ABSTRACT
Dendritic spines can undergo structural remodeling, and are the preferential site for the induction of long-term potentiation (LTP) and long-term depression (LTD). In a variant of LTP and LTD, known as spike-timing dependent plasticity (STDP), the sign and magnitude of the change in synaptic strength depends on the timing between the spikes of two connected neurons. Although STDP has been extensively studied in cortical pyramidal neurons, the precise structural organization of excitatory inputs that supports STDP, as well as the structural, molecular and functional properties of dendritic spines during STDP remain unknown. Here we developed a spine STDP protocol, in which two-photon glutamate uncaging over single or multiple spines from the basal dendrites of layer 5 pyramidal neurons, which mimics presynaptic release of glutamate (pre), was paired with somatically generated postsynaptic spikes (post). We found that the induction of STDP in single spines follows a classical Hebbian STDP rule, where pre-post pairings at timings that trigger LTP (t-LTP) produce shrinkage of the activated spine neck and a concomitant increase in its synaptic strength; and post-pre pairings that trigger LTD (t-LTD) decrease synaptic strength without affecting the activated spine shape. Furthermore, we tested whether the single spine-Hebbian STDP rule could be affected by the activation of neighboring (clustered) or distant (distributed) spines. Our results show that the induction of t-LTP in two clustered spines (< 5 μm apart) enhances LTP via a mechanism that is accompanied by local spine calcium increases that accumulates during the induction protocol, and that requires actin polymerization-dependent neck shrinkage, which permits AMPA receptor transport to the spine head and insertion into the postsynaptic density (PSD). Moreover, the induction of t-LTD is disrupted when two clustered spines are activated, with no calcium accumulation in spines or dendrites, but can be recovered if the activated spines are separated by > 40 μm. These results indicate that the induction of STDP in single, or distributed spines, follow a Hebbian STDP rule. Interestingly, synaptic cooperativity, induced by the co-activation of only two clustered spines and the local spatio-temporal summation of clustered synaptic inputs, provides local dendritic depolarization and local calcium signals that are sufficient to disrupt t-LTD and extend the temporal window for the induction of t-LTP, leading to STDP only encompassing LTP.
Dendritic spines, the main recipient of excitatory information in the brain 1, are tiny protrusions with a small head (∼1 μm in diameter and <1 fL volume) separated from the dendrite by a slender neck. Spines can undergo structural remodeling that is tightly coupled with synaptic function 1-4, and are the preferential site for the induction of long-term potentiation (LTP) 4-6 and long-term depression (LTD) 7, thought to be the underlying mechanisms for learning and memory in the brain 8. A variation of LTP and LTD has been described in pyramidal neurons that involves the pairing of pre-and postsynaptic action potentials, known as spike-timing dependent plasticity (STDP) 9,10. In this process, the timing between pre-and postsynaptic action potentials modulates synaptic strength, triggering LTP or LTD 10. The sign and magnitude of the change in synaptic strength depends on the relative timing between spikes of two connected neurons (the pre-and postsynaptic neuron 11). The STDP learning rules and their dependency on postsynaptic dendritic depolarization 12, 13, firing rate 12, and somatic distance of excitatory inputs 13-15 have been extracted from studies using connected neuronal pairs or by using extracellular stimulating electrodes, but the precise location and structural organization of excitatory inputs capable of supporting STDP at its minimal functional unit – the dendritic spine – are unknown.
Activity-dependent spine morphological changes (spine head 4, neck 2, or both 16) have been correlated with changes in synaptic strength in cortical pyramidal neurons by mechanisms involving biochemical and electrical spine changes 1, 6. Thus, here we asked what patterns of activity and structural organization of excitatory synaptic inputs support the generation of t-LTP and t-LTD, and which morphological, biophysical and molecular changes observed in dendritic spines can account for the induction of t-LTP and t-LTD?
To induce synapse-specific STDP we developed a protocol whereby two-photon (2P) uncaging of a caged glutamate (MNI-glutamate 3) at a single spine – to mimic synaptic release – is preceded or followed in time (STDP timing window 10) by a backpropagating action potential (bAP) to trigger t-LTP (Figure 1A, pre-post) or t-LTD (Figure 2A, post-pre), respectively. Postsynaptic spikes were triggered by current injection via a whole-cell recording pipette. Two-photon uncaging of a caged glutamate at a single spine triggered excitatory postsynaptic potentials (uncaging(u)EPSP) that were recorded in the soma of layer 5 (L5) pyramidal neurons before and after the induction of STDP, while the morphology of the activated spine neck and head was monitored (Figure 1A and 2A). To induce synapse-specific STDP and monitor calcium levels in the activated spines and parent dendrites we developed a protocol during which we perform nearly simultaneous 2P uncaging of glutamate and 2P calcium imaging of the activated spines and nearby dendrites.
Here, we provide evidence showing that the induction of STDP in single or distributed spines follows a bidirectional Hebbian STDP rule. Furthermore, we show that synaptic cooperativity, induced by the co-activation of only two clustered spines, disrupt t-LTD (< 40 µm distance between spines) and extend the temporal window for the induction of t-LTP (< 5 µm distance between spines) via the generation of differential local calcium signals leading to an STDP rule for clustered inputs only embracing LTP.
RESULTS
Induction of t-LTP in single dendritic spines
To induce t-LTP, we used a repetitive spike-timing protocol (40 times, 0.5 Hz) in which 2P uncaging of glutamate at a single spine was closely followed in time (+7 or +13 ms later, see Methods section) by a bAP (Figure 1A). We evaluated spine morphology and uEPSP amplitude for 40 min following STDP induction to establish the time course of STDP at individual synapses (Figure 1C.1 and D.1). In addition, the maximum uEPSP change and concomitant changes in spine morphology observed in each experiment are shown (Figure 1C.2 and D.2).
A repetitive pre-post pairing protocol of +13 ms reliably induced t-LTP (significant increase in the uEPSP amplitude over time, P < 0.001, n = 7 experiments, from 6 neurons, from 6 mice, Figure 1B.1 and C.1), and shortening of the activated spine neck within a few minutes (P < 0.001), with no significant change in spine head volume (n = 7, Figure 1B.2 and C.1). These results were also consistent when we considered the maximum uEPSP change in amplitude from each experiment and concomitant changes in spine morphology (uEPSP = 134.21 ± 3.29%, P < 0.001, n = 7; neck length = 71.88 ± 10.66%, P < 0.05, n = 7; spine head volume = 98.11 ± 7.34%, P = 0.81, n = 7) (Figure 1C.2). We obtained similar results when we instead considered the average of all the values obtained following t-LTP induction for uEPSP amplitude, neck length and head volume (Supplementary Figure 1). This effect was specific to the activated spine (Figure 1B.2), with neighbouring spines having no appreciable changes in their neck length or head volume (neck length = 98.09 ± 5.06%, P = 0.71, n = 13; head volume = 103.01 ± 3.61%, P = 0.42, n = 14). Control experiments showed that there was no significant change in uEPSP amplitude or spine morphology following the STDP protocol when either action potentials or synaptic stimulation were applied in isolation, as well as when we monitored the long-term stability of these parameters without any STDP protocol (Supplementary Figure 2).
A pre-post pairing of + 7 ms showed a non-significant tendency for the induction of LTP following t-LTP induction, and a non-significant tendency for the shrinkage of the activated spine neck (n = 8 experiments, from 8 neurons, from 8 mice, Figure 1D.1). When the maximum change in uEPSP amplitude from each experiment and the concomitant changes in spine morphology were analyzed, we saw no significant changes in uEPSP amplitude and spine morphology (Figure 1D.2, uEPSP = 112.27 ± 14.19%, P = 0.42, n = 8; neck length = 88.90 ± 5.89%, P = 0.10, n = 8; head volume = 97.81 ± 4.54%, P = 0.64, n = 8) (Figure 2E). Similar results were observed when we instead considered the average of all the values obtained following t-LTP (+7 ms) induction for uEPSP amplitude, neck length and head volume (Supplementary Figure 1). Because voltage has been shown to be an important factor in the induction of t-LTP and t-LTD, we verified that the initial uEPSP amplitudes were not significantly different for pre-post pairing protocols of +13 ms versus +7 ms (uEPSP: 0.62 ± 0.14 versus 0.53 ± 0.16 mV, P = 0.68; Supplementary Figure 3).
These results indicate that there is a preferred pre-post t-LTP pairing time-window (+ 13 ms) at which activated spines in basal dendrites from L5 pyramidal neurons undergo a significant increase in synaptic strength, and a concomitant neck shrinkage (Figure 2E).
Induction of t-LTD in single dendritic spines
We then studied t-LTD in single spines by using a repetitive spike-timing protocol (40 times, 0.5 Hz) in which 2P uncaging of glutamate at a single spine was preceded in time (−15 or -23 ms) by a bAP (post-pre protocol, Figure 2A). When postsynaptic spikes preceded presynaptic firing by 15 ms (i.e., -15 ms), a significant reduction of the uEPSP amplitude occurred within a few minutes following induction of t-LTD (n= 6 experiments, from 5 neurons and 5 mice, Figure 2B.1 and C.1, P < 0.001), with no significant changes in spine neck length or head dimension (Figure 2B.2 and C.1). Furthermore, when the maximum change in uEPSP amplitude after the induction of t-LTD in single spines at pairings of -15 ms was analyzed (Figure 2C.2) we also observed a significant depression of uEPSP amplitude (uEPSP = 71.52 ± 7.07%, P < 0.01, n = 6), with no significant changes in spine morphology (Figure 2B.2 and C.2, neck length = 105.54 ± 9.85%, P = 0.62, n = 6; head volume = 103.25 ± 3.02%, P = 0.33, n = 6). Interestingly, after the induction of t-LTD in single spines when postsynaptic spikes preceded presynaptic firing by 23 ms (i.e., -23ms) there were no significant changes in the amplitude of the uEPSPs or in the spine neck length and head dimensions for the duration of the recordings (n= 7 experiments, from 7 neurons and 7 mice, Figure 2D.1). These results were also consistent with analyses of the maximal uEPSP change in each experiment and concomitant spine morphology (Figure 2D.2: uEPSP = 82.09 ± 9.89%, P = 0.12, n = 7; neck length = 84.15 ± 7.73%, P = 0.09, n = 7; head volume = 98.81 ± 5.57%, P = 0.84, n = 7). There was no significant difference between the initial EPSP amplitude for post-pre pairing protocols of -15 ms versus -23 ms (EPSP: 0.59 ± 0.07 versus 0.49 ± 0.08 mV, P = 0.42; Supplementary Figure 3). We obtained similar results when we instead considered the average of all the values obtained following t-LTD induction in single spines for uEPSP amplitude, neck length and head volume (Supplementary Figure 1). This indicates that only a post-pre t-LTD pairing time-window of -15 ms can effectively induce LTD in single dendritic spines in the basal dendrites from L5 pyramidal neurons.
Taken together these results show that the induction of t-LTP and t-LTD in single spines follows a Hebbian-STDP learning rule that is bidirectional, and favors presynaptic inputs that precede postsynaptic spikes and depresses presynaptic inputs that are uncorrelated with postsynaptic spikes at a very precise and narrow temporal window (+13 ms for the generation of t-LTP and -15 ms for t-LTD, Figure 2E-F). The single spine STDP rule we observed has a narrower post-pre LTD induction pairing time window than previously observed in connected pairs of L2/3 14, 17 and L5 pyramidal neurons 12 – where the presynaptic control of t-LTD via an mGluR and/or cannabinoid type 1 receptor-dependent mechanism 18-21 could plausibly account for these differences.
Induction of t-LTP in clustered dendritic spines
It has been suggested that STDP not only depends on spike timing and firing rate but also on synaptic cooperativity and the amount of voltage generated at the postsynaptic site 12, 13. However, a direct demonstration of synaptic cooperativity at the level of single spines in the dendrites of pyramidal neurons remains unknown. Hence, an experiment was designed to directly test if synaptic cooperativity, marked by the co-induction of t-LTP in clustered dendritic spines from basal dendrites of L5 pyramidal neurons, and the local spatio-temporal summation of inputs, can generate a local dendritic depolarization and local calcium signals, that are high enough to disrupt the single spine STDP learning rule described in Figure 2E. To test this, a two spine STDP protocol (forty 2P uncaging pulses, pulse duration 2ms, 0.5Hz, see methods section) was performed, whereby 2P uncaging of caged glutamate in clustered (distance between spines < 5 µm) spines was followed in time by a bAP to trigger t-LTP (Figure 3A). With this protocol, we investigated whether activating clustered spines extended the pre-post timing window capable of generating LTP by increasing the degree of depolarization immediately before the postsynaptic spike at timings where plasticity was not reliably generated. Specifically, we induced t-LTP in two clustered spines at pre-post timings of +7 ms, and surprisingly found that this protocol was in fact capable of effectively and significantly generating increases in uEPSP amplitude (Figure 3B.1) and the concomitant shrinkage of the activated spine neck, with no apparent changes in its spine head size (Figure 3B.2). Pooled data showed that significant increases in uEPSP amplitude and shrinkage of the spine neck of the activated spines occurs only a few minutes post t-LTP induction and lasted for the duration of the recordings (Figure 3C.1, P < 0.01, n = 10 spines from 8 experiments, 6 neurons and 6 mice), with no significant changes in the spine head size (Figure 3C.1, n = 16 spines from 8 experiments, 6 neurons, 6 mice). Similar results were observed when we analyzed the maximal change in uEPSP amplitude in each experiment and the concomitant spine neck length and head size of the two clustered spines after induction of t-LTP at pairings of + 7 ms (Figure 3C.2; uEPSP = 130.86 ± 8.18%, P < 0.01, n = 8; neck length = 73.22 ± 5.84%, P < 0.01, n = 10; head volume =102.00 ±2.58%, P = 0.45, n = 16). We obtained similar results when we instead considered the average of all the values obtained following t-LTP induction in two clustered spines for uEPSP amplitude, neck length and head volume (Supplementary Figure 1). In control experiments, no significant change in uEPSP amplitude or spine morphology was observed when we monitored the long-term stability of these parameters without any STDP protocol (Supplementary Figure 2). These results indicate that synaptic cooperativity – shown by the induction of t-LTP in only two clustered spines (< 5 µm apart) – is sufficient to significantly trigger synaptic potentiation and shrinkage of the activated spine necks at a pre-post timing that is otherwise ineffective at generating significant morphological changes and synaptic potentiation when only one spine is being activated (for comparison between one versus two cluster spines see Supplementary Figure 4). Hence, the synaptic cooperativity of only two neighbouring synaptic inputs onto spines (< 5 µm apart) in the basal dendrites of L5 pyramidal neurons extends the pre-post timing window that can trigger potentiation (Figure 3C.2, and compare Figure 3C.1 with Figure 1D.1).
Molecular mechanisms responsible for the generation of t-LTP in dendritic spines
The results led us to consider the possible mechanisms underlying the generation of t-LTP at the level of single spines. Specifically, we asked why the induction of t-LTP in single and clustered spines is associated with the shrinkage of the activated spine neck. We and others have reported that the induction of LTP can trigger activity dependent changes in neck length 2, 16 and spine head size 4, 6, 16, and that the amplitude of uEPSP recorded at the cell soma is inversely proportional to the length of the activated spine neck 2, 22, 23. However the mechanisms by which the t-LTP-induced neck shrinkage is associated with synaptic plasticity remains unknown. 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 2. Nevertheless, solutions that rely exclusively on the passive electrical attenuation of synaptic inputs through the spine neck assume very high (> 2 GOhm) neck resistance 2, which are at odds with recent spine neck resistance estimations 24, 25. These results suggest that the control of AMPA receptor content in spines could contribute significantly to the observed t-LTP-dependent changes in synaptic strength. To experimentally study the contribution of AMPA receptors to these phenomena, we performed t-LTP experiments in two clustered spines from L5 pyramidal neurons recorded via patch pipettes loaded with intracellular solution containing 200 µM PEP1-TGL – a peptide that inhibits AMPA receptor incorporation to the postsynaptic density (PSD) by blocking GluR1 C-terminus interaction with PDZ domains at the PSD 26 (Figure 4A). PEP1-TGL incubation by itself did not trigger a run-down of uEPSP amplitude or changes in spine morphology over time (Supplementary Figure 5A). Pooled data from experiments where a repetitive pre-post pairing protocol of + 7 ms was used to activate clustered spines in the presence of PEP1-TGL show that the peptide completely inhibited t-LTP for the duration of the experiment (Figure 4B.1 and C.1), but had no effect on the t-LTP-induced shrinkage of the activated spine necks (Figure 4B.2 and C.1, P < 0.001, n = 6 spines, from 5 experiments, 5 neurons and 5 mice) or in modifying spine head size (n = 10 spines, from 5 neurons and 5 mice, Figure 4C.1). Furthermore, when we analyzed the maximum change in uEPSP amplitude and the concomitant spine morphology after the induction of t-LTP in the presence of PEP1-TGL, we also found an inhibition of t-LTP, but no effect on the t-LTP-induced shrinkage of the spine neck (Figure 4B.2-C.2, uEPSP = 94.82 ± 14.82%, P = 0.74, n = 5 experiments, from 5 neurons and 5 mice; neck length = 83.92 ± 5.35%, P < 0.05, n = 6; head volume = 100.08 ± 3.23%, P = 0.98, n = 10). We obtained similar results when we instead considered the average of all the values obtained following t-LTP induction in two clustered spines in the presence of PEP1-TGL for uEPSP amplitude, neck length and head volume (Supplementary Figure 1). No significant difference was observed between the initial uEPSP amplitude for pre-post pairing protocols of +7 ms with versus without PEP1-TGL (uEPSP: 1.06 ± 0.2 versus 1.16 ± 0.28 mV, P = 0.81; Supplementary Figure 3). These results indicate that GluR1 receptor incorporation into the PSD - via its interaction with PDZ domains - is required for the induction of t-LTP in spines. However, the role of the spine neck shrinkage in AMPA receptor incorporation into the PSD and ultimately on the induction of t-LTP remains open.
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 27-30. 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) 27 – which is supported by studies that show reduced diffusion of membrane proteins located in spine necks 31. 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 29, even when their interaction with scaffolding cytoskeletal proteins is neglected 30. Actin is highly enriched in the spine neck and head 32, and plays an important role in anchoring AMPA receptors in the spine 33 and AMPA receptor trafficking 34, being instrumental for synaptic transmission and plasticity 35-37. 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 focused on actin dynamics. We used the actin polymerization inhibitor latrunculin A (Lat-A) 33, 35, 37 (Figure 4D) ? which did not trigger any run-down of uEPSP amplitude or changes in spine morphology over time in the absence of STDP induction (Supplementary Figure 5B) ? to test the potential role of actin dynamics on the spine induction of t-LTP, and on the neck shrinkage and AMPA receptor incorporation into the PSD in the activated spines (Figure 4D). The induction of t-LTP at pre-post pairings of +7 ms in two clustered spines in the presence of 100 nM Lat-A completely blocked the shrinkage of the activated spine necks and the induction of t-LTP (Figure 4E and F.1, n = 8 spines, from 3 neurons and 2 mice), inducing instead a significant reduction in uEPSP amplitude over time (Figure 4F.1, P < 0.001, n = 4 experiments, from 3 neurons and 2 mice). These observations were also consistent with analyses of the maximal change in uEPSP amplitude in each experiment and concomitant spine morphology post t-LTP induction in the presence of Lat-A (Figure 4F.2, uEPSP= 55.45 ± 7.13%, P < 0.01, n = 4 experiments; neck length = 87.20 ± 9.15%, P = 0.21, n = 8 spines; head volume = 101.25 ± 10.99%, P = 0.91, n = 8 spines).
We obtained similar results when we instead considered the average of all the values obtained following t-LTP induction in two clustered spines in the presence of 100 nM Lat-A for uEPSP amplitude, neck length and head volume (Supplementary Figure 1).
No significant difference was observed between initial uEPSP amplitudes for pre-post pairing protocols of +7 ms with versus without Lat-A (uEPSP: 0.94 ± 0.20 versus 1.16 ± 0.28 mV, P = 0.61; Supplementary Figure 3). The lack of run-down of uEPSP amplitude over time in neurons treated with Lat-A in the absence of STDP induction (Supplementary Figure 5B), but the significant depression in uEPSP after the induction of t-LTP suggests that the induction of plasticity, and the rearrangement of actin filaments de-stabilized AMPA receptors, leading to removal from the PSD.
In summary, these results show that actin polymerization is required for the t-LTP-dependent neck shrinkage and the induction of plasticity. Our findings further suggest that the induction of t-LTP occurs via a mechanism that involves a neck-shrinkage-dependent facilitated diffusion of GluR1 subunits from the spine neck to the head, and subsequent incorporation into the PSD. We hypothesize that a shorter and wider neck facilitates the transport of AMPA receptors into the spine head (Figure 4D), a mechanism that is required for the induction of t-LTP.
Induction of t-LTD in clustered and distributed dendritic spines
We then studied whether the induction t-LTD in single spines observed at pairings of -15 ms could be affected by synaptic cooperativity. Our reasoning was based on two previous observations which suggest that 1) t-LTP induction in the distal dendrites of L5 pyramidal neurons (layer 3-L5 pyramidal neuron pairs) triggers LTD instead of LTP, and 2) that LTD can be converted into LTP by increasing the local voltage 13. We hypothesised that the induction of t-LTD in single spines depends on the degree of local depolarization and hence, LTD can be disrupted by the activation of neighboring spines. To test this, we performed repetitive spike-timing protocol (40 times, 0.5 Hz) in which 2P uncaging of glutamate at two spines (separated by up to 100 µm) was preceded in time (−15 ms) by a bAP (Figure 5D and Supplementary Figure 6A). Surprisingly, we found that this t-LTD protocol failed to induce any change in uEPSP amplitude or spine head volume with only a slight but significant reduction in spine neck length (Supplementary Figure 6B). When we analyzed the maximum uEPSP change in amplitude in each experiment and the concomitant morphological alterations in the activated spines after t-LTD induction, we observed a complete inhibition in the induction of t-LTD, and no change in spine morphology (Supplementary Figure 6C; uEPSP = 93.22 ± 6.29%, P = 0.30, n = 17 experiments from 14 neurons and 14 mice; neck length = 88.56 ± 5.69%, P = 0.06, n = 23 spines; head volume = 102.41 ± 6.10%, P = 0.69, n = 34 spines). To more precisely characterize the effect of activating two spines on the induction of t-LTD, we correlated the inter-spine distance and the uEPSP change following STDP induction (see Methods). We found that as the two activated spines were further away from each other, the more t-LTD was recovered (Supplementary Figure 6D). Specifically, the uEPSP change decayed exponentially as a function of inter-spine distance with a length constant (λ) of 43.5 µm. Therefore, we used this value as a boundary between clustered (< 40 µm) and distributed (> 40 µm) spines. Using this classification, clustered spines were located in the same dendrite (n = 11/12 pairs) or in sister branches emanating from the same bifurcation point (n = 1/12 pairs), while distributed spines were always located on separate dendrites (n = 5/5 pairs). When we separated our data in this manner, the t-LTD protocol in two clustered spines (Figure 5A) failed to induce LTD (Figure 5B.1) or changes in spine head size at all the times tested post t-LTD induction (Figure 5B.2 and C.1, n = 12 experiments, n = 24 spines, from 11 neurons and 11 mice), with only a slight but significant induction of shrinkage of the spine neck at some time points (Figure 5C.1, P < 0.05, n = 19 spines, from 11 neurons and 11 mice). For comparison between the activation of one versus two clustered spines with a post-pre timing of -15 ms see Supplementary Figure 7. When we analyzed the maximum uEPSP change in amplitude in each experiment and the concomitant morphological alterations in the activated spines after t-LTD induction in two clustered spines, we also observed a complete inhibition in the induction of LTD, a slight but significant reduction in spine neck length, and no changes in spine head size (Figure 5C.2; uEPSP = 101.70 ± 7.02%, P = 0.81, n = 12 experiments; neck length = 85.28 ± 5.96%, P < 0.05, n = 19 spines; head volume = 102.98 ± 8.86%, P = 0.73, n = 24 spines, from 12 experiments performed in 11 neurons from 11 mice). We obtained similar results when we instead considered the average of all the values obtained following t-LTD induction in clustered spines for uEPSP amplitude, neck length and head volume (Supplementary Figure 1). These results were surprising since not only did we not observe t-LTD in clustered spines, but we also observed significant neck shrinkage with no LTP (see Figure 1 and 3). To account for this observation, we explored if there was a correlation between the induction of plasticity in these experiments and both the shrinkage of the spine neck and the distance between the activated clustered spines – since the local voltage, and hence the induction of plasticity, could be affected by the distance between the activated clustered spines. Indeed, we found that the distance between the activated spines under these experimental conditions (t-LTD induction protocol in clustered spines) is correlated with the induction of plasticity and the shrinkage of the activated spine necks (Equation 1 in Methods; P < 0.01; Supplementary Figure 6E). This analysis suggests that during t-LTD induction the structural arrangement of clustered spines (< 40 µm) determines the sign and magnitude of the change in synaptic strength and concomitant neck shrinkage.
We next investigated the mechanisms underlying the disruption of t-LTD by activating spines separated by increasingly larger distances (Figure 5D). Interestingly, the induction of t-LTD in spines separated by more than 40 µm (distributed spines) was capable of recovering the generation of LTD (Figure 5E.1 and Supplementary Figure 6D). Pooled data from all experiments demonstrate that the activation of distributed spines reliably induces t-LTD (Figure 5F.1, significant reduction in uEPSP amplitude, P < 0.01, n = 5 experiments from 5 neurons and 5 mice), without triggering changes in neck length or spine head size (Figure 5E.2 and F.1). When we analyzed the maximal change in uEPSP amplitude in each experiment and concomitant spine morphological changes, we saw a significant induction of t-LTD and no change in spine morphology (Figure 5F.2, uEPSP = 72.86 ± 8.08%, P < 0.05, n = 5 experiments, neck length = 97.85 ± 15.47%, P = 0.89, n = 6 spines; head volume = 101.06 ± 4.59%, P = 0.82, n = 10 spines) as what was found in experiments where t-LTD was generated at pairing times of -15 ms in single dendritic spines (Figure 2B-C). We obtained similar results when we instead considered the average of all the values obtained following t-LTD induction in distributed spines for uEPSP amplitude, neck length and head volume (Supplementary Figure 1). No significant difference was observed between the initial uEPSP amplitude for clustered versus distributed spines activated with post-pre pairings of -15 ms (EPSP: 1.06 ± 0.13 versus 1.31 ± 0.19 mV, P = 0.25; Supplementary Figure 3). For comparison between the activation of clustered versus distributed spines after post-pre pairings of -15 ms, see Supplementary Figure 8.
In summary, this data shows that the induction of t-LTD at pairing times of -15 ms was completely disrupted when only two clustered spines (< 40 μm apart) were activated in the basal dendrites of L5 pyramidal neurons, but could be recovered if the activated spines are distributed (> 40 μm) in the dendritic tree.
Spine calcium transients during the induction of t-LTP and t-LTD in single and clustered dendritic spines
Calcium is critical for the induction of synaptic plasticity 38-42, and high or low local concentration difference in dendrites and spines are thought to be associated with gating LTP or LTD, respectively 43-45. Therefore, to investigate the different mechanisms – with respect to local calcium accumulations – underlying the induction of t-LTP and t-LTD in single versus two clustered spines, we performed 2P calcium imaging in a region of interest (ROI) of the activated spines and their parent dendrites during STDP induction protocols throughout each of the 40 pre-post or post-pre repetitions (see Methods). The “before” images correspond to the calcium signals observed in the ROI right before the pairing in each repetition – uncovering the lack or presence of local calcium accumulation during the 40 pairing repetitions. The “after” images correspond to the calcium signals observed in the ROI right after the pairing in each repetition – uncovering a proxy for the amplitude and local calcium accumulation during the 40 pairing repetitions.
To dissect potential differences in local calcium signals and accumulation that can account for the presence or absence of t-LTP and t-LTD induction in clustered versus distributed spines, we imaged 2P calcium activity during four different STDP induction protocols: (1) pre-post pairing of +7 ms in one spine; (2) pre-post pairing of +7 ms in two clustered spines; (3) post-pre pairing of -15 ms in one spine; (4) post-pre pairing of -15 ms in two clustered spines.
During the pre-post (+7 ms) pairing protocol in single spines we found that, across the 40 repetitions, there was little to no calcium accumulation in the spine or dendrite (Figure 6A-B and left panels in Figure 6C, n = 7 experiments, from 6 neurons, and 4 mice). As expected, there was, however, a significant increase in calcium immediately following the stimulation (left panels in Figure 6F) that due to the lack of accumulation throughout the 40 repetitions, did not build up a local calcium signal in the activated dendrites and spines. In contrast, when we applied the exact same pairing protocol (pre-post + 7ms) in two clustered spines, there was a striking calcium accumulation in both the activated spines and dendrite that was evident when we analyzed the images taken before (Figure 6D-E and middle panels in Figure 6C) and after stimulation (middle panels in Figure 6F, n = 6 experiments, from 4 neurons, and 4 mice). Thus, activating just one additional spine using the same pairing protocol alters the calcium dynamics (compare black and green traces in right panels of Figure 6C and F), possibly through a mechanism that is incapable of extruding calcium increases in spines in between pre-post repetitions, leading to its build up in spines and parent dendrites, which ultimately guide the induction of plasticity.
We performed the same experiments with a post-pre (−15 ms) pairing protocol in both single and clustered spines. In single spines, we observed moderate calcium increases (Figure 7A-B, left panels in Figure 6C and F, n = 5 experiments, from 4 neurons, and 4 mice) that were observed when we analyzed images taken before and after the post-pre stimulation. Surprisingly, we found similar results to those observed with single spine t-LTD induction protocols, when we applied the same pairing protocol in two clustered spines (Figure 7D-E, middle panels in Figure 7A and F, n = 6 experiments, from 4 neurons, and 3 mice) even though no plasticity is induced in this condition. As suggested by previous studies 46, we hypothesize that the range of spine calcium levels required for the induction of t-LTD is relatively narrow, and that the resolution with our current experimental set-up is not sufficient to tease apart significantly different calcium dynamics in one versus two clustered spines during a post-pre pairing protocol of -15 ms. Moreover, modeling STDP provide evidence that, in addition to overall calcium levels, the detailed time course of calcium levels in the postsynaptic cell during a pairing protocol also guide the induction of plasticity 47. Nonetheless, these results suggest that the induction of t-LTD does not require significant calcium accumulations during the 40 repetitions, and most likely depends on the amplitude of calcium signals right after the stimulation. An interesting observation is that when we fit the calcium signals with a linear regression, the fits from all the different induction protocols have different slopes, which goes in line with our hypothesis (Supplementary Figure 9). More specifically, in single spines, a pre-post pairing protocol of +7 ms induced a relatively low calcium signal with a shallow slope (black lines in Supplementary Figure 9) whereas in clustered spines this same protocol caused a robust increase in calcium with a steep slope (blue lines in Supplementary Figure 9). A post-pre pairing protocol of -15 ms, caused a modest increase in calcium in both single and clustered spines before the stimulation (red and green lined in left panels of Supplementary Figure 9), whereas after the stimulation the calcium increase is more prominent in clustered spines (red and green lined in right panels of Supplementary Figure 9). These results provide evidence that the calcium levels needed to induce of t-LTD are restricted to a narrow range and that surpassing this range biases towards the induction of weak levels of LTP.
DISCUSSION
We uncovered the STDP rules for single, clustered and distributed dendritic spines in the basal dendrites of L5 pyramidal neurons. Our results show that the induction of STDP in single spines follows a classical Hebbian STDP learning rule that is bidirectional, in which presynaptic input leading postsynaptic spikes generates t-LTP and postsynaptic spikes preceding presynaptic activation of single dendritic spines results in t-LTD. Furthermore, we found that the induction of t-LTP triggers the shrinkage of the activated spine neck, without any significant changes in the spine head size. Our results indicate that the induction of t-LTP requires 1) the incorporation of new GluR-1 receptors with PDZ-domain containing proteins in the PSD and, 2) an actin polymerization-dependent neck shrinkage of the activated spine neck (Figure 4). We showed that the induction of t-LTP triggers actin-dependent neck shrinkage, which is likely required for the lateral diffusion of GluR-1 receptors from the spine neck to the spine head, and its incorporation to the PSD – generating plasticity. In support of this spine mechanism of LTP induction is a recent report showing that AMPA receptor surface diffusion is fundamental for the induction of hippocampal LTP and contextual learning 48. In addition, 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 4, 7, 49. The discrepancy between our results and those observed previously after the induction of t-LTP (head enlargement 49), LTP 4, or LTD (head shrinkage, 7) 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 a magnesium-free ACSF 4, 7, low calcium extracellular solution for the induction of LTD 7, 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 49.
Nonetheless, it has been shown in vivo that a spike-timing protocol triggers receptive field plasticity in layer 2/3 pyramidal neurons is correlated with spine head volume changes (enlargement and shrinkage) observed after 1.5-2 hours 50. Taken together, our data suggest that there is a new form of structural spine plasticity during t-LTP that involves rapid neck shrinkage without head volume enlargements. In addition, we show that the induction of t-LTD does not require structural spine changes. Although spines have the machinery and do undergo structural head changes, we propose that our results represent a stage during memory formation that occurs before structural head volume changes, a process likely linked with memory consolidation. Importantly, our data suggest that during STDP, the use of spine volume changes as the sole proxy for LTP or LTD 50 is not a complete representation of plasticity in spines from dendrites in cortical pyramidal neurons.
We then explored the functional consequences of synaptic cooperativity on STDP. Our results show that the induction of t-LTP in two clustered spines - separated by less than 5 µm - is sufficient to induce LTP and shrinkage of the activated spine necks at a pre-post timing that is otherwise ineffective at triggering significant morphological changes and synaptic potentiation when only one spine is being activated. These results show that the activation of clustered spines extends the pre-post timing window that can trigger potentiation. On the other hand, the induction of t-LTD in two clustered spines disrupts the generation of LTD leading to a STDP learning rule that is incapable of supporting LTD, but only encompasses LTP (Figure 8A). We next investigated the dendritic mechanisms responsible for the disruption of t-LTD, and found that the induction of t-LTD is fully recovered when the activated spines are separated by more than 40 µm (Figure 5, Figure 8A and Supplementary Figure 6). Interestingly, the effective length constant (λ), that represents the length at which the electrotonic potential decays to a value of 37% of that at the point of origin, in the basal dendrites of L5 pyramidal neurons has been reported to be 50 µm 51. This value of λ supports the idea that significant voltage attenuations – capable of recovering LTD – can be expected when the t-LTD induction protocol is triggered in spines that are separated by more than 40 µm in the basal dendrites of L5 pyramidal neurons (Figure 5 D-F). However, we cannot discard that other mechanisms, such as the diffusion of active molecules 5, could contribute to the switch from LTD to no-LTD induction observed in distributed/single spines and clustered spines, respectively. These results are in discrepancy with observations showing that in connected pairs of L5-L5 pyramidal cells, t-LTD is reliably generated after post-pre pairing protocols 12. A likely explanation for this apparent controversy is that the synaptic inputs from one L5 pyramidal neuron to another are distributed 52. Importantly, clustered and distributed excitatory inputs have been described in the dendrites of pyramidal neurons both in vitro and in vivo 1, 53-55. Our results clearly show the functional importance that the structural and temporal organization of excitatory synaptic inputs have on the induction of t-LTP and t-LTD, and how just two clustered excitatory synaptic inputs are capable of altering the STDP learning rule in the basal dendrites of L5 pyramidal neurons (Figure 8A).
How the synaptic activation of just one extra clustered spine is capable of (1) inducing t-LTP at a pre-post timing that is otherwise ineffective in inducing potentiation and (2) disrupting the induction of t-LTD? To explore the mechanisms that may be responsible for these observations we imaged local calcium signals in the activated spines and parent dendrites before and after each of the 40 pairings performed during t-LTP and t-LTD induction protocols. Our reasoning for performing these experiments was based on findings that different levels of depolarization gate local calcium signals, which depending on its magnitude and kinetics, can generate LTP (high calcium) or LTD induction (sustained but moderate calcium signals) 9, 43, 56. In addition, calcium-based modeling studies of STDP have shown that different calcium dynamics mediate the induction of t-LTP versus t-LTD 46, 47. Specifically, the calcium control hypothesis indicates that large levels of calcium (above a plasticity threshold, Θp) are thought to lead to t-LTP whereas more moderate, prolonged levels (between the depression threshold, ΘdSTART, and ΘdSTOP)) give rise to t-LTD and a mid-level range in which t-LTD does not occur (below ΘdSTART) (Figure 8B) 47, 57, 58. A major assumption of these models is infinite time constants for synaptic variables at resting calcium levels so that the synaptic changes do not to decay after the presentation of the stimulus 46 - a significant constraint for the stabilization of synaptic changes. A potential solution to this problem is the degree of local calcium accumulation observed in the activated spines throughout the t-LTP or t-LTD induction protocol. In fact, these models are consistent, fundamentally, with our results which show that a pre-post pairing (+7 ms) protocol in two clustered spines gives rise to t-LTP accompanied by a substantial increase in the intracellular calcium levels following each pairing repetition, and a significant accumulation of local calcium levels throughout the induction protocol – likely mediated by the inability of the two clustered activated spines to efficiently extrude the local calcium signals in between each pre-post pairing (Figure 6). We propose that the local spine calcium accumulation we observe provides a new and key variable for the induction of plasticity, which reduces the constraints imposed by calcium-base models for the stabilization of synaptic changes 46, 47, 57, 58.
These changes in local spine and dendritic calcium signals (Figure 6) suggest that perhaps Θp can be reached only with ∼ 10 pre-post pairings (∼20-30 seconds). In contrast this same protocol in one spine induces no plasticity, producing calcium signals right after the pairing stimuli that are effectively extruded in between pairings leading to no calcium accumulation during the induction protocol (reaching levels below Θp and ΘdSTART; Figure 8B). These results suggest that it is not only the amplitude of the local calcium signals after each pairing, but also the local calcium accumulation during the induction protocol (40 pairings, ∼80 seconds) in spines and dendrites that are required to reach Θp for the induction of t-LTP in clustered spines. As mentioned before, recently it has been demonstrated in vivo that spike timing-induced receptive field plasticity, with millisecond time delays between visual stimulus (pre) and optogenetic stimulation in layer 2/3 pyramidal neurons (post), is correlated with increases in synaptic strength 50. These results together with evidence from other in vivo studies showing that layer 5 pyramidal neurons can spike up to frequencies of 20 Hz during movement 59, suggest that our pairing protocol, and findings, are likely present under in vivo conditions and are relevant for plasticity of networks and ultimately behaviour.
Our results further show that a post-pre protocol of -15 ms in a single spine induces t-LTD and moderate intracellular calcium signals in spines and parent dendrites after each pairing, without an evident increase in local calcium accumulation. These results possibly reflect that the calcium signal generated during the induction protocol passed ΘdSTART and remain for several seconds in this permissive calcium concentration window – between ΘdSTART and ΘdSTOP – generating LTD (Figure 8B). Activating two clustered spines with the same protocol, however, does not induce plasticity and gives rise to an apparent smaller initial calcium accumulation than that observed with the activation of a single spine but with a slow build-up of calcium. These results possibly reflect that the spine calcium levels crossed ΘdSTART only after > 20 repetitions and then crossed ΘdSTOP and ΘpSTART after a few (<10) repetitions reaching slightly higher local calcium levels. This calcium control hypothesis of t-LTD induction is based on the average linear fits of each experiment, and a tendency, that although clear, is not statistically significant with our measurements (Figure 8B and Supplementary Figure 9).
These findings presented here are quite remarkable since stimulating just one additional spine during a STDP protocol can completely alter the calcium dynamics and the induction of t-LTP and t-LTD. To our knowledge, this is the first demonstration of the functional relevance that the structural organization and simultaneous subthreshold activation of only a few clustered inputs in the dendrites of pyramidal neurons have on plasticity. We propose the term micro clusters to describe this structural and functional modality of synaptic connectivity. 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 53, L5 pyramidal neurons from somatosensory cortex 54 and in the distal apical tuft dendrites in stratum lacunosum-moleculare of hippocampal CA1 pyramidal neurons 55. 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, recently 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 60, and that functional clusters of dendritic spines separated by less than 10 µm share similar spatial receptive field properties, spontaneous and sensory-driven activity 61. Taken together these reported findings and our data suggest that the functional specificity and structural arrangement of synaptic inputs, distributed or forming micro clusters in the dendrites of pyramidal neurons, are fundamental for guiding the rules for sensory perception, affecting the STDP learning rule, learning and memory, and ultimately cognition.
METHODS
Brain slice preparation and electrophysiology
Brains from postnatal day 14-21 C57B/6 mice - anesthetized with isoflurane - were removed and immersed in cold (4°C) oxygenated sucrose cutting solution containing (in mM) 27 NaHCO3, 1.5 NaH2PO4, 222 Sucrose, 2.6 KCl, 1 CaCl2, and 3 MgSO4. Coronal brain slices (300-μm-thick) of visual cortex were prepared as described 22. Brain slices were incubated for 1/2 hour at 32°C in artificial cerebrospinal fluid (ACSF, in mM: 126 NaCl, 26 NaHCO3, 10 Dextrose, 1.15 NaH2PO4, 3 KCl, 2 CaCl2, 2MgSO4) and then transferred to a recording chamber. Electrophysiological recordings were performed in whole-cell current-clamp configuration with MultiClamp 700B amplifiers (Molecular Devices) in L5 pyramidal neurons with a patch electrode (4-7 MΩ) filled with internal solution containing (in mM) 0.1 Alexa Fluo 568, 130 D-Gluconic Acid, 2 MgCl2, 5 KCl, 10 HEPES, 2 MgATP, 0.3 NaGTP, pH 7.4, and 0.4% Biocytin. All experiments were conducted at room temperature (∼20-22°C). We did not extend our experiments to include voltage-clamp recordings since recent evidence indicates that the high spine neck resistance can prevent the voltage-clamp control of excitatory synapses and that these measurements can be significantly distorted in spiny neurons62.
Two-photon imaging and two-photon uncaging of glutamate
Two-photon imaging was performed with a custom-built two-photon laser scanning microscope, consisting of 1) a Prairie scan head (Bruker) mounted on an Olympus BX51WI microscope with a 60X, 0.9 N.A. water immersion objective; 2) a tunable Ti-Sapphire laser (Chameleon Ultra-II, Coherent, >3 W, 140-fs pulses, 80 MHz repetition rate), 3) two photomultiplier tubes (PMT) for fluorescence detection. Fluorescence images were detected with Prairie software (Bruker).
Fifteen minutes after break-in, two-photon scanning images of basal dendrites were obtained with 720 nm and low power (∼5 mW on sample (i.e., after the objective)) excitation light and collected with a PMT. Two-photon uncaging of 4-methoxy-7-nitroindolinyl (MNI)-caged L-glutamate (2.5 mM; Tocris) was performed as previously described 63. This concentration of MNI-glutamate completely blocked IPSCs 64, thus, our results represent excitatory inputs only. Uncaging was performed at 720 nm (∼25-30 mW on sample). Note that the laser power used for imaging is not sufficient to result in any partial uncaging of glutamate (Supplementary Figure 10). Activated spines were mostly on the second and third branch of the basal dendrites and were on average ∼ 40 µm away from the soma (Supplementary Figure 11). Only spines with a clear head contour and that were separated by >1 µm from neighboring spines were selected. The location of the uncaging spot was positioned at ∼ 0.3 μm away from the upper edge of the selected spine head (red dot in figures), which has a spatial resolution of 0.71 and 0.88 µm for one and two spines respectively (Supplementary Figure 12). Care was taken maintain the position of the uncaging spot. After each uncaging sequence, the spot was repositioned to keep the same distance of ∼ 0.3 μm from the edge of the soma and to avoid artificial potentiation or depression. The uncaging-induced excitatory postsynaptic potentials (uEPSP) were recorded at the soma of L5 pyramidal neurons. The kinetics of uEPSPs from the present study are not significantly different (10/90 rate of rise: 0.07 ± 0.014 mV/ms, p=0.92; duration: 115.5 ± 15.3 ms, p=0.65) from those triggered at 37°C 22.
Spike timing-dependent plasticity (STDP) induction protocol
To induce t-LTP in single spines, we used two-photon uncaging of MNI-glutamate (40 times every 2 seconds, with each uncaging pulse lasting 2 ms), which, after 7 or 13 ms, was followed by a backpropagating action potential (bAP) (triggered by a brief (10 ms) current injection (0.4 -0.6 nA) in the soma). To induce t-LTD in single spines, two-photon uncaging of MNI-glutamate (40 times every 2 seconds, with each uncaging pulse lasting 2 ms) was preceded for -15 or -23 ms by bAP. When we evaluated t-LTP and t-LTD in two spines, we used similar protocols to those described above, but the spines were activated with two-photon uncaging of MNI-glutamate sequentially with an onset delay of ∼2.1 ms for the second spine. No significant difference was observed in the in 10/90 rise time of the uEPSPs triggered when one versus two spines were activated (9.05 ± 1.19 ms versus 9.49 ± 0.54 ms, respectively; p=0.71).
To evaluate the morphological and synaptic strength of the activated spines before and after the STDP induction protocol, we performed 2P imaging, and low frequency 2P uncaging (0.5 Hz) in single or multiple spines. To establish the time course of the changes in uEPSP amplitude, neck length and head volume following STDP induction, for each experiment, we interpolated the data taken at different time points using the interp1 function in MATLAB (MathWorks) with the pchip option, which performs a shape-preserving piecewise cubic interpolation. Note that we constrained this fit so that it terminated with a slope of zero following the last data point. Then, for each condition, we averaged the uEPSP amplitude, neck length and head volume temporal traces. The length of the x-axis was set as the time at which the last data point was obtained for those sets of experiments. Shaded area represents ±SE. To determine at which time the EPSP amplitude, neck length and head volume temporal traces are significantly different from baseline, we binned the temporal traces every 5min and tested whether it was significantly different from baseline (100%). The time at which the maximal change in uEPSP was observed after t-LTP and t-LTD induction was used to calculate the percent change from control, and the percent changes in neck length and head volume. These analyses are displayed in Figures 1 to 5.
Experimental checkpoints and data analysis
Electrophysiological data were analyzed with Wavemetrics software (Igor Pro) and MATLAB. The resting membrane potential of the recorded L5 pyramidal neurons was -58.27 ± 2.08 mV (n = 58 neurons). After taking this measurement, pyramidal neurons were maintained at – 65 mV in current clamp configuration throughout the recording session. Only neurons for which the injected current to hold the cell at – 65 mV was < 100 pA were included in this study. For the generation of bAP, only action potentials with amplitude of > 45 mV were considered for analysis. In most experiments, two control tests (each consisting of 10 uncaging pulses at 0.5 Hz), spaced by 5 min were performed to assess the reliability of the uEPSP amplitude. Only experiments for which uEPSP amplitudes were not significantly different before and after 5 minutes in control conditions were considered for analysis (less than 10% variation).
Synaptic plasticity was assessed by two parameters: the uEPSP amplitude and the spine morphology (neck length and head volume). The peak uEPSP amplitude was measured from each individual uEPSP by taking the peak value and averaging 2 ms before and after using Wavemetrics (Igor Pro). Only uEPSPs that were >0.1 mV in the control condition (i.e., before the induction of plasticity) were included in the analysis.
Synaptic plasticity was determined by the relative change of uEPSP amplitude (average of 10 uEPSP) measured before and after the STDP protocol. For each experiment, we evaluated whether the STDP protocol generated potentiation or depression by determining how many uEPSP data points fell above or below baseline values over the course of the experiment. Potentiation was defined as the majority of uEPSP amplitude data points measured over time increasing relative to baseline, and the maximum uEPSP increase was used for statistical test. Depression was defined as the majority of uEPSP amplitude data points measured over time decreasing relative to baseline, and the maximum uEPSP decrease was used for statistical test. The spike timings +7 ms and +13 ms (pre leads post) or -23 ms and -15 ms (post leads pre) correspond to the delta time offset between the beginning of the uncaging pulse (pre) and the beginning of the bAP pulse (post) repeated 40 times.
The analysis of spines morphology was performed from z-projections of the whole spine using ImageJ 65 (neck length) and MATLAB (MathWorks) (head volume). The neck length was measured from the bottom edge of the spine head to the edge of its parent dendritic shaft using the segmented line tool in ImageJ. We selected mostly spines with a spine neck longer than 0.2 µm. For those with a shorter neck, we did not report their length for analysis and statistics due to the diffraction limited resolution of our images. For spines whose necks shrunk after the STDP protocol below the diffraction limited resolution of our microscope, we set their length as the minimal measurement of spine neck length reported by Tonnesen et al., using STED microscopy (0.157 µm) 16. We estimated the relative spine head volume using the ratio of the maximum spine fluorescence and the maximum fluorescence observed in the dendrite measured from z-projections of the whole spine 66, 67. To obtain the spine volume, we then multiplied this ratio by the PSF of our microscope (0.11 fL) 68. Linear optimization techniques were used to determine the correlation between EPSP change, neck length change and distance between 2 activated spines. Specifically, the change in EPSP amplitude was modeled using the following equation: Where uEPSP and NL are the percent change in uEPSP and neck length, respectively, following the STDP protocol, D is the distance between the 2 spines, and c1, c2 and c3 are constant coefficients. These parameters were estimated using a least squares technique to obtain an optimal fit of the data that minimized the sum of the residuals squared. The relationship between inter-spine distance and the percent change in uEPSP was fit with the following exponential equation: where α, β and λ are constants, y represents the change in uEPSP, and x is the inter-spine distance.
Calcium imaging
During calcium imaging experiments, we performed whole-cell current-clamp recordings of L5 pyramidal neurons with a patch electrode containing calcium indicator Fluo-4 (300 μM; Thermo Fisher) and Alexa-594 (100 μM) diluted in an internal solution containing (in mM) 130 D-Gluconic Acid, 2 MgCl2, 5 KCl, 10 HEPES, 2 MgATP, 0.3 NaGTP, pH 7.4, and 0.4% Biocytin. To perform sequential 2P calcium imaging and 2P uncaging of caged glutamate in selected spines at one wavelength (810 nm), we used ruthenium-bipyridine-trimethylphosphine caged glutamate (RuBi-glu, Tocris) 64, diluted into the bath solution for a final concentration of 600 µM. Uncaging of Rubi-glu was performed at 810 nm (∼25-30 mW on sample). The location of the uncaging spot was positioned at ∼ 0.3 μm away from the upper edge of the selected spine head (red dot in Figures 6-7). Changes in calcium were monitored by imaging 2P calcium signals and detecting the fluorescence with 2 PMTs placed after wavelength filters (525/70 for green, 595/50 for red). We performed 2P calcium imaging during 4 different STDP induction protocols triggered at 0.5Hz: (1) pre-post pairing of +7 ms in one spine; (2) pre-post pairing of +7 ms in two clustered spines; (3) post-pre pairing of -15 ms in one spine; (4) post-pre pairing of -15 ms in two clustered spines. We restricted the image acquisition to a small area (∼150 x 150 pixels) which contained the spine(s) that we uncaged and the shaft. Images were acquired at ∼ 30 Hz, averaged 8 times, with 8 µs dwell time. Calcium signals were imaged 500 ms before STDP induction protocol and right after (4ms) the stimulation for more than 600 ms. We focused our analysis on the images obtained before and immediately after the stimulation in each pairing repetition. ROI drawing was performed using custom algorithms (MATLAB; MathWorks). For spine heads, the ROI was a circle whereas for dendrites it was a polygon. Fluorescence was computed as the mean of all pixels within the ROI. We quantified the relative change in calcium concentration using the following formula: where G is the fluorescence from the Fluo-4 dye and R is the fluorescence from the Alexa-594 dye. Gbaseline is the mean of all pixels of Fluo-4 signal within the ROI taken from the first image at the first stimulation. We estimated the calcium signal during each condition and using the following equation: where is the estimated change in calcium signal, x is the repetition (binned every 5 repetitions), a the slope and b a constant coefficient.
Statistics
Statistics were performed with GraphPad Prism 5. Statistical significance was determined using two-tailed Student’s paired t-test when we analyzed the maximum change in uEPSP amplitude after the induction of t-LTP or t-LTD in each experiment and the concomitant changes in the activated spine morphology. Statistical significance was determined using one-way repeated measures ANOVA when we analyzed the time course of the uEPSP amplitude and spine morphological changes after induction of t-LTP or t-LTD with post-hoc pairwise comparisons using Dunnett’s test. *P<0.05; **P<0.01; ***P<0.001.
Pharmacology
Latrunculin A (Lat-A, Tocris Bioscience) was dissolved in DMSO at 1/1000 and added to the recording chamber containing the brain slice at 100 nM for 15 min before starting the STDP protocol. PEP1-TGL (Tocris Bioscience) was added in the pipette at 200 µM; after 15 min in whole cell condition, electrophysiological recording and synaptic plasticity experiments were started. MNI-glutamate (Tocris Bioscience) was diluted in ACSF from stock solution and bath applied at 2.5 mM. Fresh vials of MNI-glutamate were used for each experiment.
ETHICS
Animal experimentation
these studies were performed in compliance with experimental protocols (13-185, 15-002, 16-011 and 17-012) approved by the Comité de déontologie de l’expérimentation sur les animaux (CDEA) of the University of Montreal.
Author contributions
R.A. conceived the project. S.T. and D.E.M. performed the experiments. S.T. and D.E.M. performed data analyses. S.M-R. performed control experiments. R.A., S.T., and D.E.M. designed experiments. R.A. and D.E.M. wrote the manuscript. R.A. supervised the project. All authors read and approved the contents of the manuscript.
Competing financial interests
The authors declare no competing financial interest.
Materials & Correspondence
Correspondence and material requests should be addressed to R.A.
Acknowledgements
We thank A. Kolta and P.J Sjöström for critical discussion and reading of the manuscript, and are grateful to all other members of Roberto Araya’s laboratory for kind support. We also thank members of the Groupe de Recherche sur le système nerveux central (GRSNC) for support and equipment shearing. This work was funded by the Canadian Institutes of Health Research (CIHR) grant MOP-133711 to R.A., a Canada Foundation for Innovation (CFI) equipment grant Fonds des leaders 29970 to R.A., and a Natural Sciences and Engineering Research Council of Canada (NSERC Discovery Grant) grant application No. 418113-2012 (NSERC PIN 392027) to R.A. S.T. was supported in part by a salary support from the GRSNC at Université of Montréal. D.E.M. was supported in part by a postdoctoral fellowship from the Fonds de recherche du Québec – Santé (FRQS).