Hippocampal place code plasticity in CA1 requires postsynaptic membrane fusion

Abstract

Rapid delivery of glutamate receptors to the postsynaptic membrane via vesicle fusion is a central component of synaptic plasticity. However, it is unknown how this process supports specific neural computations during behavior. To bridge this gap, we combined conditional genetic deletion of a component of the postsynaptic membrane fusion machinery, Syntaxin3 (Stx3), in hippocampal CA1 neurons of mice with population in vivo calcium imaging. This approach revealed that Stx3 is necessary for forming the neural dynamics that support novelty processing, spatial reward memory and offline memory consolidation. In contrast, CA1 Stx3 was dispensable for maintaining aspects of the neural code that exist presynaptic to CA1 such as representations of context and space. Thus, manipulating postsynaptic membrane fusion identified computations that specifically require synaptic restructuring via membrane trafficking in CA1 and distinguished them from neural representation that could be inherited from upstream brain regions or learned through other mechanisms.

Figure 1.

Stx3-dependent postsynaptic membrane fusion in CA1 is critical for novelty preference and reward anticipation. A) Model of Stx3-dependent AMPA receptor trafficking. Stx3(+) SNARE complexes insert GluA1(+) AMPA receptors into the postsynaptic membrane adjacent to spines in an activity dependent manner. Receptors laterally diffuse into the synapse, leading to potentiation. B-C) CA1 Stx3 is necessary for novel environment preference B) Novel environment preference assay. During training trials, animals can explore one arm of a T-maze (familiar arm “F”), while the other arm is blocked (novel arm, “N”). During the test trial, the block is removed and animals are free to explore the full maze. ITI - inter-trial-interval. C) CA1 Stx3 is necessary for novel environment preference. Top - Representative test trial from a single Cre-injected and a single ΔCre mouse. Heatmaps indicate relative occupancy of each location. Bottom - Novel arm preference score [difference score = (time spent in the novel arm) - (time spent in the familiar arm)]. Dotted line indicates equal time spent in the two arms [N=19 control mice, 24 Cre mice. Unpaired t-test: t=2.83, p=0.007] D) Schematic of the head-fixed virtual reality (VR) and two photon (2P) apparatus. E) Task design of the VR novel arm Y-Maze. Left - Track schematic. Arrows indicate the animals’ trajectory on left or right trials. Shaded regions indicate reward zones. Right - Training protocol. Days 1–5: 5 blocks of “familiar arm” trials (20 trials or 5 minutes, black), 1 block of randomly interleaved “familiar” and “novel arm” trials (40 trials or 10 minutes, orange). Animals are forced to take left or right trajectories on each trial. The animals cannot control the yaw angle in the virtual environment. Day 6: familiar and novel arms randomly interleaved in all blocks. F-G) CA1 Stx3 is necessary for reward anticipation behaviors. F) Control animals (gray) display more anticipatory licking than Cre-injected animals (red). Left - peri-reward lick rate on familiar arm trials as a function of position relative to the reward zone on day 5 (magenta - reward zone, shaded regions indicate across animal mean ± sem). Right - Average familiar trial peri-reward lick rate on each day. Dots indicate the across trial average for each mouse. Shaded bars indicate across animal mean. [N=9 control mice, 8 Cre mice, 6 days. Mixed effects ANOVA: virus main effect F(1,15)=10.66 p=5.22×10⁻⁴, day main effect F(5,75)=7.55 p=8.61×10⁻⁶, interaction F(5,75)=.801 p=.55]. G) Same as (F) for novel arm trials. [N=9 control mice, 7 Cre mice, 6 days. Mixed effects ANOVA: virus main effect F(1,15)=8.40 p=.011, day main effect F(5,75)=2.01 p=.087, interaction F(5,75)=1.33 p=.263] See also Figures S1 & S2.

Figure 2.

CA1 Stx3 deletion does not affect gross measures of place code quality and stability. A-D) Accurate neural coding of position does not require CA1 Stx3 A) Example place cell timeseries (“Population Timeseries”, Left) and spatial decoding (“Spatial Decoding”, Right) from a single control animal (Ctrl-3; gray). Population Timeseries: Top- Activity rate of co-recorded place cells over time from a subset of familiar arm trials on day 5. Cells are sorted by their location of peak activity in the average familiar trial activity rate map. Bottom - The position of the animal on the virtual track over time (blue). Shaded regions indicate inter-trial intervals. Red dashes are times of licks, and the orange dot indicates when a reward was dispensed. Rewards were randomly omitted on 10% of trials. Note the tight correspondence between the position of the animal on the track and the sequence of neural activity. Spatial Decoding: Left - Posterior distributions from leave-one-trial-out naive Bayes decoding of position from familiar trials on day 5 from the same example animal (Ctrl-3, N=1208 cells, 120 trials). Right - Same as (Left) for novel trials on day 1 (N=1568 cells, 20 trials). Note accurate decoding even on the initial novel trials. B) Same as (A) for a single Cre animal (Cre-3; red). Note similar place cell sequences and decoding accuracy between (A) and (B). (Day 5 familiar trial decoding: N=1087 cells, 17 trials, Day 1 novel trial decoding: N=936 cells, 17 trials) C) Average leave-one-trial out naive Bayes spatial decoding error on familiar trials on each day. Dots indicate the average error of 50 models trained with a random subset of neurons chosen with replacement (n=128 cells per model) for each mouse. Shaded bars indicate across animal mean. [N=9 control mice, 8 Cre mice, 6 days. Mixed effects ANOVA: virus main effect F(1,14)=1.75 p=0.207, day main effect F(5,70)=30.2 p=3.25×10⁻¹⁶, interaction F(5,70)=1.70 p=0.145; Posthoc paired t-test (Holm-corrected p-value) comparing decoding error on each day: day 1 vs day 2 - t=8.17 p=7.96 × 10⁻⁶; day 1 vs day 3 - t=9.26 p=1.76×10–6; day 1 vs day 4 - t=10.2 p=5.79×10⁻⁷; day 1 vs day 5 - t=6.39 p=1.33×10⁻⁴; day 1 vs day 6 - t=9.68 p=1.07×10⁻⁶; day 2 vs day 3 - t=3.35 p=0.039; day 2 vs day 6 - t=3.64 p=.024; all other comparisons p>.05] D) Same as (C) for novel trials. [N=9 control mice, 8 Cre mice, 6 days. Mixed effects ANOVA: virus main effect F(1,14)=.576 p=0.460, day main effect F(5,70)=14.6 p=8.74×10⁻¹⁰, interaction F(5,70)=2.96 p=0.017; Posthoc paired t-test (Holm-corrected p-value) comparing decoding error on each day: day 1 vs day 2- control t=5.60 p=6.61×10–4; day 1 vs day 3 - t=5.37 p=9.28×10⁻⁴; day 1 vs day 4 - t=8.64 p=5.00×10⁻⁶; day 1 vs day 5 - t=3.38 p=.037; day 1 vs day 6 - t=6.52 p=1.35×10⁻⁴; day 2 vs day 4 - t=3.67 p=.023; day 2 vs day 6 - t=3.75 p=.021; all other comparisons p>.05] E-H) Comparable stability of spatial coding between Control and Cre animals. E) Activity of place cells tracked over all 6 days from Control animals. Left - Position-binned activity rate of a single place cell in familiar trials from all days (mouse: Ctrl-5, cell 1). Right - Position-binned trial-averaged activity rate (z-scored) of all place cells tracked over days 1–5 (top row - familiar trials, bottom row - novel trials). Cells are plotted in the same order for each plot within a row and are sorted by their location of peak activity on odd-numbered trials from day 3. Day 3 heatmaps indicate average activity on even-numbered trials. All other heatmaps indicate the average of all trials. These heatmaps show a stable tiling of place cells across days. Far right - (top) novel trial activity with cells sorted by familiar trial activity. (bottom) Familiar trial activity with cells with cells sorted by novel trial activity. Place cells that code for the stem of the Y-Maze are visible at the beginning of the track in both conditions. Place cells remap on the arms of the maze as indicated by the disorganized activity rate maps. F) Same as (E) for place cells tracked over all experimental days in Cre animals. G) Across animal average day x day population vector correlation (PV corr.) for familiar trials (Left: Control animals, Middle: Cre animals, Right: difference between groups). Each animal’s population vector is calculated using the trial-averaged spatial activity rate map for each cell on each day. [N=9 control mice, 7 Cre mice. Right: control vs Cre unpaired t-test of average across day PV correlation t=.333 p=.745] H) Same as (G) for novel trials. [N=9 control mice, 7 Cre mice. Right: control vs Cre unpaired t-test of average across day PV correlation t=1.27 p=.223] See also Figures S3 & S4.

Figure 3.

Spatial coding in Cre animals may be passively inherited from upstream activity. A-E) Computational model of the CA3-CA1 circuit. A) Schematic for K-Winners-Take-All (KWTA) model of place field inheritance. A set of place selective cells (“CA3”) projects randomly to a set of output neurons (“CA1”). The set of active neurons at the output layer at each position is determined by a KWTA threshold (the K most excited neurons are active [blue], all other neurons are silent [black]). In the “Plasticity” model, synaptic weights are updated according to a noisy Hebbian rule. In the “No-Plasticity” model, synaptic weights randomly fluctuate. B-C) Both the Plasticity and No-Plasticity models produce stable place cells at the CA1 layer. B) Plasticity model spatial coding. Left - Position-binned activity of an example place selective unit from the model. Each row is a single training iteration of the model, akin to a trial in the real experiment. Columns indicate virtual position. Middle & Right - The activity of all place selective units is plotted for the first (Middle) and last trial (Right). Model cells are sorted by their location of peak activity on the last trial. Each row indicates the z-scored activity rate of a unit as a function of position. C) Same as (B) for the No-Plasticity model. Note similar spatial coding and stability in the two models. D) Place fields are wider in the Plasticity model. Histogram of place field widths (black - Plasticity, red - No-Plasticity). Inset - mean ± std field width from each model. E) The Plasticity model has fewer place fields per cell than the No-Plasticity model. Histogram of the number of place fields per cell (black - Plasticity, red - No-Plasticity). Inset - mean ± std number of fields per cell from each model. F-I) Differences between Control and Cre animal place field properties mirror differences between Plasticity and No-Plasticity models. F-G) Cre animals have narrower place fields. F) Familiar trials. Left - Histogram of place field widths for combined data from day 5 (black - Control, red - Cre). Right - Within animal average field width on each day. Dots indicate the across cell average for each mouse. Shaded bars indicate across animal means. [N=9 control mice, 7 Cre mice, 5 days (day 6 excluded). Mixed effects ANOVA: virus main effect F(1,14)=10.6 p=.006, day main effect F(4, 56)=4.57 p=0.003, interaction F(4,56)=2.81 p=0.034; day 6 unpaired t-test control vs Cre: t=−3.90 p=.002] G) Same as (F) for novel trials. [N=9 control mice, 7 Cre mice, 5 days (day 6 excluded). Mixed effects ANOVA: virus main effect F(1,14)=10.2 p=.007, day main effect F(4,56)=2.11 p=.091, interaction F(4,56)=.758 p=.557; day 6 unpaired t-test control vs Cre: t=−2.38 p=.036] H-I) Cre animals have more place fields per cell. H) Familiar trials. Left - Histogram of number of place fields per cell for day 5 data combined across all mice (Control - black, Cre-red). Right - Within animal average number of place fields per cell on each day. Dots indicate the across cell average for each mouse. Shaded bars indicate across animal means. [N=9 control mice, 7 Cre mice, 5 days (day 6 excluded). Mixed effects ANOVA: virus main effect F(1,14)=25.5 p=1.76×10⁻⁴, day main effect F(4,56)=.804 p=.528, interaction F(4,56)=3.29 p=.017; day 6 unpaired t-test control vs Cre: t=5.30 p=1.19×10⁻⁴]. I) Same as (H) for novel trials [N=9 control mice, 7 Cre mice, 5 days (day 6 excluded). Mixed effects ANOVA: virus main effect F(1,14)=17.5 p=9.12×10⁻⁴, day main effect F(4,56)=.158 p=.959, interaction F(4,56)=2.96 p=0.027; day 6 unpaired t-test control vs Cre: t=3.52 p=.008] See also Figure S5.

Figure 4.

CA1 Stx3 is necessary for novelty-induced increases in neural activity. A) Normalized population activity rate as a function of position on each trial during block 6 for an example control mouse (Ctrl-1). We divide each cell’s activity by its mean activity on the 10 trials preceding the start of block 6 (which occurs at Trial 0). We take the log of the resulting quantity for each cell, so that increases in activity are positive and decreases in activity are negative. This value is plotted as a function of position on each trial. Left - Normalized population activity on day 1. Familiar and novel trials are also plotted separately. Right - Data from day 5. B) Same as (A) for an example Cre mouse (Cre-1). C) CA1 neural activity increases relative to baseline for both familiar and novel trials during block 6. Cre animals demonstrate a blunted response. Activity rate on each novel trial for days 1 and 5. Data are shown as across animal mean +/− SEM. Dashed line indicates the start of block 6 (Top-Familiar trials, Bottom-Novel trials). [N=9 control mice, 7 Cre mice, posthoc unpaired t-test of trial-averaged activity rate control vs Cre (Holm-corrected p-value) for mixed effects ANOVA in (E): Familiar Day 1 - t=−3.36 p=.020; Familiar Day 5 - t=−2.73 p=.040; Novel Day 1 - t=−4.04 p=.004; Novel Day 5 - t=−4.69 p=.003] D) Activity rate increases are shared across large portions of the population. Cumulative histograms of trial averaged activity rates across cells on day 5 (Left - familiar trials, Right - novel trials). Lines indicate histograms from each animal (black - Control, red - Cre). See (C) for information on statistical tests E) Within animal averaged activity rate on block 6 trials (black - Control, red - Cre, dots - familiar trials, “x” - novel trials). Activity rates from Cre animals are significantly lower than activity rates from Control animals for both familiar trials [N=9 Control mice, 7 Cre mice, 5 days (day 6 excluded). Mixed effects ANOVA: virus main effect F(1,14)=16.6 p=.001, day main effect F(4,56)=.944 p=.445, interaction F(4,56)=1.01 p=.412] and novel trials [N=9 Control mice, 7 Cre mice, 5 days (day 6 excluded). Mixed effects ANOVA: virus main effect F(1,14)=37.5 p=2.60×10⁻⁵, day main effect F(4,56)=4.13 p=.005, interaction F(4,56)=1.60 p=.186]. The difference between novel trial activity and block 6 familiar trial activity is larger in Control animals than Cre animals [N=9 Control mice, 7 Cre mice, 5 days (day 6 excluded). Mixed effects ANOVA: virus main effect F(1,14)=8.13 p=.013, day main effect F(4,56)=6.31 p=2.89×10⁻⁴, interaction F(4,56)=.443 p=.777]. See also Figure S6.

Figure 5.

CA1 Stx3 is necessary for reward location memory. A-D) Stx3 is necessary for recruitment of place cells to reward locations. A) Left - Day 6 trial-averaged place cell activity (even trials only, odd trial sorted, z-scored) from all control animals. Arrows indicate place cell overrepresentation of rewarded locations. Blue line shows diagonal as a reference. Right - Joint normalized histogram of locations of peak activity on left and right trials for place cells on day 6. Rewarded locations are highlighted by the shaded regions (blue - left reward zone, green-right reward zone). Cells that do not remap across trial types will lie on the diagonal. Cells that remap specifically to match shifts in the reward locations will lie on the off-diagonal near the intersection of the reward zones (“reward cells”, dashed box). B) Same as (A) for Cre animals. C) The difference between control and Cre histograms highlights fewer reward cells in the Cre mice. D) Proportion of the population defined as “reward cells” shown for each mouse on each day. Statistical tests performed on logit transformed proportions. [N=9 control, 7 Cre mice, 6 days, Mixed effects ANOVA: virus main effect F(1,14)=8.49 p=.011, day main effect F(5,70)=10.2 p=2.40×10⁻⁷, interaction F(5,70)=1.46 p=.214]. Dashed line indicates the proportion of reward coding neurons expected by random remapping between novel and familiar arms with constant coding for the stem of the maze. Across-day average fraction of reward cells is greater than chance in control but not Cre animals [One sample t-test: control - t=16.7 p=1.64×10⁻⁷, Cre - t=2.11 p=.080] E-G) After moving reward locations, Cre animals extinguish licking faster at previously rewarded locations. E) Schematic of the reward reversal task. On day 7, the animals performed the first 2 blocks of trials as in day 6 (randomly interleaved left and right trials). At the beginning of block 3, the reward locations on the two arms were switched. The rewards stayed in this new location for the remaining trials on day 7 and all trials on day 8. F) Lick rate as a function of position for binned left arm trials following the reward reversal (N=9 control mice, 6 Cre mice). Data are shown as across animal mean ± SEM. Solid blue shaded region indicates the current reward zone. Hatched blue region indicates the previous reward zone. G) Normalized lick rate as a function of position for binned trials following the reward reversal. We divide each animal’s lick rate by the mean lick rate on the 10 trials preceding the reversal (N=9 control mice, 6 Cre mice). Data are shown as across animal mean ± SEM. Cyan arrow highlights slower licking extinction at the old reward zone for control mice. H-I) A mixed effects binomial regression was used to quantify the probability of licking more than once in the previous reward anticipatory zone during left arm extinction trials. H) Log loss of model fit (red) vs shuffled distribution (blue). Shuffled distribution was generated by randomly swapping mouse labels of control vs Cre (p=.002). I) Model predictions. Probability of licking more than once in the previous reward anticipatory zone is plotted as a function of trial (black-control, red-Cre). Shaded region shows the best fit model parameters ± estimated standard deviation of the parameters. See also Figure S7.

Figure 6.

CA1 Stx3 contributes to experience dependent place field shifts in novel environments. A) Left - Example co-recorded place cells form a control mouse (Ctrl-2) during novel arm trials on day 1. Cells display an abrupt backwards shift of their place fields after the field first appears. Green dashed line indicates location of peak activity on the place field induction lap. Middle - Cross-correlation of population vector activity quantifies population backward shift. For illustration, we show the across animal average population vector (PV) spatial cross-correlation between the first novel trial and the next 4 novel trials on day 1. Inset scatterplot (above x-axis, underneath cross-correlation plot) shows the center of mass (COM) of the cross-correlation (“PV COM shift”) for each trial for both control (greyscale colormap) and Cre (red colormap). Control PV shifts are to the left of the Cre PV shifts indicating a larger magnitude backward shift for control trials. Right - Across animal average PV COM shift on day 1 for each pair of initial novel trials. B) Same as (A) for data from Cre animals (example cells Cre-4). Note that smaller abrupt shifts are visible in example cells. C) Average initial novel trial PV shift for each mouse on each day. Dots are the average PV shift for all pairs of initial novel trials on each day for each mouse. Shaded bars are across animal means. [N=9 control mice, 7 Cre mice, 5 days (day 6 excluded). Mixed effects ANOVA: virus main effect F(1,14)=6.41 p=0.024, day main effect F(4,56)=6.09 p=3.85×10⁻⁴, interaction F(4,56)=2.95 p=0.028. Posthoc unpaired t-test: control vs Cre for day 1-t=−4.63 p=0.002 (Holm-corrected), all other days p>0.05. Posthoc paired t-test (Holm-corrected p-value) for PV shift different from 0: control: day 1 t=−6.66 p=.001, all other days p>.05; Cre: all days p>.05]. D) Individual place fields shift abruptly following their formation lap in novel trials from Day 1. Control animal place fields shift to a greater extent. The COM of place cell activity on each trial, relative to the average place field COM, following the place field’s formation lap. Shaded regions indicate across place field mean +/− SEM (gray-control, red-Cre). After controlling for formation lap running speed through the place field, COM shifts were significantly greater on day 1 for control animals (Linear mixed effects regression, Day 1 - χ² =6.37, p=.016). E) The place field width is correlated with running speed through the place field on the formation lap in both groups. Heatmaps show the joint histogram of place field widths and place field formation lap speed through the place field (Left - Control, Right - Cre). Blue line shows the best fit regression for each condition from mixed effects linear regression. Neither the intercept nor the slope of the regression differs between Cre and control (Linear mixed effects regression, Wald test p>.05) See also Figure S8.

Figure 7.

CA1 Stx3 is necessary for offline consolidation of place codes. A) Left - Example highly synchronous event (HSE) from a novel trial from a control animal (Ctrl-1) on day 3. Top - Activity rate of cells that participate in the HSE over time on a single trial. Middle - Running speed. Orange dot indicates the time of reward delivery. Bottom - Number of simultaneously active neurons in the population during the post-reward stopping period. Orange line indicates threshold for defining a HSE from shuffled data. Green shading indicates post-reward stopping periods eligible for HSE analysis. Right - Example novel arm place cell that is reactivated during a HSE for the first time on day 3 from a control animal (Ctrl-1). The cyan arrow highlights the trial of the first HSE in which the cell participates. Note that the place field appears more stable and reliable on subsequent days. B) Same as (A) for a Cre animal (Cre-1). Right - Note that the place cell destabilizes on subsequent days. C-E) For control but not Cre animals, cells that are reactivated during HSEs are more stable across days than cells that are not reactivated during HSEs. This effect is greater in novel trials. C) Difference in across day spatial rate map correlation between reactivated and non-reactivated cells for control animals in familiar and novel trials. Data shown as across animal average. D) Same as (C) for Cre animals. E) Average difference in across day correlation between reactivated and non-reactivated cells. Each point is the average difference in correlation for each mouse (dots - familiar trials, “x” - novel trials, black - control, red - Cre). Reactivated cells in control animals have higher across day correlation than non-reactivated cells for both familiar (one-sample t-test: t=4.67 p=.005, all p-values are Holm-corrected for multiple comparisons) and novel trials (one-sample t-test: t=5.23 p=.003). In Cre animals, reactivated cells have higher across day correlation for novel trials (one-sample t-test: t=3.23 p=.036) but not for familiar trials (one-sample t-test: t=−.367 p=.726). The increase in across day correlation for reactivated cells is greater for control animals than Cre animals. This increase in stability of the neural representation is larger for novel trials than familiar trials [N=9 control mice, 7 Cre mice. Mixed effects ANOVA: virus main effect F(1,14)=18.1 p=7.94×10⁻⁴, familiar vs novel main effect F(1,14)=20.7 p=4.54×10⁻⁴, interaction F(1,14)=4.91 p=0.044]. Furthermore, the effect of novelty is greater for control animals [Novelty effect control vs Cre: unpaired t-test t=2.22 p=.044]. F-H) For control but not Cre animals, cells that are reactivated during HSEs have higher trial by trial correlation than cells that are not reactivated. This effect is greater in novel trials. F) Difference in trial-by-trial spatial rate map correlation between reactivated and non-reactivated cells for control animals in familiar and novel trials. Data shown as across animal mean. G) Same as (J) for Cre animals. H) Average difference in trial-by-trial correlation between reactivated and non-reactivated cells. Each point is the average difference in correlation for each mouse (dots - familiar trials, “x”- novel trials, black - control, red - Cre). Reactivated cells in control animals have higher trial x trial correlation than non-reactivated cells for both familiar (one-sample t-test: t=4.46 p=.006, all p-values Holm-corrected for multiple comparisons) and novel trials (one-sample t-test: t=5.44 p=6.12×10⁻⁵). In Cre animals, reactivated cells do not have higher trial x trial correlation than non-reactivated cells for either trial type (Holm-corrected p>.05). The increase in trial-by-trial correlation for reactivated cells is greater for control animals than Cre animals. This increase in stability of the neural representation is larger for novel trials than familiar trials [N=9 control mice, 7 Cre mice. Mixed effects ANOVA: virus main effect F(1,14)=25.9 p=1.66×10⁻⁴, familiar vs novel main effect F(1,14)=20.5 p=4.76×10⁻⁴, interaction F(1,14)=4.47 p=0.053]. See also Figure S9.