Location-dependent synaptic plasticity rules by dendritic spine cooperativity

Nonlinear interactions between coactive synapses enable neurons to discriminate between spatiotemporal patterns of inputs. Using patterned postsynaptic stimulation by two-photon glutamate uncaging, here we investigate the sensitivity of synaptic Ca2+ signalling and long-term plasticity in individual spines to coincident activity of nearby synapses. We find a proximodistally increasing gradient of nonlinear NMDA receptor (NMDAR)-mediated amplification of spine Ca2+ signals by a few neighbouring coactive synapses along individual perisomatic dendrites. This synaptic cooperativity does not require dendritic spikes, but is correlated with dendritic Na+ spike propagation strength. Furthermore, we show that repetitive synchronous subthreshold activation of small spine clusters produces input specific, NMDAR-dependent cooperative long-term potentiation at distal but not proximal dendritic locations. The sensitive synaptic cooperativity at distal dendritic compartments shown here may promote the formation of functional synaptic clusters, which in turn can facilitate active dendritic processing and storage of information encoded in spatiotemporal synaptic activity patterns.


Supplementary
. Cooperative spine Ca 2+ signaling is not due to extracellular glutamate diffusion or release from intracellular Ca 2+ stores.
(a-c) Spatial precision of NMDAR signaling by 2P glutamate uncaging. Experiments were performed in 0.1 mM Mg 2+ and 1 µM TTX. a) Single focal plane image of a distal spine with color coded uncaging locations at various distances from the spine head. (b) GluEPSPs and spine Ca 2+ signals evoked by uncaging at the locations indicated in a. (c) Summary of spine Ca 2+ signals as a function of the distance of the uncaging spot center (n=8 spines in 2 cells). Zero µm distance corresponds to uncaging spot center at the border of the spine head. (d-f) Extracellular glutamate accumulation cannot explain nonlinear spine Ca 2+ signaling in distal dendrites. Experiments were performed in 1 mM Mg 2+ and 1 µM TTX. (d) Single focal plane image of a distal dendrite. Ca 2+ signals in s1 (red uncaging spot) were measured alone or either with synchronous coactivation of three neighbor spines (yellow spots) or with uncaging at laterally placed uncaging spots at similar distances but >2 µm away from spines (mock uncaging, blue spots). Glutamate diffusion to s1 is expected to be similar in the two scenarios. (e) Cooperative spine Ca 2+ signaling was detected in s1 only when uncaging onto neighbor spines, but not with mock uncaging in the experiment shown in d. (f) Summary data (n=5 experiments in 3 cells, p<0.05, Wilcoxon test). Inset: total distance of all coactive uncaging spots were similar between the 4-spine and 1-spine+3mock conditions (p=0.500, Wilcoxon test). (g) Left, calculated (black) and measured (red) spine Ca 2+ signals from representative experiments using the 4S protocol (averaged data from the four spines) at distal dendritic segments under control conditions (in the presence of the vehicle DMSO, 0.03%) and after >20 min treatment with the smooth endoplasmic reticulum Ca 2+ pump inhibitor CPA (30 µM), which depletes intracellular Ca 2+ stores 5 . Right, summary of cooperative spine Ca 2+ nonlinearity (n=6/7 dendrites in 3/3 control and CPA-treated cells, p=0.353, Mann-Whitney test).

Supplementary Figure 4. Cooperative spine Ca 2+ signaling does not depend on the type of Ca 2+ indicator or the presence of TTX.
(a) 4S protocol (as in Fig. 2a-b) on a set of four spines in a proximal (left) and a distal (right) dendritic segment using the Ca 2+ indicator Fluo-5F (300 µM). Changes in green Fluo-5F fluorescence were normalized to the red signal of Alexa Fluor 594 (10 µM; see Methods). (b) Comparison of cooperative spine Ca 2+ signaling (left) and EPSP nonlinearity (right) measurements using OGB-1 (open black circles) and Fluo-5F (red filled circles). TTX was present in the ACSF in both experiments. Each circle represents an individual spine set (results of all four spines averaged). To directly compare results with the two dyes we used the ratio of the measured and the calculated Ca 2+ signals. Data with OGB-1 are the same as (expressed as F/F difference) in Fig. 2a. Note that some data points on the graph overlap. No difference was found between Ca 2+ nonlinearity (two-way ANOVA: p=0.436 for dye type, p<0.001 for location, p=0.500 for interaction) or EPSP nonlinearity (two-way ANOVA: p=0.935 for dye type, p=0.039 for location, p=0.669 for interaction).
(c) 4S protocol on a set of four spines in a distal dendritic segment in the absence of TTX, measured using OGB-1. Right panel shows the rising phase of the EPSP, expanded from the dashed box. Note that no spikelet (as a sign of dendritic Na + spike) is evident on the rising phase (compare with Fig. 6b). (d) Cooperative spine Ca 2+ nonlinearity (left) and somatic EPSP nonlinearity (right) evoked by four coactivated spines at different relative locations along individual branches in the absence of TTX. Open circles represent individual spine sets (results of all four spines averaged). Filled symbols and error bars represent grouped data for proximal (relative location, RL<0.33, light blue, n=8), middle (RL: 0.33-0.67, light green, n=8) and distal (RL>0.67, orange, n=10) locations. Group data obtained in the presence of TTX (from Fig. 2ab) are also shown for comparison (proximal: dark blue, middle: green, distal: red). No difference was found between Ca 2+ nonlinearity (two-way ANOVA: p=0.653 for TTX treatment, p<0.001 for location, p=0.657 for interaction) whereas a slightly higher EPSP nonlinearity was observed without TTX (two-way ANOVA: p=0.045 for TTX treatment, p=0.658 for location, p=0.132 for interaction).

Supplementary Figure 5. Computational modelling of clustered input integration
(a) Image of the perisomatic dendrites of the reconstructed and modelled neuron 6 , with the proximal (blue, 15% of branch length) and distal (red, 75% of branch length) stimulation sites on a branch indicated. Distal apical dendrites are not shown. (b) Ca 2+ concentration change in one stimulated spine in response to the stimulation of 4 spines heads in proximal (15% of branch, blue) and distal (75% of branch, red) clusters, and the expected response based on linear summation of the responses to separate stimulations (black). Parameters were: 2 μm inter-spine distance, 0.3 ms delay. (c) Somatic voltage response amplitudes for 1-4 clustered input stimuli from proximal, middle and distal stimulation sites.  (compare with b). For the simulations we used linear gradients and kept the parameters for synapses at the middle of the branch constant (hence there is no change in the integration, when the middle of the branch is stimulated, green). The gradient is quantified by the conductance ratio, that is, the ratio between the maximal conductance of the proximal and the distal synapses. The conductance ratio 1.5 between synapses at the 15% and 75% distance from the branch point is equivalent to a 2-fold reduction in the synaptic strength from the branch point to the tip, suggested by Katz et al. 9 . We applied this gradient to the maximal conductance of the AMPA and NMDA synapses as well as to the diameter of the dendritic spine heads along the branch. (e) Ca 2+ nonlinearity factor (measured response/expected response) measured for 4 inputs (as in panel b) as the function of the conductance ratio between proximal and distal synapses. The gradients consistent with the data of Katz et al. 9 are shown with grey shading. To achieve similar spine Ca 2+ nonlinearity in proximal versus distal synapses a conductance ratio of ~7 would be required, i.e., distal synapses should be 7-times weaker than proximal synapses. Similar results were obtained from two other apical oblique dendrites in the modeled neuron.
Computational methods: Simulations were carried out using the NEURON simulation environment 10 , using a morphologically detailed reconstruction of a CA1 pyramidal neuron 6 . The model included a membrane capacitance of 1 µF/cm 2 , axial resistivity of 150 Ωcm and membrane resistivity of 20,000 Ωcm 2 . Dendritic spines were modelled using a spine neck length of 1.58 µm, neck diameter of 0.077 µm, while the diameter of the spherical spine head was 0.5 µm 8 . To reproduce somatic responses to proximal and distal stimulations in the presence of the NMDA receptor blocker AP5 (data not shown), we modelled AMPA synapses as a double-exponential conductance function with rise time 0.1 ms, decay time 15 ms, reversal potential 0 mV and maximal conductance of 0.1 nS. After fixing the parameters of the AMPA receptors, we fitted the parameters of the NMDA currents to somatic responses without AP5 (not shown). The NMDA currents were modelled with double exponential kinetics using rise time 1 ms and decay time 50 ms, maximal conductance of 0.6 nS and a reversal potential 0 mV. The maximal conductances of the AMPA and NMDA receptors were set to achieve a ~ 0.2 mV depolarization at the soma in response to the stimulation of a single synapse. The voltage dependence of the NMDA receptor was modelled by multiplying the maximal conductance with a voltage dependent factor 11 : g(V) = 1/(1 + exp(−0.08 V [Mg 2+ ] / 5)) where the membrane potential V is measured in mV, and [Mg 2+ ] = 1 mM. The flow of Ca 2+ ions through the NMDA receptors was modelled as a separate process governed by the reversal potential of the Ca 2+ ions and the Ca 2+ permeability of the NMDA receptors, which was set to 0.03. This was the only source of intracellular Ca 2+ in our model. We modelled the decay of the intracellular Ca 2+ after Graham et al. (2014) 12 which included an instantaneous buffer capacity of b=17. We set the time constant of the Ca 2+ extrusion to 14 ms and used 0.1 µM baseline intracellular, and 2 mM extracellular Ca 2+ concentration.