A spike-timing-dependent plasticity rule for dendritic spines

The structural organization of excitatory inputs supporting spike-timing-dependent plasticity (STDP) remains unknown. We performed a spine STDP protocol using two-photon (2P) glutamate uncaging (pre) paired with postsynaptic spikes (post) in layer 5 pyramidal neurons from juvenile mice. Here we report that pre-post pairings that trigger timing-dependent LTP (t-LTP) produce shrinkage of the activated spine neck and increase in synaptic strength; and post-pre pairings that trigger timing-dependent LTD (t-LTD) decrease synaptic strength without affecting spine shape. Furthermore, the induction of t-LTP with 2P glutamate uncaging in clustered spines (<5 μm apart) enhances LTP through a NMDA receptor-mediated spine calcium accumulation and actin polymerization-dependent neck shrinkage, whereas t-LTD was dependent on NMDA receptors and disrupted by the activation of clustered spines but recovered when separated by >40 μm. These results indicate that synaptic cooperativity disrupts t-LTD and extends the temporal window for the induction of t-LTP, leading to STDP only encompassing LTP.

showing the relationship between uEPSP change (color coded) and neck length change and distance between two clustered spines following a post-pre t-LTD induction protocol. Note that when a pairing protocol of -15 ms is performed in two adjacent spines that display neck shrinkage, the result is potentiation (increase in uEPSP amplitude, more than 100%). On the other hand, when the induction protocol is performed in two spines that are further apart, without neck length changes, the result is depression (decrease in uEPSP amplitude, less than 100%). The change in uEPSP amplitude was modeled using equation 1 (described in methods). Shaded area and error bars represent SEM.

Induction of t-LTP-mediated spine morphological changes
We found that the induction of t-LTD was not accompanied with spine neck or head changes, which is at odds with previous findings suggesting structural changes in spine head volume during the induction of LTP or LTD 1, 2, 3 . The discrepancy between our results and those observed previously after the induction of t-LTP (head enlargement 1, 4 ), LTP 2 , or LTD (head shrinkage 3 ) using glutamate uncaging are likely explained by methodological differences. While our data was obtained using ACSF with physiological concentrations of magnesium and calcium, those from other reports were done in low or a magnesiumfree ACSF 2, 3 , low calcium extracellular solution for the induction of LTD 3 , or in a magnesium-free ACSF and an intracellular solution containing 5 µM actin that was required for the t-LTP-mediated spine head enlargements 1 . In fact, inducing t-LTP with a pre-post timing of +7 ms in slices that were perfused with ACSF containing no magnesium ions resulted in significantly increased uEPSP amplitudes, shrinkage of the activated spine necks, and importantly an increase in spine head size (Supplementary Figure 12).

Molecular mechanisms responsible for t-LTP in dendritic spines
What are the mechanisms responsible for the generation of t-LTP in spines? Why t-LTP induction in single and clustered spines is associated with spine neck shrinkage?
We and others have reported that LTP induction can trigger activity-dependent changes in spine neck length 5, 6 and head size 2, 6, 7 , and that the amplitude of somatically recorded uEPSP is inversely proportional to the spine neck length 5,8,9 . Numerical simulations show that the EPSP amplitude/neck length correlation can be explained by variations in synaptic conductance, electrical attenuation through the neck, or a combination of the two 5 . Nevertheless, these models rely exclusively on the passive electrical attenuation of synaptic inputs through the spine neck assume very high (> 2 GOhm) neck resistance 5 , which is at odds with recent spine neck resistance estimations 10,11 . Furthermore, these simulations suggest that if the neck resistance is low, changes in synaptic conductance mediated by an increase in the number of AMPA receptors could contribute significantly to t-LTP-dependent changes in synaptic strength 5 . Furthermore, AMPA receptor content is one of the major mechanisms underlying LTP (for review see 12 ). Hence, we studied the contribution of AMPA receptors to this phenomenon, and our results showed that GluR1 receptor incorporation into the PSD is required for t-LTP induction in spines.
What is the role of spine neck shrinkage on the incorporation of AMPA receptors into the PSD and ultimately on t-LTP induction in spines?
Experimental and theoretical studies have indicated that lateral diffusion of AMPA receptors into and out of the spine head can be restricted by the spine neck geometry 13,14,15,16 . In particular, lateral diffusion of AMPA receptors into and out of mushroom spines (long-necked spines) has been shown to be significantly slower than that observed in stubby spines (small-necked spines) 13 , which is supported by studies showing reduced diffusion of membrane proteins located in spine necks 17 . In addition, quantitative models using realistic spine morphologies indicate that decreasing the radius and increasing the spine neck length increases the retention of AMPA receptors at the synapse 15 , even when their interaction with scaffolding cytoskeletal proteins is neglected 16 . Actin is highly enriched in the spine neck and head 18 , and plays an important role in anchoring AMPA receptors in the spine 19 and AMPA receptor trafficking 20 , being instrumental for synaptic transmission and plasticity 21,22,23 . Hence, to address the role that t-LTP-induced neck shrinkage has on AMPA receptor lateral trafficking to the PSD, and the generation of t-LTP in the activated spines we study actin dynamics. Our results showed that actin polymerization is required for the t-LTP-dependent shrinkage of the activated spine necks and increase in uEPSP amplitude, suggesting that the induction of t-LTP in spines involves a neck-shrinkage-dependent facilitated diffusion of GluR1 subunits to the spine head where they are incorporated to the PSD.

Micro clusters: a structural and functional modality of synaptic connectivity and plasticity
We found the remarkable result that a micro cluster of just two spines during a STDP protocol alter the calcium dynamics and the induction of t-LTP and t-LTD. In fact, the relevance of synaptic micro clusters on the input/output properties of pyramidal neurons is also supported by three dimensional electron microscopy and neuronal reconstruction studies that have shown the presence of postsynaptic innervation of the same axon spaced at less than 10 µm in the basal dendrites of L2/3 pyramidal neurons from the medial entorhinal cortex 24 , L5 pyramidal neurons from somatosensory cortex 25 and in the distal apical tuft dendrites in stratum lacunosum-moleculare of hippocampal CA1 pyramidal neurons 26 . In addition to having spines innervated by the same axon, it is likely that functional synaptic micro clusters can be gated by the convergence of different axons, which could increase the computational power of cortical circuits through a multi-neuronal control of synaptic cooperativity and ultimately the implemented STDP learning rule. Furthermore, it has been shown that orientation selectivity in visual cortex is correlated with the degree of spatial synaptic clustering of co-tuned synaptic inputs within the dendritic field 27 , and that functional clusters of dendritic spines separated by less than 10 µm share similar spatial receptive field properties, spontaneous and sensory-driven activity 28 . Interestingly, is has been recently shown in the mouse visual cortex that a single axon can contact clustered synapses in the postsynaptic neuron, but that local clustering is not favoured over widespread spacing of synaptic inputs 29 . In addition, recently it has been demonstrated in L2/3 pyramidal neurons from mouse primary visual cortex that synaptic micro clusters reflects the interaction of functionally similar inputs -with similar orientation preference -from different sourcescallosal and non-callosal inputs 30 , providing functional and anatomical data for the presence of multi-neuronal control of synaptic cooperativity in a short dendritic segment. Moreover, it has been shown that coactive neighboring synapses drive the maturation of cluster synapses in CA1 pyramidal neurons 31 , suggesting that micro clusters also play an important role in the development and shaping of network connectivity. Hence, our data and these findings indicate that the presence of synaptic micro clusters in the dendrites of pyramidal neurons affect the STDP learning rule, likely providing an efficient strategy for guiding learning and memory, and cognition. Furthermore, the functional consequences of synaptic micro clusters in the dendrites of pyramidal neurons and their role in plasticity rules could inspire new strategies for the representation of learning and data efficiency in supervised and unsupervised deep learning algorithms.