Figure 2. Between- versus within-session population coding.
(A) Upper left: illustration of rat wearing MiniLFOV. Upper right: cell contours identified during pre-training (green) and training (red) sessions in one rat; regions of overlap between contours that recurred in both sessions appear yellow. Lower left: target position of GRIN lens over the CA1 layer. Lower right: fluorescence image of lens position from the example rat. (B) Top: rastergram shows normalized calcium fluorescence traces of place cells (one per row, sorted by preferred firing location and direction) during several traversals of the short path during an example session. Bottom: rat’s position (black) with running epochs colored by direction of travel (‘LR,’ left to right, in blue; ‘RL,’ right to left, in red). (C) Individual rat data (lines and symbols) and session means (bars) for the number of subsampled beeline trials (left) and median beeline running speed (right) during pre-training (pre) and training (trn) sessions; subsampled beeline trial counts were lower for trn than pre sessions because only trials from the first 10 min of trn (prior to shock delivery) were included; running speeds were lower for trn than pre sessions because scopolamine (SCP) reduced running speeds, resulting in preferential subsampling of slower beeline trials from drug-free (DF) sessions by the algorithm that minimized running speed differences between training conditions (see ‘Methods’). (D) Bar/line graphs show number of place cells imaged per rat (left) and percentage of all imaged cells classified as place cells (right) during pre and trn sessions. (E) Pie graphs show percentages of place cells imaged per condition (‘n’ gives total number summed over rats) that were spatially tuned in the LR only, RL only, or both LR and RL running directions. (F) Tuning curve properties of place cells imaged during pre and trn sessions. (G) Place cell recurrence ratios (RR) between pre and trn sessions. (H) Top: diagram shows timeline for pre and trn sessions given 48 hr apart. Bottom: tuning curve heatmaps for recurring place cells (from all rats combined, co-sorted by peak locations from the trn1 session) that were spatially tuned in the LR (top) or RL (bottom) running directions; separate heatmaps are shown for pre, trn part 1 (trn1), and trn part 2 (trn2) sessions. (I) Between- (B) and within- (W) session population vector correlation matrices are shown for DF and SCP shock training conditions; middle bar graph shows median place tuning stability (S) for each rat (lines and symbols) and mean over rats (bars) for B and W heatmap pairs. Asterisks in (C) and (D) denote significance for main effect of DF vs. SCP or uncorrected t-test comparing pre vs. trn sessions; asterisks in (E) and (G) denote significance for uncorrected t-tests. †p<0.1; *p<0.05; **p<0.01; ***p<0.001.
Figure 2—figure supplement 1. GRIN lens placements and cell contours for imaged neurons in CA1.
The left image in each pair shows a photomicrograph of the GRIN lens placement for one of the rats in the study; arrows point to the CA1 layer, scale bar = 1 mm. The right image in each pair shows cell contours (green) identified by CNMFe superimposed over cropped images of the MiniLFOV’s view through the 1.8 mm GRIN lens; the image shown for each rat was obtained from the day of the rat’s drug-free shock training session. Figure shows results from 14 rats that were included in imaging data analysis (one rat that was included in the behavior analysis but dropped from image analysis is not shown).
Figure 2—figure supplement 2. Beeline trial counts and running speeds prior to downsampling.
To control for any possible confounding effects of behavior upon imaging results, analyses presented in the main text were performed on behavioral data that was downsampled to equalize the number of beeline trials and median running speed across imaging sessions (see ‘Methods’). Graphs plotted here show the rats’ behavior prior to downsampling beeline trials in the drug-free shock condition for rats that retained avoidance learning (n = 9), scopolamine shock condition for rats that failed to retain avoidance learning (n = 7), and barrier training condition (n = 6). (A) Number of beeline trials during pre-training (pre; data from full 15’ session), training (trn; data from 10’ pre-shock portion of session), and post-extinction (pst; data from full 15’ session) sessions prior to downsampling; a 3 × 3 mixed ANOVA yielded a marginal main effect of training condition (drug-free shock, scopolamine shock, barrier: F2,19 = 3.5, p=0.0509), a significant main effect of session (pre, trn, pst: F2,38 = 51.5, p=1.49e-11), and a significant interaction (F4,38 = 13.9, p=4.37e-7). In all training conditions there were fewer trials during ‘trn’ sessions because data only came from the first 10 min (prior to shock or barrier introduction) of the training sessions, compared with 15 min of data for other session types. For the drug-free shock condition, the mean number of beeline trials is marginally lower during ‘pst’ than ‘pre’ sessions (paired t8 = 2.1, p=0.0676; uncorrected) because despite having reached the extinction criteria during the ‘pst’ session, rats persisted in showing some avoidance of the short path and thus earned fewer short path rewards during the ‘pst’ session. (B) Median beeline running speed during per session prior to downsampling; a 3 × 3 mixed ANOVA yielded a marginal main effect of training condition (drug-free shock, scopolamine shock, barrier: F2,19 = 3.07, p=0.0698), a significant main effect of session (pre, trn, pst: F2,38 = 5.08, p=0.0111), and a significant interaction (F4,38 = 20.8, p=3.74e-9). Uncorrected post hoc paired t-tests found that for the drug-free shock condition, beeline running speed was significantly lower during ‘pst’ than ‘pre’ (t8 = 4.4, p=0.0022) or ‘trn’ (t8 = 4.7, p=0.0015) sessions; this was because despite having reached the avoidance extinction criteria during the ‘pst’ session, rats persisted in running more a bit slowly on the short path after avoidance extinction. In addition, during ‘trn’ sessions given on scopolamine, rats ran more slowly during ‘trn’ sessions than during ‘pre’ (t6 = 5.5, p=0.0015) or ‘pst’ (t6 = 5.2, p=0.0019) sessions (both of which were given drug free) because the drug acutely reduced running speeds on the maze, as reported in the main text.
Figure 2—figure supplement 3. Relationship between place tuning versus size and eccentricity of imaged cell contours.
(A) Scatter plot shows contour area (in pixels, x-axis) versus eccentricity of contour from center of the GRIN lens (in pixels, y-axis) for all place cells and non-place cells imaged during the pre-training session for the drug-free and scopolamine shock conditions; histograms on perimeter show distributions of values plotted on each axis. Independent t-test results show that non-place cells were slightly but significantly smaller and less eccentric from the lens center than place cells, possibly because residual post-correction motion artifact may have caused the smallest cell contours to jitter occasionally out of their ROI and therefore meet the criterion for spatial tuning stability slightly less often. (B) Same as (A) except graphs show contour area and eccentricity for all place cells imaged during the pre-training session prior to drug-free shocks; place cells that recurred during the subsequent shock session are shown in blue, whereas place cells that did not recur are shown in orange. Independent t-test results show that non-recurring cells were slightly but significantly larger and more eccentric than recurring place cells, possibly because compared with contours near the center of the lens, between-session overlap of contours at eccentric locations (near the edge of the lens) was slightly more sensitive to small between-session changes in the mounted position of the miniscope camera. (C) Same as (B) except data is shown for place cells imaged during the pre-training session prior to shocks on scopolamine (rather than drug free); as in the drug-free condition, non-recurring cells were slightly but significantly larger and more eccentric than recurring place cells; hence the effect of eccentricity on recurrence probability was similar on the drug-free and scopolamine conditions. Confirming this, independent t-tests found no significant difference between the drug-free versus scopolamine conditions in the size of recurring (t1860 = 0.18, p=0.86) or non-recurring (t882 = 0.02, p=0.98) place cells, nor in the eccentricity of recurring (t1860 = 1.62, p=0.11) or non-recurring (t882 = 1.01, p=0.31) place cells.
Figure 2—figure supplement 4. Distributions of place cell tuning properties.
Histograms at top compare pre-training and training session distributions of firing properties measured from tuning curves that were spatially tuned in the LR (top row, blue) or RL (second from top row, red) direction during one or both sessions. Boxplots below histograms show each rat’s median and range of tuning property values for LR and RL tuning curves combined; boxplot medians for each rat are the same values plotted for each rat in Figure 2F. (A) Peak Af/s rates. From the histograms, it can be seen that scopolamine acutely reduces the proportion of tuning curves peaks that are <5 Af/s while having little effect on the frequency of peaks that are >5 Af/s. (B) Out-of-field Af/s rates. Similar to peak Af/s rates, scopolamine acutely reduced the incidence of low out-of-field rates (<1 Af/s), while having little effect on the incidence of higher rates (>1 Af/s). (C) Field widths. Scopolamine slightly narrows the width of place fields; since the field width is defined as the number of spatial bins that exceed 50% of the peak rate, this slight narrowing of place fields may be related to the fact that peak Af/s rates are higher on scopolamine (see A). (D) Spatial information. Distributions of bits/Af showed little change on scopolamine.
Figure 2—figure supplement 5. Pre-training and training session data from individual rats.
As in Figure 2H, heatmaps are shown for the pre, trn1, and trn2 sessions, with between- (pre vs. trn1) and within- (trn1 vs. trn2) session correlation matrices below heatmaps. For brevity, LR and RL tuning curves are co-sorted within heatmaps so that cells meeting spatial selectivity criteria in both running directions contribute two rows to each heatmap (one per direction), whereas cells meeting selectivity criteria in only one direction contribute only row to each heatmap. (A1) Rats (n = 9) from the drug-free shock condition that showed 48 hr avoidance retention. (A2) Rats (n = 2) from the drug-free shock condition that failed to show 48 hr avoidance retention. (B1) Rats (n = 7) from the scopolamine condition that failed to show 48 h avoidance retention. (B2) Rats (n = 2) from the scopolamine shock condition that showed 48 hr avoidance retention.
Figure 2—figure supplement 6. Between- and within-session place coding in male versus female rats.
Statistical analyses reported in Figure 2 were performed to compare results from male versus female rats. 2 × 2 ANOVAs were performed with sex (M, F) and drug (drug-free, scopolamine) as independent factors. (A) There were no significant effects of sex upon the number of place cells imaged per rat (analysis shown for training session), recurrence ratio (RR), or spatial tuning properties (peak rate, out-of-field rate, field width, spatial information); main effect of sex is shown above each graph; however, sex-by-drug interactions also were not significant (not shown). (B) As reported in the main text, there was no significant main effect of drug upon within-session population vector correlations; however, there was a significant main effect of sex arising from the fact that females exhibited lower within-session population vector correlations than males in both drug conditions; this was the only significant effect of sex that was found in any of the analyses. (C) As reported in the main text, there was a significant main effect of drug upon between-session population vector correlations; however, there was no significant main effect of sex and no sex-by-drug interaction.
Figure 2—figure supplement 7. Shuffled analysis of between- and within-session population coding.
Results shown are similar to those plotted in Figure 2, except that prior to analysis, the deconvolved spike train from every beeline trial was circularly shifted by a random amount against position tracking data from the trial. (A) Bar/line graphs show number (left) and percentage (right) of cells per rat that were classified as place cells after the random circular shift; the number of place cells was greatly reduced from that shown in Figure 2. There was a significant tendency for more shuffled trains to meet place criteria in trn than pre sessions, attributable to the fact that all detections of place cells in shuffled data were artifactual and the likelihood of artifactual detection was higher for trn than pre sessions because fewer trials were analyzed in trn sessions (from which only the first 10 min pre-shock were analyzed) than pre sessions (from which all trials were analyzed across the 15 min session). (B) Pie graphs show percentages of ‘place cells’ detected in shuffled data (‘n’ gives total number summed over rats) that were spatially tuned in the LR only, RL only, or both LR&RL running directions. (C) Tuning curve properties of ‘place cells’ detected in shuffled data; tuning property differences between trn and pre sessions are attributable to differences in behavior sampling. (D) Recurrence ratios (RR) between pre and trn sessions for ‘place cells’ detected in shuffled data. (E, F) Results plotted in panels (E, F) are from analyses of shuffled trains performed on cells that were classified as place cells based on their unshuffled trains (the same population of place cells analyzed in Figure 2). (E) Top: diagram shows timeline for pre and trn sessions given 48 hr apart. Bottom: tuning curve heatmaps of shuffled trains for recurring place cells (from all rats combined, co-sorted by peak locations from the trn1 session) whose unshuffled trains were spatially tuned in the LR (top) or RL (bottom) running directions; separate heatmaps are shown for pre, trn part 1 (trn1), and trn part 2 (trn2) sessions. (F) Between- (B) and within- (W) session population vector correlation matrices are shown for DF and SCP shock training conditions; middle bar graph shows median place tuning stability (S) derived from shuffled trains for each rat (lines and symbols) and mean over rats (bars) for B and W heatmap pairs. p<0.1; *p<0.05; **p<0.01; ***p<0.001. Higher correlation values were uniformly distributed throughout the within- than between-session correlation matrices, indicating that baseline firing rates tended to be better correlated within than between sessions.
Figure 2—figure supplement 8. Pre-training and training session results from drug-free (DF) rats that failed to retain avoidance and scopolamine (SCP) rats that successfully retained avoidance.
Panels (A–G) are similar to those shown in panels (C–I) of Figure 2, except here, results are shown for DF rats that failed to retain avoidance (n = 2) and SCP rats that successfully retained avoidance (n = 2) 48 hr after training. The sample size of rats was not sufficient to perform ANOVAs, so independent t-tests were performed to compare values included in each graph bar against values in corresponding bars from Figure 2. The only comparison to beat p<0.05 (uncorrected) between-session place tuning stability (S) in the SCP condition (‘*’ and up arrow over center bar graph in panel G; compare with panel I of Figure 2). Hence, in rats for which SCP failed to block avoidance, SCP also failed to impair between-session stability of place cell population vectors. Further supporting this conclusion, a marginally significant correlation was observed (see regression line in panel H, upper right) between 48 hr avoidance retention scores and between-session S values in the SCP condition (n = 11 rats, combining those that retained and forgot the avoidance response). Hence, lower between-session place tuning stability was associated with more severely impaired retention of avoidance. No such correlations were observed in the DF condition or for within-session stability scores.