstrap mean = 21.86% error bar: 2.5th percentile = 16.37%, and 97.5th percentile = 27.64%, N = 13,123 trials, 7 mice; p < 0.05). F, The detection false alarm rates were not different between enriched and control animals (control bootstrap mean = 28.79%, error bar: 2.5th percentile = 24.15% and 97.5th percentile = 33.92%, N = 7829 trials, 6 mice; enriched bootstrap mean = 27.39%, 2.5th percentile = 19.93% and 97.5th percentile = 33.44%, N = 15 073 trials, 7 mice; p > 0.05). G, An example of a large cage with objects of various shapes and textures used for tactile enrichment. *p < 0.05.
- Figure 2.
EE increases whisker adjacent-tuning in vS1. A, An example two-photon calcium image of vS1 L2/3 response to a single-whisker stimulation in an awake, head-fixed mouse. Right, The response of an example cell (outlined in red) to two adjacent whisker stimulation in separate, interleaved trials. Thin colored lines are single trials of calcium trace, thick black lines are trial-averaged. B, Whisker tuning calculation: the maximum response from all trials of whisker 1 stimulation form a distribution while those from whisker 2 stimulation trials form the other. ROC analysis is used to quantify the separation of the two distributions. C, Two example ROC curves from two individual cells. An AUROC above 0.5 signifies tuning to whisker 1, and below 0.5 to whisker 2. D, Examples of L2/3 cell whisker tuning in vS1 (left: vS1 of a control mouse, right: vS1 of an enriched mouse). The AUROC is normalized by subtracting 0.5, so that positive values signify tuning to whisker 1, and negative values to whisker 2. Warm-colored dots: cells tuned to whisker 1 at its physical location; cold-colored dots: cells tuned to an adjacent whisker 2. Red outline: approximate functional space of whisker 1 principal barrel, identified using wide-field imaging. Blue outline: approximate functional space of whisker 2 principal barrel. E, Adjacent whisker tuning increases in enriched animals. Adjacent tuning is defined as the total absolute value of normalized AUROC of all the cells tuned to the adjacent whisker within the approximate functional boundary of a given barrel (control bootstrap mean 1.07, error bar: 2.5th percentile 0.53, and 97.5th percentile 1.50, N = 60 AT cells, 5 mice; enriched bootstrap mean 2.47, error bar: 2.5th percentile 1.61, and 97.5th percentile 3.42, N = 108 AT cells, 6 mice; p < 0.05). *p < 0.05.
- Figure 3.
Function of adjacent-tuned cells explains behavior in enriched animals. A, An example session of S1 L2/3 cellular activity in an awake animal performing a whisker discrimination task. Images are averaged over all trials within a performance criterion. B, For a single cell, decision encoding is quantified by ROC analysis between the response distributions during the trials where the animal licked versus those where the animal did not lick. It is calculated under conditions where go whisker was stimulated versus no-go whisker was stimulated. C, S1 cells encode stimulus feature more than decision. The information about the stimulus and decision is averaged (mean of absolute value of normalized AUROC over all cells). Because the determining factor in discrimination performance is the false alarm rate, all decision coding in this study refers to the decision between false alarm and correct rejection trials (bottom of B). Stimulus encoding bootstrap mean = 0.090, error bar: 2.5th percentile = 0.077, and 97.5th percentile = 0.11; decision encoding bootstrap mean = 0.065, error bar: 2.5th percentile = 0.056, and 97.5th percentile = 0.077, N = 7092 cells, 11 mice; p < 0.05. D, In PT cells, enrichment does not improve their decision-coding capacity (control bootstrap mean 3.34, error bar: 2.5th percentile 1.10, and 97.5th percentile 4.96, N = 361 PT cells, 5 mice; enriched bootstrap mean 4.09, error bar: 2.5th percentile 2.59, and 97.5th percentile 6.78, N = 400 PT cells, 6 mice; p > 0.05). E, AT cells encode more decision information with enrichment (control bootstrap mean 0.51, error bar: 2.5th percentile 0.25, and 97.5th percentile 0.97, N = 60 AT cells, 5 mice; enriched bootstrap mean 1.31, error bar: 2.5th percentile 0.80, and 97.5th percentile 1.97, N = 108 AT cells, 6 mice; p < 0.05). F, Decision information encoded in PT cells cannot predict behavior (control animals r = 0.54, p > 0.05, bootstrap portion of samples that have significant correlation was 23.2%, the mean of significant correlation value is 0.62, N = 361 PT cells, 5 mice; enriched animals r = −0.37, p > 0.05, bootstrap portion of samples that have significant correlation was 24.3%, the mean of significant correlation value is −0.75, N = 400 PT cells, 6 mice). G, In enriched animals, but not control animals, the average decision information encoded in AT cells predicts false alarm rate in discrimination task (enriched: r = −0.86, p < 0.05, in multilevel hierarchical bootstrapped samples, portion of samples that have significant correlation was 55%, the mean of significant correlation value is −0.82, N = 108 AT cells, 6 mice; control: r = 0.02, p > 0.05, bootstrap portion of samples that have significant correlation was 7.4%, the mean of significant correlation value is −0.28, N = 60 AT cells, 5 mice). H, Using populations of either PT cells or AT cells, Fisher LDA is used to classify false alarm and correct rejection trials. For any given session, the number of PT cells and the number of AT cells are kept the same. The decoder predicts a single trial more accurately using AT cells than PT cells in enriched animals, but not in control animals (control PT bootstrap mean 56%, error bar: 2.5th percentile 35%, and 97.5th percentile 71%, AT bootstrap mean 56%, error bar: 2.5th percentile 36% and 97.5th percentile 68%, N = 60 cells, 5 mice; enriched PT bootstrap mean 67%, error bar: 2.5th percentile 62%, and 97.5th percentile 74%, AT bootstrap mean 72%, error bar: 2.5th percentile 66% and 97.5th percentile 78%, N = 108 cells, 6 mice). *p < 0.05.
- Figure 4.
Premotor cells encode decision. A, An example two-photon calcium image of vM2 L2/3 cells in an awake, head-fixed mouse. Some cells respond vigorously to a single whisker stimulation (left), but exhibits little whisker tuning, while others exhibit some moderate extent of whisker preference (right). Thin colored lines are single trials of calcium trace, thick black lines are trial-averaged. B, An example of L2/3 cell whisker tuning in vM2. Unlike vS1, cells tuned to a given whisker do not exhibit any topographical organization. C, In control animals, vM2 cells on average do not encode the animal’s decision relative to stimulus information more than vS1 cells do (control vS1 cells bootstrap mean 0.66, error bar: 2.5th percentile 0.46, and 97.5th percentile 0.81, N = 2826 cells; 5 mice; vM2 cells bootstrap mean 0.71, error bar: 2.5th percentile 0.55 and 97.5th percentile 0.84, N = 1842 cells, 4 mice). D, In enriched animals, vM2 cells on average encode the animal’s decision more than vS1 cells do (enriched vS1 cells: bootstrap mean 0.88, error bar: 2.5th percentile 0.79, and 97.5th percentile 0.97, N = 4266 cells, 6 mice; vM2 cells bootstrap mean 1.13, error bar: 2.5th percentile 0.94 and 97.5th percentile 1.34, N = 3972 cells, 7 mice, p < 0.05). E, In control animals, the overlap between cellular populations encoding stimulus and decision is not different between vM2 and vS1 (control vS1 cells bootstrap mean 5.1%, error bar: 2.5th percentile 2.9%, and 97.5th percentile 7.7%, N = 2826 cells; 5 mice; vM2 cells bootstrap mean 2.3%, error bar: 2.5th percentile 1.3% and 97.5th percentile 3.3%, N = 1842 cells, 4 mice, p > 0.05). F, In enriched animals, the cellular population encoding stimulus and decision become more separated in vM2 than in vS1, quantified as % of overlap between stimulus-encoding cells and decision-encoding cells (enriched vS1 cells: bootstrap mean 8.0%, error bar: 2.5th percentile 5.9%, and 97.5th percentile 10%, N = 4266 cells, 6 mice; vM2 cells bootstrap mean 3.1%, error bar: 2.5th percentile 1.7% and 97.5th percentile 4.5%, N = 3972 cells, 7 mice, p < 0.05). *p < 0.05.
- Figure 5.
EE increases decision coding in vM2 cells. A, The total amount of information encoded about the animal’s decision in vM2 cells is greater in enriched animals (control bootstrap mean 2.36, error bar: 2.5th percentile 1.46 and 97.5th percentile 3.25, N = 1842 cells, 4 mice; enriched bootstrap mean 7.25, error bar: 2.5th percentile 4.04 and 97.5th percentile 11.37, N = 3972 cells, 7 mice, p < 0.05). B, In enriched animals, the total amount of information encoded about the animal’s decision in vM2 cells predict the animal’s false alarm rate in discrimination task (control r = −0.62, p > 0.05, in multilevel hierarchical bootstrapped samples, portion of samples that have significant correlation was 25%, the mean of significant correlation value is −0.7, N = 1842 cells, 4 mice; enriched r = −0.7, p < 0.05, in multilevel hierarchical bootstrapped samples, portion of samples that have significant correlation was 54%, the mean of significant correlation value is −0.81, N = 3972 cells, 7 mice). C, Using either randomly sampled vM2 cells or decision-encoding cells only, Fisher LDA is used to classify false-alarm and correct rejection trials. For any given session, the number of each type of cells are kept the same. Using either cell type, the decoder predicts a single trial more accurately in enriched animals (control: random cells bootstrap mean 65%, error bar: 2.5th percentile 59%, and 97.5th percentile 73%, decision cells bootstrap mean 73%, error bar: 2.5th percentile 68% and 97.5th percentile 78%, N = 131 cells, 4 mice; enriched random cells bootstrap mean 74%, error bar: 2.5th percentile 68%, and 97.5th percentile 80%, decision cells bootstrap mean 80%, error bar: 2.5th percentile 77% and 97.5th percentile 84%, N = 510 cells, 7 mice). In general, the decoder performs better using decision cells than using randomly sampled cells. *p < 0.05.
In this issue
