The subcellular organization of neocortical excitatory connections (original) (raw)

Nature. Author manuscript; available in PMC 2009 Sep 17.

Published in final edited form as:

PMCID: PMC2745650

NIHMSID: NIHMS143295

Janelia Farm Research Campus, Howard Hughes Medical Institute, Ashburn VA 20147, USA

Author contributions

LP and KS conceived and designed the experiments. LP performed the experiments with help from TM (viral injections in M1 and related recordings). SS provided critical reagents. LP and KS analyzed the data and wrote the paper.

Supplementary Materials

Supplementary Methods

In utero electroporation. Experimental protocols were conducted according to the National Institutes of Health guidelines for animal research and were approved by the Institutional Animal Care and Use Committee at Janelia Farm Research Campus. Venus 34 (gift from A. Miyawaki) was fused to the C-terminus of the first 315 amino acids of channelrhodopsin-2 (gift from G. Nagel). The constructs were inserted into pCAGGS vector modified for in utero electroporation . DNA was purified and concentrated using Qiagen plasmid preparation kits and dissolved in 10 mM Tris–HCl (pH 8.0).

L2/3 progenitor cells were transfected via in utero electroporation 28, as previously described 8. E16 timed-pregnant C57BL / 6J mice (Charles River, Wilmington, MA) were deeply anesthetized using an isoflurane-oxygen mixture (1% vol isoflurane / vol O2) delivered by an anesthesia regulator (SurgiVet, Waukesha, Wisconsin). The uterine horns were exposed and ~ 1 μl of DNA solution with Fast Green (Sigma) was pressure injected (Picospritzer, General Valve) through a pulled glass capillary tube (Warner Instruments, Hamden, CT) into the right lateral ventricle of each embryo. The DNA solution contained a mixture of plasmids encoding ChR2-Venus and mCherry in a 3:1 molar ratio, at final concentration of 2 μg/μl. The head of each embryo was placed between custom-made tweezer-electrodes, with the positive plate contacting the right side of the head. Electroporation was achieved with 5 square pulses (duration = 50 ms, frequency = 1Hz, 40V). mCherry fluorescence was used to screen for positive animals under a fluorescence dissecting scope (MVX10, Olympus), 1 to 2 days after birth. The transfected cortical region in electroporated animals was always restricted to L2/3 in the electroporated hemisphere. It usually encompassed most of the barrel cortex and in some cases included parts of auditory, visual and secondary somatosensory areas.

Virus preparation and injections. The ChR2-Venus fusion was cloned into an adeno-associated viral cassette (serotype 2/1) containing the CAG promoter, a woodchuck posttranscriptional regulatory element (WPRE) and the SV40 polyadenylation site. Viral vectors were packaged and purified to 1.8×1012 gc/ml (University of Pennsylvania Vector Core). C57BL / 6J mice (Charles River, Wilmington, MA) were deeply anesthetized using an isoflurane-oxygen mixture (1 % vol isoflurane / vol O2) at postnatal day 12-15 and placed in a custom stereotactic apparatus. A small incision was made in the scalp to expose the skull. We injected bilaterally in the ventral posteriomedial nucleus (VPM) of the thalamus (1.45 mm posterior to bregma, 1.6 mm lateral ; 3.1 mm deep from pia), the posterior medial (POm) thalamus (1.45 mm posterior to bregma, 1.6 mm lateral; 2.9 mm deep), or unilaterally into vibrissal primary motor cortex (1 mm anterior to bregma, 0.6 mm lateral; 0.4 and 0.7 mm deep). 40 nl of viral suspension at 4.5×1011 gc/ml were injected over ~ 1 min using a pulled glass micropipette (Drummond, Broomall, PA). To label L4 neurons, 40 nl of a Cre recombinase-dependent AAV virus encoding ChR2-H134R 35 fused with mCherry (3.4×1013 gc/ml) 29 was injected in the barrel cortex of P15 Six3-Cre#69 mice 30, 36. ChR2-mCherry expression was dense within a few barrels in L4 (Fig. 2,3 and Fig. S5). Only a few scattered cells were labeled in L3, L5 and L6 (laminar distribution of labeled cells; L3, 0.9 ± 0.8 %; L4, 96.6 ± 2.5 %; L5, 1.5 ± 0.9 %; L6, 1 ± 1.3 %; n=4) (Fig. S5). It is thus unlikely that these cells contribute significantly to our measurements.

Experiments started after 10 days of expression. To control for possible retrograde labeling we inspected labeled somata in barrel cortex after AAV virus injections in M1, POm and VPM. Even after immunohistochemical amplification we did not observe any labeled somata in barrel cortex.

Slice preparation. P26 to P34 mice were anesthetized with an intraperitoneal injection of a ketamine/xylazine mixture (0.13 mg ketamine/0.01 mg xylazine/g body weight) and perfused through the heart with a small volume (~ 5 ml) of ice cold ACSF containing in mM (in mM): 127 NaCl, 25 NaHCO3, 25 D-glucose, 2.5 KCl, 1 MgCl2, 2 CaCl2, and 1.25 NaH2PO4, aerated with 95% O2 / 5% CO2. The brain was removed and placed into ice-cold cutting solution containing (in mM): 110 choline chloride, 25 NaHCO3, 25 D-glucose, 11.6 sodium ascorbate, 7 MgCl2, 3.1 sodium pyruvate, 2.5 KCl, 1.25 NaH2PO4, and 0.5 CaCl2. 300 μm thick coronal slices of the right barrel cortex were cut with a vibrating slicer (Microm, Walldorf, Germany) and incubated in oxygenated ACSF for 45 min at 37°C before the recordings.

Identification and definition of cortical layers. L1 was identified by its low cell density. Based on functional mapping studies 21, L2 was defined as a 75 μm thick band immediately below L1. L3 was between L2 and the top of the barrels. L4 was identified by the presence of barrels (Fig. 1 d). L5A was identified as a light band under brightfield illumination (Fig. 1 d) below the barrels (~ 100 μm thick). L5B was a darker band immediately below L5A and contained the largest pyramidal cells (~ 200 μm thick). L6 extended from the bottom of L5B to the white matter.

Morphological reconstructions and analysis. During the recordings cells were loaded with biocytin (3mg/ml) and fixed overnight on 4% paraformaldehyde (PFA). For biocytin staining, after a quenching step ( 30 min in 1% H2O2 ), samples were incubated overnight in ABC solution (ABC kit, Vector Labs) in the presence of 4% Triton × ( Sigma). After rinsing for 8 hs in PBS, sections were reacted to DAB and mounted in AquaMount (Polysciences). Dendrites were reconstructed in Neurolucida (MicroBrightField) using a 40x objective and analyzed using custom Matlab (Mathworks ) routines. Neurolucida tracings were corrected for shrinkage. Dendritic length density 5 was calculated in 50 μm bins. For group averages, peak normalized dendritic length density maps were aligned to the soma (Fig. 2 and S7) or pia position (Fig. S6 and S8) and linearly interpolated for display.

Simulation of ChR2-mediated photoexcitation of axons. Non-myelinated axons (0.1 μm diameter) were simulated in NEURON 6.137. The passive membrane properties were: axial resistance, 150 Ω-cm; membrane capacitance, 0.75 μF/cm2. The membrane conductances were: fast Na+ current (INa), voltage-activated K+ current (IKv) 20, dendrotoxin-sensitive K+ current (IKd) 11, leak current with a reversal potential at −70mV. The reversal potentials for Na+ and K+ were 60 mV and −90 mV respectively. Channel densities were homogeneous throughout the axon: INa, 1000 pS/μm2; IKv, 20 pS/μm2; IKd, 2500 pS/μm2; leak , 6.67e-5 S/cm2. The ChR2 conductance was modeled as an alpha conductance with τ = 2.5 ms, gmax= 0.000154 μS and a reversal potential of 0 mV. The temperature of the model was 23 °C.

Supplementary discussion:

sCRACM compared to other methods for circuit mapping

Most cortical wiring diagrams describe the connectivity between populations of neurons, based on reconstructions of dendrites and axons and the supposition that synaptic connections are made in proportion to the overlap of dendrites and axons (i.e. ‘Peters’ rule’) 13, 15, 19, 38. Peters’ rule ignores specificity in cortical connections beyond the overlap of axons and dendrites 3-5, 39, 40, necessitating functional measurements or measurements with synaptic resolution, such as electron microscopy.

sCRACM is a novel method for mapping functional neural circuits with subcellular resolution. sCRACM measures connections between presynaptic cells, defined by ChR2 expression, and postsynaptic cells, defined by intracellular recording. sCRACM is therefore well-poised to probe functional connections between genetically defined neurons. Because photostimulation is performed with action potentials blocked, sCRACM does not suffer from ambiguities due to excitation of polysynaptic pathways. Neuronal circuits have been functionally mapped in the past using laser scanning photostimulation (LSPS) of caged glutamate 4, 21, 31, 41 or paired recordings 15, 19, 22. These methods require the circuit studied to be preserved in the slice and are therefore limited to the study of local connections. In contrast, because ChR2-labeled axons are photoexcitable, even when severed from their parent somata, functional connectivity can be mapped across all spatial scales, including long-range connections 8.

Electron microscopy (EM) methods can map inputs within the dendritic arbors of postsynaptic neurons with submicrometer resolution 3, 15, 19, 42. However, EM is typically used to analyze only short dendritic segments. It is possible that recently developed super-resolution optical techniques might allow more efficient mapping of the spatial distribution of synapses 43. However, structural techniques by themselves do not provide information about synaptic strength. sCRACM provides a lower resolution view of the distribution of functional input within entire dendritic arbors of single cells. sCRACM is therefore complementary to EM methods. sCRACM is applicable to any system in which afferent axons can be labeled.

The electrotonic spread of depolarization within photostimulated axons is likely a key determinant of sCRACM resolution. Simulations show that the length constant of thin axons (diameter, 0.1 μm) is ~ 70 μm, on the same order as the sCRACM resolution. The spread of excitation light within the tissue might also contribute, but likely plays only a minor role. Before entering the tissue, the diameter of the excitation beam was smaller than the sCRACM resolution. Furthermore, varying the beam diameter by almost a factor of three (from 6 to 16 μm) did not change sCRACM resolution. We conclude that the beam diameter by itself does not determine the sCRACM resolution. Within the tissue, the excitation light spreads due to light scattering. However, the EPSCsCRACM amplitude is a steeply supralinear function of the intensity of the excitation light (Fig. 1c), suggesting that scattered light might be too dilute to contribute appreciably to the EPSCsCRACM amplitude. The non-linear relationship between EPSCsCRACM amplitude and light intensity suggests that three-dimensional sectioning in sCRACM mapping could be achieved without two-photon excitation.

Impact of filtering on sCRACM maps

sCRACM maps are based on somatic measurements of synaptic currents generated in the dendrites signal attenuation due to dendritic filtering influences the structure of the maps. For example, input on the apical tufts of L5 neurons could be reduced several-fold compared to more proximal input 12. sCRACM maps thus represent a “somatocentric” view of the dendritic distribution of synaptic input, where electrotonically distant synapses will appear relatively weak (Fig. S8).

To approximate the relative input strength at the source (Fig. S8) we compensated the sCRACM maps for dendritic attenuation of synaptic charge. The dendritic attenuation, 1/g(r), was estimated as an exponentially increasing function with distance from the soma:

Q(x, y)~ _ρ_a(x, y) _ρ_b(x, y)f q g(L(x, y)).

where r(x,y) is the distance from the soma of the photostimulation location. We use λ = 270 μm 12, measured in rat L5 pyramidal neurons with K+-based internal solution 12. These measurements were obtained from the main apical tuft, whereas most synapses impinge on secondary and tertiary dendrites in the apical tuft. In addition, the measurements are from rat neurons with larger dendritic diameters compared to our mouse neurons. As a consequence, the attenuation-compensated maps (Fig. S8) likely still underestimate dendritic filtering by distant inputs.

Dendritic input domains

L3 pyramidal neurons received specific types of input on single, spatially distinct domains (Fig. 2). Remarkably, the position of VPM, L4, L2/3, and M1 input along the apical axis of L3 neurons mirrors the flow of excitation within the cortical circuit: more peripheral (central) input impinges on lower (higher) parts of the dendritic arbor. After stimulation of the principal whisker, lower basal dendrites are the first to receive input (directly from VPM, ~ 8 ms post-stimulus), followed by middle basal dendrites (from L4, ~ 10 ms post-stimulus), followed by upper basal and apical dendrites (from L2/3, ~13-16ms post-stimulus) 44, followed by the apical tuft in L1 (feedback from M1, ~10 ms later). Input for POm is expected to activate synapses sparsely over most of the dendritic tree (~20 ms post-stimulus) 45.

What determines the locations of the dendritic input domains? To asses the input from different presynaptic sources per unit length of dendrite we divided and peak-normalized vertical profiles of synaptic input by the mean vertical profile of dendritic length-density (Fig. S6a and Fig. S8). Unsurprisingly, the peaks in this profile overlapped roughly with axonal labeling (Fig. S5), with some exceptions (discussed below and in Fig. 3 c & d).

In some cases the overlap of axons with dendritic arbors seems to specify the location of the dendritic input domain. The axons of L2/3 neurons arborize in a thin (thickness ~ 200 um) lamina at the border of L5A and L5B (Fig. 4). The peak of the input from L2/3 onto L5 dendrites was closely aligned with the peak of axonal density (Fig. 4b). Cells with somata that were superficial to the band of L2/3 axons received L2/3 input on the bottom of the basal dendrites, whereas deep L5B cells received input on the upper half of the basal dendrites and in the apical oblique dendrites. In other cases, functional connections do not seem to be predicted by the product of axonal and dendritic density. The overlap of L2/3 axons and L3 dendrites is centered on the basal dendrites, dominated by the great abundance of dendritic branches 15. However, we found that the majority of L2/3→L3 input was above the soma and most of the basal dendrites (Fig. 2); this finding was independent of cortical depth of the postsynaptic neurons (Fig. S6). Thus, unlike L2/3→L5 input, the subcellular domain for L2/3→L3 input cannot be explained by the presence of presynaptic axons in a restricted lamina. L2/3→L3 inputs might avoid the denser basal branches, or get preferentially strengthened in the upper basal and apical dendrites. Since EPSCsCRACM amplitudes depend on both the fraction of axons that make synapses with the recorded neuron and the strength of these synapses, sCRACM can not distinguish these two alternatives.

Supplementary References:

34. Nagai, T. et al. A variant of yellow fluorescent protein with fast and efficient maturation for cell-biological applications. Nat Biotechnol 20, 87-90. (2002).

35. Nagel, G. et al. Light activation of channelrhodopsin-2 in excitable cells of Caenorhabditis elegans triggers rapid behavioral responses. Curr Biol 15, 2279-84 (2005).

36. Furuta, Y., Lagutin, O., Hogan, B. L. & Oliver, G. C. Retina- and ventral forebrain-specific Cre recombinase activity in transgenic mice. Genesis 26, 130-2 (2000).

37. Hines, M. L. & Carnevale, N. T. The NEURON simulation environment. Neural Comp. 9, 1179-1209 (1997).

38. Braitenberg, V. & Schutz, A. Anatomy of the Cortex (Springer Verlag, Berlin, 1991).

39. Song, S., Sjostrom, P. J., Reigl, M., Nelson, S. & Chklovskii, D. B. Highly nonrandom features of synaptic connectivity in local cortical circuits. PLoS Biol 3, 1-13 (2005).

40. Yoshimura, Y. & Callaway, E. M. Fine-scale specificity of cortical networks depends on inhibitory cell type and connectivity. Nat Neurosci 8, 1552-9 (2005).

41. Shepherd, G. M. G. & Svoboda, K. Laminar and columnar organization of ascending excitatory projections to layer 2/3 pyramidal neurons in rat barrel cortex. J. Neurosci. 25, 5670 (2005).

42. Somogyi, P., Tamas, G., Lujan, R. & Buhl, E. H. Salient features of synaptic organisation in the cerebral cortex. Brain Res Brain Res Rev 26, 113-35 (1998).

43. Micheva, K. D. & Smith, S. J. Array tomography: a new tool for imaging the molecular architecture and ultrastructure of neural circuits. Neuron 55, 25-36 (2007).

44. de Kock, C. P., Bruno, R. M., Spors, H. & Sakmann, B. Layer- and cell-type-specific suprathreshold stimulus representation in rat primary somatosensory cortex. J Physiol 581, 139-54 (2007).

45. Masri, R., Bezdudnaya, T., Trageser, J. C. & Keller, A. Encoding of Stimulus Frequency and Sensor Motion in the Posterior Medial Thalamic Nucleus. J Neurophysiol (2008)

Supplementary Table 1: Ratio of total sCRACM input per map in pairs of cells in different layers recorded in the same column with the same photostimulation power (mean ± S.D.). The total sCRACM input per map is a quantitative measure of the input from defined projections; it is defined as the sum of all sCRACM map pixels.

Figure S1: Summary of the main results. a, Schematic of the laminar position of the axonal arborizations from the five axonal projections studied. The sources of the projections are indicated by the boxes, stars and triangles. Horizontal bars indicate the layers of primary arborization. The relative thickness of the bars is a qualitative indication of the strength of the arborization. b, Schematic of the location of the different inputs (colors correspond to panel a) on the dendrites of L3, L5A and L5B pyramidal neurons. Vertical bars indicate the locations of functional input. The relative thickness of the bars is a measure of the strength of the input.

Figure S2: sCRACM responses as a function of power and distance to the soma. a, Onset of L2/3-->L5A EPSCsCRACM for four different neurons as a function of laser power. b, Onsets of EPSCsCRACM for three different axonal populations (from POm, L4 and L2/3) recorded in L5A pyramidal neurons. Each circle represents the mean value of the onset of the responses on the perisomatic area (<107 μm from the soma) for one cell.c Unitary EPSCsCRACM traces obtained by repeatedly photostimulating a single spot at the minimal laser power eliciting a response. The blue tick indicates the laser pulse. d, Group average of the 10-90% rise time of EPSCsCRACM responses of L2/3 inputs mapped onto L5 dendrites, aligned by soma position. e, Risetime vs. distance to the soma of the photostimulation site for basal dendrites. Responses at different distances from to soma in locations corresponding to the basal dendrites were binned and averaged (white semi-circles in d). f, Risetime vs. distance to the soma of the photostimulation site for apical dendrites. Pixels in the two columns above the soma (white box in d) were used, as they are more likely to correspond to the main trunk of the apical dendrite. Error bars, SD. (a,b) or s.e.m. (e,f).

Figure S3: sCRACM point-spread-function and resolution. a, Measurement of the sCRACM point-spread-function. Line scan over the apical dendrite of a L5A neuron receiving L2/3 input (mCherry fluorescence). Locations with non-zero EPSCsCRACM are color coded to represent mean amplitude. b, Responses shown in Panel a (blue) and similar responses obtained with twice the laser power normally used for sCRACM mapping (red). c, Full-width-at-half-maximum (FWHM) of the Gaussian fits of the responses obtained on apical dendrites of different L5A pyramidal neurons. Experiments performed at the same locations with different laser powers are connected by lines. d, L2/3-->L5 input on an apical dendrite. e, Histogram of the distance (‘d’ in panel d) between the apical dendrite and locations where non-zero EPSCsCRACM were evoked by photostimulation of L2/3 (n=6) , L4 (n=5) or VPM axons (n=3).

Figure S4: Reproducibility of sCRACM maps. a, Left, sCRACM map overlaid on a brain slice containing ChR2-Venus-positive L2/3 neurons. Right, four repetitive EPSCsCRACM responses recorded from a L5A pyramidal neuron. The stimulus occurred at the beginning of each trace. Traces correspond to the pixels surrounded by the dashed boxes. b, sCRACM maps of L2/3 input on a L5A pyramidal neuron at different laser powers (white triangles, soma position). c, Same maps as in b, normalized to their peak. d, Four repetitions of individual maps and corresponding average map (same cell as in a). e, Two-dimensional correlation coefficients between map repetitions (n=8 cells, 3 repetitions per cell), and between peak-normalized maps obtained with different laser powers (n=6 cells, average of 3 repetition per power level). Error bars, SD.

Figure S5: Targeting ChR2 expression to specific neuronal populations. a, mCherry fluorescence from L2/3 cells transduced by in utero electroporation. b, Detail of the barrel cortex, showing mCherry and ChR2-Venus fluorescence. c, ChR2-mCherry expression after injection of Cre-recombinase-dependent ChR2-mCherry AAV in the barrel cortex of a Six3Cre mouse. d, Detail of mCherry expression in barrel cortex. e, AAV virus expressing ChR2-Venus in the VPM nucleus of the thalamus. f, Detail of the barrel cortex, showing ChR2-Venus labeled VPM axons.g, GFP-immunostained ChR2-Venus labeled VPM axons in layers 1-3 of barrel cortex. h, AAV virus expressing ChR2-Venus in the POm nucleus of the thalamus. i, Detail of the barrel cortex, showing ChR2-Venus labeled POm axons.j, GFP-immunostained ChR2-Venus labeled POm axons in layers 1-3 of barrel cortex. k, AAV virus expressing ChR2-Venus in M1. l, Detail of the barrel cortex showing ChR2-Venus positive axons. m, Confocal image of GFP-positive M1 axons in layers 1-3 of barrel cortex.

Figure S6: Inputs to L3 pyramidal neurons. a, Vertical profile of the laminar distribution of the dendritic input density obtained by dividing the pia-aligned synaptic input maps (from Fig 2a) by the dendritic length-density of L3 pyramidal neurons (right). Error bars, s.e.m. b, Vertical profiles of the relative subcellular distribution of L2/3-->L2/3 inputs. Each column represents an individual cell, ordered by increasing cortical depth. Cells were aligned by pia position. The white vertical lines indicate the boundaries of the three cortical depth bins used for averaging in c. c, Average relative subcellular location of L2/3 inputs, aligned by soma (top) or pia position (bottom) , of groups of L2/3 cells at increasing distances from the pia (corresponding to b). d, Vertical profiles of the relative subcellular distribution of L4-->L2/3 inputs. e, Average relative subcellular location of L4 inputs, aligned by soma(top) or pia position (bottom) , of groups of L2/3 cells at increasing distances from the pia (corresponding to d).

Figure S7: Subcellular distribution of specific inputs onto L5A pyramidal neurons. a, Examples of sCRACM maps overlaid on reconstructed dendritic arbors and fluorescence images. b, Group averages aligned by pia position. White triangles, soma position. c, top: Group averages aligned by soma position ; bottom: vertical profiles of the distribution of synaptic input (red) and the dendritic length density (green; from panel d). d, Average normalized dendritic length density of L5A pyramidal neurons. Error bars, s.e.m.

Figure S8: Inputs to L5 pyramidal neurons, corrected for dendritic attenuation and dendritic density. a, Top, Group averages corrected for the estimated dendritic attenuation of inputs onto L5B pyramidal neurons aligned by pia position. White triangles, soma position. Bottom, Vertical profiles of the distribution of synaptic input as recorded from the soma (red) or corrected for the estimated dendritic attenuation (blue). Vertical profiles of the dendritic length density (green; from panel b). Error bars, s.e.m. b, Average normalized dendritic length density of L5B pyramidal neurons. c Vertical profiles of the distribution of synaptic input divided by the dendritic length-density of L5B pyramidal neurons, as recorded from the soma (red), or corrected for the estimated dendritic attenuation (blue). d, Top, Group averages corrected for the estimated dendritic attenuation of inputs onto L5A pyramidal neurons aligned by pia position. White triangles, soma position. Bottom, Vertical profiles of the distribution of synaptic input as recorded from the soma (red) or corrected for the estimated dendritic attenuation (blue). Vertical profiles of the dendritic length density (green; from panel e). Error bars, s.e.m.e, Average normalized dendritic length density of L5A pyramidal neurons. f, Vertical profiles of the distribution of synaptic input divided by the dendritic length-density of L5A pyramidal neurons, as recorded from the soma (red), or corrected for the estimated dendritic attenuation (blue).

Figure S9: Simulations of Chr2-mediated photostimulation of axons. ChR2-mediated depolarization of the axons, with the transient Na+, fast voltage activated K+ and dendrotoxin (IKd) sensitive K+ currents (black), without the Na+ current (blue) or without the Na+ and IKd currents (red).

GUID: 04759FBB-36B6-465F-A77F-3DF359F61F07

Abstract

Understanding cortical circuits will require mapping the connections between specific populations of neurons 1, as well as determining the dendritic locations where the synapses occur 2. The dendrites of individual cortical neurons overlap with numerous types of local and long-range excitatory axons, but axodendritic overlap is not always a good predictor of actual connection strength 3-5. Here we developed an efficient Channelrhodopsin-2 (ChR2)-assisted method 6-8 to map the spatial distribution of synaptic inputs, defined by presynaptic ChR2 expression, within the dendritic arbors of recorded neurons. We expressed ChR2 in two thalamic nuclei, the whisker motor cortex and local excitatory neurons and mapped their synapses with pyramidal neurons in layers (L) 3, 5A, and 5B in the mouse barrel cortex. Within the dendritic arbors of L3 cells, individual inputs impinged onto distinct single domains. These domains were arrayed in an orderly, monotonic pattern along the apical axis: axons from more central origins targeted progressively higher regions of the apical dendrites. In L5 arbors different inputs targeted separate basal and apical domains. Input to L3 and L5 dendrites in L1 was related to whisker movement and position, suggesting a role of these signals in controlling the gain of their target neurons 9. Our experiments reveal exquisite specificity in the subcellular organization of excitatory circuits.

We recorded from pyramidal neurons in neocortical brain slices containing ChR2-expressing axons 7(Fig. 1a). To map the dendritic locations of input from ChR2-positive axons (Fig. S1) we used a laser to depolarize these axons only in the vicinity of the laser beam (i.e. with action potentials blocked), triggering local glutamate release (subcellular Channelrhodopsin-2 Assisted Circuit Mapping, sCRACM). We blocked Na+ channels (1 μM tetrodotoxin,TTX ) and the K+ channels that are critical for repolarization of the axon (200 nM α-dendrotoxin or 100 μM 4-aminopyridine, 4-AP) (Fig. 1 b)10. Under these conditions, photostimulation with short (1 ms) light pulses (< 2 mW) triggered robust excitatory postsynaptic currents (EPSCsCRACM) (Fig. 1 b). Higher light intensities caused larger EPSCsCRACM amplitudes (Fig. 1 c) and shorter onsets (Fig. S2). As the cylindrical laser beam was scanned over the dendrites of a recorded neuron (map pattern: 12 × 24 grid; 50 μm spacing), EPSCsCRACM were detected only when the laser beam overlapped with the dendritic arbor of the recorded cell and with ChR2-positive axons (Fig. 1 b, d-f), indicating that under these conditions light depolarizes ChR2-positive axons to cause local release of neurotransmitter. Converting EPSCsCRACM into pixel values (EPSCsCRACM averaged over a time window 0 – 75 ms post-stimulus) provides a two-dimensional ‘image’ of the distribution of specific input within the dendritic arbors of the recorded cell (Fig. 1 e,f). Measurements of the point-spread-function revealed that sCRACM maps specific types of input with ~ 60 μm spatial resolution (Fig. S3).

Subcellular Channelrhodopsin-2-Assisted Circuit Mapping (sCRACM)

a, Confocal image showing L2/3 neurons expressing mCherry (red) and ChR2-Venus (green) in the barrel cortex. b, Left, Schematic of the photostimulation geometry. Right, EPSCs evoked by photostimuli corresponding to the locations indicated in the schematic. Blue ticks indicate the laser pulse. Laser power is indicated on top. c, EPSCsCRACM evoked by photostimulation with increasing laser powers (right). d, Brightfield image of a brain slice showing the recording pipette and the photostimulation grid (blue dots). e, sCRACM map overlayed on a fluorescence image, showing ChR2-positive neurons, and the reconstructed dendrite of the recorded neuron (same experiment as d). Non-zero EPSCsCRACM are color-coded to represent mean amplitude. f, EPSCsCRACM recorded in the boxed regions in Panel e.

EPSCsCRACM amplitudes depend on the density of ChR2-positive axons, the fraction of axons that make synapses with the recorded neuron, the strength of the synapses, and their electrotonic distance from the soma 11 (see Methods). Since the density of ChR2-positive axons varies between preparations, sCRACM maps were normalized to the largest pixels within a map and thus represent the relative strength of input within the dendritic tree. Repeated sCRACM maps where reproducible at the level of single pixels (Fig. S4), and the structure of peak-normalized maps was similar across a large range of light intensities (Fig. S4).

Multiple types of axon overlap with the dendrites of cortical neurons 5, 12-17. Some axons arise locally 5, 14, 18, whereas others ascend from the thalamus 15, 16 or descend from higher cortical areas 17. Which axons in a target region connect with a particular cell type? And what are the spatial distributions of input within the dendritic tree? To answer these questions we expressed ChR2 in five distinct axonal populations (in separate experiments) that overlap with pyramidal cell dendrites in the barrel cortex (Fig. S5). The spatial distribution of labeled axons in the barrel cortex was largely in agreement with previous anatomical studies (Fig. S5). Projections from the ventral posterior medial nucleus (VPM) of the thalamus were focused in L4 and at the border of L5 and L6 (Fig. 2 a and Fig. S5) 16, but diffuse axons were found throughout all cortical layers 15. L4 axons arborized in L4 and ascended into L2/3, up to the lower edge of L1 18; a weaker projection descended into L5 and L6 18. Axons from L2/3 pyramidal cells arborized within L2/3 and on the border of L5A and L5B 7, 14. Axons from the primary whisker motor cortex (M1) arborized densely in L1, and more diffusely in L5 and L6 17. A dense bundle of ascending M1 axons was often apparent next to the most medial barrels (Fig. 2 a, arrow head). Axons from the posterior medial nucleus (POm) of the thalamus were focused in L5A, and more weakly in L1 15.

Subcellular distribution of inputs onto L3 pyramidal neurons

a, Examples of sCRACM maps overlaid on reconstructed dendrites and fluorescence images showing ChR2-positive axons (VPM, M1 and POm) or axons and dendrites (L2/3 and L4). White arrow head, bundle of ascending axons from M1. b, Group averages aligned by pia position (White triangles, soma position). c, top: Group averages aligned by soma position ; bottom: vertical profiles of the distribution of synaptic input (red) and the dendritic length density (green; from d). d, Average normalized dendritic length density of L3 pyramidal neurons. Error bars, s.e.m.

We mapped specific types of input within the dendritic trees of individual L3 cells (Fig. 2a). Maps were then averaged across cells either aligned on the pia (Fig.2 b), to visualize the laminar location of the inputs, or aligned on the soma, to measure the location of the inputs relative to the soma (Fig.2 c). L3 cells received input from all five projections. Each input overlapped with a single contiguous dendritic subregion. Ascending VPM→L3 input was focused on the bottom part of the basal dendritic arbors. Input from ascending L4→L3 axons was centered on the soma and basal arbor, above the input from VPM. Input from recurrent L2/3→L3 axons was mostly in the upper basal dendrites and the apical oblique dendrites, above the input from L4 (see also Fig.S6). Feedback from M1 targeted the tuft branches in L1, above the input from L2/3. The positions of VPM, L4, L2/3, and M1 input along the apical axis of L3 neurons mirrors the flow of excitation within the cortical circuit: more peripheral (central) input impinges on lower (higher) parts of the dendritic arbor. During somatosensation L3 neurons thus receive an ascending wave of excitatory input. POm→L3 input was weighted towards L1, although it was distributed relatively broadly, spanning most of the dendritic arbor.

We next mapped the same group of five inputs within the dendrites of L5 pyramidal neurons. Both L5A and L5B pyramidal cells received input from L4, L2/3, M1 and VPM (Fig.3 a,b and S7) . To quantify the strength of input from defined axonal projections across postsynaptic cells in different layers, we recorded from pairs of cells in the same column with identical laser powers (Supplemental table 1). L5B cells received 62-fold less input from POm compared to L5A cells, despite pronounced overlap between L5B dendrites and POm axons (Fig. 3c,d, Fig. S7 and Supplemental table 1) (dendritic length: L5A pyramids (n=12), in L1, 871±546 μm , in L5A, 2158±899 μm; L5B pyramids (n=13), in L1, 1609 ±732 μm, in L5A, 761 ± 350 μm). L5A, but not L5B, pyramidal cells received significant input from POm. This confirms that average cortical connectivity between populations of neurons cannot always be deduced from the structure of axons and dendrites alone 3, 4, 5 . Because L5B pyramidal neurons constitute the main projection from barrel cortex to POm 19, there appears to be no disynaptic loop between these two areas.

Subcellular distribution of inputs onto L5B pyramidal neurons

a, Examples of sCRACM maps overlaid on reconstructed dendritic arbors and fluorescence images. b, Group averages aligned by pia position. White triangles, soma position. c, sCRACM map of POm input onto a L5A pyramidal cell (blue). No responses were detected on the L5B neuron (green). d, EPSCsCRACM recorded on the L5A neuron (blue) or the L5B neuron (green) when photostimuating the boxed regions in Panel c . The stimulus occurred at the beginning of each trace.

In contrast to L3 cells, the inputs on L5 cells were not limited to a single compartment, but were split into basal and apical domains (Fig.3 and Fig. S7), reinforcing the view that large pyramidal neurons consist of multiple, weakly coupled compartments 20. Here we describe the inputs to L5B neurons (Fig. 3), and then highlight the differences with L5A neurons (Fig. S7). VPM→L5B input was distributed along most of the dendritic arbor, but was most prominent on the basal dendrites and in L4. L4→L5B input was centered on the basal dendrites, overlapping with VPM input; weak input was also detected along the apical dendrite up to the edge of L1. L2/3→L5B input was focused on the upper basal and apical oblique dendrites, as well as on the apical tuft in L2. M1→L5B input was on the basal dendrites and on the apical tuft in L1.

Inputs to L5A pyramidal neurons similarly targeted dual dendritic compartments (Fig. S7), with some differences. POm →L5A input was prominent, both on the basal dendrites and the apical tuft in L1. Unlike for L2/3→L5B input, L2/3→L5A input was centered on the basal dendrites (Fig. 4 a and Fig. S7). For all L5 neurons taken together, there was a monotonic relationship between the laminar position of the recorded cell and the location of L2/3 input relative to the soma position (Fig. 4 b-d). The axodendritic overlap of L2/3 axons and L5 dendrites is likely an important factor determining the subcellular location of L2/3→L5 input.

The laminar position of L5 pyramidal neurons determines the dendritic location of L2/3 inputs

a, Examples of the subcellular distribution of L2/3 input on superficial (L5A) (left) or deep (L5B) (right) pyramidal neurons. b, Vertical profiles of the subcellular distribution of L2/3→L5 inputs. Each column represents one cell, ordered by cortical depth. Cells were aligned by pia position. The relative density of L2/3 axons in the deep layers is indicated to the left of the panel. c, Average subcellular location of L2/3 input (aligned by soma position) of L5 cells grouped by increasing distance from the pia (groups correspond to the white lines in b).d, Plot of the vertical distance from the soma to the center of mass of L2/3 input on the perisomatic region ( < 285 μm from the soma) of L5 pyramidal neurons vs. cortical depth. The line is a regression fit.

sCRACM maps functional neural circuits with subcellular resolution. Since sCRACM maps are based on somatic measurements of synaptic currents generated in the dendrites signal attenuation due to dendritic filtering influences the structure of the maps. For example, input on the apical tufts of L5 neurons could be reduced several-fold compared to more proximal input 11. sCRACM maps thus represent a “somatocentric” view of the dendritic distribution of synaptic input, where electrotonically distant synapses will appear relatively weak (Figure S8).

We mapped input from VPM, L4, L2/3, M1 and POm, within the dendritic arbors of L3 and L5 neocortical pyramidal cells. L3 and L5 cells received input from most axonal populations, with the exception of POm→L5B. Some connections (VPM→L3, VPM→L5A, VPM→L5B, POm→L5A, L4→L3, L4→L5A, L2/3→L3, L2/3→L5A, L2/3→L5B) have been previously characterized 3, 14, 18, 21, 22, whereas others (M1→L3, M1→L5A, M1→L5B, POm→L3, L4→L5B,) were previously unknown.

We identified several connections in L1: M1→L3, M1→L5A, M1→L5B, POm→L3 and POm→L5A. Axons from VPM, L2/3 and L4 did not contribute significantly to L1 input. POm neurons are thought to encode aspects of whisker position 23 and whisker M1 carries signals related to voluntary whisker control 24. Our findings suggest that synapses in L1 carry signals related to whisker movement and position.

The spatial segregation of specific types of input within dendritic arbors might subserve several functions. Segregated inputs are less likely to interact at the level of synaptic plasticity 25. Spatially clustered coactive synapses are more efficacious in driving postsynaptic neurons than spatially distributed synapses 26, 27. For a fixed number of synapses, spatial segregation of different axonal populations within dendritic arbors might thus serve to strengthen the effective coupling between pre- and postsynaptic populations.

Supplementary Material

1