Homeostatic Plasticity Scales Dendritic Spine Volumes and Changes The

bioRxiv preprint doi: https://doi.org/10.1101/308965; this version posted April 26, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

1 Homeostatic plasticity scales dendritic spine volumes and 2 changes the threshold and specificity of Hebbian plasticity 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Authors: 17 Anna F. Hobbiss1, Yazmin Ramiro Cortés1,2, Inbal Israely1,3 18 19 20 1. Champalimaud Centre for the Unknown, 1400-038 Lisbon, Portugal. 21 2. Instituto de Fisiología Celular, Universidad Nacional Autónoma de México, Ciudad Universitaria, 22 Circuito exterior s/n, Ciudad de México 04510, México. 23 3. Department of Pathology and Cell Biology in the Taub Institute for Research on Alzheimer's 24 Disease and the Aging Brain, Department of Neuroscience, College of Physicians & Surgeons, 25 Columbia University, New York, New York, 10032, USA. 26 27 Corresponding author: 28 Inbal Israely. Electronic address: [email protected] 29

30 Abstract

31 Information is encoded within neural networks through synaptic weight changes. Synaptic 32 learning rules involve a combination of rapid Hebbian plasticity with slower homeostatic 33 synaptic plasticity (HSP) that regulates neuronal activity through global synaptic scaling. While 34 Hebbian plasticity has been extensively investigated, much less is known about HSP. Here we 35 investigate the structural and functional consequences of HSP at dendritic spines of mouse 36 hippocampal neurons. We find that prolonged activity blockade induces spine growth, 37 paralleling synaptic strength increases. Following activity blockade, glutamate uncaging- 38 mediated long-term potentiation at single spines leads to size-dependent structural plasticity: 39 smaller spines undergo robust growth, while larger spines remain unchanged. Moreover, we 40 find that neighboring spines in the vicinity of the stimulated spine exhibit volume changes 41 following HSP, indicating that plasticity has spread across a group of synapses. Overall, these 42 findings demonstrate that Hebbian and homeostatic plasticity shape neural connectivity 43 through coordinated structural plasticity of clustered inputs.

44 Introduction

45 Hebbian synaptic plasticity, widely regarded as the leading biological mechanism for 46 information storage, involves activity-dependent changes in synaptic connectivity (Malenka 47 and Bear, 2004). Notably, these changes have a physical component and excitatory 48 postsynaptic current is highly correlated with dendritic spine volume (Matsuzaki et al., 2001). 49 However, left unchecked activity would lead to a positive feedback loop in which changes in 50 synaptic weight are further reinforced by future events. This has been proposed to result in 51 loss of functionality of the system, by either driving synapses towards saturation, or through 52 silencing the population (Miller and MacKay, 1994). How neurons maintain stability in the 53 face of destabilizing cellular and circuit events has been a longstanding question (LeMasson 54 et al., 1993). One method neurons employ to resolve this is Homeostatic Synaptic Plasticity 55 (HSP) (Turrigiano et al., 1998), a feedback mechanism through which a population of synapses 56 can be maintained to function within optimal bounds. During HSP, a decrease in global activity 57 drives a counteracting scalar increase in synaptic strengths to bring the network into an 58 optimal functioning range, and conversely, network activity increases will promote synaptic 59 weakening (Turrigiano, 2011). This scaling is instantiated in part by AMPA receptor trafficking

60 to and from the post-synaptic density, as well as through presynaptic changes which modify 61 neurotransmitter release, leading to concomitant changes in synaptic strength (Murthy et al., 62 2001; O’Brien et al., 1998). HSP has been observed both in vitro and in vivo (Desai et al., 2002), 63 supporting a fundamental role for this mechanism in the proper functioning of a neural 64 system. 65 It is well established that HSP modulates synaptic function; however, it is currently unknown 66 how this process also correlates with spine structural changes. Moreover, how HSP affects 67 the induction and longevity of subsequent forms of Hebbian plasticity at individual inputs 68 remains to be determined. In this study, we examine how the induction of HSP affects the 69 structure and function of hippocampal pyramidal neuron synapses, and what are the 70 consequences of these changes for future Hebbian plasticity. We find that the induction of 71 HSP through activity blockade leads to an overall increase in the size of spines, corresponding 72 to a structural scaling that matches the functional scaling of synapses. Through precise 2- 73 photon mediated glutamate uncaging, we further investigate how these homeostatic 74 functional and structural plasticity changes impact the ability of individual inputs to undergo 75 subsequent plasticity. We demonstrate that after HSP, spines express increased longevity of 76 LTP, an increased growth rate after stimulation, and a reduced plasticity threshold. We find 77 that HSP enhances the magnitude of synaptic potentiation through the preferential 78 modulation of small spines and by promoting structural plasticity at clustered inputs following 79 Hebbian activity at single synapses. Together, these changes provide a mechanism by which 80 homeostatic plasticity can modulate synaptic efficacy and enhance future learning without 81 compromising previously stored information. 82

83 Results

84 Structural correlates of homeostatic plasticity 85 Hebbian plasticity at an individual input is linearly correlated with volume changes in the 86 corresponding spine (Govindarajan et al., 2011; Harvey and Svoboda, 2007; Matsuzaki et al., 87 2004). Since HSP is known to change the functional properties of the spine, we reasoned that 88 homeostatic modifications of efficacy would be accompanied by structural plasticity of inputs. 89 We induced HSP using prolonged activity blockade through Tetrodotoxin (TTX) inhibition of

90 sodium channels, for either 0 h, 24 h, 48 h or 72 h (Figure 1a), as this has been shown to 91 induce scaling of synaptic strengths in a variety of systems (Karmarkar and Buonomano, 2006; 92 Turrigiano et al., 1998). We chose to conduct our experiments in mouse hippocampal 93 organotypic slice cultures, which maintain a physiologically relevant tissue architecture and 94 are amenable to genetic manipulation. To verify that functional synaptic scaling occurred, we 95 recorded spontaneous miniature excitatory post-synaptic currents (mEPSCs) from both TTX 96 treated and Control CA1 hippocampal pyramidal neurons at 48 h after the beginning of the 97 HSP induction period (Figure 1b-d). All recordings were performed in the presence of acute 98 TTX to block action potentials. As expected, we found a significant increase in mEPSC 99 amplitude following 48 h of activity blockade (Figure 1b,c; Control = 18.6 ± 0.25 pA, TTX = 100 20.72 ± 0.28 pA, p = 1.55e-8 Mann Whitney). The distribution of the TTX mEPSCs scaled 101 linearly to overlay with the control distribution (Figure 1d), in accordance with the synaptic 102 scaling theory. We next investigated whether homeostatic plasticity leads to structural 103 modification of synapses by live, 2-photon imaging of GFP-labelled CA1 dendrites (Figure 1e). 104 The volumes of all visible spines within the image were measured using SpineS, a custom in- 105 house developed Matlab toolbox (Erdil et al., 2012). We found that spines from chronically 106 TTX treated cells were significantly bigger than those from control cells beginning at 48 h 107 (Figure 1e,f; Control = 0.136 ± 0.0062 µm³, TTX = 0.211 ± 0.0123 µm³). Surprisingly, as opposed 108 to the linear functional scaling of mEPSCs seen in response to HSP, we found that structural 109 scaling of spine volumes at 48 h are best fit to controls with a second order equation (Figure 110 1g). This superlinear scaling resulted from a preponderance of large spines after TTX 111 treatment, for which correspondingly large mEPSCs were not observed. 112 113 After 72 h of activity blockade, synapses maintained the enhancement of mEPSC size (Figure 114 1h; Mean ± SEM: control 72 h = 20.35 ± 0.27 pA, TTX 72 h= 22.18 ± 0.239 pA, p = 1.88e-15, 115 Mann-Whitney) and spine volume increases (Figure 1i; Mean ± SEM: control 72 h= 0.16 ± 116 0.0073 µm³, TTX = 0.19 ± 0.0093 µm³, p = 0.0008 Mann-Whitney). However, at this time, 117 scaling was instantiated in a linear fashion at both the level of mEPSC amplitude and spine 118 volumes (Figure 1j,k). This suggests that the expression of functional and structural scaling 119 evolves dynamically over time in response to prolonged activity blockade. 120 121

122 Reversibility of homeostatic plasticity mediated structural changes 123 Synaptic strength modifications that occur in response to activity blockade are reversible 124 upon re-exposure of a circuit to activity (Desai et al., 2002; Wallace and Bear, 2004). We 125 tested whether the structural changes that resulted following activity blockade were also 126 reversible when activity is restored. After 48 h of homeostatic plasticity induction, by which 127 time significant structural and functional scaling have occurred (Figure 1), we removed TTX 128 allowing slices to resume activity and measured spontaneous firing using whole-cell patch 129 clamp recordings (Figure 2a,b). Activity blockade for 48 h led to higher firing rates compared 130 to untreated cells (Figure 2c; Control (median) = 1.33, IQ range = 0.11:18.83 spikes/min; TTX 131 (median) = 13.33, IQ range = 11.5:21.06 spikes/min. p=0.03, Kruskal-Wallis post-hoc Dunn). 132 To exclude the possibility that the higher firing rates we observed were due to a rebound after 133 the acute withdrawal of TTX from the system, a set of control neurons were briefly incubated 134 in TTX (2-4h, named “Acute TTX condition”). This manipulation did not significantly alter firing 135 rates, which were similar to untreated controls (Figure 2c; Acute-TTX: med = 4.44, IQ range = 136 2.0:29.33 spikes/min vs Control no-TTX: Median = 1.33, IQ range = 0.11:18.83 spikes/min, 137 p>0.9 Kruskal-Wallis post-hoc Dunn). Thus, the firing rate changes we observed were the 138 result of HSP and did not arise from the acute withdrawal of TTX. We then investigated 139 whether the circuit would return to its original level of activity once neurons were released 140 from blockade. Indeed, after having removed TTX for 48 h, firing rates were indistinguishable 141 from controls (Figure 2c; Control (median) = 1.33, IQ range = 0.11:15.56. TTX reversal 142 (median) = 1.67, IQ range = 0.194:11.76. p>0.05, Kruskal-Wallace post-hoc Dunn). We then 143 examined whether structural modifications accompanied these firing rate changes, and 144 observed a similar reversal in spine volumes (Figure 2d,e). TTX treated spines significantly 145 reduced in size 48 h after TTX removal (Figure 2e; mean volume ± SEM: TTX 48 h = 0.21 ± 146 0.0063 µm³; TTX 48 h reversal = 0.16 ± 0.0064 µm³; p=4.1852e-07). These findings support 147 the idea that functional plasticity is matched by structural plasticity, and that spines sizes 148 reversibly and accurately reflect the activity landscape. 149 150 Enhanced Hebbian potentiation of small spines after HSP 151 Functional plasticity can be enhanced upon homeostatic modulation in hippocampal slices 152 (Arendt et al., 2013; Félix-Oliveira et al., 2014), but occlusion of LTP may also occur (Soares et 153 al., 2017). Given our finding that robust growth of spines occurs during homeostatic plasticity,

154 we wanted to determine whether individual inputs were able to undergo further activity- 155 dependent structural plasticity. We therefore induced HSP in hippocampal slices for 48 h and 156 followed this with synaptic potentiation at visually identified dendritic spines through 2- 157 photon mediated glutamate uncaging (Figure 3a,b). This paradigm elicits potentiation and 158 growth of only the stimulated input (Govindarajan et al., 2011; Harvey and Svoboda, 2007). 159 We followed the growth of spines relative to their average baseline volumes, in both the TTX 160 and control conditions (Figure 3c,d). During the initial 45 minutes post-stimulation, both TTX- 161 treated and control spines grew to a similar extent, indicating that homeostatic structural 162 scaling does not occlude activity-dependent structural plasticity (mean ± SEM; Control 163 stimulated 45’= 1.45 ± 0.10; Control neighbors 45’= 1.08 ± 0.04; TTX stimulated 45’= 1.59 ± 164 0.14; TTX neighbors 45’= 1.14 ± 0.04; p > 0.99) (Figure 3d). However, after 120 minutes, only 165 the stimulated inputs in the TTX-treated neurons remain significantly larger than their 166 neighbors (Figure 3d; mean M:± SE TTX stimulated 120’= 1.53 ± 0.13; TTX neighbors 120’= 167 1.16 ± 0.04), as control spines begin to return to their initial volume (Figure 3d; Control 168 stimulated 120’= 1.30 ± 0.01; Control neighbors 120’= 1.14 ± 0.04). Although the TTX and 169 control spines are not significantly different at 120’ (p = 0.1345, Mann Whitney), they begin 170 to diverge, as seen relative to their respective neighbors. Thus, at scaled synapses, activity 171 that would otherwise lead to short lived structural plasticity now elicits long lasting structural 172 plasticity. 173 174 Structural plasticity at individual inputs varies with activity and spine size. While weak 175 stimulations preferentially affect small spines (Matsuzaki et al., 2004), strong stimuli can elicit 176 structural plasticity across a variety of spine sizes (Govindarajan et al., 2011; Oh et al., 2013; 177 Ramiro-Cortés and Israely, 2013). Our stimulated spines spanned a range of sizes, allowing us 178 to test whether size interacted with prior expression of HSP when expressing synaptic 179 potentiation and further spine growth (Figure 3e). We examined the amount of structural 180 plasticity expressed by spines of different initial sizes after activity blockade. We found a 181 negative correlation between a spine’s initial average value and its normalized final volume 182 when the spine population had first undergone homeostatic plasticity, but not in the control 183 population, indicating that spine volume is an important modulator of the potential for 184 plasticity after HSP ( Figure 3ef; r2 = 0.48 for TTX, p < 0.01, r2 = 0.06 for control, p > 0.3). To 185 further examine how HSP modulates the size dependence of structural plasticity, we classified

186 spines as ‘large’ if their initial volume was more than 150% of the median initial volume of all 187 spines, and the remainder were placed in the ‘small’ category. Among the small spines, we 188 observed a significant increase in the magnitude of LTP in the TTX condition compared to the 189 controls (Figure 3g; p=0.046, 2-way repeated measures ANOVA). On the other hand, large 190 spines expressed a short-lasting growth that quickly decayed to baseline and was not 191 dependent on whether they had been subjected to activity blockade (Figure 3h; p=0.70 2-way 192 repeated measures ANOVA). Therefore, small spines preferentially underwent long lasting 193 structural plasticity after HSP. We further observed that small spines, which had undergone 194 HSP, tended to show a greater degree of initial growth in response to the induction of 195 plasticity (Figure 3g, time point 0). To quantify the dynamic growth of these spines, we 196 analyzed high-speed images of the spine head taken throughout the stimulation period 197 (approximately 20 Hz) and did not observe a significant difference in the growth curves 198 between the control and TTX treated spines (Figure 4a,b). When we examined the rate of 199 growth that each spine expressed in the first minute after the stimulation however, we found 200 that TTX treated small spines grew more than control small spines (Figure 4c; p = 0.018, 1- 201 way ANOVA with post-hoc Tukey’s), while large spines showed no difference between the 202 two conditions. Therefore, glutamate stimulation leads to a sustained growth of 203 homeostatically modified small spines, which may reflect prolonged signaling at these 204 synapses. Together, these data indicate that HSP facilitates Hebbian plasticity, and that this 205 is accomplished at the individual spine level through preferential structural plasticity of small 206 inputs. 207 208 Homeostatic plasticity facilitates the induction of Hebbian structural plasticity 209 Having observed that small spines showed enhanced responses to stimulation after 210 homeostatic plasticity, as reflected by a higher rate of growth and longer lasting structural 211 plasticity (Figure 3g and 4c), we reasoned that these inputs may respond more robustly to a 212 weaker stimulation. This is also supported by previous findings that loss of activity at 213 individual inputs lowers their threshold for subsequent plasticity (Lee et al., 2010). To test this 214 possibility, we applied a sub-threshold stimulation that utilizes a shorter laser pulse of 1 ms 215 (instead of 4 ms) during the glutamate uncaging protocol, since this does not induce structural 216 plasticity at spines (Govindarajan et al., 2011; Harvey and Svoboda, 2007). Indeed, control 217 spines that underwent this sub-threshold stimulation showed only a transient post-stimulus

218 growth that lasted 5 min, resulting in no long term plasticity (Figure 4d; Control spines, 1 ms 219 stimulated vs neighbors: p > 0.05, 2-way repeated measures ANOVA, post hoc Bonferroni). 220 These results were significantly different from what we had observed in response to the 4ms 221 stimulation (Figure 4d; 1 ms stimulated vs 4 ms stimulated; p = 0.04, 2-way repeated 222 measures ANOVA; compared to data from Figure 3c). In contrast to this, HSP modified spines 223 grew significantly in response to the subthreshold stimulation and remained larger than their 224 neighbors (Figure 4e; TTX spines, 1 ms stimulated vs neighbors p = 0.01 2-way repeated 225 measures ANOVA, post hoc Bonferroni). Following TTX treatment, both weak and strong 226 stimulations elicited similar levels of long lasting growth of spines (Figure 4e; 1ms stimulated 227 vs 4 ms stimulated: p = 0.27, 2-way repeated measures ANOVA). Thus, homeostatic plasticity 228 facilitates structural plasticity by lowering the threshold for stimulation, without affecting the 229 magnitude of plasticity. In this way, HSP acts locally, in conjunction with activity, to increase 230 the likelihood that an individual synapse will participate in the encoding of information. 231 232 HSP influences structural plasticity of neighboring spines 233 In addition to modifying synaptic inputs through scaling, homeostatic plasticity can also 234 influence cellular firing rates by modulating the intrinsic excitability of neuronal membranes 235 (Desai et al., 1999). We reasoned that such alterations could facilitate the expression of 236 structural plasticity not only at stimulated spines but also at nearby inputs within the dendritic 237 branch. To test this idea, we examined whether homeostatic plasticity altered neighboring 238 spine volume dynamics following the potentiation of single inputs. We ranked neighboring 239 spines according to their distance from the stimulated spine (with 1 representing the closest 240 neighbor to a stimulated spine) and plotted their growth dynamics for two hours following 241 the stimulation (Figure 5a). We found that spines located in close proximity to the stimulated 242 spine increased in volume only in neurons that had first undergone homeostatic plasticity 243 (compare the first 20 spines, from 0 to 60 minutes after stimulation, Figure 5a). We classified 244 the unstimulated neighbors as “near” or “far” - located either within or beyond 5 µm of the 245 stimulated spine respectively - and quantified spine volume changes over time (Figure 5b-d). 246 As expected, none of the neighbors of stimulated spines changed significantly from their 247 original size in untreated neurons (Figure 5b,c). However, after HSP, neighbors that were 248 within 5 µm of the target spine exhibited significant growth in the 5 minutes following the 249 stimulation (Figure 5b) and remained significantly larger than more distant spines (Figure

250 5b,d; p = 0.001, 2-way repeated measures ANOVA). Interestingly, farther neighbors (located 251 up to 20 µm away from the site of stimulation) tended to decrease in size for several minutes 252 following activity, although this was not significantly different from baseline (Figure 5d). We 253 next investigated whether there was a correlation between the structural dynamics of each 254 stimulated spine and its neighbors. We calculated correlation coefficients between the 255 volume change of the stimulated spine and those of its neighbors across the time-course of 256 the experiment. While fluctuations in the volumes of neighboring spines in control conditions 257 were independent of the stimulated input (Figure 5e; r2 = 0.00263, p = 0.61), spine volumes 258 in TTX treated neurons changed in accordance with the stimulated spine (Figure 5f; r2 = 0.156, 259 p < 0.001). Specifically, near TTX neighbors had a positive correlation with the stimulated 260 input, reflecting growth, which eventually became a negative correlation at further distances 261 from the stimulated spine. These findings indicate that HSP reduces the input specificity of 262 activity in a distance dependent manner that is at once both cooperative with near, and 263 competitive with far, inputs. This allows for homeostatic modulation to effect structural 264 plasticity within a dendritic region through activity at a single spine, which favors clustered 265 plasticity of synaptic inputs. 266

267 Discussion 268 Synaptic networks must balance the need to enhance efficacy during learning without 269 saturating their capacity for further changes. Homeostatic plasticity can achieve this 270 regulatory role by effecting a gain modulation on neurons in order to maintain synaptic 271 activity within an optimal target range. It is unclear what are the consequences of 272 implementing HSP for the network, and specifically, how its interaction with Hebbian 273 plasticity impacts the subsequent encoding of information at the level of single inputs. Due 274 to the linear relationship between synaptic structure and function, we used live 2-photon 275 imaging and glutamate uncaging to probe the physical consequences of inducing homeostatic 276 plasticity on a neuron, and to determine how this impacts the ability of single synapses to 277 undergo further changes in efficacy. We show that homeostatic synaptic plasticity, induced 278 by 48 hours of activity blockade, leads to the reversible growth of dendritic spines. We 279 demonstrate that this form of modulation facilitates subsequent structural plasticity at single 280 synapses and that this effect preferentially occurs at smaller inputs. After HSP, activity elicits

281 a faster growth rate at these spines, and converts an otherwise subthreshold stimulation into 282 one that is now capable of eliciting structural plasticity. Interestingly, we find that the 283 induction of Hebbian plasticity on a background of homeostatic plasticity leads to 284 compromised input specificity, as neighboring spines grow when they are located in close 285 proximity to a stimulated synapse. Taken together, our results show that homeostatic 286 plasticity can modulate a neuron’s response to activity by facilitating the sensitivity of smaller 287 inputs and inducing structural plasticity at neighboring synapses. 288 The structural scaling that we observe after 48 hours of activity blockade results in a 289 non-linear upscaling of spines and an increased number of large spines (Figure 1g). By 72 290 hours, this structural scaling becomes linear (Figure 1k), suggesting that the initial changes 291 may represent a physical overshooting of the target size. This has also been observed on short 292 timescales following glutamate stimulation of single spines, where an initial large volume 293 change is followed by stabilization of the spine at a more modest size (Matsuzaki et al., 2004). 294 Upon the reinstatement of activity, we find that spines return to their original size as firing 295 patterns normalize by 48 hours (Figure 2c-e), indicating that homeostatic structural changes 296 are plastic and reversible, similarly to homeostatic responses to fluctuating activity levels. 297 We demonstrate that activity blockade increases spine sizes across the population, 298 raising the question of whether further enhancement of potentiation and spine growth could 299 be achieved across inputs of different sizes. Both enhancement and occlusion of plasticity 300 have been reported following HSP at the population level (Arendt et al., 2013; Félix-Oliveira 301 et al., 2014; Soares et al., 2017), but it is unclear how individual inputs are modulated. Further, 302 while larger synapses have been shown to undergo functional homeostatic modulation 303 (Thiagarajan et al., 2005), inputs show a size dependent variation in their ability to undergo 304 Hebbian structural plasticity, large spines being more stable and requiring stronger 305 stimulation in order to be potentiated (Govindarajan et al., 2011; Matsuzaki et al., 2004). We 306 probed plasticity with glutamate uncaging and found that HSP leads to structural plasticity 307 lasting over two hours preferentially at small spines, while control spines decay to baseline 308 after one hour (Figure 3g,h). This suggests that a long lasting, protein synthesis dependent 309 form of Hebbian plasticity was induced at these inputs (Govindarajan et al., 2011), supported 310 by recent findings that TTX-mediated activity blockade leads to the specific production of 311 plasticity related proteins (Schanzenbacher et al., 2018). Closer examination of the structural 312 plasticity we induced revealed that the population of small spines express the majority of the

313 growth, while large spines remain stable (Figure 3g,h). This result implies that plasticity at 314 large spines is saturated after homeostatic modifications, and that their threshold for 315 structural plasticity is not altered by HSP. 316 Upon stimulation, we did not observe a significant difference in the magnitude of the 317 potentiation that spines expressed, nor in their immediate response to the stimulation itself, 318 but rather we found that spines post-HSP express a significantly faster growth rate compared 319 to their counterparts in the first minute after stimulation (Figure 4a-c). This enhanced 320 response may be due to amplified signaling cascades shared between these forms of plasticity 321 (Fernandes and Carvalho, 2016), which could facilitate the induction of long lasting structural 322 plasticity. It was therefore not surprising to find that a subthreshold stimulation elicits long 323 lasting growth only at inputs that have undergone homeostatic plasticity (Figure 4 d,e), 324 without affecting the initial magnitude of the response between the HSP modified synapses 325 and the control populations. Changes to NMDA receptor composition that occur following 326 activity blockade could provide a means by which to enhance the calcium permeability of 327 synapses and thus would allow for prolonged signaling responses (Barria and Malinow, 2005; 328 Lee et al., 2010; Paoletti et al., 2013). Taken together, these data show that the effects of 329 homeostatic plasticity preferentially impact small spines during Hebbian regulation of 330 synaptic strength by predisposing them to undergo long lasting structural changes. Such 331 modulation may result in the conversion of weaker stimuli into more salient forms, and 332 proposes a more global role for HSP in information encoding beyond the optimization of 333 neuronal activity. 334 A hallmark of activity dependent changes during Hebbian plasticity is input specificity 335 (Barrionuevo and Brown, 1983). In contrast, homeostatic plasticity is generally expressed over 336 a wider synaptic range, and how this modulation impacts Hebbian learning rules at individual 337 inputs is unknown (Vitureira and Goda, 2013). Computational studies have postulated that 338 homeostatic plasticity can influence previous Hebbian events at a synapse (Rabinowitch and 339 Segev, 2008). Additionally, several lines of evidence indicate that activity can lead to local 340 homeostatic changes within dendrites (Branco et al., 2008; Ju et al., 2004; Liu, 2004; Sutton 341 et al., 2006). We therefore considered the possibility that HSP may function to influence 342 subsequent Hebbian plasticity. Indeed, we find that HSP induces the spillover of Hebbian 343 structural plasticity within the dendrite, as seen by the growth of neighboring unstimulated 344 spines following the activation of one input (Figure 5). Thus, homeostatic plasticity drives the

345 local expression of Hebbian plasticity and reduces input specificity. While cooperative 346 interactions have been shown to occur between inputs when multiple synapses are 347 stimulated (Govindarajan et al., 2011; Harvey and Svoboda, 2007), in the absence of such co- 348 activation, structural changes at neighbors have thus far served to counterbalance the 349 direction of plasticity (Oh et al., 2015). Our observations that HSP promotes activity-driven 350 growth of a group of synapses within a five micron area is in alignment with the observed 351 distance over which plasticity related proteins can spread between co-active synapses 352 (Harvey et al., 2008). Thus, the interaction between homeostatic and Hebbian plasticity may 353 coordinately delimit a region of clustered plasticity, which has been proposed to increase the 354 memory capacity of a neural circuit (Chklovskii et al., 2004; Govindarajan et al., 2011, 2006; 355 Ramiro-Cortés et al., 2014). Although clustered activity and its structural correlates have 356 begun to emerge (Fu et al., 2012; Kleindienst et al., 2011; Makino and Malinow, 2011; Yadav 357 et al., 2012), the mechanisms which drive the physical organization of inputs are unknown. 358 HSP may provide a basis for such organization by reducing the threshold for cooperativity 359 between inputs, allowing primed synapses located within several microns of an active spine 360 to be more easily potentiated. This may serve as a first step in the physical arrangement of 361 synapses into clusters. The broader consequence of this could be the binding together of 362 information of varying saliencies, within a delimited region, into the same engram. 363 Our finding that synaptic threshold modulation after HSP is implemented at small, 364 rather than at large spines, and in a reversible manner, suggests that synaptic scaling 365 mechanisms are separable from plasticity mechanisms. However, by promoting long-lasting 366 plasticity at select spines, and potentiating unstimulated neighboring spines within a 367 delimited dendritic region, homeostatic plasticity may interact with and reduce the input 368 specific nature of Hebbian plasticity, enhance the clustering of synaptic inputs, and shape the 369 long-term organization of neural circuits. In this way, despite the global nature of homeostatic 370 plasticity, inputs may be locally modulated in an activity-dependent manner.

371 Materials and Methods

372 1. Slice culture preparation and biolistic transfection

373 Mouse hippocampal organotypic slice cultures were prepared using p7-10 C57BL/6J mice as 374 previously described (Govindarajan et al., 2011). Briefly, hippocampi were dissected and 350 375 µm slices were cut with a tissue chopper in ice-cold artificial cerebral spinal fluid (aCSF)

376 containing 2.5 mM KCl, 26 mM NaHCO3, 1.15 mM NaH2PO4, 11 mM D-glucose, 24 mM

377 sucrose, 1 mM CaCl2 and 5 mM MgCl2. The slices were cultured on membranes (Millipore), 378 and maintained at an interface with the following media: 1x MEM (Invitrogen), 20% horse 379 serum (Invitrogen), 1 mM GlutaMAX (Invitrogen), 27 mM D-glucose, 30 mM HEPES, 6 mM

380 NaHCO3, 2mM CaCl2, 2mM MgSO4, 1.2% ascorbic acid, 1 µg/ml insulin. The pH was adjusted 381 to 7.3, and the osmolarity adjusted to 300–310 mOsm. All chemicals were from Sigma unless 382 otherwise indicated. Media was changed every 2-3 days. 383 Biolistic transfection of slice cultures was accomplished using a Helios gene gun (Bio-Rad) 384 after 4–7 days in vitro (DIV). Gold beads (10 mg, 1.6 µm diameter, Bio-Rad) were coated with 385 100 mg AFP-plasmid DNA (a GFP-expressing plasmid driven by the β-actin promoter (Inouye 386 et al., 1997) according to the manufacturer’s protocol and delivered biolistically to the slices, 387 using a pressure of 160-180 psi.

388 2. Homeostatic plasticity induction by activity block

389 TTX (1 µM) was added to the culture media at 7-9 DIV. The day of application was then 390 designated day 0 for experiments. Control experiments were maintained in normal culture 391 media, and were age- and animal-matched to treated slices for experiments.

392 3. Patch Clamp Electrophysiology

393 Hippocampal slice cultures were perfused continuously with aCSF (as above, with the addition 394 of 0.5 µM TTX for all mEPSC recordings) for a pre-incubation period of 15 to 30 min. Whole 395 cell voltage-clamp recordings were performed in CA1 pyramidal neurons, using 7–8 MΩ 396 electrodes. For mEPSC recordings, the internal solution contained 135 mM Cs-

397 methanesulfonate, 10 mM CsCl, 10 mM HEPES, 5 mM EGTA, 2 mM MgCl2, 4 mM Na-ATP and 398 0.1 mM Na-GTP, with the pH adjusted to 7.2 with KOH, at 290-295 mOsm. Cells were voltage 399 clamped at -65 mV. Cellular recordings in which series resistance was higher than 25 mV were

400 discarded. Stability was assessed throughout the experiment, with cells whose series 401 resistance changed more than 30% being discarded. mEPSCs recordings were started 3 402 minutes after break-in and continued for 10 minutes. Signals were acquired using a 403 Multiclamp 700B amplifier (Molecular Devices), and data was digitized with a Digidata 1440 404 at 3 kHz. mEPSC events were detected off-line using Mini-Analysis Program (Synaptosoft). 405 Events smaller than 15 pA fell within the range of noise in the system and were not included 406 in the analysis. For spontaneous activity recordings, slices were perfused continuously with 407 aCSF without the addition of TTX for a pre-incubation period of 5 to 10 min. The internal 408 solution for the electrodes contained 136.5 mM K-Gluconate, 9 mM NaCl, 17.5 mM KCl, 10 409 mM HEPES, 0.2 mM EGTA, and 0.025 mM Alexa 594, with the pH adjusted to 7.2 with KOH, 410 at 280-290 mOsm. In current clamp with no external current applied, an IV curve was first 411 recorded to check for spike frequency accommodation in order to validate the identity of 412 pyramidal neurons. Spiking events were then recorded for a period of 6-9 minutes. The 413 addition of Alexa-594 allowed cells to be imaged post-recording.

414 4. Two-photon Imaging

415 Two-photon imaging was performed on a BX61WI Olympus microscope, using a 416 galvanometer-based scanning system (Prairie Technologies /Bruker) with a Ti:sapphire laser 417 (910 nm for imaging AFP; Coherent), controlled by PrairieView software (Prairie 418 Technologies). Slices were perfused with oxygenated aCSF containing 127 mM NaCl, 2.5 mM

419 KCl, 25 mM NaHCO3, 1.25 mM NaH2PO4, 25 mM D-glucose, 2 mM CaCl2, 1 mM MgCl2 and 0.5

420 µM TTX (equilibrated with O2 95%/CO2 5%) at room temperature, at a rate of 1.5 ml/min. 421 Secondary or tertiary apical dendrites of CA1 neurons (where the apical trunk is counted as 422 the primary branch) were imaged using a water immersion objective (60x, 1.0 NA, Olympus 423 LUMPlan FLN) with a digital zoom of 8x. For each neuron, 2-3 dendrites were imaged. Z-stacks 424 (0.5 µm per section) were collected at a resolution of 1024 x 1024 pixels, resulting in a field 425 of view of 25.35 x 25.35 µm. 3 images were taken per dendrite at 5-minute intervals, with the 426 reported spine volume being the average of the 3 images. Images were taken at the highest 427 possible fluorescence intensity without image saturation in order to accurately quantify spine 428 volumes (see below).

429 5. Glutamate uncaging

430 Uncaging experiments (Pettit et al., 1997) and caged glutamate calibration were carried out 431 as previously described (Govindarajan et al., 2011), and briefly as follows. MNI-caged-L-

432 glutamate (MNI-Glu) (Tocris) was reconstituted in the dark in aCSF lacking MgCl2 or CaCl2 to 433 make a 10 mM stock solution. Individual aliquots were diluted to the working concentration 434 of 2.5 mM MNI-Glu in uncaging aCSF (see below), in 3 ml volumes. We tested each batch of 435 reconstituted MNI-Glu as previously described (Govindarajan et al., 2011). Briefly, five 436 uncaging test pulses of 1 ms were delivered to single spines and evoked EPSCs were measured 437 by whole cell patch clamp recordings. We compared these to spontaneous mEPSCs, and 438 determined the power needed (60mW at the back aperture) to produce an uncaging EPSC of 439 comparable size to that of an average spontaneous mEPSC. During plasticity experiments, 440 slices were incubated for 30-45 minutes in normal aCSF as described above. Uncaging was 441 performed using a Ti:sapphire laser (720 nm; Coherent), controlled by PrairieView software 442 (Prairie Technologies, Bruker). All stimuli were carried out in uncaging-aCSF containing: 2.5

443 mM MNI-Glu, 0 mM MgCl2, and 4 mM CaCl2. The uncaging train consisted of 30 pulses at 0.5 444 Hz, with a pulse width of either 4 ms (standard protocol) or 1 ms (sub-threshold protocol), 445 using 30mW power as measured at the back aperture (half the power as determined in the 446 calibration step, as previously described for an uncaging LTP protocol (Govindarajan et al., 447 2011)). The uncaging point was positioned 0.6 µm from the end of the spine head, away from 448 the parent dendrite. Each experiment began with a sham stimulation in uncaging-aCSF lacking 449 MNI-Glu on a control dendrite of the experimental neuron, to rule out the possibility of non- 450 specific structural changes due to phototoxicity or poor neuronal health. Subsequently, 451 experimental spines on new dendrites were identified, uncaging-aCSF containing 2.5mM 452 MNI-glutamate was recirculated for 5 minutes, and the uncaging protocol was delivered. Fast 453 imaging (approximately 20 Hz) of a region of interest (ROI) of the stimulated spine head and 454 neck was performed throughout the 60 second stimulation, to record the growth dynamics 455 for this time period. After the stimulation, the perfusion was returned to normal aCSF for the 456 remainder of the experiment (1-2 hours). The first image was taken immediately after 457 returning to normal aCSF and is designated as time = 0 post stimulation.

458 6. Spine Volume Determination

459 To quantify spine volume, we used the custom built Matlab plug-in SpineS, which uses semi- 460 automatic detection, automatic alignment and segmentation of spine heads (Erdil et al., 2012; 461 Ghani et al., 2017). Spine volume was calculated using the summed fluorescence intensity of 462 the spine, normalized to the median fluorescence intensity of the parent dendrite, in order 463 to correct for changes in overall fluorescence levels within a cell. Fluorescence intensity was 464 converted to real volumes by taking Full Width Half Max (FWHM) measurements of spines 465 and calculating a conversion factor between the two measures (Matsuzaki et al., 2004). For 466 the homeostatic plasticity analysis (Figures 1 & 2), all spines with a discernible head within 467 the field of view were included in the analysis, unless obstructed by other structures. For the 468 uncaging analysis (Figures 3, 4 & 5), between 2 and 8 neighboring spines were analyzed from 469 the same dendrite, within the whole field of view. Normalization of spine volumes throughout 470 the experiment was performed using the mean of three baseline images. 471

472 7. Statistical analyses

473 All statistical analyses were carried out in Graphpad Prism. Stars (*) represent degrees of 474 significance as follows: (*) = p < 0.05; (**) = p < 0.01; (***) = p < 0.001).

475

476 Acknowledgements

477 We thank Ali Ozgur Argunsah for his advice and help with data analysis. We thank Steven 478 Kushner, Rui Costa, and members of the Israely lab for their support and critical reading of 479 the manuscript. This work was supported by grants from the Bial Foundation (to II), Fundação 480 para a Ciência e a Technologia (to II and AH), and CONACYT 254878 (to YRC). 481 482 Author Contributions: 483 AFH, YRC and II designed the experiments. AFH performed the experiments and analyzed 484 the data. AFH, YRC and II wrote the manuscript.

485

486 Competing interests 487 The authors declare no competing interests

488 Bibliography 489 490 Arendt, K.L., Sarti, F., Chen, L., 2013. Chronic inactivation of a neural circuit enhances LTP by 491 inducing silent synapse formation. J. Neurosci. 33, 2087–2096. 492 https://doi.org/10.1523/JNEUROSCI.3880-12.2013

493 Barria, A., Malinow, R., 2005. NMDA receptor subunit composition controls synaptic 494 plasticity by regulating binding to CaMKII. Neuron 48, 289–301. 495 https://doi.org/10.1016/j.neuron.2005.08.034

496 Barrionuevo, G., Brown, T.H., 1983. Associative long-term potentiation in hippocampal 497 slices. Proc. Natl. Acad. Sci. 80, 7347–7351. https://doi.org/10.1073/pnas.80.23.7347

498 Branco, T., Staras, K., Darcy, K.J., Goda, Y., 2008. Local Dendritic Activity Sets Release 499 Probability at Hippocampal Synapses. Neuron 59, 475–485. 500 https://doi.org/10.1016/j.neuron.2008.07.006

501 Chklovskii, D.B., Mel, B.W., Svoboda, K., 2004. Cortical rewiring and information storage. 502 Nature 431, 782–788. https://doi.org/10.1038/nature03012

503 Desai, N.S., Cudmore, R.H., Nelson, S.B., Turrigiano, G.G., 2002. Critical periods for 504 experience-dependent synaptic scaling in visual cortex. Nat. Neurosci. 5, 783–789. 505 https://doi.org/10.1038/nn878

506 Desai, N.S., Rutherford, L.C., Turrigiano, G.G., 1999. Plasticity in the intrinsic excitability of 507 cortical pyramidal neurons. Nat. Neurosci. 2, 515–520. https://doi.org/10.1038/9165

508 Erdil, E., Yagci, A.M., Argunsah, A.O., Ramiro-Cortes, Y., Hobbiss, A.F., Israely, I., Unay, D., 509 2012. A Tool for Automatic Dendritic Spine Detection and Analysis . Part I : Dendritic 510 Spine Detection Using. Proc. Int. Conf. Image Process. Theory, Tools {&} Appl.

511 Fé lix-Oliveira, A., Dias, R.B., Colino-Oliveira, M., Rombo, D.M., Sebastião, A.M., 2014. 512 Homeostatic plasticity induced by brief activity deprivation enhances long-term 513 potentiation in the mature rat hippocampus. J. Neurophysiol. 112, 3012–3022. 514 https://doi.org/10.1152/jn.00058.2014

515 Fernandes, D., Carvalho, A.L., 2016. Mechanisms of homeostatic plasticity in the excitatory 516 synapse. J. Neurochem. 139, 973–996. https://doi.org/10.1111/jnc.13687

517 Fu, M., Yu, X., Lu, J., Zuo, Y., 2012. Repetitive motor learning induces coordinated formation 518 of clustered dendritic spines in vivo. Nature 483, 92–95. 519 https://doi.org/10.1038/nature10844

520 Ghani, M.U., Mesadi, F., Kanik, S.D., Argunsah, A.O., Hobbiss, A.F., Israely, I., Unay, D., 521 Tasdizen, T., Cetin, M., 2017. Dendritic spine classification using shape and appearance 522 features based on two-photon microscopy. J. Neurosci. Methods 279, 13–21. 523 https://doi.org/https://doi.org/10.1016/j.jneumeth.2016.12.006

524 Govindarajan, A., Israely, I., Huang, S.-Y., Tonegawa, S., 2011. The dendritic branch is the 525 preferred integrative unit for protein synthesis-dependent LTP. Neuron 69, 132–46. 526 https://doi.org/10.1016/j.neuron.2010.12.008

527 Govindarajan, A., Kelleher, R.J., Tonegawa, S., 2006. A clustered plasticity model of long- 528 term memory engrams. Nat. Rev. Neurosci. 7, 575–583. 529 https://doi.org/10.1038/nrn1937

530 Harvey, C.D., Svoboda, K., 2007. Locally dynamic synaptic learning rules in pyramidal neuron 531 dendrites. Nature 450, 1195–1200. https://doi.org/10.1038/nature06416

532 Harvey, C.D., Yasuda, R., Zhong, H., Svoboda, K., 2008. The spread of Ras activity triggered 533 by activation of a single dendritic spine. Science 321, 136–140. 534 https://doi.org/10.1126/science.1159675

535 Inouye, S., Ogawa, H., Yasuda, K., Umesono, K., Tsuji, F.I., 1997. A bacterial cloning vector 536 using a mutated Aequorea green fluorescent protein as an indicator. Gene 189, 159– 537 162. https://doi.org/10.1016/S0378-1119(96)00753-6

538 Ju,., W Morishita, W., Tsui, J., Gaietta, G., Deerinck, T.J., Adams, S.R., Garner, C.C., Tsien, 539 R.Y., Ellisman, M.H., Malenka, R.C., 2004. Activity-dependent regulation of dendritic 540 synthesis and trafficking of AMPA receptors. Nat. Neurosci. 7, 244–53. 541 https://doi.org/10.1038/nn1189

542 Karmarkar, U.R., Buonomano, D. V, 2006. Different forms of homeostatic plasticity are 543 engaged with distinct temporal profiles. Eur. J. Neurosci. 23, 1575–84. 544 https://doi.org/10.1111/j.1460-9568.2006.04692.x

545 Kleindienst, T., Winnubst, J., Roth-Alpermann, C., Bonhoeffer, T., Lohmann, C., 2011.

546 Activity-Dependent Clustering of Functional Synaptic Inputs on Developing 547 Hippocampal Dendrites. Neuron 72, 1012–1024. 548 https://doi.org/10.1016/j.neuron.2011.10.015

549 Lee, M.-C., Yasuda, R., Ehlers, M.D., 2010. Metaplasticity at single glutamatergic synapses. 550 Neuron 66, 859–870. https://doi.org/10.1016/j.neuron.2010.05.015

551 LeMasson, G., Marder, E., Abbott, L.F., 1993. Activity-dependent regulation of conductances 552 in model neurons. Science 259, 1915–1917. https://doi.org/10.1126/science.8456317

553 Liu, G., 2004. Local structural balance and functional interaction of excitatory and inhibitory 554 synapses in hippocampal dendrites. Nat. Neurosci. 7, 373. 555 https://doi.org/10.1038/nn1206

556 Makino, H., Malinow, R., 2011. Compartmentalized versus global synaptic plasticity on 557 dendrites controlled by experience. Neuron 72, 1001–1011. 558 https://doi.org/10.1016/j.neuron.2011.09.036

559 Malenka, R.C., Bear, M.F., 2004. LTP and LTD: an embarrassment of riches. Neuron 44, 5–21. 560 https://doi.org/10.1016/j.neuron.2004.09.012

561 Matsuzaki, M., Ellis-Davies, G.C.R., Nemoto, T., Miyashita, Y., Iino, M., Kasai, H., 2001. 562 Dendritic spine geometry is critical for AMPA receptor expression in hippocampal CA1 563 pyramidal neurons. Nat. Neurosci. 4, 1086–1092. https://doi.org/10.1038/nn736

564 Matsuzaki, M., Honkura, N., Ellis-Davies, G.C.R., Kasai, H., 2004. Structural basis of long-term 565 potentiation in single dendritic spines. Nature 429, 761–766. 566 https://doi.org/10.1038/nature02594.1.

567 Miller,.D., K MacKay, D.J.C., 1994. The Role of Constraints in Hebbian learning. Neural 568 Comput. 6, 100–126. https://doi.org/https://doi.org/10.1162/neco.1994.6.1.100

569 Murthy, V.N., Schikorski, T., Stevens, C.F., Zhu, Y., 2001. Inactivity produces increases in 570 neurotransmitter release and synapse size. Neuron 32, 673–682. 571 https://doi.org/10.1016/S0896-6273(01)00500-1

572 O’Brien, R.J., Kamboj, S., Ehlers, M.D., Rosen, K.R., Fischbach, G.D., Huganir, R.L., 1998. 573 Activity-dependent modulation of synaptic AMPA receptor accumulation. Neuron 21, 574 1067–1078. https://doi.org/https://doi.org/10.1016/S0896-6273(00)80624-8

575 Oh, W.C., Hill, T.C., Zito, K., 2013. Synapse-specific and size-dependent mechanisms of spine 576 structural plasticity accompanying synaptic weakening. Proc. Natl. Acad. Sci. U. S. A. 577 110, E305-12. https://doi.org/10.1073/pnas.1214705110

578 Oh, W.C.C., Parajuli, L.K.K., Zito, K., 2015. Heterosynaptic structural plasticity on local 579 dendritic segments of hippocampal CA1 neurons. Cell Rep. 10, 162–169. 580 https://doi.org/10.1016/j.celrep.2014.12.016

581 Paoletti, P., Bellone, C., Zhou, Q., 2013. NMDA receptor subunit diversity: impact on 582 receptor properties, synaptic plasticity and disease. Nat. Rev. Neurosci. 14, 383–400. 583 https://doi.org/10.1038/nrn3504

584 Pettit, D.L., Wang, S.S.H., Gee, K.R., Augustine, G.J., 1997. Chemical two-photon uncaging: A 585 novel approach to mapping glutamate receptors. Neuron 19, 465–471. 586 https://doi.org/10.1016/S0896-6273(00)80361-X

587 Ramiro-Cortés, Y., Hobbiss, A.F., Israely, I., 2014. Synaptic competition in structural plasticity 588 and cognitive function. Philos. Trans. R. Soc. Lond. B. Biol. Sci. 369, 20130157. 589 https://doi.org/10.1098/rstb.2013.0157

590 Ramiro-Cortés, Y., Israely, I., 2013. Long Lasting Protein Synthesis- and Activity-Dependent 591 Spine Shrinkage and Elimination after Synaptic Depression. PLoS One 8.

592 Schanzenbacher, C.T., Langer, J.D., Schuman, E.M., 2018. Time- and polarity-dependent 593 proteomic changes associated with homeostatic scaling at central synapses. Elife 1–20. 594 https://doi.org/10.7554/eLife.33322

595 Soares, C., Lee, K.F.H., Béïque, J.-C., 2017. Metaplasticity at CA1 Synapses by Homeostatic 596 Control of Presynaptic Release Dynamics. Cell Rep. 21, 1293–1303. 597 https://doi.org/10.1016/j.celrep.2017.10.025

598 Sutton, M. a, Ito, H.T., Cressy, P., Kempf, C., Woo, J.C., Schuman, E.M., 2006. Miniature 599 neurotransmission stabilizes synaptic function via tonic suppression of local dendritic 600 protein synthesis. Cell 125, 785–799. https://doi.org/10.1016/j.cell.2006.03.040

601 Thiagarajan, T.C., Lindskog, M., Tsien, R.W., 2005. Adaptation to synaptic inactivity in 602 hippocampal neurons. Neuron 47, 725–737. 603 https://doi.org/10.1016/j.neuron.2005.06.037

604 Turrigiano, G., 2011. Too Many Cooks? Intrinsic and Synaptic Homeostatic Mechanisms in 605 Cortical Circuit Refinement. Annu. Rev. Neurosci. 34, 89–103. 606 https://doi.org/10.1146/annurev-neuro-060909-153238

607 Turrigiano, G.G., Leslie, K.R., Desai, N.S., Rutherford, L.C., Nelson, S.B., 1998. Activity- 608 dependent scaling of quantal amplitude in neocortical neurons. Nature 391, 892–896. 609 https://doi.org/10.1038/36103

610 Vitureira, N., Goda, Y., 2013. The interplay between hebbian and homeostatic synaptic 611 plasticity. J. Cell Biol. 203, 175–186. https://doi.org/10.1083/jcb.201306030

612 Wallace, W., Bear, M.F., 2004. A Morphological Correlate of Synaptic Scaling in Visual 613 Cortex. J. Neurosci. 24, 6928–6938. https://doi.org/10.1523/JNEUROSCI.1110-04.2004

614 Yadav, A., Gao, Y.Z., Rodriguez, A., Dickstein, D.L., Wearne, S.L., Luebke, J.I., Hof, P.R., 615 Weaver, C.M., 2012. Morphologic evidence for spatially clustered spines in apical 616 dendrites of monkey neocortical pyramidal cells. J Comp Neurol 520, 2888–2902. 617 https://doi.org/10.1002/cne.23070

618

619 Figure 1

620 Figure 1. Spines undergo structural scaling following activity blockade 621 a) Experimental protocol to induce HSP in hippocampal organotypic slices. b) Example mEPSC 622 recording traces from control and TTX-treated conditions at the 48 h time point. Bottom: 623 representative individual mEPSCs. c) Quantification of mEPSC amplitudes after 48 h of activity 624 blockade, represented as mean amplitudes ± SEM. ***p = 1.553e-08, Mann-Whitney. For all 625 data presented, total number of spines or mEPSCs are represented by the first number (n) 626 and followed by the number of independent neurons (N = in parentheses) from which they 627 were collected. Number of mEPSCs (n) from independent neurons (N): Control = 552 (8), TTX 628 = 742(8). d) Ranked control mEPSC amplitudes plotted against TTX condition at 48 h. Best fit 629 line to the data: TTX = Control x 1.45 – 6.06. e) Representative images of dendrites at either 630 0 or 48 h after TTX treatment or in control media. Asterisks indicate all the spines that were 631 analyzed. Scale bar = 5µm. f) Quantification of spine volumes for the indicated conditions. 632 ***p = 0.91e-10, Two-way ANOVA, post-hoc Bonferroni. Number of spines (n) from 633 independent neurons (N) for the 0 h, 24 h and 48 h timepoints: Control = 103(7), 245(8), 285 634 (7). TTX = 172 (7), 232 (5), 208 (6). g) Ranked control spine volumes plotted against TTX 635 condition at 48 h. Best fit line to the data: TTX = Control2 x 5.44 + Control x 0.25 + 0.55. h) 636 Quantification of mEPSC amplitudes after 72 h in TTX. ***p = 1.883e-15, Mann-Whitney. 637 Number of mEPSCs (n) from independent neurons (N): Control = 1164 (10), TTX = 1302 (12). 638 i) Quantification of spine volume after 72 h in TTX. ***p = 0.0008, Mann-Whitney. Number of 639 spines (n) from independent neurons (N): Control = 258 (6), TTX = 214 (6). j) Ranked control 640 72 h mEPSC amplitudes plotted against TTX mEPSC amplitudes. Best fit line to the data: TTX 641 = Control x 1.10 – 0.01. k) Ranked control 72 h spine volumes plotted against TTX 72 h spine 642 volumes. Best fit line to the data: TTX = Control x 1.26 – 0.004). 643

644 Figure 2

645

646 Figure 2 Structural and functional changes after HSP are reversible 647 a) Experimental timeline for reversal experiments. b) Example traces showing cell firing 648 following 48 h of activity block. Spikes are demarcated with black circles. c) Quantification of

649 spikes per minute throughout experiment. *p48 control – 48 TTX = 0.03, Kruskal-Wallis post hoc

650 Dunn. **p48 TTX – 48 post TTX = 0.010, Kruskal-Wallis post hoc Dunn. Number of neurons: 48h TTX 651 condition: 48 h Control = 17, 48 h Acute TTX = 20, 48 h TTX = 26; 48h ‘post-TTX’ Control = 15, 652 48 h post-TTX Reversal = 18. d) Representative images from dendrites after 48 h of HSP 653 induction and at 48 h post TTX. Scale bar = 5µm. e) Quantification of spine sizes throughout 654 the experiment. Graph is plotted as mean ± SEM. All p values calculated using Kruskal-Wallis

655 test, post-hoc Dunn’s. ***p48 TTX – 48 post TTX = 4.1852e-07. Number of spines (n = first number) 656 of Neurons (N = in parentheses): 48 h Control = 285 (7), 48 h TTX = 208 (6), 48 h ‘post-TTX’ 657 Control = 514 (14), 48 h post-TTX = 430 (15). 658

659 Figure 3 660 661

662 Figure 3. Potentiation of single inputs is enhanced at smaller spines following HSP 663 a) Timeline for uncaging experiments. b) Representative images of structural plasticity 664 following single spine glutamate uncaging. Arrowheads and dots indicate stimulated and 665 neighboring non-stimulated spines respectively. Scale bar = 5µm (main image), 1 µm (inset). 666 c) Normalized volume of stimulated and neighboring spines for control and TTX-treated 667 conditions over 2 hours of LTP induction. Volumes were calculated from Z-stacks taken in 5 668 minute bins from time point 0 (thus the stack represented as 0 was collected between 0-5 669 minutes, for example). For all data, the total number of spines are represented by the first 670 number (n), and followed by the number of independent neurons (N = in parentheses) from 671 which they were collected. Control Stimulated = 17 (17), TTX Stimulated = 18 (18), Control 672 neighbors = 85 (17), TTX neighbors = 102 (18). d) All analyzed stimulated and neighboring 673 spine volumes at time points 45 minutes and 120 minutes post stimulation. Significance was

674 calculated using one-way ANOVA post-hoc Bonferroni. Control condition: **pstimulated45 –

675 neighbors45 = 8.4246e-04. pstimulated120 – neighbors120 = 0.6881. TTX condition: ***pstimulated45 – neighbors45

676 = 2.9383e-04. **pstimulated120 – neighbors120 = 0.0046. (n same as in panel c). e) Initial absolute spine 677 size vs. normalized spine volume at 120 minutes. TTX spines linear regression: r2 = 0.48, **p 678 = 0.002. Control linear regression: r2 = 0.06, p = 0.346. f) Distribution of all initial volumes of 679 stimulated spines, with example spine images. The cut-off for defining ‘large’ spines was the 680 median volume off all spines multiplied by 1.5. Scale bar = 1 µm. g) Time-course of the 681 structural LTP experiment plotting only the small spines. *p = 0.0458, 2-way repeated 682 measures ANOVA). Number of spines (n) from neurons (N = in parentheses): Control = 13 (13), 683 TTX = 11 (11). h) Time-course of structural LTP plotting only large spines. p = 0.699, 2-way 684 repeated measures ANOVA. Number of spines (n) from neurons (N = in parentheses): Control 685 = 4 (4), TTX = 17 (17). 686

687 Figure 4

688 Figure 4. HSP increases the efficacy and reduces the threshold for induction of LTP at 689 individual spines 690 a) Representative image of a single spine during the course of uncaging. Scale bar = 5µm (main 691 image), 1µm (inset). b) Individual traces of spine growth during glutamate uncaging 692 stimulation. Control spine growth vs TTX spine growth: p = 0.198, 2-way repeated measures 693 ANOVA. Number of spines (n): Control = 16, TTX = 18. c) Growth rates of stimulated spines

694 for 1 minute after the end of stimulation. *psmall control–small TTX = 0.018, 1-way ANOVA with post-

695 hoc Tukey’s. plarge control–large TTX < 0.99, 1-way ANOVA with post-hoc Tukey’s. Number of spines 696 (n): Control small = 13, TTX small = 11, Control large = 4, TTX large = 7. d) Quantification of 697 normalized spine volume with different stimulation pulse durations in the control condition.

698 p1ms stimulated–1ms neighbors = 0.292, 2-way repeated measures ANOVA post-hoc Bonferroni. 699 Number of spines (n) from Neurons from which they were collected (N = in parentheses):

700 stimulated 1ms = 11 (11), neighbors 1ms = 52 (11), stimulated 4ms = 17 (17). *p1ms stimulated–

701 4ms stimulated = 0.039 2-way rm ANOVA. Note 4ms data is the same as shown in Figure 3. e) 702 Quantification of normalized spine volume with different stimulation pulse durations in the

703 TTX condition. **p1ms stimulated–1ms neighbors = 0.007, 2-way rm ANOVA. p1ms stimulated– 4ms stimulated = 704 0.27, 2-way rm ANOVA). Number of spines (n) from Neurons (N): stimulated 1ms = 11 (11), 705 neighbors 1ms = 52 (11), stimulated 4ms = 18 (18). Note 4ms data is the same as in Figure 3.

706 Figure 5

707 Figure 5. Stimulation of single inputs leads to plasticity of clustered neighbors after HSP 708 a) Heat maps of growth dynamics of neighboring spines over time. Neighbors from all 709 experiments were pooled and ranked in order distance from the stimulated spine. Index 1 is 710 the spine nearest to its stimulated spine, and 80 is the furthest away. Heat map = normalized 711 spine volume. Number of spines (n) from Neurons from which they were collected (N = in 712 parentheses): Control = 80 (17), TTX = 80 (18). b) Mean volumes for the first 10 minutes

713 following stimulation for the different neighboring conditions. ***pTTX near – TTX far = 1.0674e- 714 05, Kruskal-Wallis. Number of spines (n) from Neurons from which they were collected (N = 715 in parentheses): Control near = 22 (17), control far = 63 (17), TTX near = 32 (18), TTX far = 70 716 (18). c) Normalized volumes of near and far neighboring spines in the control condition. p = 717 0.987, 2-way rm ANOVA. Number of spines – same as above. d) Normalized volumes of 718 neighboring spines in the TTX condition. *p = 0.0124, 2-way repeated measures ANOVA. 719 Number of spines – same as above. e) & f) Correlation coefficients of volume change of 720 stimulated spines vs volume change of neighboring spine, plotted against the distance 721 between the neighboring spine and the stimulated spine. e) Linear regression control: r2 = 722 0.003, p = 0.641. f) Linear regression TTX: r2 = 0.156, ***p = 3.91e-05).