Synaptic basis of feature selectivity in hippocampal neurons

Abstract

A central question in neuroscience is how synaptic plasticity shapes the feature selectivity of neurons in behaving animals¹. Hippocampal CA1 pyramidal neurons display one of the most striking forms of feature selectivity by forming spatially and contextually selective receptive fields called place fields, which serve as a model for studying the synaptic basis of learning and memory. Various forms of synaptic plasticity have been proposed as cellular substrates for the emergence of place fields. However, despite decades of work, our understanding of how synaptic plasticity underlies place-field formation and memory encoding remains limited, largely due to a shortage of tools and technical challenges associated with the visualization of synaptic plasticity at the single-neuron resolution in awake behaving animals. To address this, we developed an all-optical approach to monitor the spatiotemporal tuning and synaptic weight changes of dendritic spines before and after the induction of a place field in single CA1 pyramidal neurons during spatial navigation. We identified a temporally asymmetric synaptic plasticity kernel resulting from bidirectional modifications of synaptic weights around the induction of a place field. Our work identified compartment-specific differences in the magnitude and temporal expression of synaptic plasticity between basal dendrites and oblique dendrites. Our results provide experimental evidence linking synaptic plasticity to the rapid emergence of spatial selectivity in hippocampal neurons, a critical prerequisite for episodic memory.

Extended Data Fig. 1 |. Spatial tuning of excitatory inputs.

a, Average iGluSnFr ΔF/F across space for all tuned spines sorted by peak location (Top: opsin-expressing (+), n = 484 spines from 15 induction experiments, 12 cells, 8 mice; Bottom: opsin-free (–), n = 370 spines from 16 control experiments, 11 cells, 6 mice). Heat maps for even and odd laps are shown separately. Spines are ordered according to their peak location during all laps (see Fig. 3f). b, Distribution of synaptic iGluSnFr even–odd lap spatial tuning curve correlations (Top: opsin-expressing (+), median Pearson’s correlation coefficient = 0.3923; Bottom: opsin-free (–), median Pearson’s correlation coefficient = 0.3488). Average Pearson’s correlation coefficient value is not different between opsin-expressing and opsin-free cells (Kruskal-Wallis tests, NS = not significant). c, Example fluorescence traces of synaptic glutamate–calcium and dendritic calcium signals during navigation. Glutamate-evoked and non-evoked spine calcium responses are highlighted. Grey traces represent the animal’s position on the virtual track. d, Activity heatmaps for two spines receiving spatially tuned iGluSnFr input and associated spine and dendritic calcium activity profiles. Using spine calcium signals to extract input timing leads to erroneous time measurements because spine calcium signals can be strongly contaminated by global dendritic events (bottom row, grey). In other spines, using either spine glutamate or spine calcium signals leads to the correct measurement of input timing (top row, black).

Extended Data Fig. 2 |. Local and global spine voltage-calcium dynamics.

a, Peri-stimulus time histograms (PSTHs) triggered on the onset of spine voltage events. Average voltage and calcium responses in each compartment are shown. PSTHs were generated by averaging every 150 detected spine-voltage events (with a 50 spine-voltage event overlap) sorted based on peak somatic activity. b, Left: Plots of the area under the curve (AUC) of all spine-voltage events versus AUC of corresponding spine-calcium events (n = 1939 spine events; Spearman r: two-tailed t-test, R = 0.5774, P < 0.0001). Right: Average AUC of all spine voltage events versus corresponding AUC of spine calcium per spine (n = 129 spines from 3 cells; Spearman r: two-tailed t-test, R = 0.6368, P < 0.0001). Spines are colour-coded based on the parent cell they were recorded from (black: cell 1; grey: cell 2; white: cell 3). c, Binned plots of the data shown in panel b left. Far left: Average AUC of all spine-calcium events in each binned spine-voltage category (data binned by quartiles). Top left: Plots of the first quartile (Q1) of AUC spine-voltage events versus AUC of corresponding spine-calcium events (n = 485 spine events; Spearman r: two-tailed t-test, R = 0.1328, P = 0.0034). Top right: Same as top left, except looking at the second quartile (Q2) of spine-voltage events (n = 485 spine events; Spearman r: two-tailed t-test, R = 0.1394, P = 0.0021). Bottom left: Same as top left, except looking at the third quartile (Q3) of spine-voltage events (n = 484 spine events; Spearman r: two-tailed t-test, R = 0.1612, P = 0.0004). Bottom right: Same as top left, except looking at the fourth quartile (Q4) of spine-voltage events (n = 485 spine events; Spearman r: two-tailed t-test, R = 0.2844, P < 0.0001).

Extended Data Fig. 3 |. Similar magnitude and distribution of synaptic plasticity recorded in optogenetically induced place cells and non-induced cells.

a, Top: Average amplitude of isolated spine calcium events across all spines in opsin-expressing (magenta, n = 19 induction experiments, 12 cells, 8 mice) and opsin-free (grey, n = 16 control experiments, 11 cells, 6 mice) cells before place field induction (Kruskal-Wallis tests). Bottom: Average frequency of isolated spine calcium events across all spines in opsin-expressing (magenta) and opsin-free (grey) cells before place field induction (Kruskal-Wallis tests). b, Same as a, except looking at all spine calcium events (local and global). c, Top: Cumulative distribution of potentiation plasticity events observed at tuned and untuned spines of optogenetically induced place cells (magenta) and non-induced cells (grey) (Kolmogorov-Smirnov tests). Bottom: Quantification of the potentiation plasticity events occurring at tuned and untuned spines of optogenetically induced place cells (magenta) and non-induced cells (grey) (Kruskal-Wallis tests). d, Same as c, except looking at depression plasticity events.

Extended Data Fig. 4 |. Place field induction selectively drives larger changes in synaptic weight at tuned spines compared to untuned spines.

a, Average magnitude of plasticity event (regardless of direction, i.e., potentiation or depression) happening at tuned and untuned spines in induced (left, magenta) and non-induced (right, grey) cells (Wilcoxon matched-pairs signed rank two-tailed test: (Induced) n = 13 induction experiments, 11 cells, 7 mice, P = 0.0105; (Control) n = 11 control experiments, 8 cells, 5 mice, P = 0.2061)). Data were thresholded only to include the upper 25^th percentile of all plasticity events occurring at tuned and untuned spines.

Extended Data Fig. 5 |. Temporal profile of experimentally measured synaptic plasticity kernel.

a, Fitted exponentials to rise and decay phases of the plasticity kernel (tau_b = 3.1 seconds; tau_f = 1.0 seconds; see Methods for exponential fitting procedure). b, Temporal profile of average plasticity occurring around the time of optogenetic place field (PF) induction (bin = 0.5 seconds). Graded colour code: Pooled data across all spines, n = 226 spines from 15 induction experiments (12 cells, 8 mice); Only looking at the upper 75^th percentile of all plasticity events occurring at tuned spines; Only looking at the upper 50^th percentile of all plasticity events occurring at tuned spines; Only looking at the upper 25^th percentile of all plasticity events occurring at tuned spines. c, Same as b, except comparing the top 25^th percentile to the bottom 25^th percentile of all plasticity events occurring at tuned spines.

Extended Data Fig. 6 |. Spine-localized glutamate-calcium dynamics.

a, Fluorescence glutamate-calcium responses for three separate spines. Top row of peri-stimulus time histograms (PSTHs) corresponds to a spine undergoing potentiation (pot.). Middle row of PSTHs correspond to a spine undergoing no plasticity (n.p.). Bottom row of PSTHs corresponds to a spine undergoing depression (dep). Base of the arrow is set to the peak PRE amplitude, and the tip of the arrowhead is set to the peak POST amplitude. b, Change in peak calcium-glutamate amplitude PRE-POST induction (Spearman r: n = 480 spines, two-tailed t-test, R = 0.3029, P < 0.0001). c, Peak fluorescence amplitude for spines undergoing potentiation (Wilcoxon matched-pairs signed rank two-tailed test: (jRGECO1a) n = 254 spines, P < 0.0001, PRE mean = 39%ΔF/F, POST mean = 53%ΔF/F; (iGluSnFr) n = 254 spines, P < 0.0001, PRE mean = 14%ΔF/F, POST mean = 19%ΔF/F). d, Peak fluorescence amplitude for spines undergoing depression (Wilcoxon matched-pairs signed rank two-tailed test: (jRGECO1a) n = 226 spines, P < 0.0001, PRE mean = 51%ΔF/F, POST mean = 39%ΔF/F; (iGluSnFr) n = 226 spines, P = 0.0236, PRE mean = 18%ΔF/F, POST mean = 15%ΔF/F).

Extended Data Fig. 7 |. No detectable changes in spine head size between potentiated and depressed spines following place cell induction.

a, Magnified and cropped images of spines undergoing potentiation and depression (see Fig. 3d) but no detectable changes in spine head size. All ROIs were drawn on motion-corrected, maximum-intensity projected dendritic segments using the membrane-bound glutamate sensor, SFVenus.iGluSnFr. A184S, as the reference channel (see Methods). b, Schematic illustrating strategy for removing imaging frames containing activity-dependent fluctuations in the iGluSnFr channel. The remaining imaging frames were averaged to generate the images in c. c, Same as a, except with all activity-dependent imaging frames removed. d, Change in spine-ROI area of potentiated and depressed tuned spines active inside the plasticity kernel (−4.5 to +2.0 seconds, n = 226 spines from 15 induction experiments (12 cells, 8 mice)). Percent change was calculated by measuring the difference in the area of the manually drawn ROI following induction (area after induction minus area before induction) divided by the area before induction (Mann-Whitney two-tailed unpaired t-test: Pot., Δw > 0, n = 102 spines; Dep., Δw ≤ 0, n = 124 spines; P = 0.0803). e, Same as d, except all imaging frames containing activity-dependent changes in the iGluSnFr channel were removed (see Methods), and all ROIs were re-drawn on these static images (Mann-Whitney two-tailed unpaired t-test: Pot., Δw > 0, n = 102 spines; Dep., Δw ≤ 0, n = 124 spines; P = 0.2574). All box plots depict the median (central line) and interquartile range (25th and 75th percentile). Whiskers extend to the min-max data points.

Extended Data Fig. 8 |. Spatial distribution of synaptic plasticity.

a, Fraction of dendrites with spines undergoing significantly correlated changes in synaptic weight (Δw). Branches were categorized as housing spines undergoing significantly correlated potentiation (red; POT.) or depression (blue; DEP.) by comparing the average weight change across all spines on the dendrite with the 95% confidence intervals (CI) generated from 1,000 data shuffles of the recorded weight changes across all spines (see panel b). Branches with spines not undergoing significantly correlated changes in synaptic weight are labelled in grey (no correlation; N.C.). Analysis was restricted to the branches with spines active inside the plasticity kernel (see Fig. 4a, centre). b, Histogram of shuffled average weights for all spines on branches containing spines undergoing significantly correlated depression (left) or potentiation (right). Dashed vertical grey lines indicate the 95% CI. Dashed vertical blue and red lines indicate the true measured mean of synaptic weights on each branch. c, Fraction of spines undergoing strong plasticity (upper 25^th percentile of all $\left|\Delta W\right|$) in dendrites showing significantly correlated (sig.) or non-correlated (Non-Sig.) changes in synaptic weight (two-tailed unpaired t-test: Corr., n = 8 dendrites; Non-Corr., n = 14 dendrites; P < 0.0001). Branches are colour-coded based on categorization in a. d, Spatial dispersion of synaptic plasticity centered on the most strongly potentiated (upper 25^th percentile of all plasticity events occurring at tuned spines inside the plasticity kernel; n = 23 spines) or weakly potentiated (bottom 75^th percentile of all plasticity events occurring at tuned spines; n = 79 spines) spine on the dendritic branch. Shuffles are indicated by white circles. e, Same as d, except looking at spatial dispersion of synaptic plasticity centered on the most strongly (n = 39 spines) or weakly (n = 85 spines) depressed spine on the dendritic branch.

Extended Data Fig. 9 |. Intracellular comparison of spatial tuning properties and plasticity events occurring in basal and oblique spines of induced and non–induced cells.

a, Induction experiments (12 cells, 8 mice). Left: fraction of basal and oblique spines receiving spatially tuned synaptic iGluSnFr input (orange) and spines receiving no tuned input (grey). Basal vs. Oblique, P = 0.0007 (Fisher’s exact test). Right: fraction of tuned basal and oblique spines undergoing potentiation (Δw > 0) and depression (Δw ≤ 0). Basal vs. Oblique, P = 0.9234 (Fisher’s exact test). b, Control experiments (11 cells, 6 mice). Left: fraction of basal and oblique spines receiving spatially tuned synaptic iGluSnFr input (orange) and spines receiving no tuned input (grey). Basal vs. Oblique, P = 0.0010 (Fisher’s exact test). Right: fraction of tuned basal and oblique spines undergoing potentiation (Δw > 0) and depression (Δw ≤ 0). Basal vs. Oblique, P = 0.1947 (Fisher’s exact test). c, Top: Cumulative distribution of potentiation plasticity events observed at tuned basal and oblique spines of optogenetically induced place cells (magenta) and non-induced cells (grey) (Kolmogorov-Smirnov tests). Bottom: Quantification of the potentiation plasticity events (Kruskal-Wallis tests). d–f, Same as c, except looking at depression plasticity events at tuned spines (d), potentiation plasticity events at untuned spines (e), and depression plasticity events at untuned spines (f). All box plots depict the median (central line) and interquartile range (25^th and 75^th percentile). Whiskers extend to the min-max data points.

Extended Data Fig. 10 |. Differences in layer-specific plasticity cannot be explained by a gradient in LED power along the basal–apical axis.

a, Schematic illustrating method for measuring the distance between the soma and the recorded dendritic segment. We hypothesized that if the LED stimulations differentially excite basal and oblique dendrites due to a gradual reduction in power along the basal–apical axis, this would result in a plasticity gradient across all the layers. For all scatterplots, Spearman r and n values are specified in the figure. Data points are fit by linear equation (all spines, orange; basal spines, green; oblique spines, purple). Spines located on the same dendritic branch were assigned the same distance from the soma. Left: Distribution of plasticity events observed at all spines of optogenetically induced place cells (12 cells, 8 mice) as a function of distance from the soma (all spines, P = 0.2207; basal spines, P = 0.4476; oblique spines, P = 0.0002). Centre: Same as the left graph, except only looking at tuned spines (all spines, P = 0.6652; basal spines, P = 0.2787; oblique spines, P = 0.0001). Right: Same as the centre graph, except only looking at tuned spines inside the plasticity kernel (Fig. 4a, left, Fig. 5d) (all spines, P = 0.1713; basal spines, P = 0.2091; oblique spines, P = 0.0005). b, Same as a, except using a different method for measuring the distance between the soma and the recorded dendritic segment. Left: (all spines, P = 0.5158; basal spines, P = 0.0288; oblique spines, P = 0.0001). Centre: (all spines, P = 0.2271; basal spines, P = 0.0074; oblique spines, P = 0.0002). Right: (all spines, P = 0.0512; basal spines, P = 0.0187; oblique spines, P = 0.0017).

Fig. 1 |. In vivo subcellular-resolution imaging of presynaptic glutamate release and postsynaptic spine calcium activity before and after optogenetic PF induction in single CA1 pyramidal neurons.

a, Schematic of the in vivo 2P-guided SCE setup and functional readout of electroporated plasmids. Cells express a depolarizing opsin (bReaChes) to induce PFs and sensors to measure input timing (using the postsynaptic membrane-bound glutamate sensor iGluSnFr) and changes in synaptic weight (using the postsynaptic cytosolic calcium sensor jRGECO1a). Obj., objective. b, Schematic of the combined 2P and VR system. Mice perform a head-restricted spatial navigation task for fixed water rewards. ITI, intertrial interval. c, Glass pipettes containing plasmids and Alexa 488 were used to guide the electroporation of single cells in the pyramidal layer (top left). Cells reached peak expression 48–72 h after SCE (bottom left). Dendrites spanning the entire dendritic tree were accessible (right). s.o., stratum oriens; s.p., stratum pyramidale; s.r., stratum radiatum; s.l.m., stratum lacunosum moleculare. d, Magnification of the areas indicated in c, showing maximum-intensity projections of single-plane time-series imaging of iGluSnFr-expressing basal and oblique dendrites, along with associated fluorescence traces of a subset of spines (yellow circles). The grey traces represent the animal’s position on the virtual track. e, Synaptic activity heat maps for two spines receiving spatially tuned iGluSnFr input. f, Maximum-intensity projection of single-plane time-series imaging of a jRGECO1a-expressing basal dendrite (dend.). Associated fluorescence traces acquired from the dendritic (white outline) and synaptic (yellow circle) regions of interest are shown. The grey traces represent the animal’s position on the virtual track. g, A single pyramidal cell expressing a red-shifted excitatory opsin (bReaChes-mRuby3) and GCaMP7b. Spatially restricted optogenetic stimulations (LED) for five consecutive laps evoke strong somatic responses and induce long-lasting PFs (+) (magenta). These effects are lost in cells electroporated without the opsin plasmid (−) (grey). Scale bars, 75 μm (c (left) and g), 50 μm (c (right)) and 5 μm (d and f).

Fig. 2 |. In vivo simultaneous dual-colour two-photon imaging of membrane potential and calcium dynamics in dendritic spines.

a, Maximum-intensity projection image of an ASAP6.1- and jRGECO1a-expressing CA1 pyramidal neuron (CA1PN) next to a H2B–tdTomato-expressing reference cell used for real-time motion correction (RT MC) (left). Increased magnification of recorded dendritic segments (white box, centre) and 3D orientation of line-scan recordings placed over laterally protruding spines and dendritic shaft (yellow lines, right) are shown. Ref., reference. b, Associated fluorescence traces acquired from dual-colour (990 nm, ASAP6.1; 1,035 nm, jREGOC1a) high-speed line-scan recordings (1.1 kHz per line per wavelength) of single dendritic spines, their associated dendritic shaft and conjugate soma, along with 3D corrected movement artifacts in the x, y and z axes. The black vertical lines denote detected voltage events. c, Example 20 s optical recording of voltage dynamics from a subset of spines and conjugate soma in an awake behaving mouse. The black vertical lines denote global events (that is, events present in all compartments). The magenta arrowheads denote spine-restricted voltage events (that is, events present only at spines and not the soma). d, Peri-stimulus time histograms triggered on the onset of global voltage events (left) or spine-restricted voltage events (right). The average voltage–calcium signals from spines (top row), dendritic shaft (middle row) and soma (bottom row) are shown. Norm., normalized. e, The AUC of spine-restricted voltage events versus the AUC of corresponding spine-restricted calcium events (left). n = 989 spine-restricted events. Spearman r was determined using a two-tailed t-test; R = 0.3257, P < 0.0001. Right, the average AUC of all spine-restricted voltage events versus the corresponding AUC of spine-restricted calcium per spine. n = 108 spines from 3 cells. Spearman r was calculated using a two-tailed t-test; R = 0.3044, P = 0.0014. Spines are colour coded based on the parent cell that they were recorded from (black, cell 1; grey, cell 2; white, cell 3). Scale bars, 60 μm (a (left)) and 5 μm (a (right)).

Fig. 3 |. Optical measurement of synaptic plasticity associated with induction of hippocampal BTSP.

a, Maximum-intensity projection of a jRGECO1a-, iGluSnFr- and bReaChes-expressing cell. Fluorescence traces of synaptic glutamate–calcium (yellow circle) and dendritic calcium signals during navigation are shown. b, Before and after somatic activity heat maps for induced and control cells (left). The red boxes indicate the stimulation (stim.) site. The dashed lines indicate the expected PF location. Middle, optogenetic stimulation increases somatic activity within the LED zone (n = 19 inductions, n = 16 controls; two-tailed unpaired t-test; P < 0.0001). Right, the strong relationship between PF width and velocity (Spearman r: R = 0.7009, P = 0.0008). Datapoints were fit with a linear equation (blue line). c, Activity heat maps of spines receiving spatially tuned iGluSnFr input and a schematic of the extraction of input timing. d, Synaptic calcium signals for a subset of spines undergoing potentiation and depression (yellow circles). e, The fraction of spines (n = 906) receiving spatially tuned synaptic iGluSnFr input to one, two or three locations, and spines receiving no tuned input. f, The average (avg.) iGluSnFr ΔF/F across space for all tuned spines sorted by peak location (n = 484). The fraction of tuned spines versus PF location (bin = 30 cm) is shown. Data are mean ± s.e.m. Excitatory inputs are homogeneously distributed (χ² test of independence, d.f. = 9, P = 0.451). g, The fraction of tuned spines undergoing potentiation or depression per experiment (n = 15 inductions; two-tailed unpaired t-test; P = 0.0998). h, The magnitude of potentiation and depression events at tuned spines ($\Delta W>0$, n = 213; $\Delta W\le 0$, n = 271; two-tailed unpaired t-test; P = 0.0810). The box plots show the median (centre line), interquartile range (box limits, and the whiskers extend to the minimum–maximum datapoints. The grey traces in a and d represent the animal’s position on the virtual track. Scale bars, 50 μm (a (left)) and 5 μm (a (right) and d).

Fig. 4 |. Coordinated bidirectional changes in synaptic weight underlie behavioural timescale plasticity.

a, Temporal profile of the average plasticity after PF induction (bin = 0.5 s; left, induced, n = 226 spines (12 cells); centre, same as induced, except looking at the upper 25th percentile of all plasticity events, n = 62 spines; right, control, n = 224 spines (9 cells)). The blue minus and red plus symbols depict time bins in which depression or potentiation dominates (bootstrapped 95% confidence intervals (CIs)). b, Temporal profile of the fraction of spines undergoing potentiation (pot.) and depression (dep.) after PF induction (top, induced, see a (centre)); bottom, control, see a (right). The blue minus (depression) and red plus (potentiation) symbols indicate the time bins that lie outside the 95% CIs generated from 1,000 data shuffles. c, Spatial profile of synaptic plasticity events (top row, induced; bottom row, control; bin = 30 cm). Left, the distribution of all excitatory inputs (χ² test of independence, d.f. = 9; P = 0.114 (IND.); P = 0.804 (control)). Centre, the distribution of potentiated excitatory inputs. Right, the distribution of depressed excitatory inputs. Data were thresholded to include only the upper 25th percentile of all plasticity events. The up and down arrowheads indicate position bins that lie above or below the 95% CIs generated from 1,000 data shuffles, respectively. d, The fraction of tuned spines undergoing potentiation after PF induction as a function of PF width (top; Spearman r: n = 13 inductions (12 cells), two-tailed t-test, R = 0.6267, P = 0.0244) and animal velocity (bottom; Spearman r: n = 13 inductions (12 cells), two-tailed t-test, R = 0.7675, P = 0.0031). Spines undergoing no plasticity (bottom 25th percentile of all plasticity events) were removed from the analysis. Data are fitted with a linear equation (blue line).

Fig. 5 |. Compartment-specific expression of synaptic plasticity.

a, The fraction of tuned basal and oblique spines undergoing potentiation $\left(\Delta W>0\right)$ and depression $\left(\Delta W\le 0\right)$ after PF induction (–4.5 s to +2.0 s around LED onset; see Fig. 4a (left)). Statistical analysis was performed using a Fisher’s exact test; P = 0.1281 (basal versus oblique). b, The cumulative distribution of initial weights of tuned basal (n = 142, 9 cells, green) and oblique (n = 84, 6 cells, purple) spines. Statistical analysis was performed using Kolmogorov–Smirnov tests. Insets: the average initial weight (a.u.) of tuned spines. Data are mean ± s.e.m. Statistical analysis was performed using a two-tailed unpaired t-test; P < 0.0001. c, The cumulative distribution of plasticity events observed at tuned basal (n = 142, green) and oblique (n = 84, purple) spines. Statistical analysis was performed using Kolmogorov–Smirnov tests. Insets: the average $\Delta W$ (a.u.) of tuned spines. Data are mean ± s.e.m. Statistical analysis was performed using a two-tailed unpaired t-test; P = 0.0476. d, Temporal profile of plasticity after PF induction (bin = 0.5 s; top, basal, n = 142; bottom, oblique, n = 84). The blue minus and red plus symbols depict the time bins in which depression or potentiation dominates (bootstrapped 95% CI). e, Temporal profile of the fraction of spines undergoing potentiation and depression after PF induction (bin = 0.5 s; top, basal, n = 142; bottom, oblique, n = 84). The blue minus (depression) and red plus (potentiation) symbols indicate the time bins that lie outside the 95% CI generated from 1,000 data shuffles. f, Spatial profile of synaptic plasticity events (top row, basal spines; bottom row, oblique spines; bin = 30 cm). Left, the distribution of potentiated excitatory inputs. Right, the distribution of depressed excitatory inputs. Data were thresholded to include only the upper 25th percentile of all plasticity events occurring at tuned spines. The up and down arrowheads indicate position bins that lie above or below the 95% CI generated from 1,000 data shuffles.