Laminar Analysis of Excitatory Local Circuits in Vibrissal Motor and Sensory Cortical Areas

Optical and electrophysiological tools were used to map out the neural circuits within and between cortical layers in three different brain regions, and the results suggest regional specializations for sensory versus motor information processing.


Introduction
Sensation in the rodent vibrissal system relies on active whisking for interactions with the environment [1,2]. Motor circuits control whisker movement, while sensory afferents collect information about contact with objects. Interactions between motor and sensory systems are necessary for object localization and identification [3][4][5].
In rodents, some of the cytoarchitectonic features of vM1, vS1, and S2 are area-specific, such as the presence of ''barrels'' in layer (L) 4 of vS1, and others are not, such as the presence of most cortical layers, including L1, L2/3, L5A, L5B, and L6 [12]. Here, to explore the synaptic organization of cortical circuits in these three areas, we used glutamate uncaging and laser scanning photostimulation (LSPS) to map the local sources of excitatory synaptic input to individual excitatory neurons in vM1, vS1, and S2. We recorded from postsynaptic neurons distributed across L2-6 (i.e., all the cortical layers that contain excitatory neurons) and, for each one, stimulated presynaptic neurons also distributed across L2-6. The collection of synaptic input maps for each area was analyzed to extract a laminar connectivity matrix representing the local pathways between excitatory neurons in each area [13,14]. These connectivity matrices provide a quantitative survey of the interlaminar organization of local excitatory networks in each of these three cortical areas.

Mapping Local Excitatory Pathways with Laser Scanning Photostimulation (LSPS)
We prepared coronal brain slices containing vM1, vS1, or S2 ( Figure 1A-C) and used LSPS with glutamate uncaging [23][24][25] to map excitatory inputs to excitatory neurons ( Figure 1D-G). We excited small clusters of neurons at each site in an array of locations while recording from individual excitatory neurons ( Figure 1D,E), obtaining maps of local intracortical sources of excitatory input ( Figure 1F,G).
To calibrate LSPS, we recorded in cell-attached mode from excitatory neurons, while uncaging glutamate on a grid around the cell (Figure 2A,B). The spatial distribution of action potentials (APs) evoked by uncaging (the ''excitation profile'') provides a measure of the effective spatial resolution of photostimulation (Figure 2A,B). These data were used to estimate neuronal photoexcitability ( Figure 2C) and the spatial resolution of LSPS ( Figure 2D) for photostimulating neurons in different cortical layers and areas. Photostimulation-evoked APs always occurred in perisomatic regions ( Figure 2B, Figure S2) with short latencies, and almost always as singlets. Stimulation of strong synaptic pathways, such as L4RL3 in vS1, did not cause APs in the target location ( Figure S2), indicating that synaptic activity did not cause APs in neurons that were not directly photostimulated. Ultraviolet (UV) attenuation in scattering tissue causes photoexcitation to decline as a function of depth in the slice; consistent with this, excitation was not observed for neurons deeper than 100 mm ( Figure S3) [26]. The total number of neurons excited per stimulus, estimated from the excitation profiles and measured densities of neurons ( Figure 2 and Figure S4), was in the range 50-200, consistent with previous results [13,26]. Only a small fraction of these neurons were synaptically connected to the recorded postsynaptic neuron [27].
An input map represents the aggregate functional synaptic connectivity between small clusters of presynaptic excitatory neurons at the stimulus locations and individual postsynaptic neurons. Pixels in input maps do not represent the strengths of unitary connections; rather, they measure average monosynaptic excitatory responses to a single uncaging event (see Text S1, Equations 1-4) [26]: where r cell is the neuronal density at the point of uncaging (neurons/mm 3 ), V exc is the volume of excited neurons (mm 3 ), and S AP is number of APs fired per presynaptic neuron (AP/neuron). The average strength of a synaptic connection (q con ) is calculated from equation (1). The collection of q con for different neuronal populations defined by laminar location is the basis of connectivity matrices. We first present the mapping data for each area in the more familiar form of average input maps. In subsequent sections we summarize connectivity in laminar connectivity matrices, which take into account the parameters in equation (1).

vM1 Maps
Unlike vS1, vM1 lacks a distinct granular L4. The superficial layers L2/3 and L5A are compressed, and deeper layers L5B and L6 are expanded, consistent with vM1's location at the crest of a cortical convexity [28]. In addition, L1 was thicker than in the other areas (Table 1). Both superficial and deep L5 neurons had dense basal dendrites and a single apical dendrite extending to L1, and L6 neurons had apical dendrites that did not extend to L1; in some cases, these were inverted pyramids ( Figure 3A).
We recorded from 95 excitatory neurons located in all layers (i.e., from upper L2 to lower L6) and mapped the local sources of excitatory synaptic input with LSPS using a stimulus grid that spanned vM1 ( Figure 3B; Figure S5). We pooled neurons into groups by dividing the cortex into 10 equal distance bins; the topmost bin was empty, because L1 lacks excitatory neurons. We averaged the maps in each bin ( Figure 3C). The strongest pathway was a descending projection, L2/3R upper L5. Weaker ascending projections, within L5 and L5ARL2/3, were also found ( Figure 3C). On average, neurons in the lower one-third (0.7-1.0) of vM1 showed weak inputs. However, individual neurons in this deeper range received strong inputs, but these tended to be spatially dispersed and sparse ( Figure S5).

vS1 Maps
We recorded from 80 excitatory neurons in vS1, using a different stimulus grid matched to the cortical thickness ( Figure 4; Figure S6). In vS1, laminar boundaries were distinct, allowing pooling of cytoarchitectonically defined groups for binning (Table 1; Figure S8). The ascending L4RL3 pathway and the descending L2/3RL5 pathway were both prominent ( Figure 4; Figure S6). Similar to vM1, L6 neurons had relatively weak inputs (mainly from L4). L4 neurons also showed little intracortical interlaminar input [29]. In addition, we further distinguished sublayers within L2/3 and L5B based on patterns of connectivity observed in the input maps. For example, L2 constituted a narrow superficial layer of neurons lacking strong input from L4, but with input from L5A [30]. Binning with a simple three-layer scheme ('supragranular-granular-infragranular'; Figure 4) conveyed the main feedforward local excitatory connections in vS1.

Author Summary
The neocortex of the mammalian brain is divided into different regions that serve specific functions. These include sensory areas for vision, hearing, and touch, and motor areas for directing aspects of movement. However, the similarities and differences in local circuit organization between these areas are not well understood. The cortex is a layered structure numbered in an outside-in fashion, such that layer 1 is closest to the cortical surface and layer 6 is deepest. Each layer harbors distinct cell types. The precise circuit wiring within and between these layers allows for specific functions performed by particular cortical regions. To directly compare circuits from distinct cortical areas, we combined optical and electrophysiological tools to map connections between layers in different brain regions. We examined three regions of mouse neocortex that are involved in active whisker sensation: vibrissal motor cortex (vM1), primary somatosensory cortex (vS1), and secondary somatosensory cortex (S2). Our results demonstrate that excitatory connections from layer 2/3 to layer 5 are prominent in all three regions. In contrast, strong ascending pathways from middle layers (layer 4) to superficial ones (layer 3) and local inputs to layer 6 were prominent only in the two sensory cortical areas. These results indicate that cortical circuits employ regional specializations when processing motor versus sensory information. Moreover, our data suggest that sensory cortices are elaborations on a basic motor cortical plan involving layer 2/3 to layer 5 pathways.

S2 Maps
S2 abuts the lateral edge of vS1, where the barrel pattern terminates ( Figure S1). The cytoarchitectonic layers appeared similar in S2 and vS1, except that the cortex was thinner and L5A thicker. L4 included neurons with a sparse apical dendrite, and neurons lacking an apical dendrite ( Figure 5A). L5 neurons had many basal dendrites and an apical dendrite that ramified in L1; L6 neurons' apical dendrites did not extend above L4.
We recorded input maps for 100 excitatory neurons in S2 (Figure 5B,C; Figure S7). Similar to vS1, an ascending pathway to more superficial layers (L4RL3) was present but was not the strongest projection. Instead, the descending projection L2/3RL5 was predominant. L5 also received substantial ascending input from L6.

Derivation of Connectivity Matrices
Connectivity matrices represent local circuits in a compact manner [13,14,27,31,32]. Each element (i, j) in the matrix (W i,j ) corresponds to the strength of a connection (q con ; Equation 1) from the j th presynaptic location (along the rows) to the i th postsynaptic location (along the column). Distance is measured in normalized units along the radial directions (pia, 0; white matter, 1). Because of the curvature of vM1 at the cortical flexure ( Figure 6A,B), we converted map data from the coordinates of the slice image (x, y) to coordinates corresponding to an unfolded cortex (h, r), where h is the horizontal distance along the laminar contour and r is the distance along the radial axis. Figure 6 provides a graphical illustration of the process of converting the pixels in an input map from x-y coordinates ( Figure 6A), using a spatial transform defined on the basis of the radial structure of the cortex ( Figure 6B), into r-h coordinates ( Figure 6C,D).
This approach allowed us furthermore to convert input maps into vectors, by averaging input across the horizontal dimension (h) at a given presynaptic radial distance (r) into bins (Figure 6E-G; a similar analysis in the horizontal dimension is given in Figure S9). This is identical to averaging along the rows of input maps, except that it takes into account the curvature of the cortex. One neuron's input vector ( Figure 6G) thus represents the inputs to one neuron from different laminar locations; i.e., the horizontal dimension has been collapsed. Each neuron was also assigned a postsynaptic radial distance. This allowed us to group all the input vectors and then sort them by the postsynaptic neuron's depth in the cortex ( Figure 6H). Stacking the vectors on top of each other, sorted by depth, provided a raw connectivity matrix, W raw (r post , r pre ), describing connectivity between neurons at different locations along the radial axis ( Figure 6H, Figure 7A). The rows in such a connectivity matrix represent synaptic input to a particular laminar location, and the columns represent synaptic output from that laminar location. Intralaminar connections lie along the main diagonal. We note that intralaminar connectivity was undersampled because of direct excitation of the postsynaptic neurons' dendrites.
In addition to deriving matrices based on the collections of input vectors ( Figure 7A,D,G), we further analyzed the data in terms of the excitation parameters given in equation (1). To compute the average connectivity matrix at the level of individual neurons (W neuron ), we binned the data and applied correction factors to derive the strength of input per presynaptic neuron per AP. We divided the connection strength in the raw connectivity matrix by the mean number of APs per uncaging event at the presynaptic region ( Figure 7B,E,H; Figure S10; Text S1) and the number of   Numbers correspond to lower borders of the corresponding layer. Cortical thickness was normalized (0, pia; 1, white matter) and presented 6SD. Italic numbers preceded by tildes indicate approximate borders where cytoarchitectonic distinctions were not strongly evident. We note that these measurements for vM1 were made at the mid-flexure ( Figure 6B). The upper layers are relatively more expanded at more lateral locations. doi:10.1371/journal.pbio.1000572.t001 presynaptic neurons stimulated. The number of stimulated neurons was obtained from measurements of r cell ( Figure S4) and V exc . To compute the connectivity matrix at the level of cortical layers (W layer ) we multiplied the neuronRneuron connections by the number of presynaptic and postsynaptic neurons per layer ( Figure 7C,F,I; a detailed calculation is illustrated in Figures  S11 and S12). Values for all connectivity matrices are provided in Table S1 and Dataset S1.

Discussion
We used glutamate uncaging and LSPS to map local synaptic connections among excitatory neurons in mouse vM1, vS1, and S2, three cortical areas centrally involved in vibrissa-based somatosensation. From single cell input maps recorded at different cortical depths, we derived connectivity matrices that compactly describe the local network. Our main findings were that vM1 contains a strong pathway from L2/3 to upper L5; that vS1 and S2 contain two strong pathways, corresponding to L4RL3 and L2/3RL5; and that S2 contains these plus pathways between L6 and L5B.

Connectivity Matrix Descriptions of Cortical Circuits: NeuronRNeuron and LayerRLayer
The connectivity matrix description allows us to directly contrast local circuits in different cortical regions. The elements (pixels) in the neuronRneuron connectivity matrices, W neuron ( Figure 7B,E,H), represent the mean strength of postsynaptic response in a single neuron extrapolated to a single presynaptic AP in a single cell of the indicated layer (q con ). Pixel values were 10-100 times lower than typical unitary EPSCs, reflecting both the generally low probability of connections between excitatory neurons in cortical circuits (typically 0.1-0.2) [27,[33][34][35], and the fact that the current amplitude in the maps represents a mean over 50 ms rather than the peak of the EPSC.
In contrast, the elements in the layerRlayer connectivity matrices, W layer ( Figure 7C,F,I), represent the average strength of connections extrapolated to the entire projection from one layer to another. The W layer matrices differ from the W neuron matrices in that they enhance thicker and more neuron-dense layers and diminish thinner and less neuron-dense layers. For example, because in vS1 the L5A is thin (Table 1) and both L5A and L5B are low in neuronal density ( Figure S4), the projections to and from L5, such as L5ARL2/3 and L2/3RL5B, are relatively strong at the level of neuronRneuron connectivity ( Figure 7E) but relatively weak at the level of layerRlayer connectivity ( Figure 7F). Interestingly, in rat vS1 the L4RL2/3 projection is functionally weak compared to the structural density of L4 axons and L2/3 dendrites, while the converse holds for the L5ARL2/3 projection [36]. Our results here show how weak neuronR neuron connections may be strong in aggregate at the layerRlayer level. Further structure-function analyses will be required to determine whether it is generally the case that larger and more neuron-dense layers have weaker neuronRneuron but stronger layerRlayer projections.

Major Features of Connectivity Matrices in the Three Areas
The connectivity matrix representations of vM1 show strong descending projections from L2/3Rupper L5 ( Figure 7A-C), similar to the forelimb area of mouse M1 [13,14,37]. This input straddled the L5A/B border. L5B received an additional hotspot from itself, which appeared strong when considered as an entire layer ( Figure 7C). The deepest one-third of vM1 (consisting mostly of L6) had weak inputs and outputs.
The vS1 excitatory circuits were more complex ( Figure 7D-F). The major ascending pathway from L4RL3 was paralleled by an ascending component from L5A. The high cell density in L4 made the L4RL3 connection prominent in the laminar analysis ( Figure 7F). Another prominent projection was from L2 and L3 to L5A and L5B; inputs originating in more superficial regions of L2/3 targeted relatively more superficial regions of L5A/L5B (note the diagonal shape of the L2/3RL5 hotspot in Figure 7E). On a neuronRneuron basis, the L3RL5B connection was stronger than L4RL3, although the layerRlayer analysis showed a reduction in cell density relative to L4. L2 received input from L3, and weaker input from L5A. However, L2 was thin and thus contributed little to W layer . As in vM1, deep layers had weak inputs and outputs.
In S2 ( Figure 7G-I), an ascending L4RL2/3 pathway and descending L3RL5 pathway were present. Neurons on the L5A/ L5B border also showed strong intralaminar connections. The L6 output evident in the input maps ( Figure 5; Figure S7) also supplied potent input to L5B. Although not as strong at the single cell level, the entire L6 excited L5B as much as L3 ( Figure 7I). L6 was enhanced in S2 relative to other regions as both a source of synaptic output and a recipient of synaptic input, due to the relatively high density of neurons ( Figure S4) and their relatively low photoexcitability ( Figure 2C-E). The functional connectivity in the local excitatory circuits of all three regions is simplified into quantitative laminar wiring diagrams ( Figure 8).

Limitations in the Derivation of Connectivity Matrices
LSPS with glutamate uncaging simultaneously excites a group of presynaptic neurons, while the postsynaptic response is measured. To derive average connection strength per neuron (q con ), the number of excited neurons needs to be estimated, based on the excitability (S AP ), neuron density (r cell ), and excitation volume (V exc ) at the uncaging location (Equation 1). The accuracy of the estimate of q con is limited by our measurement of r cell and neuronal excitation (S AP , V exc ; Text S1 Equations 3-4): Measurements of neuronal density vary by a factor of two [27,38,39]. Although excitation profiles give a direct measure of evoked APs in brain slices under the relevant recording conditions (Figure 2), excitation varies across neurons and somewhat across cortical areas, and decreases with depth in the slice; these effects together introduce uncertainty roughly on the order of a factor of two ( Figure S3). Despite these uncertainties, our estimates of q con are broadly consistent with those derived from pair recordings ( Figure S13).
Because LSPS excites many neurons, this strong stimulus allows weak pathways to be detected. However, the average connection strength, q con , reflects both the connection probability and unitary connection strength: It is therefore not possible to separate connection probability and unitary connection strength directly. Furthermore, p con is inversely related to the horizontal separation between cell pairs [34]. LSPS averages inputs from a range of presynaptic locations with varied horizontal offset. For each cell class, a broad distribution of p con values contributes to LSPS maps.
In addition, by computing the average connection strength, we average out the underlying distribution of unitary connection strength, which is a skewed distribution of numerous weak and a few strong connections [27,33]. This inherent averaging also makes LSPS insensitive to certain non-laminar aspects of cell-type specificity in cortical connectivity [14,33,35,[40][41][42][43].

Comparisons with Previous Studies of vS1 Connectivity
Comparison of our neuronRneuron connectivity matrix with a pair-recording study [27] reveals qualitative similarities ( Figure  S13). After both methods are corrected to similar units (peak amplitude in pA/AP), the general shape of the connectivity matrix and values for neuronRneuron connectivity are similar. The major interlaminar pathways are L2/3RL5 and L4/5ARL2/3. However, local intralaminar connections are underestimated in our data set due to direct responses to uncaging. Furthermore, descending projections from L4RL5A and from L5ARL5B may be underestimated in LSPS relative to pair recording due to exclusion of direct responses along the apical dendrite of the postsynaptic neuron (see L5A and L5B maps in Figure S6). Undersampling of connected pairs in low-p con pathways, such as L4RL6, may account for differences from LSPS, where many L4 neurons are excited during each L6 recording. Lastly, L2 connectivity differs in part because of differences in the definition of this layer.

Inter-Areal Comparisons: Ascending Pathways to Supragranular Layers
We compared the matrices for the four areas so far studied, vM1 (present study), the forelimb region of somatic M1 [13], vS1 (also the present study) [27], and S2 (present study). Overall, the main differences are attributable to the presence of a distinct granular layer in somatosensory cortex. Specifically in vS1, L4 outflow contributed strongly to the connectivity matrix. L4RL2/3 is also a major pathway in rodent V1 [44]. In S2, the local excitatory circuit differs from vS1 most prominently in that the L4RL3 pathway is reduced. LSPS analyses of auditory cortex circuits have found L4RL2/3 inputs [45,46], which is adjacent to S2. However, ascending pathways were not unique to vS1, as a similar but weaker L3/5ARL2/3 pathway was prominent in forelimb M1, and present but weaker still in vM1 ( Figure 3C and Figure S5C, leftmost panels). The upward compression of layers in vM1, typical of cortical convexities [28], may be why L3/5ARL2 was less distinct in vM1 than in forelimb M1 (e.g., it was more prone to masking by dendritic responses of L2 neurons). However, inspection of individual maps and traces ( Figure S5C) showed that these ascending pathways were present for some L2 neurons.

Inter-Areal Comparisons: Descending Pathways to Deep Layers
A second main interlaminar hotspot in vS1 was the descending pathway(s) L2/3RL5, which was the predominant hotspot in the two motor areas. We noted that this pathway was present in all three cortical regions studied here and was similarly prominent in somatic M1 [13]. Indeed, it was the predominant pathway in S2. Thus, a strong supragranular to infragranular descending connection emerged as a common element of local cortical circuits examined here. Superficial L5B neurons and deep L5A neurons at the laminar border were most strongly activated, suggesting that the cytoarchitectonic boundaries identified do not correspond well with functional gradient within L5. Perhaps an alternative molecular marker, such as Etv1 ( Figure S8), better denotes this functional division.

Inter-Areal Comparisons: Involvement of Deep Neurons in Local Circuit Function
In three of the four areas, L6 neither received nor sent strong projections (but vS1 neurons in L6 received a weak projection from L4). L6 output is provided by an ascending connection to L4 in cat visual cortex [31,32,47] but was absent or reduced in all vibrissal areas we studied. L6RL4 projections studied in mouse somatosensory and auditory cortical areas have ''modulator'' rather than ''driver'' properties, including paired pulse facilitation [48]. Although deeper neurons tend to have relatively small dendritic arbors [49], which may account for a reduction (but not absence) of inputs, this difference in arbor size is not of sufficient magnitude to account for the paucity of inputs. Similarly, the paucity of outputs was not due to lack of photoexcitability of these neurons. Channelrhodopsin-assisted circuit mapping experiments [50] have shown that the supragranular layers indeed connect preferentially to upper rather than lower infragranular neurons. Thus, the lack of inputs was not due simply to slice-related artifacts such as severing of pathways. Consistent with weak local inputs, in vivo recordings in cat motor cortex suggest that a large number of L6 neurons are virtually silent, even during motor activity [51]. Thus, the sources and modes of excitation for L6 neurons remain to be determined [49,52]. However, L6 was more engaged in local circuits in S2, supplying a measurable output to L5A and L5B and to other L6 neurons. In addition to input from L5B, L6 neurons in S2 collected inputs from a wide horizontal distance, sometimes .300 mm ( Figure S7B,C at right). Thus, S2 may be better suited for studying L6 function.

Quantitative Comparison of Cortical Microcircuits
One major difficulty in making a comparison of connectivity between two cortical areas is selecting the laminar position of preand postsynaptic neurons for the comparison. Is it better to compare identical relative laminar depths between cortical areas, not accounting for the decreased thickness of superficial layers, and increased thickness of deep layers, in motor areas? How shall we treat the presence or apparent absence of a distinct layer 4? We present a direct quantitative comparison of three major areas identified in our study, based on cytoarchitectonic laminar These are used to measure stimulus grid locations (x, y) to transform them into (h, r) coordinates, where h is the horizontal arc distance (in mm) from the center spoke, and r is the normalized radial distance from the pia. Rainbow-like plots show interpolated maps of radial distance (C) and horizontal distance (D) for given points in an input map. (E-H) We selected points at a given presynaptic radial distance (two white boxes shown for adjacent superficial regions in E), within a limited horizontal range from the postsynaptic neuron. These regions were used to select points in the input map for binning purposes. By averaging the selected points in the input map at the given presynaptic depth (within the white boxes in F, for example), we converted input maps to input vectors (G). The postsynaptic radial distance for each recorded neuron was then used to place the input vectors in order, with vectors from superficial neurons in the top rows and deeper neurons in lower rows. By stacking the input vectors for every cell in a given cortical region, ordered by postsynaptic radial distance, a rough outline of the connectivity matrix can be presented (H  Figure S8) (Figure 9). In vS1 [53] and vM1 (Tianyi Mao, BMH, GMGS, KS, unpublished observations) these layers correspond to distinct cell types with different projection patterns. The descending projection from L2/3RL5A/B was prominent in all areas, but the strength of the pathways at a neuronRneuron level varied by a factor of four between the areas. Ascending projections from middle layers to superficial ones (L4RL2/3 in vS1 and S2; L5ARL2/3 in vM1 for comparison) were also present in all regions but were the least prominent in agranular vM1. Lastly, the L6RL5 projection identified in S2 was more than twice as strong at the neuronRneuron level than in vS1 (and the difference was greater with vM1). Our approach provides a defined framework for measuring similarities and differences between cortical microcircuits in a quantitative manner.

Terminology for Cortical Axes
We use the term radial to refer to the axis defined by the apical dendrites of pyramidal neurons; this axis is approximately normal to the cortical surface. Normalized radial distance is along the radial axis, bounded by the pia and the white matter, where pia = 0 and the L6/white matter border = 1. Vertical is synonymous with radial. Horizontal, or lateral, refers to planes normal to the radial axis, approximately parallel to layers, or laminae ( Figure 6). Oblique refers to off-axis interlaminar connections.

Slice Preparation
Mice were decapitated at postnatal day 20-25 under isofluorane anesthesia, and the brain rapidly placed in ice cold choline solution (in mM: 110 choline chloride, 25 NaHCO 3 , 25 D-glucose, 11.6 sodium ascorbate, 7 MgCl 2 , 3.1 sodium pyruvate, 2.5 KCl, 1.25 NaH 2 PO 4 , 0.5 CaCl 2 ). Coronal brain slices (300 mm) were cut (Microm HM 650V), incubated 30 min at 37uC in oxygenated ACSF (in mM: 127 NaCl, 25 NaHCO 3 , 25 D-Glucose, 2.5 KCl, 2 CaCl 2 , 1.25 NaH 2 PO 4 , 1 MgCl 2 ), and maintained in a holding chamber at 22uC for up to 5 h during recording. For vM1 slices, the brain was pitched upward ,10u to optimize alignment with the radial axis of vM1, and slices ,0.7-1.3 mm anterior to bregma were used; for vS1 and S2 slices, the brain was cut coronally, and slices ,1-2 mm posterior to bregma were used ( Figure 1A,B). To determine the optimal slice angle for each area, we used the appearance of the intact apical trunk at high magnification to select slices for recording and avoided those sections where the apical dendritic trunk was at an angle with respect to the slice plane. Thus, only one or two sections per animal could be used for recording. Separate experiments in our laboratory using the photostimulation methods in vS1 [42] and vM1 (unpublished data) measure input to L1 dendrites of L5 pyramidal neurons, confirming the apical trunk is intact using this slice angle. We added biocytin to visualize a subset of dendritic arbors, some of which are reconstructed in Figure 3 and Figure 5. These neurons appeared radially symmetric, with arbors ranging from 300-500 mm in diameter. Since the neurons were 50-100 mm deep in the slice, a portion of the apical and basal dendrites are truncated by slicing and the deep half of the arbor is intact.
Neurons were selected based on pyramidal appearance, or in the case of L4 recordings in vS1, either pyramidal or stellate appearance. In vS1, recordings were generally made in the middle of the barrel field and not a specific whisker barrel. Following patching, a family of current steps was presented to determine firing properties. Neurons with narrow APs and high firing rates were rejected for analysis as presumed interneurons.

Laser Scanning Photostimulation (LSPS)
Methods followed published procedures [13,26]. MNI-glutamate (0.2 mM; Tocris, MO [55]) was added to a recirculating bath. Photolysis was performed by shuttering (1.0 ms pulse) the beam of an ultraviolet (355 nm) laser (DPSS Lasers, San Jose, CA), ,20 mW in the specimen plane, set by a combination of a gradient neutral density filter wheel and Pockels cell (electrooptical modulator; Conoptics). A 16616 standard stimulus grid for input maps had row and column spacing of 110 mm for vM1 and 90 mm for vS1 and S2 recordings. Maps were recorded in voltage clamp at 270 mV. Inhibitory input amplitude was minimized by recording near the chloride reversal potential. The 256 grid sites were visited in a sequence that optimized the spatiotemporal separation between sites [25]. The sequence was repeated 2-4 times per neuron. In vS1 and S2, the map was aligned to the top of the pia and centered on the soma. In vM1, the medial edge of the map was aligned to To convert each map's set of traces into an array of pixels that represent response amplitudes, we calculated the average current over a 50 ms post-stimulus window. Direct dendritic responses were excluded on the basis of temporal windowing [56], rejecting traces with events (detected as .3 SD above baseline) with onset latencies of ,7 ms. At locations where some maps had direct responses at a given pixel while others did not, the average of the non-direct responses was used; pixels were excluded from the average raw input map for a given neuron if all traces had direct responses.
We measured excitation profiles using loose-seal recordings with the amplifier in voltage-follower mode, to gauge the efficacy of photostimulation for neurons in the different layers in the three areas. Excitation profiles were recorded and analyzed following previous methods [13,25,26,30]. To characterize the size of the excitation profile, we calculated the mean weighted distance from the soma of AP generating sites as: S(APs 6 distance from soma)/ S(APs).

Connectivity Matrix Analysis
Procedures build on [13]. A transformation step was added, to account for cortical curvature, which was especially strong in vM1. As described in Results, we assigned each point in the stimulus grid a normalized radial distance and horizontal offset ( Figure 6). Individual recorded neurons were also assigned a postsynaptic radial distance based on the same criteria. Individual input maps for a given neuron could then be averaged together based on postsynaptic radial distance. Furthermore, when computing the input to a given neuron for the purpose of determining the connectivity matrix, a presynaptic point would be averaged into a bin appropriate to its location. Most aspects of local connectivity were robust to changes in binning. Subsequent corrections to the connectivity matrices were made to account for variations in excitability between layers, and the number of neurons in pre-and post-synaptic layers; these were then presented as neuronRneuron connectivity matrices and layerRlayer connectivity matrices (see Results and Text S1).

Quantitative Comparisons of Connectivity
To make quantitative comparisons between the strength of pathways in different areas, we determined both the average strength of pathways and their variability using a bootstrap-based analysis (Figure 9). After selecting the pre-and postsynaptic neuron populations by relative laminar depth, the strength of corresponding pixels in the input map (limited to maps from neurons in the postsynaptic layer, and pixels in the presynaptic layer within 300 mm horizontal distance of the soma) were averaged for each map. We resampled the individual map

Analysis of Cortical Lamination
In vS1 and S2, we performed morphometric measurements of cortical landmarks in video images of brain slices. Along a radial line, we marked the locations of the soma, pia, white matter, and major laminar boundaries and calculated the absolute and normalized radial distances of these locations. The bottom extent of cortex was marked at the border between L6 (including the subplate zone) and white matter [57]. The distances to lower borders of layers (6SD) are given in Table 1. The division between L2 and L3 in vS1 was drawn between groups of neurons that did not receive appreciable L4 input (L2 [58]) and those that did. Since this functional division was not clear in S2, L2/3 was divided in half. These values are bracketed in the table. vM1 appearance was similar to somatic motor cortex, with a prominent clear zone in the upper middle part of the cortex, corresponding to L5A [13]. Thus, landmarks indicating the border between L1, a compressed L2/3, and the bottom of L5A were apparent in video images and used to measure laminar boundaries in vM1. The division between L5B and L6 was estimated as the radial distance where cell density increased ( Figure S4; Table 1), as a clear border was not apparent based on image contrast. Alternative methods of determining cortical layers in motor and sensory cortex were performed on images of gene expression patterns from the Allen Brain Atlas ( Figure S8).

Supporting Information
Dataset S1 Connectivity matrix values for neuron-and layer-based connectivity matrices. Matlab file containing the values for all neuron-and layer-based connectivity matrices of  Figure S10, but bins are determined based on boundaries between cortical layers instead of even spacing. Given the individual input vectors to a given neuron ( Figure 6A) averaged in laminar specific bins, these data are presented for all neurons as an uncagingRneuron matrix (left). Cell-type specific excitability was accounted for by dividing each presynaptic bin by the average number of AP per region. Corrections are shown as a 2D matrix; corrections are the same for all columns along the presynaptic orientation. Furthermore, input was divided by presynaptic cell density to correct for the number of neurons activated per uncaging event, resulting in a neuronRneuron connectivity matrix (right). (B) Data presented as in (A), but with postsynaptic neurons binned into cortical layer specific bins. Found at: doi:10.1371/journal.pbio.1000572.s012 (0.39 MB TIF) Figure S12 Construction of layerRlayer connectivity matrices in cortical layer bins. (A) Example of how layerRlayer connectivity matrices are constructed from neu-ronRneuron connectivity matrices; vS1 is used for this example. The corrected neuronRneuron matrix of Figure S11B (right) is used as a starting point. For the purpose of determining total number of neurons per layer, instead of density, a cortical column of 3006300 mm was used in the plane normal to the radial axis from pia to white matter. The thickness of each layer along the radial axis was based on cytoarchitectonic measurements (Table 1); density was based on Figure S4. Correction to the neuronR neuron matrix involved multiplication by both the number of presynaptic (bottom; columns) and postsynaptic (top; rows) neurons. Connectivity matrices in this style are presented in Figure 7. Matrices of neuronRneuron connectivity based on pair recordings [27] and LSPS (Hooks et al., this work) plotted on the same scale. Single cell connectivity for pair recordings is converted from peak amplitude in mV to pA using layer specific input resistance, and multiplied by connection probability. Single cell connectivity for LSPS is converted from mean amplitude in pA to peak amplitude using a conversion factor of 0.2 (based on ratio of mean/peak amplitude in LSPS recordings). Text S1 Laminar analysis of excitatory local circuits in vibrissal motor and sensory cortical areas. Supplemental information to the article. Sections include: Supplemental Methods (relationship of pixel values in input maps to average synaptic connection strength (q con ); converting input maps to connectivity matrices; methods for optical microstimulation), Supplemental Discussion (comparison and limitations of circuit mapping techniques; horizontal connectivity), and References. Found at: doi:10.1371/journal.pbio.1000572.s019 (0.32 MB RTF)