Local Order within Global Disorder: Synaptic Architecture of Visual Space

Article

Local Order within Global Disorder: Synaptic Architecture of Visual Space Highlights

Authors

Spine receptive fields are diverse and do not show global dendritic organization

Benjamin Scholl, Daniel E. Wilson, David Fitzpatrick

Neighboring spines separated by less than 10 mm share receptive field properties

Correspondence

d

Local spines exhibit spontaneous and sensory-driven coactivity

In Brief

d

Dimensionality reduction reveals functional synaptic clusters in spine populations

d

d

Scholl et al., 2017, Neuron 96, 1–12 December 6, 2017 ª 2017 Elsevier Inc. https://doi.org/10.1016/j.neuron.2017.10.017

[email protected]

Scholl et al. report that the dendritic spines of pyramidal neurons in ferret visual cortex are functionally clustered at a scale of 5–10 mm, exhibiting common visual response properties and correlation in their spontaneous and sensory driven activity.

Please cite this article in press as: Scholl et al., Local Order within Global Disorder: Synaptic Architecture of Visual Space, Neuron (2017), https:// doi.org/10.1016/j.neuron.2017.10.017

Neuron

Article Local Order within Global Disorder: Synaptic Architecture of Visual Space Benjamin Scholl,1 Daniel E. Wilson,1,2 and David Fitzpatrick1,3,* 1Max

Planck Florida Institute for Neuroscience, 1 Max Planck Way, Jupiter, FL 33458, USA Biology and Neuroscience Graduate Program, International Max Planck Research School for Brain and Behavior, Florida Atlantic University, Jupiter, Florida, USA 3Lead Contact *Correspondence: [email protected] https://doi.org/10.1016/j.neuron.2017.10.017 2Integrative

SUMMARY

Substantial evidence at the subcellular level indicates that the spatial arrangement of synaptic inputs onto dendrites could play a significant role in cortical computations, but how synapses of functionally defined cortical networks are arranged within the dendrites of individual neurons remains unclear. Here we assessed one-dimensional spatial receptive fields of individual dendritic spines within individual layer 2/3 neuron dendrites. Spatial receptive field properties of dendritic spines were strikingly diverse, with no evidence of large-scale topographic organization. At a fine scale, organization was evident: neighboring spines separated by less than 10 mm shared similar spatial receptive field properties and exhibited a distance-dependent correlation in sensory-driven and spontaneous activity patterns. Fine-scale dendritic organization was supported by the fact that functional groups of spines defined by dimensionality reduction of receptive field properties exhibited non-random dendritic clustering. Our results demonstrate that functional synaptic clustering is a robust feature existing at a local spatial scale.

INTRODUCTION Dendritic compartmentalization of synaptic contacts is thought to play a critical role in how single neurons integrate synaptic inputs to perform functional computations. Specifically, the spatial clustering of synaptic contacts possessing similar functional properties is thought to enhance linear and nonlinear dendritic integration (Polsky et al., 2004; Losonczy et al., 2008; Branco et al., 2010; Lavzin et al., 2012; Palmer et al., 2014), facilitate biochemical signaling (Harvey et al., 2008; Nishiyama and Yasuda, 2015; Winnubst and Lohmann, 2012), and encourage cooperative plasticity between neighboring synaptic contacts (Harvey and Svoboda, 2007; Larkum and Nevian, 2008; Makino and Malinow, 2011; Nishiyama and Yasuda, 2015; Lee et al., 2016b). Although these studies postulate an important role for functional clustering of inputs in mechanisms of dendritic integration, a number of in vivo studies have failed to find evidence

for such clustering, raising doubts about the relevance of these dendritic mechanisms for cortical function (Jia et al., 2010b; Varga et al., 2011; Chen et al., 2011). Spontaneous activity patterns of dendritic spines have been shown to exhibit temporal correlation or co-activity at a spatial scale consistent with dendritic mechanisms (Takahashi et al., 2012; Lee et al., 2016b), but how correlations in spontaneous patterns of synaptic activity relate to the functional properties of these inputs is unclear. A departure from previous functional studies is the work by Wilson et al. (2016), which demonstrates clustering in the orientation preference of synaptic inputs that correlates with the sharpness of somatic tuning. Similar to this type of clustering in ferrets, a recent study has demonstrated synaptic clustering for retinotopy in mouse V1, but not for orientation preference (Iacaruso et al., 2017). Although both of these results are consistent with a role for functional synaptic clustering in shaping response properties of cortical neurons, whether the principle of synaptic clustering generalizes to the rich diversity of inputs a neuron receives remains unclear. Even individual inputs with similar orientation preference can differ in a number of other features, such as optimal location in visual space, sign of contrast (ON versus OFF), direction selectivity, and ocular preference (Lennie, 1998). Here we assess several different receptive field properties that characterize individual synaptic inputs, evaluate evidence for synaptic clustering, and determine whether clustering exists on a scale consistent with the proposed mechanisms of dendritic integration (Larkum and Nevian, 2008; Major et al., 2013). To address these issues, we used a simple visual stimulus configuration—an edge stimulus at the cell’s preferred orientation—to map the receptive field (RF) center and polarity (ON versus OFF) of the synaptic inputs to large numbers of spines on individual layer 2/3 neurons using in vivo two-photon imaging of a genetically encoded calcium indicator, GCaMP6s (Chen et al., 2013). Spines on individual neurons exhibited a high degree of diversity in spatial organization and polarity of their receptive fields, and there was no apparent global organization within the dendritic field. Nevertheless, within this heterogeneity, fine-scale synaptic clustering exists for spatial location and polarity: spines separated by less than 10 mm were more likely than chance to share similar properties and exhibit significant distance-dependent correlation in their activity patterns. This fine-scale clustering was evident in patterns of spontaneous synaptic activity and trial-to-trial activity evoked by drifting gratings. Furthermore, we found that functional groups of spines defined by dimensionality reduction of spine populations from Neuron 96, 1–12, December 6, 2017 ª 2017 Elsevier Inc. 1

Please cite this article in press as: Scholl et al., Local Order within Global Disorder: Synaptic Architecture of Visual Space, Neuron (2017), https:// doi.org/10.1016/j.neuron.2017.10.017

Figure 1. Measuring 1D Spatial Receptive Fields of Individual Somata and Spines (A) Two-photon projection of a cortical cell soma and an example DF/F fluorescence trace during visual stimulation (left). Stimulus-averaged responses for each bar location in visual space revealed spatial and polarity selectivity (right). Stimulus onset is shown at the beginning of each response. Black lines and gray shading indicate mean and SE, respectively. Temporal specificity of responses to individual bars allows separation of ON (e.g., increase in luminance) or OFF (e.g., decrease in luminance) responses, shown as red and blue, respectively. Peak DF/F responses extracted from stimulus-triggered cycles were used to generate ON and OFF spatial RFs (right). Data were fit to a Gaussian curve. (B) The same as (A) for an individual spine with strong OFF responses to somatically-distal spatial locations. Spine residual DF/F fluorescence trace (green) shows large, independent, isolated calcium events. (C) The same as (B) for a spine (from the same cortical cell) with strong ON responses. (D) The same as (B) for a spine exhibiting ON-OFF responses. (E) The same as (B) for a spine with segregated ON and OFF responses. See also Figures S1–S3.

individual neurons were predictive of spatial clustering. Taken together, these results provide strong support for fine-scale functional clustering as a fundamental organizing principle of the spatial arrangement of synaptic inputs within the dendritic fields of cortical neurons. RESULTS Mapping Spatial Receptive Fields of Dendritic Spines To map the synaptic organization of visual space within the dendritic field, we used in vivo two-photon imaging of calcium dynamics following sparse adeno-associated virus (AAV) expression of a genetically encoded sensor (GCaMP6s) in layer 2/3 neurons of the ferret visual cortex (Wilson et al., 2016; STAR Methods). We measured the RF in the visual space of both soma and dendritic 2 Neuron 96, 1–12, December 6, 2017

spines (Figures 1A and 1B) for individual neurons using sparse one-dimensional (1D) bars of light (Movshon et al., 1978; Lampl et al., 2004). After determining the basic RF properties of the soma, we presented an array of light or dark bars aligned to the preferred orientation and centered on the somatic RF center. By design, stimuli were optimized for individual neurons and arranged so that only a subset of bars elicited calcium responses from the soma (Figure 1A, center), allowing us to probe locations in visual space driving both suprathreshold and subthreshold input extending beyond the spiking RF (Priebe and Ferster, 2008). Probing subthreshold input from the surrounding visual space restricts sensory-driven dendritic calcium signals in the form of back-propagating action potentials. Following characterization of the somatic RF, we serially imaged dendritic segments with visible spines. Spine calcium fluorescence was isolated from dendritic

Please cite this article in press as: Scholl et al., Local Order within Global Disorder: Synaptic Architecture of Visual Space, Neuron (2017), https:// doi.org/10.1016/j.neuron.2017.10.017

A

Example 1

B

Example 2

C

Distribution Spine Responses Center of Mass

Stimulus

25 μm -9°

25 μm

84% Overlap Position = -6o

0

F

-9°

G

On Off f

1 Overlap

y-position (μm)

E

Center of Mass Overlap 20

8°

0.5

Position = 3o

-20

Stimuli distance = 9o

-20

0 20 x-position (μm)

0

Norm. Response

9°

H

I

0.4 Proportion

D

25 μm

On Spines

J Distal Proximal

0.2

0 0

10 20 30 0 10 20 30 0 Bar Position Bar Position o Distance from Soma Distance ( ) Preference (o)

10 20 30 Bar Position o Distance ( )

0.4

Off ff Spines

Proportion

-8°

9°

Distal Proximal

0.2

0 0

10 20 30 0 10 20 30 Absolute Spatial Preference Difference Between Spine and Soma (o)

0

0.5

1

On-Off Ratio

Figure 2. Lack of Global Dendritic Organization for Spine Spatial Preferences (A and B) Example neuron with all serially imaged visually responsive spines and corresponding soma colored by spatial preference (scale bar, 50 mm). Spatial preferences are shown for dominant polarity for each spine. (B) The same as (A) for another cortical cell. (C) Distribution of normalized spine responses for a single bar location is shown with a dendritic center of mass (gray ellipse). Ellipse axes are two SDs of spatial patterns of synaptic responses. Inset: center of mass ellipses for each stimulus spatial location. (D) An example of the overlap of the center of mass ellipses calculated for two different spatial locations. (E) The relationship between bar position distance and center of mass overlap. (F) The relationship between bar position distance relative to the soma preference and center of mass overlap. (G) The relationship between bar position distance and center of mass overlap for ON (red) and OFF (blue) spatial patterns is shown separately. (H) The distribution of spine ON spatial preferences, relative to the soma, is shown at distal (>75 mm) and proximal (<75 mm) dendritic locations. (I) The same as (H) for OFF spatial preferences. (J) Distribution of spine ON-OFF ratios at distal and proximal dendritic locations. See also Figure S3.

shaft signals, and dendritic spine responses were extracted using published methods (Wilson et al., 2016). Spines used for subsequent analysis fulfilled several criteria (STAR Methods). Across all neurons imaged (n = 18), we recovered a total of 1,018 spines with reliable activity driven by bar stimuli (mean = 56.6 ± 37.9 SD spines per cell, median = 47.5 spines per cell). To quantify the spatial RF of spines and soma, we calculated the peak change in fluorescence driven by each stimulus. The onset and offset of bar stimuli were used to analyze ON and OFF responses evident in stimulus-averaged responses in both the soma (Figure 1A, center) and individual spines (Figure 1B, center). Peak ON and OFF responses were plotted as a function of stimulus location relative to the soma and fitted to a Gaussian curve to characterize RF properties (STAR Methods). Average peak responses evoked with a dendritic spine’s preferred stimulus were large in amplitude (mean DF/F = 86% ± 64% SD, median DF/F = 68%), and dendritic spines often displayed calcium events independent of parent dendrites. Individual dendritic spines were strongly selective for locations in visual space, and spines occupying the same dendritic branch frequently responded to a diverse set of locations, including

those not effective in evoking a somatic response (Figure S1; Movie S1). Many spines even responded selectively to a particular bar size (Figure S2). Compared with the somatic RF center, spine spatial preferences spanned a wide region of visual space (absolute value of spatial preference difference: ON range = 0
–36.0
, ON median = 4.8
; OFF range = 0
–33.6
, OFF median = 3.4
; Figure 2H). Although the population of single-spine RF centers spanned broad regions of visual space, synaptic aggregates for individual neurons could roughly capture somatic spatial preference, as measured by the pairwise difference between spatial preferences (ON: mean difference = 1.72
± 6.47
SD, OFF mean = 0.82
± 7.70
SD; Figure S3). However, predictably, synaptic aggregate RFs were consistently larger than their somatic counterparts (ON: spine mean full width half max [FWHM] = 13.6
± 13.5
SD, soma mean FWHM = 5.8
± 4.5
SD; OFF: spine mean FWHM = 17.8
± 15.8
SD, soma mean FWHM = 5.4
± 5.1
SD; Figure S3). Dendritic spines also differed greatly in their response to stimulus polarity and arrangement of ON and OFF spatial domains. Some spines were selective for a single polarity (Figures 1B and 1C), both polarities with spatial overlap (Figure 1D), or Neuron 96, 1–12, December 6, 2017 3

Please cite this article in press as: Scholl et al., Local Order within Global Disorder: Synaptic Architecture of Visual Space, Neuron (2017), https:// doi.org/10.1016/j.neuron.2017.10.017

both with spatial segregation (Figure 1E). This wide variety of ON and OFF responses was evident in both spine ON-OFF ratios (range = 0–1, median = 0.50; Figure 2J) and ON-OFF spatial overlap (range = 0.11–1.0, median = 0.53) (Lee et al., 2016a). Given this distribution of ON and OFF response properties, synaptic aggregates for individual neurons did not match somatic polarity sensitivity because synaptic aggregates were generally balanced in ON and OFF contribution (soma ON-OFF ratio: range = 0.10–0.92, median = 0.49, synaptic aggregate ON-OFF ratio: range = 0.42–0.64, median = 0.47; Figure S3). Because the offset of white bars may not drive spines equivalent to the onset of opposing black bars (Movshon et al., 1978), in a subset of cells, we presented both polarities (298 spines from six cells). Although distributions of spine ON-OFF ratios and ON-OFF overlap between stimulus sets were similar (single polarity: ratio range = 0–1.0, overlap range = 0.11–1.0; both polarities: ratio range = 0–1.0, overlap range = 0.16–1.0), these distributions were statistically different from each other (ON-OFF ratio: single polarity median = 0.49 with an interquartile range (IQR) of 0.27, both polarities median = 0.58 with an IQR of 0.35, p = 1.80e9; ON-OFF overlap: single polarity median = 0.49 with an IQR of 0.28, both polarities median = 0.67 with an IQR of 0.30, p = 3.97e32; Wilcoxon rank-sum test). Absence of Large-Scale Dendritic Organization Before investigating the fine-scale organization of dendritic spine spatial preferences, we sought to determine whether order was present at larger scales: within individual dendritic branches or across the dendritic field. A significant degree of regional homogeneity in dendritic spine spatial preference would limit the usefulness of exploring the representation of visual space as a means to assess fine-scale dendritic organization and clustering diversity. Although, in the mouse visual cortex, synaptic inputs are reported to be retinotopically organized within the dendritic field of individual neurons (Iacaruso et al., 2017), in the ferret visual cortex we could not discern any obvious order in the large-scale topology of the representation of visual space when considering individual spine spatial preferences and corresponding locations within the dendritic arbors (Figures 2A and 2B). Spines responding to spatially disparate locations appeared to be scattered throughout the dendritic field, and individual branches exhibited spines representing the full range of spatial locations stimulated (Figure S1). To systematically test for large-scale order or dendritic topology for visual space, we examined the spatial organization of spine responses at each stimulus location (Figure 2C). For simplicity, we began by averaging ON and OFF responses for each spine to exclude stimulus polarity and then computed a center of mass with respect to spine locations within the dendritic arbors and their normalized responses for each visuotopic location (Figure 2C). If there exists an orderly global representation of visuotopic space in the synaptic inputs to single neurons, then stimuli at opposing ends of the somatic spatial RF should evoke distinct spatial patterns of synaptic activity, spatially offset within the dendrites. Thus, we computed the percentage of overlap between spatial patterns of synaptic input relative to their stimulus distances, using center of mass ellipses (Figure 2D). There was no relationship between the center of mass overlap and the dis4 Neuron 96, 1–12, December 6, 2017

tance between bar positions (slope = 0.003, r2 = 0.041, 1,162 comparisons from 18 cells; Figure 2E). We then tested the possibility that the representation of visual space in synaptic populations was relative to somatic selectivity by examining overlap as a function of distance from somatic preference. Here we found no relationship (slope = 0.009, r2 = 0.042, 168 comparisons from 18 cells; Figure 2F). Because these data excluded polarity, we also measured the center of mass for ON and OFF responses separately and found no relationship (ON: slope = 0.004, r2 = 0.006; OFF: slope = 0.003, r2 = 0.006; 1,162 comparisons from 18 cells for both; Figure 2G). It is important to note that we cannot rule out the possibility of an orderly topology along axes of visual space other than orthogonal to a cell’s preferred orientation, as measured here. In an effort to determine whether some large-scale order in the arrangement of functional inputs escaped detection with this analysis, we asked whether spine response properties exhibited systematic order as a function of distance from the soma. For example, spines closer to the soma might exhibit a greater degree of similarity to the soma than those located at more peripheral locations in the dendritic field (Iacaruso et al., 2017). For each spine we measured the shortest dendritic path distance to the soma and then compared the RF properties for proximal (<75 mm) and distal (>75 mm) spine populations. We found no significant difference between these two populations in the degree to which their responses resembled the somatic response, either for ON or OFF spatial preferences relative to the soma (ON: proximal median = 4.8 with an IQR of 7.3, n = 223, distal median = 4.6 with an IQR of 6.5, n = 287, p = 0.287; OFF: proximal median = 3.6 with an IQR of 5.4, n = 190, distal median = 3.2 with an IQR of 5.0, n = 318, p = 0.131; Wilcoxon rank-sum test; Figures 2H and 2I). We then considered that there might be proximal/distal asymmetry in the spatial distribution of ON and OFF inputs and found a small but statistically significant difference between ON-OFF ratios (proximal median = 0.52 with an IQR of 0.30, n = 413, distal median = 0.49 with an IQR of 0.28, n = 605, p = 6.4e3; Wilcoxon rank-sum test; Figure 2J). Together, these observations demonstrate a global, distributed synaptic representation of visual space within the dendritic arbors of cortical neurons, an organization similar to that described for other cortical response properties, such as orientation (Jia et al., 2010b; Chen et al., 2013; Wilson et al., 2016), tonotopy (Chen et al., 2011), and somatotopy (Varga et al., 2011). Given the lack of large-scale order, we turned our attention to probing the fine-scale synaptic arrangement of visual space within individual dendrites. Local Dendritic Organization of Synaptic Spatial Receptive Fields Despite the large amount of diversity in the synaptic inputs to single branches, we frequently observed nearby dendritic spines with similar preferences in visual space and stimulus polarity (Figure 3A, left). In some cases, spines were located adjacent to one another on the dendritic shaft; spines 1 and 2, for instance, had a strong OFF response 6
from the somatic RF center (Figure 3A). In other cases, spines were located opposite each other on the dendritic shaft, such as spines 3 and 4, which preferred white bars presented 9
from the somatic RF center (Figure 3A). Furthermore, although we did observe shared

Please cite this article in press as: Scholl et al., Local Order within Global Disorder: Synaptic Architecture of Visual Space, Neuron (2017), https:// doi.org/10.1016/j.neuron.2017.10.017

5

6

4 3

5 μm -9°

9°

Spine 3

1.6

0

0 Spine 2

0

0.3

Spine 4

1

1

0.1 0 5 10 15 Dendritic Distance (μm)

Spine 6

C 0.4

0 -9-6-3 0 3 6 9 Spatial Location

λ = 5.3 μm

0.2

0

0 -9-6-3 0 3 6 9 Spatial Location

1.4

Spine 5

Tuning Correlation

1.2

Spine 1

B ON OFF

Proportion

1

Response (∆F/F)

2

Soma

0.6

Response (∆F/F)

A Local Spatial Preference and Tuning Correlation

-9-6-3 0 3 6 9 Spatial Location

Clustered Spine Pairs All Spines

s

0

-0.5 0 0.5 1 Correlation with Soma RF

Figure 3. Local Clustering of Spine Spatial RF Properties (A) Example dendritic branch with spines sharing similar spatial preference and RF properties. Spines are colored based on spatial preference for ON and OFF. Spines with multiple colors indicate strong ON and OFF responses with different spatial preference. Spatial ON and OFF DF/F responses (mean and SE) are shown with a fitted Gaussian curve. (B) Tuning correlation between spines is dependent on dendritic distance. Data shown are mean and SE with an exponential curve fit (black line). A bootstrapped shuffled correlation is also shown (gray). (C) The distribution of the mean correlation of clustered spine pairs (correlation > 0.5, distance < 10 mm) with soma RF (black outline). Also shown is the distribution of RF correlation between all individual spines and somata (gray). See also Figure S4.

synaptic RF properties between nearby dendritic spines, these spines were intermixed with those exhibiting a wide range of spatial selectivity (e.g., spines 5 and 6). To systematically characterize functional clustering, we computed a pairwise cross-correlation of spine spatial RFs (STAR Methods) and distance along the dendritic shaft. Across all contiguous dendritic branches imaged (933 spines from 116 segments, mean = 8.04 ± 5.9 SD per dendrite), we found that spines less than 5 mm apart exhibited greater similarity in their functional responses than those farther away (Figure 3B). Importantly, local spines exhibited greater tuning correlation than explained by randomly shuffling spines within their respective contiguous dendritic segments (0–2 mm: mean correlation = 0.28 ± 0.03 SE, mean shuffled correlation = 0.17 ± 0.01 SE; Figure 3B). An exponential fit to these data (STAR Methods) revealed that the length constant of functionally defined synaptic clusters was 5 mm, similar to recent measurements of 2D RF correlation in mouse V1 (Iacaruso et al., 2017). Tuning correlation, which accounts for multiple aspects of spine RFs, was also strongly related to the difference in spatial preferences (Figure S4). Because we previously demonstrated synaptic clustering aligned to somatic orientation preferences (Wilson et al., 2016), clusters might preferentially represent somatic properties. By examining only nearby, correlated spine pairs with similar functional responses (distance, <10 mm; correlation, >0.5; n = 271 pairs), we computed the average RF correlation of the somatic RF with spine pairs (Figure 3C). The majority of spine pairs, potentially comprising synaptic clusters, did not strongly reflect somatic RF properties (median correlation = 0.25, IQR = 0.54, n = 271), and this distribution was not statistically different than

the RF correlation between all individual spines and somata (median correlation = 0.17, IQR = 0.51, n = 933, p = 0.11, Wilcoxon rank-sum test; Figure 3C). These data provide compelling evidence that local synaptic clustering spans a range of functional spatial RF properties. Functional similarity between neighboring spines suggests that an appropriate visual stimulus co-activates spines and that this co-activation could show a similar distance dependence. Because our stimulus generates brief synaptic calcium transients (Figures 1B–1E), we were able to observe co-active synaptic calcium events on a trial-by-trial basis (Figure 4A). Nearby spines with matching spatial preferences were sometimes strongly co-activated by repeated presentations of the same bar (Figure S5A; Movie S2). Synaptic events, depicted in Figure 4A, were identified as single-trial events with amplitudes greater than five SDs of the baseline calcium fluorescence (Figure S5B; STAR Methods). Across dendritic branches imaged simultaneously (n = 921 spines from 123 segments, mean = 7.5 ± 4.9 SD spines per dendritic segment) we found a strong distance-dependent relationship between spine trial-to-trial correlation and dendritic distance (Figure 4B). As for RF similarity, nearby spines were significantly correlated compared with a within-segment shuffle (0–2 mm: mean correlation = 0.23 ± 0.02 SE, n = 200 spine comparisons, mean shuffled correlation = 0.14 ± 0.004 SE, n = 1,000 iterations, p = 1.38e32, Welch’s t test); the spatial length constant was 5 mm. This relationship was observed even when excluding synaptic trials correlated with the dendritic signal (exclusion criteria: >0.15 correlation; 0–2 mm: mean correlation = 0.17 ± 0.02 SE, n = 200 spine comparisons, mean shuffled correlation = 0.07 ± 0.002 SE). We also computed the probability of co-active synaptic calcium events Neuron 96, 1–12, December 6, 2017 5

Please cite this article in press as: Scholl et al., Local Order within Global Disorder: Synaptic Architecture of Visual Space, Neuron (2017), https:// doi.org/10.1016/j.neuron.2017.10.017

A Local Spatial-Temporal Correlation Spatial Preferences

Trials

1

2

3

4

5

6

7

8

Synaptic Event Stimulus

Visual Space

C Co-Active Event Probability (10-3)

Trial-to-Trial Correlation

0.3 λ = 4.1 μm

0.2

0.1 0

5

10

15

Dendritic Distance (μm)

D

E 1

6 Trial-to-Trial Correlation

B

4

Number of Cells

5 μm

0.5 0

2

Mean λ = 5.4

6

3

0 0

5

10

15

Dendritic Distance (μm)

-0.5

0

0.5

Tuning Correlation

(Figure S5B; STAR Methods) and found a significantly greater probability of co-activity between spine pairs less than 5 mm apart (Figure 4C), although the spatial dependency for this relationship was weak. It is perhaps not surprising that we observed similar local synaptic clusters with both tuning and trial-to-trial correlation because there was a strong relationship between the two metrics for dendritic spines (bootstrapped principalcomponent analysis [PCA], slope = 0.29 ± 0.01 SE, r2 = 0.28; Figure 4D). Different cells might exhibit functional clustering at different spatial scales, possibly resulting from differences in measurement (sampling biases) or from genuine differences in physiological properties. Thus, we computed spatial length constants for each cell’s dendritic spines using spatiotemporal correlations. Of 18 cells, 13 had statistically significant length constants, as measured with a bootstrapped fit of an exponential function (STAR Methods). Of this subset, there was surprising consistency in extracted spatial length constants (Figure 4E); overall, these values matched our previous measurements across all spines (mean l = 5.4 mm ± 1.6 SD). Taken together, these data strongly support fine-scale local synaptic clustering within functionally diverse synaptic inputs arranged throughout the dendritic arbors. Local Clustering of Spontaneous Activity and Drifting Grating Responses Because activation patterns of dendritic spines during spatial mapping exhibited local clustering similar to that reported for spontaneous activity in the rodent visual (Winnubst et al., 2015) and somatosensory (Takahashi et al., 2012) cortex, we wondered whether dendritic spines on cortical cells in ferret V1 also display this innate organization. We collected long bouts of spontaneous activity (5–10 min) from neighboring dendritic spines containing events both independent and coincident with global dendritic calcium signals (Figures 5A and 5B). Across a population of 244 spines from 13 dendritic segments 6 Neuron 96, 1–12, December 6, 2017

1

0

5 10

Figure 4. Synaptic Clusters Exhibit Spatiotemporal Correlation (A) Example dendritic branch with responsive dendritic spines, colored by spatial preference. Coactive synaptic events occurred across repeated presentations of a bar at a single spatial location. (B) The trial-to-trial correlation between spines was dependent on dendritic distance. Data shown are mean and SE with an exponential curve fit (black line). A bootstrapped shuffled correlation is also shown (gray). See also Figure S5. (C) The probability of co-active synaptic events is plotted relative to spine dendritic distance. Data shown are mean and SE, and a bootstrapped shuffled correlation is shown in gray. (D) The relationship between spine tuning correlation and trial-to-trial correlation. Individual points (gray squares) are shown with mean and SE (black circles). (E) The distribution of spatial length constants was measured across cells.

Spatial Length Constants (λ)

(median = 19 spines/segment), we found a strong relationship between the temporal correlation of dendritic spine activity and the spine’s distance along the dendritic shaft, exhibiting a spatial scale of 5 mm (Figure 5C). The similarity in scale of synaptic clustering between mammalian species and for both spontaneous and driven activity suggests that fine-scale synaptic clustering might be a fundamental property. If so, then sensory inputs beyond simple, single bars of light should drive spatially confined patterns of spine activity. In a separate set of experiments, we examined trial-to-trial responses of dendritic spines to presentations of drifting gratings at multiple directions (Figure 5D). Responses during individual trials revealed that neighboring dendritic spines could exhibit strong co-activity that could occur independent of global dendritic calcium events (Figure 5E). Across a population of visually responsive dendritic spines (n = 127 from 15 dendritic segments, median = 7 spines/segment), we found that trial-to-trial correlations between spines were distance-dependent at a spatial scale of 5–10 mm (Figure 5F). A similar relationship was also observed when excluding synaptic trials correlated with the dendrite (exclusion criteria: >0.15 correlation; 0–2 mm: mean correlation = 0.26 ± 0.02 SE, n = 52 spine comparisons, mean shuffled correlation = 0.18 ± 0.006 SE). Taken together with spontaneous activity patterns and spatiotemporal activity driven by 1D spatial mapping, our data strongly support fine-scale local synaptic clustering as a pervasive feature of synaptic inputs onto cortical cells. Functionally Defined Synaptic Clusters with Dimensionality Reduction A strong test of the significance of fine-scale functional synaptic clustering is to start with an objective classification of a neuron’s spines into groups based solely on functional measures and ask whether these groups exhibit spatial clustering within the dendritic field at the spatial scale we have described. To identify functional synaptic groups within our synaptic populations, we

Please cite this article in press as: Scholl et al., Local Order within Global Disorder: Synaptic Architecture of Visual Space, Neuron (2017), https:// doi.org/10.1016/j.neuron.2017.10.017

A

C

B 1

0.3

6

7

11

14

7

13

8

2

5 9

1

λ = 4.6 μm

0.2

11

9 4

0.1

13

3

15 Dendrite

10 μm

D

E 4

5

100% ∆F/F 10 sec

F

Single Trial Responses to drifting grating (0o)

9

6 78 1

3

2

3

4

5 100% ∆F/F 2 sec

2 1

10 μm

6

7

8

9

Dendrite

Trial-to-Trial Correlation

5

12

10

Correlation

3 15

0 5 10 15 Dendritic Distance (μm)

0.28 λ = 6.9 μm

0.18

0.08 0 5 10 15 Dendritic Distance (μm)

Figure 5. Local Clustering of Spontaneous Activity and Drifting Grating Evoked Co-activity (A) Two-photon projection of an example dendritic branch with dendritic spines. (B) Time series traces of DF/F activity from a subset of dendritic spines (black) and corresponding global dendritic activity (gray). Note that spine time series traces are back propagating action potential (bAP)-subtracted, and an exponential filter has been applied (STAR Methods). (C) The relationship between spine correlation of spontaneous activity and distance along the dendritic shaft. Data are shown as mean and SE with an exponential curve fit (black line). A bootstrapped shuffled correlation is shown in gray. (D) Two-photon projection of an example dendritic segment. (E) Individual dendritic spine responses (black) to the presentation of a single oriented grating (0
) as well as the corresponding dendritic response (gray). (F) The trial-to-trial correlation between spines during drifting grating presentation is dependent on dendritic distance. Data are shown as mean and SE with an exponential curve fit (black line). A bootstrapped shuffled correlation is shown in gray.

turned to dimensionality reduction using PCA and k-means clustering. For this analysis, we only used cells with more than 30 spines (n = 10 cells, total spines = 731, mean = 73.1 ± 43.4 SD spines/cell) and chose the minimum number of principal components (PCs) explaining 60%–70% of the variance in each respective dataset (mean = 66.3% ± 4.2% SD). We first calculated Z-scored spine spatial ON and OFF responses (Figure 6A) and used PCA to generate a set of coefficients describing each spine in a low-dimensional subspace. Although PCA reduces the dimensionality of synaptic populations, it can be difficult to visualize more than three PCs at a time. For example, the population in Figure 6A is best described using five PCs, but only three can be readily visualized at once. To visualize PC coefficients in a 2D coordinate system, we employed the t-distributed stochastic neighboring embedding algorithm (t-SNE) (Maaten and Hinton, 2008; Ziemba et al., 2016), as shown in Figure 6B, left, which differentiates groupings of spines. We then ran a supervised k-means clustering algorithm on the synaptic PCA coefficients to cluster spines (STAR Methods), where the optimal number of clusters
ski and was chosen using the Calinski-Harabasz score (Calin Harabasz, 1974) as the criterion (Figure 6B, center; STAR Methods). The t-SNE visualization further revealed functional groupings after each spine was labeled based on its identified k-means cluster (Figure 6B, right). With these functional clusters identified, we could project the labels derived from k-means clustering onto spine locations throughout the dendritic tree (Figure 6C). Finally, we computed the probability of

nearby spines falling into the same functional cluster compared with random chance (STAR Methods). Across our population of cells, neighboring spines separated by less than 5 mm along dendrites were more likely to fall into the same functional cluster (probability same group: mean = 0.31 ± 0.05 SE, median = 0.26 with an IQR of 0.17; shuffled probability: mean = 0.18 ± 0.02 SE, median = 0.16 with an IQR of 0.03; p = 0.002, Wilcoxon signed-rank test; Figure 6D, left). At larger distances (5–10 mm, 10–15 mm), the probability of spines originating from the same functional clusters became no different than by chance (p = 0.70 and p = 0.77, respectively; Wilcoxon signed-rank test) (Figure 6D, center and right). Cross-validation of our analysis (STAR Methods) showed that this spatial relationship was robust. These analyses provide corroborating evidence that populations of dendritic spines conveying similar functional properties are clustered within the dendritic arbors on a local spatial scale. Having identified functional ensembles in our spine populations, we wondered how ensembles relate to the somatic RF. Synaptic clusters might build the somatic RF or aid in providing strong input co-tuned with the somatic output (Cossell et al., 2015; Wilson et al., 2016; Lee et al., 2016c). To explore this possibility, we generated spatial RFs for identified clusters by averaging spine spatial RFs within the same k-means cluster. We found that the spatial RF for functional clusters in a single cell were highly selective for visual space and stimulus polarity and could be distinct from each other (Figure 6E). Functional cluster RF properties could even be distinct from the soma (Figure 6E, Neuron 96, 1–12, December 6, 2017 7

Please cite this article in press as: Scholl et al., Local Order within Global Disorder: Synaptic Architecture of Visual Space, Neuron (2017), https:// doi.org/10.1016/j.neuron.2017.10.017

B

A Z-Scored Responses

Principal Component Analysis and K-Means Clustering

Off

Score

4

2

Probability Neighboring Spines in Same Cluster

D

Dendritic Organization 4 2 6 6 6 66

1

2 2

6 2

5 μm

24 5

2 4 4

5 μm

5

1

2

1

4

6

Distance: 0 - 5 μm

0.6

1

8 10

Number of Clusters

Dimension 1

66

Cluster Label 6

Dimension 1

Distance: 5 - 10 μm

0.3

0

0

0.3

0.6 0

Shuffled Probability

0.3

0.6 0

Shuffled Probability

Normalized Amplitude

E Spatial Receptive Fields of Functional Clusters Cluster 1

1

Cluster 5

Cluster 6

Soma

0.5 0 0

-9 -9

0

-9 -9

0

-9 -9

0

-9 -9

0

-9

Spatial Location ( ) Spatial Location ( ) Spatial Location ( ) Spatial Location ( ) Spatial Location ( ) o

o

0.3

0.6

Shuffled Probability

F

Cluster 4

On Off f

-9

Distance: 10 - 15 μm

o

o

o

Proportion of Spines

9 -9 9 Stimulus Position (o)

C

75

50

-1 -9

Dimension 2

100

Dimension 2

Spines

On

0.5

0.25

0

-0.5

0

0.5

1

Cluster and Soma RF Correlation

Figure 6. Synaptic Clusters Convey Functionally Distinct Spatial RF Properties (A) A population of spines from an example cell and Z-scored spatial RF responses for both polarities. (B) Left: 2D visualization of the principal-component coefficients for the synaptic population in (A) (computed using t-SNE; Maaten and Hinton, 2008). Each point represents one spine. Center: the k-means cluster evaluation score for these coefficients. Right: 2D visualization of principal-component coefficients for the synaptic population in (A). Colors indicate group labels from k-means clustering. (C) Dendritic locations of a subset of spines from (A). Each spine is colored to indicate the group label from k-means clustering; the group label from k-means clustering is also indicated next to each spine. (D) Neighboring spines less than 5 mm distant are more likely to found in the same functional cluster compared with a bootstrapped shuffle (left). Neighboring spines within a 5- to 10-mm or 10- to 15-mm distance are no more likely to be from the same functional cluster than chance (center and right). Shown are points from individual cells (white) and the population (black) in mean and SE. (E) Spatial RFs were generated by averaging spine spatial RFs within the same defined cluster. Data are shown as mean and SE for ON (red) and OFF (blue) responses, normalized within each functional cluster. Also shown is a soma RF (far right). (F) The relationship between the correlation of functional clusters and soma RFs and the proportion of spines in a corresponding functional cluster.

right), suggesting that presynaptic populations are comprised of groups of spines conveying a multitude of features. We then examined the RF correlation between functional clusters and the soma with respect to the proportion of spines in each functional cluster. If synaptic populations are responsible for shaping only the soma properties, then we would expect to observe a greater number of spines contributing to functional clusters resembling the soma. Instead, we observed large fractions of spines in functional clusters sharing little similarity to soma RFs (Figure 6F). There was also no relationship between spatial RF correlations and the proportion of spines in each functional cluster (bootstrapped PCA slope = 0.03 ± 0.04 SE, r2 = 0.05, n = 57 8 Neuron 96, 1–12, December 6, 2017

functional clusters from ten cells). From these analyses, synaptic clusters appear to be selective for features both resembling and distinct from the somatic output. DISCUSSION Here we demonstrated the local organization of dendritic synaptic inputs by probing the 1D spatial RFs of individual synaptic inputs onto cortical neurons in the ferret visual cortex using in vivo two-photon imaging of a genetically encoded calcium indicator. Building upon our previous study of orientation preference and synaptic homogeneity on dendritic branches (Wilson et al.,

Please cite this article in press as: Scholl et al., Local Order within Global Disorder: Synaptic Architecture of Visual Space, Neuron (2017), https:// doi.org/10.1016/j.neuron.2017.10.017

2016), we focused on evoking dendritic spine activity within these branches by mapping visual space with bars of preferred orientation. We found no evidence of large-scale topographic organization of visual space across dendritic arbors. However, RFs of neighboring spines within 5 mm were significantly correlated in their spatial location and were co-activated. Co-activation on this spatial scale was also observed in spontaneous synaptic calcium activity and spine responses driven by drifting gratings. As a further test for understanding the distribution of synaptic inputs within the dendritic field, we employed a method of dimensionality reduction (PCA) and a clustering algorithm (k-means) to identify groups of spines based on functional similarity alone. Functional groups were selective for different spatial locations and stimulus polarity while also exhibiting local dendritic spatial clustering. Altogether, these data suggest that local functional synaptic clustering is a fundamental principle of dendritic organization applying to the diverse set of inputs processed by individual cortical cells. Comparison with Mouse V1 A similar study that probes the synaptic representation of visual space in mouse V1 corroborates our measurements of spine RF diversity and distance-dependent tuning correlations (Iacaruso et al., 2017). This study, in contrast to our own, presents evidence for a global organization of visual space within the dendritic field so that proximal inputs reflect somatic/dendritic RF properties and distal inputs are selective for more distal locations in visual space. They also observe a coarse topology of visual space across the dendritic arbors of individual cells. Although discrepancies between our studies might reflect differences in stimulus design (1D versus 2D sparse noise), it is possible that there are species differences in how visual space is mapped within a neuron’s dendritic field. Indeed, cortical magnification of central visual space in the ferret is several orders of magnitude larger than that of the mouse (2 3 102 mm2/deg2 and 1 3 103 mm2/deg2, respectively) (Law et al., 1988; Garrett et al., 2014), whereas the width of the dendritic arbors of layer 2/3 neurons is more comparable in size (NeuroMorph.Org; Ascoli et al., 2007). Thus, it is perhaps not surprising that a coarse visuotopic progression is evident within the dimensions of the dendritic field in the mouse and less detectable in the ferret. It also appears that there is a species difference in synaptic architecture at the scale of synaptic clustering. Although Iacaruso et al. (2017) provide evidence for synaptic clustering based on visuotopy, they find no evidence for clustering of inputs with similar orientation preference. In the ferret, synaptic clustering is found not only for visuotopy and polarity but also for orientation preference, stimulusevoked temporal co-activity, and spontaneous activity. Further studies will be necessary to better understand the factors contributing to the spatial organization of functional inputs within the dendritic field of cortical neurons at both global and local scales and how these vary across species. Spatial Scale of Synaptic Plasticity and Co-activity An intriguing finding from our study is the omnipresent spatial length constant of 5–10 mm. This value has been reported across a wide range of studies employing different experimental techniques in several neural circuits. Following the induction of syn-

aptic plasticity in hippocampal neurons, many biochemical signaling molecules diffuse into neighboring spines and along dendritic segments in a spatially restricted manner (Nishiyama and Yasuda, 2015). For example, activation of the guanosine triphosphatase Ras in single spines caused the spread of Ras through 10 mm of dendrites and neighboring spines (Harvey et al., 2008). Similarly, induction of plasticity in single spines causes a rapid activation of RhoA, which diffuses 5 mm along the parent dendrite (Murakoshi et al., 2011). Synaptic plasticity in single spines can also have a direct cooperative effect: neighboring spines less than 10 mm from a spine that has undergone synaptic plasticity have a reduced threshold for potentiation (Harvey and Svoboda, 2007). Further, in developing hippocampal neurons, synaptic inputs can trigger local calcium-induced calcium activity at this spatial scale (8–9 mm), a property enhanced by clustered synaptic input and dependent on ryanodine receptor activation (Lee et al., 2016b). Because local plasticity mechanisms might shape synaptic arrangements on dendritic segments, studies have explored the spatial scale of spine ensemble activity. Studies by both Takahashi et al. (2012) and Kleindienst et al. (2011) reached a similar conclusion from in vitro measurements: spines within 10 mm exhibited higher rates of co-activation. Takahashi et al. (2012) also observed co-active spines in vivo and arrived at a similar, albeit slightly smaller, spatial length constant for spine co-activations (6 mm), which closely resembled our own measurements (Figures 5A–5C). A recent study in the mouse visual cortex demonstrated that spontaneous synaptic co-activity exhibits a spatial scale of 10 mm, with the greatest probability of co-activity occurring between neighboring spines less than 5 mm apart (Winnubst et al., 2015). The similarity between our in vivo functional measurements and these studies provide an indirect link, suggesting that functional synaptic clusters arise through spatially restricted mechanisms of plasticity (Larkum and Nevian, 2008; Winnubst and Lohmann, 2012). Source of Clustered Synaptic Inputs Although our observations of synaptic clusters suggest that wiring specificity underlies synaptic architecture, the source of clustered inputs is unclear. Co-active and spatially clustered spines could originate from multiple synaptic contacts of single presynaptic axons. Anatomical reconstructions of a small dendritic volume, comparable with the spatial extent of our dendritic spine imaging, reveal single axons terminating on two or three spines of the same dendritic segment, albeit a minority of the total axonal contacts (Kasthuri et al., 2015). Recent work reconstructing functional excitatory networks in mouse V1 revealed that connected neurons with multiple synaptic contacts occur more often than by chance and that multiple synaptic contacts between neuron pairs are arranged close together on dendritic branches at distances comparable with the distance-dependent correlations we observed (Lee et al., 2016c). However, co-active spine clusters might also reflect inputs from different neurons belonging to functionally similar presynaptic networks that exhibit synchronized activity (Takahashi et al., 2012). Additional experiments that combine functional imaging of dendritic spines with ultrastructural reconstruction of presynaptic axons will be necessary to distinguish these possibilities. Neuron 96, 1–12, December 6, 2017 9

Please cite this article in press as: Scholl et al., Local Order within Global Disorder: Synaptic Architecture of Visual Space, Neuron (2017), https:// doi.org/10.1016/j.neuron.2017.10.017

Contribution of Synaptic Clustering to Somatic Responses The prevailing view of input organization on dendritic structures is that clustered inputs amplify functional features contributing to spike generation at the soma and, in the case of the measurements presented here, the RF center (Polsky et al., 2004; Losonczy et al., 2008; Branco et al., 2010; Lavzin et al., 2012; Palmer et al., 2014). Although an attractive hypothesis, our measurements suggest that the role of synaptic clustering in shaping somatic response is more complex; functional clustering was not restricted to spines correlated with the somatic RF (Figure 3). Additionally, synaptic clusters revealed with PCA were not functionally restricted to the exact somatic RF properties (Figure 6), suggesting that each cluster potentially represents a collection of synaptic inputs conveying a specific feature or comprising a functional subunit underlying the full RF. In fact, methods estimating the full extent of V1 neuron RFs can reveal a rich set of spatial filters or RFs, rather than a single one, from spiking responses (Rust et al., 2005) and from subthreshold membrane potential fluctuations (Fournier et al., 2014). Our results suggest that synaptic clustering could amplify inputs contributing to both supra- and subthreshold responses and emphasize the challenges that remain in understanding the complex interplay of different factors within the dendritic field that shape a neuron’s input/output function. In this regard, it should not be surprising that the synaptic aggregate we calculated for individual neurons in our sample was not highly predictive of the somatic response properties (Figure S3). A full accounting of the input/output function would need to take into account differences in synaptic strength, the distribution of inhibitory inputs within the dendritic field, as well as the nonlinearities expected to derive from dendritic clustering. Implications for Network Organization The lack of a global dendritic topography of synaptic inputs within neurons in V1 is surprising but consistent with the distributed functional representations evident for other visual properties (Jia et al., 2010b; Chen et al., 2013; Wilson et al., 2016) as well as with topographically organized properties in other cortical areas (Varga et al., 2011; Chen et al., 2011). These data suggest that organization at the cellular level is formed through integration and processing of distributed networks at the subcellular level. Although this mechanism appears to be disadvantageous in forming a systematic retinotopic map in the visual cortex, it is perhaps amenable to the formation of more selective and larger spatial RFs of higher-order cortical areas and allows for flexibility in forming multidimensional or multimodal RFs. This wiring allows local integration of synaptic inputs from the entirety of a cell’s RF (center and surround), potentially providing a basis for contextual interactions within the dendritic field (Major et al., 2008; Branco et al., 2010; Behabadi et al., 2012). Functional intermingling of inputs from different sources could also allow for facilitative nonlinear interactions within individual dendrites (Xu et al., 2012). At the subcellular level, functionally distributed synaptic inputs might provide a number of advantages in the operation of single neurons. Synaptic diversity on individual branches could enhance the computational power of neurons by allowing individual branches or dendritic segments to be more flexible and form specific outputs, given the wide range of sensory inputs 10 Neuron 96, 1–12, December 6, 2017

(Larkum and Nevian, 2008; Major et al., 2013; Remme and Schreiber, 2015). Depending on synaptic strength or efficacy, distributed arrangements could provide an optimal dynamic range for integrating and transforming incoming inputs. A dendritic branch receiving synaptic inputs conveying a single functional property would be susceptible to saturation of dendritic signals and limited sensitivity if those inputs were sufficiently strong; in contrast, distributing those inputs across dendrites would limit saturation susceptibility in individual dendrites. This scenario might prove difficult for neurons to operate in the face of noisy or weak synaptic inputs, but the addition of functional clustering at a local scale could aid in increasing branch sensitivity without sacrificing dynamic range, taking advantage of either dendritic nonlinearities or passive integration (Polsky et al., 2004; Losonczy et al., 2008; Lavzin et al., 2012; Palmer et al., 2014). Local spatial clustering, in combination with a heterogeneous distribution of synaptic inputs and redundancy of functional properties conveyed to individual dendrites could contribute to error tolerance of individual cortical cells in forming robust response properties. A further advantage of distributed synaptic clustering could be a balance in the trade-off between selectivity and plasticity (Remme and Schreiber, 2015): functionally homogeneous branches would provide strong somatic input but be resistant to synaptic plasticity, whereas individual spines within a fully heterogeneous branch would be more easily modifiable but would provide little, if not negligible, somatic input. Although our observations provide strong evidence of local dendritic clustering, the nature of synaptic networks within individual neurons is only recently coming into focus. For example, spatial dimensions of synaptic clusters might depend on natural stimuli or active vision, such as saccadic eye movements or attention. Continued work in this area will undoubtedly provide a more complete understanding of the subcellular operations driving the activity of individual neurons and shaping the functional properties of cortical networks. STAR+METHODS Detailed methods are provided in the online version of this paper and include the following: d d d d

d d

KEY RESOURCES TABLE CONTACT FOR REAGENT AND RESOURCE SHARING EXPERIMENTAL MODEL AND SUBJECT DETAILS METHODS DETAILS B Virus Injection B Cranial Window B Two-Photon Imaging B Visual Stimuli and Receptive Field Mapping B Analysis QUANTIFICATION AND STATISTICAL ANALYSIS DATA AND SOFTWARE AVAILABILITY B Software

SUPPLEMENTAL INFORMATION Supplemental Information includes five figures and two movies and can be found with this article online at https://doi.org/10.1016/j.neuron.2017.10.017.

Please cite this article in press as: Scholl et al., Local Order within Global Disorder: Synaptic Architecture of Visual Space, Neuron (2017), https:// doi.org/10.1016/j.neuron.2017.10.017

AUTHOR CONTRIBUTIONS B.S. designed the project. B.S., D.E.W., and D.F. wrote the manuscript. B.S. and D.E.W. conducted the experiments. B.S. and D.E.W. analyzed the data. ACKNOWLEDGMENTS We are grateful to David Whitney and Kenneth Latimer for helpful discussions and analytical advice, Dominique Ouimet-Erlacher for surgical expertise and animal care, and the support of the Fitzpatrick lab and the Max Planck Florida Institute for Neuroscience. This work was supported by grants from the National Institutes of Health (NEI EY011488) and the Max Planck Florida Institute for Neuroscience. Received: June 23, 2017 Revised: September 14, 2017 Accepted: October 11, 2017 Published: November 2, 2017 REFERENCES Ascoli, G.A., Donohue, D.E., and Halavi, M. (2007). NeuroMorpho.Org: a central resource for neuronal morphologies. J. Neurosci. 27, 9247–9251. Behabadi, B.F., Polsky, A., Jadi, M., Schiller, J., and Mel, B.W. (2012). Location-dependent excitatory synaptic interactions in pyramidal neuron dendrites. PLoS Comput. Biol. 8, e1002599. €usser, M. (2010). Dendritic discrimination of Branco, T., Clark, B.A., and Ha temporal input sequences in cortical neurons. Science 329, 1671–1675.
ski, T., and Harabasz, J. (1974). A dendrite method for cluster analysis. Calin Commun. Stat. Theory Methods 3, 1–27. Chen, X., Leischner, U., Rochefort, N.L., Nelken, I., and Konnerth, A. (2011). Functional mapping of single spines in cortical neurons in vivo. Nature 475, 501–505. Chen, T.-W., Wardill, T.J., Sun, Y., Pulver, S.R., Renninger, S.L., Baohan, A., Schreiter, E.R., Kerr, R.A., Orger, M.B., Jayaraman, V., et al. (2013). Ultrasensitive fluorescent proteins for imaging neuronal activity. Nature 499, 295–300.

Lampl, I., Ferster, D., Poggio, T., and Riesenhuber, M. (2004). Intracellular measurements of spatial integration and the MAX operation in complex cells of the cat primary visual cortex. J. Neurophysiol. 92, 2704–2713. Larkum, M.E., and Nevian, T. (2008). Synaptic clustering by dendritic signalling mechanisms. Curr. Opin. Neurobiol. 18, 321–331. Lavzin, M., Rapoport, S., Polsky, A., Garion, L., and Schiller, J. (2012). Nonlinear dendritic processing determines angular tuning of barrel cortex neurons in vivo. Nature 490, 397–401. Law, M.I., Zahs, K.R., and Stryker, M.P. (1988). Organization of primary visual cortex (area 17) in the ferret. J. Comp. Neurol. 278, 157–180. Lee, K.-S., Huang, X., and Fitzpatrick, D. (2016a). Topology of ON and OFF inputs in visual cortex enables an invariant columnar architecture. Nature 533, 90–94. Lee, K.F.H., Soares, C., Thivierge, J.-P., and Be´ı¨que, J.-C. (2016b). Correlated synaptic inputs drive dendritic calcium amplification and cooperative plasticity during clustered synapse development. Neuron 89, 784–799. Lee, W.C.A., Bonin, V., Reed, M., Graham, B.J., Hood, G., Glattfelder, K., and Reid, R.C. (2016c). Anatomy and function of an excitatory network in the visual cortex. Nature 532, 370–374. Lennie, P. (1998). Single units and visual cortical organization. Perception 27, 889–935. Losonczy, A., Makara, J.K., and Magee, J.C. (2008). Compartmentalized dendritic plasticity and input feature storage in neurons. Nature 452, 436–441. Maaten, L., and Hinton, G. (2008). Visualizing data using t-SNE. J. Mach. Learn. Res. 9, 2579–2605. Major, G., Polsky, A., Denk, W., Schiller, J., and Tank, D.W. (2008). Spatiotemporally graded NMDA spike/plateau potentials in basal dendrites of neocortical pyramidal neurons. J. Neurophysiol. 99, 2584–2601. Major, G., Larkum, M.E., and Schiller, J. (2013). Active properties of neocortical pyramidal neuron dendrites. Annu. Rev. Neurosci. 36, 1–24. Makino, H., and Malinow, R. (2011). Compartmentalized versus global synaptic plasticity on dendrites controlled by experience. Neuron 72, 1001–1011. Movshon, J.A., Thompson, I.D., and Tolhurst, D.J. (1978). Receptive field organization of complex cells in the cat’s striate cortex. J. Physiol. 283, 79–99. Murakoshi, H., Wang, H., and Yasuda, R. (2011). Local, persistent activation of Rho GTPases during plasticity of single dendritic spines. Nature 472, 100–104.

Cossell, L., Iacaruso, M.F., Muir, D.R., Houlton, R., Sader, E.N., Ko, H., Hofer, S.B., and Mrsic-Flogel, T.D. (2015). Functional organization of excitatory synaptic strength in primary visual cortex. Nature 518, 399–403.

Nishiyama, J., and Yasuda, R. (2015). Biochemical computation for spine structural plasticity. Neuron 87, 63–75.

Fournier, J., Monier, C., Levy, M., Marre, O., Sa´ri, K., Kisva´rday, Z.F., and Fre´gnac, Y. (2014). Hidden complexity of synaptic receptive fields in cat V1. J. Neurosci. 34, 5515–5528.

Palmer, L.M., Shai, A.S., Reeve, J.E., Anderson, H.L., Paulsen, O., and Larkum, M.E. (2014). NMDA spikes enhance action potential generation during sensory input. Nat. Neurosci. 17, 383–390.

Garrett, M.E., Nauhaus, I., Marshel, J.H., and Callaway, E.M. (2014). Topography and areal organization of mouse visual cortex. J. Neurosci. 34, 12587–12600.

Peirce, J.W. (2007). PsychoPy—psychophysics J. Neurosci. Methods 162, 8–13.

Harvey, C.D., and Svoboda, K. (2007). Locally dynamic synaptic learning rules in pyramidal neuron dendrites. Nature 450, 1195–1200. Harvey, C.D., Yasuda, R., Zhong, H., and Svoboda, K. (2008). The spread of Ras activity triggered by activation of a single dendritic spine. Science 321, 136–140. Iacaruso, M.F., Gasler, I.T., and Hofer, S.B. (2017). Synaptic organization of visual space in primary visual cortex. Nature 547, 449–452. Jia, H., Rochefort, N.L., Chen, X., and Konnerth, A. (2010a). In vivo two-photon imaging of sensory-evoked dendritic calcium signals in cortical neurons. Nat. Protoc. 6, 28–35. Jia, H., Rochefort, N.L., Chen, X., and Konnerth, A. (2010b). Dendritic organization of sensory input to cortical neurons in vivo. Nature 464, 1307–1312. Kasthuri, N., Hayworth, K.J., Berger, D.R., Schalek, R.L., Conchello, J.A., Knowles-Barley, S., Lee, D., Va´zquez-Reina, A., Kaynig, V., Jones, T.R., et al. (2015). Saturated reconstruction of a volume of neocortex. Cell 162, 648–661. Kleindienst, T., Winnubst, J., Roth-Alpermann, C., Bonhoeffer, T., and Lohmann, C. (2011). Activity-dependent clustering of functional synaptic inputs on developing hippocampal dendrites. Neuron 72, 1012–1024.

software in Python.

Pologruto, T.A., Sabatini, B.L., and Svoboda, K. (2003). ScanImage: flexible software for operating laser scanning microscopes. Biomed. Eng. Online 2, 13. Polsky, A., Mel, B.W., and Schiller, J. (2004). Computational subunits in thin dendrites of pyramidal cells. Nat. Neurosci. 7, 621–627. Priebe, N.J., and Ferster, D. (2008). Inhibition, spike threshold, and stimulus selectivity in primary visual cortex. Neuron 57, 482–497. Remme, M., and Schreiber, S. (2015). Dendritic distribution of synaptic input creates a trade-off between input selectivity and flexibility. BMC Neurosci. 16, P220. Rust, N.C., Schwartz, O., Movshon, J.A., and Simoncelli, E.P. (2005). Spatiotemporal elements of macaque v1 receptive fields. Neuron 46, 945–956. Sage, D., Prodanov, D., Tinevez, J., and Schindelin, J. (2012). MIJ: making interoperability between ImageJ and Matlab possible. In ImageJ User & Developer Conference (IUDC’12), pp. 227. Schindelin, J., Arganda-Carreras, I., Frise, E., Kaynig, V., Longair, M., Pietzsch, T., Preibisch, S., Rueden, C., Saalfeld, S., Schmid, B., et al. (2012). Fiji: an open-source platform for biological-image analysis. Nat. Methods 9, 676–682.

Neuron 96, 1–12, December 6, 2017 11

Please cite this article in press as: Scholl et al., Local Order within Global Disorder: Synaptic Architecture of Visual Space, Neuron (2017), https:// doi.org/10.1016/j.neuron.2017.10.017

Sokal, R.R., and Rohlf, F.J. (1995). The Principles and Practice of Statistics in Biological Research (Edition).

Winnubst, J., and Lohmann, C. (2012). Synaptic clustering during development and learning: the why, when, and how. Front. Mol. Neurosci. 5, 70.

Takahashi, N., Kitamura, K., Matsuo, N., Mayford, M., Kano, M., Matsuki, N., and Ikegaya, Y. (2012). Locally synchronized synaptic inputs. Science 335, 353–356.

Winnubst, J., Cheyne, J.E., Niculescu, D., and Lohmann, C. (2015). Spontaneous activity drives local synaptic plasticity in vivo. Neuron 87, 399–410.

Varga, Z., Jia, H., Sakmann, B., and Konnerth, A. (2011). Dendritic coding of multiple sensory inputs in single cortical neurons in vivo. Proc. Natl. Acad. Sci. USA 108, 15420–15425.

Xu, N.L., Harnett, M.T., Williams, S.R., Huber, D., O’Connor, D.H., Svoboda, K., and Magee, J.C. (2012). Nonlinear dendritic integration of sensory and motor input during an active sensing task. Nature 492, 247–251.

Wilson, D.E., Whitney, D.E., Scholl, B., and Fitzpatrick, D. (2016). Orientation selectivity and the functional clustering of synaptic inputs in primary visual cortex. Nat. Neurosci. 19, 1003–1009.

Ziemba, C.M., Freeman, J., Movshon, J.A., and Simoncelli, E.P. (2016). Selectivity and tolerance for visual texture in macaque V2. Proc. Natl. Acad. Sci. USA 113, E3140–E3149.

12 Neuron 96, 1–12, December 6, 2017

Please cite this article in press as: Scholl et al., Local Order within Global Disorder: Synaptic Architecture of Visual Space, Neuron (2017), https:// doi.org/10.1016/j.neuron.2017.10.017

STAR+METHODS KEY RESOURCES TABLE

REAGENT or RESOURCE

SOURCE

IDENTIFIER

AAV2/1.hSyn.Cre

UPenn Vector Core

UPenn cat # AV-1-PV2676

AAV2/1.CAG.FLEX.GCaMP6s

UPenn Vector Core

UPenn cat # AV-1-PV2818

Marshall Farms

N/A

Marshall Farms

N/A

Cell Magic Wand Tool

This paper

https://github.com/fitzlab/CellMagicWand

MATLAB

MathWorks

https://ch.mathworks.com/products/matlab.html

FIJI/ImageJ

NIH

http://fiji.sc

Miji

Sage et al., 2012

https://imagej.net/Miji

PsychoPy

Peirce, 2007

http://www.psychopy.org

Bacterial and Virus Strains

Biological Samples Ferret Experimental Models: Organisms/Strains Ferret Software and Algorithms

CONTACT FOR REAGENT AND RESOURCE SHARING Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact David Fitzpatrick ([email protected]). EXPERIMENTAL MODEL AND SUBJECT DETAILS All procedures were approved by the Max Planck Florida Institute for Neuroscience Institutional Animal Care and Use Committee and adhered to the standards of the National Institutes of Health. Juvenile female ferrets (Mustela putorius furo, Marshall Farms) were used all experiments. Ages of acute experiments ranged from P35 to P50. Animals were housed in a vivarium under 16h light/ 8h dark cycle. METHODS DETAILS Virus Injection Juvenile female ferrets (Mustela putorius furo, Marshall Farms) aged P21-P28 (n = 18) were anesthetized with ketamine (50 mg/kg, IM) and isoflurane (1%–3%) delivered in O2, then intubated and artificially respirated. Atropine (0.2 mg/kg, SC) was administered to reduce secretions and 1:1 mixture of lidocaine and bupivacaine administered subcutaneously in the scalp. Animals were placed on a feedback-controlled heating pad to maintain internal temperature at 37
C. Under sterile surgical conditions, a small craniotomy (0.8 mm) was made over the visual cortex 7-8 mm lateral and 2-3 mm anterior to lambda. AAV2/1.hSyn.Cre (Penn Vector Core) was diluted (1:50,000 or 1:100,000) in phosphate-buffered saline (Sigma) and mixed with AAV2/1.CAG.Flex.GCaMP6s or AAV2/ 1.hSyn.Flex.GCaMP6s (Penn Vector Core) to sparsely express GCaMP6s in layer 2/3 cortical neurons. Beveled glass micropipettes (10-15 mm outer diameter, Drummond Scientific Company) were lowered into the brain, and 138-207 nL of virus were injected over 5 min (Nanoject II, Drummond Scientific Company) at 600, 400, and 200 mm below the pia. To prevent dural regrowth and attachment to the arachnoid membrane, the craniotomy was filled with 1% w/v agarose (Sigma-Aldrich). Cranial Window After two to three weeks of expression, ferrets were anesthetized with 50 mg/kg ketamine and 1%–3% isoflurane. Atropine (0.2 mg/kg, SQ) and bupivacaine were administered. Animals were placed on a feedback-controlled heating pad to maintain an internal temperature of 37 to 38
C. A tracheotomy was performed and an endotracheal tube installed to artificially respirate the animal. Isoflurane was delivered between 1 and 2% throughout the surgical procedure to maintain a surgical plane of anesthesia. An intravenous cannula was placed to deliver fluids. ECG, endtidal CO2, external temperature, and internal temperature were continuously

Neuron 96, 1–12.e1–e4, December 6, 2017 e1

Please cite this article in press as: Scholl et al., Local Order within Global Disorder: Synaptic Architecture of Visual Space, Neuron (2017), https:// doi.org/10.1016/j.neuron.2017.10.017

monitored during the procedure and subsequent imaging session. The scalp was retracted and a custom titanium headplate adhered to the skull using C&B Metabond (Parkell). A 3.5-4.0 mm craniotomy was performed at the viral injection site and the dura retracted to reveal the cortex. One to two pieces of custom coverglass (3 mm diameter, 0.7mm thickness, Warner Instruments) were adhered to a larger coverglass (8mm diameter, #1.5 thickness, Electron Microscopy Sciences) using optical adhesive (# 71, Norland Products) and placed onto the brain to gently compress the underlying cortex and dampen biological motion during imaging. The cranial window was hermetically sealed using a stainless steel retaining ring (5/16’’ internal retaining ring, McMaster-Carr), Kwik-Cast (World Precision Instruments), and Vetbond (3M). A 1:1 mixture of 1% Tropicamide Ophthalmic Solution (Akorn) and 5% Phenylephrine Hydrochloride Ophthalmic Solution (Akorn) was applied to both eyes to dilate the pupils and retract the nictating membranes. Eyes were either lubricated hourly with Silicon Oil AP 150 Wacker (Sigma-Aldrich) or contact lenses were inserted (6 - 7mm diameter, Art Optical). Upon completion of the surgical procedure, Isoflurane was gradually reduced and then vecuronium (2 mg/kg/hr) or pancuronium (2 mg/kg/hr) was delivered IV to immobilize the animal. Two-Photon Imaging The animal was placed under the microscope 25 cm from the stimulus monitor, with the monitor subtending 130 degrees in azimuth and 74 degrees in elevation. Imaging was performed using a Bergamo II (Thorlabs) running Scanimage 5 or Scanimage 2015 (Pologruto et al., 2003) (Vidrio Technologies) with dispersion compensated 950nm excitation provided by an Insight DS+ (Spectraphysics). Average excitation power after the exit pupil of the objective (16x, CFI75, Nikon Instruments) ranged from 10 to 80 mW. Individual neurons in layer 2/3 were selected for imaging based on several criteria: visible dendritic spines, nuclear exclusion, orientation-tuned responses, and a lack of large blood vessels obscuring the dendritic field. Images of dendritic segments were acquired at 30Hz (512x512, resolution: 8.42 – 12.2 pixels/mm) and images of soma were acquired at 30Hz (512x512, resolution: 4.2 – 5.9 pixels/mm). Z stacks of individual cells were acquired prior to dendritic imaging by averaging 30-100 frames per plane using 1-2 mm z-steps. Multiple dendrites, both basal and apical, across multiple depths were imaged on individual cells (178 dendritic segments from 18 cells, median = 9.5 segments/cell). This imaging method only allows visualization of a fraction of spines on a given neuron: those on in-plane dendrites. Dendritic spines can also be obfuscated by blood vessels or other dendrites and small spines might be beyond the microscope resolution. Coarse dendritic ROI position was recorded on the z stack throughout the experiment. Due to warping of the brain over long imaging sessions, dendritic imaging fields of view were registered onto individual slices of the z stack using control point registration and an affine transformation. Two-photon frame triggers from Scanimage were synchronized with stimulus information using Spike2 (CED). Throughout the experiment, dendrites were carefully monitored for indications of photodamage. Individual neurons were excluded from data analysis if they exhibited nuclear fluorescence or showed prolonged calcium responses, suggesting cytotoxicity as a result of strong GCaMP expression. Images in the paper were as follows: Figures 1A–1E Structures: 6750 frames (Standard Deviation projection) Figure 5A Structure: 6000 frames (Standard Deviation projection) Figure 5D Structure: 13000 frames (Standard Deviation projection) Figure S1 Structure: 6750 frames (Standard Deviation projection) Figure S1 Synaptic Events: 8 - 12 frames (Median projection) Figure S5A Structure: 5500 frames (Standard Deviation projection) Figure S5A Synaptic Events: 12 frames (Median projection)

Visual Stimuli and Receptive Field Mapping Stimuli were generated using Psychopy (Peirce, 2007). Somatic orientation selectivity was assessed by presenting drifting gratings. Using a small grating at the preferred orientation, the location of spatial RF for the dominant eye was mapped for the soma, aided by custom online analysis software (MATLAB). An array light or dark bars aligned to the preferred orientation were centered over the soma RF location. The number of bars used ranged from 7 to 11 and the widths were 2 – 5
, where the parameters were chosen to elicit strong, reliable responses from the soma in a spatially restricted location. We pseudorandomly presented each bar location for 8-10 trials (1 s duration, 2-3 s ISI). In a subset of imaging sessions, 8 trials of 16 full-field drifting gratings were presented (12.5% contrast, 0.06 cycles/
, 4 cycles/second, 3 s stimulus period followed by 3 s ISI, plus a blank). Cells used for analysis in this paper were required to have somatic responses to 1-D bar stimuli matching that cell’s orientation preference and whose spatial receptive field (RF) center could then be identified and targeted with the bar array. Further, cells were required to have a robust spatial RF stable throughout the course of the imaging. Cells whose RF could not be determined throughout the experiment or became unstable were not used. Typically, we acquired 1 - 2 cells per animal. Analysis Images were corrected for in-plane motion using a correlation-based approach (MATLAB). ROI drawing was performed in ImageJ (Schindelin et al., 2012). For somata and spines, ROI’s were circular or drawn using custom software (Cell Magic Wand, see Software). For dendrites, polygonal ROIs were drawn spanning the extent of a short, contiguous dendritic segment. ROI distances

e2 Neuron 96, 1–12.e1–e4, December 6, 2017

Please cite this article in press as: Scholl et al., Local Order within Global Disorder: Synaptic Architecture of Visual Space, Neuron (2017), https:// doi.org/10.1016/j.neuron.2017.10.017

from the soma were measured by reconstructing dendritic arbors using Simple Neurite Tracer (Schindelin et al., 2012) and local spine positions on dendritic segments were measured by straightening contiguous dendritic segments in MATLAB. These measurements were combined together to provide estimates of spine distance from the soma. Fluorescence time-courses were computed as the mean of all pixels within the ROI at each time point and were extracted using Miji (Sage et al., 2012). Fluorescence time courses were then synchronized with stimulus information, and visually evoked responses were computed as changes in fluorescence relative to the baseline fluorescence. Dendritic spine fluorescence signals were sometimes contaminated by regenerative dendritic events and we used a computational subtraction procedure to isolate spine calcium signals (Wilson et al., 2016) as follows: (1) perform a robust fit (MATLAB) of spine DF/F against dendritic DF/F for stimulus-evoked data and (2) subtract a scaled version of the dendritic DF/F, where the scaling factor equals the slope from the robust fit. Following subtraction, dendritic spines correlated with dendritic signals (r > 0.50) were excluded from analysis. Note that this correlation was calculated between spine and dendrite DF/F while excluding spine DF/F values less than 0. Spines used for subsequent analysis were required to have stimulus-evoked responses with an SNR > 1: m  mblank ; snr = stim sstim + sblank where m is the mean and s the standard error for stimulus conditions and blank conditions, respectively. This criterion was used for spine responses evoked by both bars and drifting gratings. Spines were also required to have at least three isolated calcium events relative to the dendritic signal, evaluated by visual examination of subtracted DF/F time-series traces. To identify spine calcium events, DF/F traces were smoothed with an exponentially weighted moving average filter (MATLAB) (Jia et al., 2010a) and locating the peaks of calcium events. Peak amplitude of calcium events were compared to the standard deviation of baseline spine fluorescence values prior to subtraction. Spines used for analysis of spontaneous activity were required to have at least 1 calcium event (peak amplitude > 3 SD background fluorescence) independent of dendritic calcium events. Peak DF/F responses to bars and gratings were computed using the Fourier analysis to calculate mean and modulation amplitudes for each stimulus presentation, which were summed together. For bars, responses were computed for stimulus onset and offset separately to extract polarity-sensitive responses. Spatial RF ON and OFF responses were fit with a Gaussian equation: ðx  xpref Þ2 2s2 + ao rðxÞ = ae Here x is the bar position, xpref is the preferred spatial location, s is the tuning width, a is the response amplitude, and ao is an unselective response component. Spatial RF full-width half-max (FWHM) was calculated using the Gaussian fits as follows: pffiffiffiffiffiffiffiffiffiffiffiffiffi fwhm = 2 2lnð2Þs The center-of-mass of spine responses was computed as follows: Pn s = 1 ri ðxi ; yi Þ ðxCoM ; yCoM Þ = i P n s i = 1 ri Here r is the normalized response of each spine (i) for a given spatial location stimulated (s) and the position of each spine within the dendritic field is represented by its (x,y) coordinates (in microns). Responses were normalized across all ON and OFF stimuli within each spine. To calculate the error within each synaptic response spatial distribution we generated bootstrapped distributions of center-of-mass for each individual stimulus within individual cells. Ellipses were extracted from these data to quantify the (x,y) center and axes lengths, set to be 2 standard deviations. Trial-to-trial correlations were computed as the correlation of peak DF/F responses to each stimulus on a per-trial basis. For spontaneous activity, the correlation was computed between spine DF/F time courses (after subtraction) in their entirety (7,828 – 11,772 frames). Distance-dependent correlations were compared to those calculated after shuffling spines and performing a bootstrap, sampling with replacement from only spines on individual dendritic segments. Distance-dependent correlations were fit with an exponential equation: y = aex=l + b: Here x is the dendritic distance in microns, a is the mean maximum correlation, b is the baseline correlation, and l is the spatial length constant. To measure spatial length constants in individual cells, we bootstrapped the data: sampling with replacement, computing the distant-dependent correlation 10,000 times and assessing whether the distribution was significantly different from a mean of 0. Functional clusters within spine populations were identified by dimensionality reduction with principal components analysis (PCA) and k-means (MATLAB) clustering on PCA coefficients. PCA was conducted on z-scored mean ON and OFF spine responses for individual cells. Only cells with greater than 30 spines were used (n = 10 cells). The number of PCs used for each cell’s spine population was chosen as the minimum number to explain 60%–70% of the population variance. As the number of PC’s ranged from 3-6, PC coefficients were visualized in 2 dimensions using the t-distributed stochastic neighboring embedding algorithm (t-SNE) (Maaten and Hinton, 2008). To visualize the data in Figure 6 we used a perplexity value of 20. PCA coefficients were clustered using k-means

Neuron 96, 1–12.e1–e4, December 6, 2017 e3

Please cite this article in press as: Scholl et al., Local Order within Global Disorder: Synaptic Architecture of Visual Space, Neuron (2017), https:// doi.org/10.1016/j.neuron.2017.10.017

clustering, running the algorithm for 1,000 iterations. The number of k-means clusters was chosen by identifying the peak of the
ski and Harabasz, 1974) after evaluating PCA coefficients. Spine populations were then sorted by Calinski-Harabasz score (Calin k-means clustering label. We cross-validated these methods and the nearest neighbor analysis on functional groups by repeatedly performing this set of analyses on a random subset of each cell’s spine population, comprising 80% of the total spines. We used this procedure to bootstrap our nearest neighbor analysis (probability of spines within a given dendritic distance originating from the same functional group), running for 10,000 iterations. QUANTIFICATION AND STATISTICAL ANALYSIS Statistical analyses were performed in MATLAB. We used two-sided non-parametric Wilcoxon ranked-sum or Wilcoxon signed-rank test for paired values. We used two-sided Welch’s t test for parametric data. Correlations were calculated as Pearson’s and exact sample sizes are included in the text. A bootstrapped PCA was used to quantify the relationships between 2 independent variables (Sokal and Rohlf, 1995). No estimates of statistical power were performed prior to experiments; animal numbers were minimized to conform to ethical guidelines while accurately measuring parameters of animal physiology. DATA AND SOFTWARE AVAILABILITY Data are available from the lead author upon reasonable request. Software Data are available from the lead author upon reasonable request. Cell Magic Wand (https://github.com/fitzlab/CellMagicWand) was used to draw ROIs for image analysis.

e4 Neuron 96, 1–12.e1–e4, December 6, 2017