Cell-Type Specific Connectivity of Whisker-Related Sensory and Motor Cortical Input to Dorsal Striatum

The anterior dorsolateral striatum (DLS) is heavily innervated by convergent excitatory projections from the primary motor (M1) and sensory cortex (S1) and considered an important site of sensorimotor integration. M1 and S1 corticostriatal synapses have functional differences in their connection strength with striatal spiny projection neurons (SPNs) and fast-spiking interneurons (FSIs) in the DLS and, as a result, exert distinct influences on sensory-guided behaviors. In the present study, we tested whether M1 and S1 inputs exhibit differences in the subcellular anatomical distribution of striatal neurons. We injected adeno-associated viral vectors encoding spaghetti monster fluorescent proteins (sm.FPs) into M1 and S1 in male and female mice and used confocal microscopy to generate 3D reconstructions of corticostriatal inputs to single identified SPNs and FSIs obtained through ex vivo patch clamp electrophysiology. We found that M1 and S1 dually innervate SPNs and FSIs; however, there is a consistent bias towards the M1 input in SPNs that is not found in FSIs. In addition, M1 and S1 inputs were distributed similarly across the proximal, medial, and distal regions of SPN and FSI dendrites. Notably, closely localized M1 and S1 clusters of inputs were more prevalent in SPNs than FSIs, suggesting that cortical inputs are integrated through cell-type specific mechanisms. Our results suggest that the stronger functional connectivity from M1 to SPNs compared to S1, as previously observed, is due to a higher quantity of synaptic inputs. Our results have implications for how sensorimotor integration is performed in the striatum through cell-specific differences in corticostriatal connections.


Introduction
Sensorimotor integration is the ability to identify important stimuli from the environment and use this information to guide behaviors essential for survival (Machado et al., 2010).Neurons in the primary motor (M1) and primary sensory (S1) cortices play an important role in this process by sending excitatory glutamatergic projections to the striatum, a subcortical region, and the main input nucleus of the basal ganglia network (Hoffer and Alloway, 2001).Generally, cortical projections from functionally related brain regions show higher density and significant overlap amongst their projection fields in the striatum (Bolam et al., 2002;Hintiryan et al., 2016;Hunnicutt et al., 2016).Notably, the dorsolateral striatum (DLS) is a site of overlap between inputs from M1 and S1 and plays a critical role in sensorimotor integration (Hoffer and Alloway, 2001;Hunnicutt et al., 2016;Makino et al., 2016;Hooks et al., 2018;Smith et al., 2022), but how these inputs are integrated by different cell types in DLS is still unclear.
The DLS mostly contains inhibitory GABAergic spiny projection neurons (SPNs) and a sparse population of various interneurons (Kawaguchi, 1993(Kawaguchi, , 1997;;Tepper et al., 2018).Activation of SPNs that express the D1 dopamine receptor constitutes the "direct" pathway and promotes selected motor actions, while SPNs that express the D2 dopamine receptor constitute the "indirect" pathway and inhibit nonselected motor actions (Kreitzer and Malenka, 2008).One major interneuron type, the parvalbumin-positive fast-spiking interneuron (FSI), is a potent inhibitor of striatal SPNs and involved in the sensorimotor transformation of behavior (Lee et al., 2019;Gritton et al., 2019;Martiros et al., 2018).For example, stimulation of S1 inputs to FSIs in the DLS suppresses responding to both rewarded and unrewarded cues in a whisker-based texture discrimination task.In contrast, stimulation of the M1 inputs, which has a different balance of functional innervation to SPNs compared to S1, promotes behavioral responses in the same task (Lee et al., 2019).SPNs and FSIs rely on excitatory inputs along the extent of their elaborate dendrites to drive action potential (AP) discharge (Kawaguchi, 1993(Kawaguchi, , 1997;;Stern et al., 1998;Tepper et al., 2018).Therefore, both the local connectivity and external synaptic inputs are important factors of integration in the DLS (Straub et al., 2016;Hjorth et al., 2020).
While M1 and S1 inputs are capable of evoking excitatory postsynaptic potentials (EPSPs) in SPNs and FSIs, recent reports have found that S1 inputs generate larger responses in FSIs compared to SPNs (Lee et al., 2019;Johansson and Silberberg, 2020).Conversely, Lee et al. (2019) found that the response to stimulation of M1 inputs was similar for SPNs and FSIs.Potential differences in synaptic efficacy of M1 and S1 corticostriatal inputs to striatal neurons could reflect presynaptic, postsynaptic, or structural differences (Fino et al., 2005;Perrin and Venance, 2019;Badreddine et al., 2022), mechanisms that could interact to produce differences in functional innervation.However, whether there are anatomical differences in the connectivity of M1 and S1 to SPNs and FSIs remains to be investigated.Two properties that have been associated with the efficacy of total synaptic input that a cell receives are the number of synaptic inputs and the distribution of synapses along the dendritic arbor (Rall, 1967;Magee, 2000).Understanding how M1 and S1 inputs are organized onto SPNs and FSIs can reveal how the DLS integrates functionally related signals and, commands downstream basal ganglia nuclei, which ultimately affects behavior.Here, we investigate the structural input to SPNs and FSIs and provide evidence for the differential innervation of striatal neurons by M1 and S1.

Materials and Methods
Animals.All procedures involving animals were approved by the Rutgers University Institutional Animal Care and Use Committee (Protocol #: 999900197).Experiments were performed on wild-type male and female mice on a C57BL/6J background of at least 3 months of age at the time of the first surgical procedures.Mice were housed in a reverse light cycle room (lights off from 08:00 to 20:00) with food and water available ad libitum.All mice underwent two simultaneous unilateral adeno-associated virus (AAV) injections between 35 and 104 d old, and all mice were euthanized after at least 3 weeks post-injection.
Whole-cell patch clamp recordings were obtained in a chamber that was constantly perfused (2-4 ml/min) with oxygenated external ACSF at 34°C.Sections and neurons were visualized using infrared differential interference contrast (IR-DIC) microscopy with an IR1000 CCD camera (Dage-MTI) mounted onto a BX51-WI upright microscope (Olympus) fitted with two switchable lenses: a 4× air lens and a 40× water-immersion lens.Patch pipettes (2-5 MΩ) were fabricated by pulling borosilicate glass micropipettes (2 mm o.d., Warner Instruments) via a P-1000 horizontal puller (Sutter Instruments).The internal pipette solution contained 130 mM K methanesulfonate, 10 mM KCl, 10 mM HEPES, 2 mM MgCl 2 , 4 mM Na 2 ATP, 0.4 mM Na 2 GTP at pH 7.25, and 290-295 mOsm/L.In addition, 2% biocytin was freshly dissolved in the internal solution on each recording day.Current clamp recordings were obtained from neurons within the anterior dorsal striatum (approximately 1.4 to 0.4 mm relative to bregma).Once a selected cell was patched, it was subjected to hyperpolarizing and depolarizing current steps (−500 pA to 500 pA, 100 pA steps, 500 ms, 11 sweeps) for post hoc electrophysiological analysis and identification.Additionally, patched neurons were held for at least 15 min to permit biocytin filling for post hoc morphological analysis and identification.Data was acquired via a HEKA EPC10USB amplifier and digitized at 20 kHz in Patchmaster Next (HEKA).We did not correct for the liquid junction potential.
Analysis of patch clamp recordings.Analysis of electrophysiological responses to hyperpolarizing and depolarizing current injections was performed using MATLAB and Python.In MATLAB, a custom script was used to import (Keine, 2022), standardize, and save the data as a .matvariable.A custom python script was used to import and analyze the data stored in the mat variable, partly using the electrophysiology feature extraction library (eFEL; Van Geit et al., 2016) of the Blue Brain Project (Blue Brain Project, 2015).Briefly, eFEL was used to mark specific sweep events including the values and indices of AP thresholds, peaks, and after hyperpolarizations (AHP) to calculate physiological parameters including halfheight width (HHW), interspike interval (ISI), instantaneous firing frequency (IFF), and max firing frequency.The HHW of each AP (from AP threshold to AHP) was calculated by setting half the maximum amplitude [(peak value − AP threshold value)/2] as a horizontal threshold, interpolating a line containing 1,000 points over the AP, identifying when the interpolated line crossed the threshold during the rising and falling phases, and subtracting the values from each other.The ISI was calculated by subtracting the latter AP threshold value from the former (e.g., ISI[i] = APthres[i + 1] -APthres[i]).The IFF was calculated by dividing the ISI values by 1 (e.g., IFF = 1/ISI).The max firing frequency was calculated by selecting the maximum frequency value [e.g., MFF = max (IFF)].Neurons that did not reach steady state firing during patch clamp recordings were excluded from the analysis of electrophysiological parameters.
Confocal imaging.Confocal Z-stack images were obtained using a Zeiss LSM 800 confocal microscope.Fixed slices containing filled neurons from ex vivo recordings were imaged using a 40× oil immersion objective [Plan-Apochromat 40×/1.4Oil DIC (UV) VIS-IR M27] set at 1.0 airy disk unit.The biocytin-filled cell was centered in the field of view, and the zoom factor was set to a minimum of 0.5× or adjusted to capture the entirety of the dendritic field.Images acquired at either 1,024 × 1,024 or 2,048 × 2,048 pixels with a 0.5× zoom factor corresponded to a minimum voxel dimension of 0.312 × 0.312 × 0.530 µm or 0.156 × 0.156 × 0.530 µm, respectively, in the X, Y, and Z dimensions.Fluorescence acquisition settings were as follows: Alexa Fluor 488 (excitation 493λ, emission 517λ, detection 490-550λ), Alexa Fluor 568 (excitation 577λ, emission 603λ, detection 565-642λ), and Alexa Fluor 647 (excitation 653λ, emission 668λ, detection 645-700λ).The pinhole size of the objective was set to 1.0 airy disk unit when capturing the longest wavelength fluorophore (Alexa Fluor 647).When imaging the other fluorophores (Alexa Fluor 405, 488, and 568), the pinhole size was adjusted to match the airy disk size of the 647 laser.This ensured that the scan with each laser excited fluorophores in the same sized area.Optical laser intensities were set individually for each channel and optimized throughout the Z-stack using the Z-stack auto-brightness correction tool.Z-stack ranges were set manually by tracking the range of the entire labeled cell.
Morphological measurements.For each neuron in our dataset, we obtained the position along the AP, ML, and DV axes by manually aligning the coronal field view of the recorded slice with the corresponding Allen brain atlas figure (Allen Reference Atlas).We were not able to obtain the ML and AP positions for one SPN and one FSI.Dendritic field measurements were obtained by creating a 2D orthogonal projection of the confocal Z-stack and then measuring the largest diameter of an elliptical sphere encompassing the neuron using ImageJ (Fiji) software.The range along the z-axis for each cell was calculated as the distance between the first and last appearance of neuronal processes in the Z-stack.
Imaris 3D reconstructions of single neurons.Carl Zeiss image files were imported into Imaris version 9.7 equipped with the Filament Tracer plugin (Bitplane).Biocytin-filled neurons were first reconstructed into 3D surface objects using the "surfaces" tool.For our surface reconstructions, the "background subtraction" option was enabled, and the "smoothing" option was disabled to avoid adding artificial curvature to the cell (Fogarty et al., 2013).We determined the largest diameter setting by measuring the largest cross-sectional diameter of the soma in the "slice view" mode.Using an interactive histogram of volumetric pixels, we manually adjusted the threshold to include as much of the neuron as possible while aiming to exclude any extraneous background fluorescence.Neurons were then reconstructed using the "filament tracer" plugin using a semiautomated tracing method.Reconstructing the neuron as a filament allowed us to segment the neuron into separate branch levels and dendrites, as well as obtain distance measurements for M1 and S1 inputs from the beginning of the filament (soma).To construct the neuron as a "filament," we manually traced the dendrites using the "autopath" tool.Once the final outline of the neuron was created, the diameter of the dendrites was adjusted using an automatic threshold.For SPNs, we used the automatic spine detection tool in the "filament" tab to detect and reconstruct dendritic spines.The diameter and length of spines were measured in "slice viewer" using min/max diameters of 0.6 µm and 2.5 µm for automatic spine detection thresholds.This step was excluded from our FSI dataset due to the lack of spines on their dendrites.We verified that there was a single filament for each reconstruction and that all the branch points and dendrites were accurately represented in comparison to the raw fluorescence images.
M1 and S1 corticostriatal synaptic puncta identification and quantification.The Imaris "surface" reconstruction enabled us to mask the raw fluorescence from M1 and S1 channels onto the neuron.Using the "edit" tab within the surface reconstructed neuron, we masked putative presynaptic contacts from M1 and S1 corticostriatal projections by setting the intensity of voxels outside of the surface to 0. This allowed us to filter putative presynaptic contacts so that only fluorescence from M1 and S1 that colocalized with the surface reconstruction remained.This step resulted in two new fluorescent channels representing masked fluorescence from M1 and S1 onto the neuron.
Fluorescence from masked M1 and S1 inputs were made into 3D spots using the "spots" tool.First, the cross-sectional diameter of putative presynaptic puncta from cortical projections was measured in "slice viewer."The "different spot sizes" option was checked because fluorescent puncta from M1 and S1 varied in size.We chose a minimum diameter of 1 µm, and the "background subtraction" option was disabled.Spots were manually thresholded using an interactive histogram and then validated by verifying that the new spot object was representing masked fluorescence from a corticostriatal projection and not from extraneous background fluorescence.The "spots region type" was set to "local contrast," and the "spots regions" diameter threshold was set to "region border."We used automatic thresholds for these two parameters and found that they gave us the best representation of our puncta as 3D objects.
In Imaris, the default "spots close to filament" function is capable of filtering and quantifying M1 and S1 spots whose center lies within a specified distance from the edge of a neuron reconstructed as a filament object.However, it was limited in that it provided us with the absolute distance of each spot from the soma instead of considering the length of the dendrites.To circumvent this issue, we used a publicly available custom python script provided by Dr. Matthew Gastinger (Gastinger, 2022).Using the "spots close to filament" function provided by this script, we filtered S1 and M1 spots so that only those with edges within 0.5 µm from the edge of the filament-constructed neuron remained, and we identified these spots as putative M1 and S1 inputs.This custom script also provided us with the quantity of inputs and the distance from the soma measured along the dendrite for each spot object.
The number of inputs to spines was determined by removing dendrites from the 3D reconstruction so that only the dendritic spines remained.We then quantified the number of M1/S1 inputs that were within 0.5 µm of the new filament containing only spines.Next, we counted the number of inputs to the soma and subtracted these combined values (spines + soma) from the total number of counts on the filament to determine the inputs that were solely on dendritic shafts.We visually confirmed that these spots were not attached directly to a spine head or neck.The density of inputs was determined by dividing the number of inputs on the soma by the membrane area of the soma, and for dendrites, this was determined using the built-in statistic in Imaris.
Statistical analyses.All statistical analyses were performed using GraphPad Prism software.For each metric, we first tested if the data was normally distributed using the Shapiro-Wilk test (Ghasemi and Zahediasl, 2012;Mishra et al., 2019).For normally distributed data, parametric comparative t tests were used.A statistical F test was used to compare variance within each group of data.Levene's test was used to compare the variance between SPNs and FSIs because of its robustness to deal with small and unequal sample sizes.If the standard deviation between two normally distributed data sets was significantly different, the unequal variance Welch's t test was used.For unpaired samples that were not normally distributed, the nonparametric Mann-Whitney U test was used.For paired data that were not normally distributed, we used the Wilcoxon matched-pair signed-rank test.For multiple group comparisons, we used the analysis of variance (ANOVA) statistical test.All data are presented as the mean ± SEM.Power analysis was performed using the statistical software G*Power (Faul et al., 2007).For all statistical tests, p < 0.05 is denoted as *, p < 0.01 is denoted as **, and p < 0.001 is denoted as ***.
Code accessibility.A custom Imaris extension written and provided by Dr. Matthew Gastinger is publicly available through GitHub (Gastinger, 2022).Scripts used for the analysis of the electrophysiological recordings are available through the MargolisLab Github (https://github.com/margolislab).

M1 and S1 corticostriatal projections converge in the DLS
We injected adeno-associated viruses (AAVs) encoding spaghetti monster fluorescent proteins (sm.FPs) GFP or Ruby into whisker M1 and whisker S1 to observe putative synaptic contacts onto striatal neurons via confocal light microscopy.We used sm.FPs because of their utility in multicolor experiments, enhanced fluorescence, and better resolution when imaging fine structures such as synaptic terminals (Viswanathan et al., 2015).To ensure there was no bias in the expression of sm.FPs due to the injection site, we counterbalanced the injection of sm.FPs in M1 and S1 throughout the experiments.In total, eight mice were injected with sm.FP-GFP in S1 and sm.FP-Ruby in M1, and six mice were injected with sm.FP-GFP in M1 and sm.FP-Ruby in S1.
After waiting at least 3 weeks to permit viral expression, coronal slices containing the anterior striatum were obtained.Ex vivo whole-cell patch clamp recordings were performed ipsilateral to the injection site to characterize the electrophysiological properties and fill individual neurons with biocytin (Fig. 1A).Recorded and filled neurons were located within extensively overlapping sm.FP-labeled M1 and S1 corticostriatal projections in the dorsolateral region of the striatum (DLS; Fig. 1B).Our final data set includes 19 striatal neurons from 14 mice (5 males, 9 females).In addition, 16 additional cells from 9 mice were excluded from reconstruction and analysis due to incomplete biocytin cell fills, inconclusive immunohistochemistry containing artifacts, excessive background fluorescence, or the inability to be identified as an SPN or FSI via their morphology or electrophysiology.

Patched SPNs and FSIs are identified by their distinct morphological and electrophysiological properties
Striatal neurons were patched without the use of genetic cell identification due to the expression of sm.FPs GFP and Ruby in cortical inputs, which limited our ability to record from SPNs or FSIs labeled with fluorescent markers such as GFP or TdTomato.Instead, patched striatal neurons were identified by their morphological and electrophysiological features, as in previous studies (Kawaguchi, 1993(Kawaguchi, , 1997;;Tepper et al., 2018).We identified two distinct populations of striatal neurons in our dataset.Since 95% of the striatal population consists of SPNs, there was a high probability of patching them relative to FSIs (Hjorth et al., 2020).These experiments did not distinguish between D1 and D2 SPNs.
Overall, the neurons in our dataset have morphological and electrophysiological properties that are consistent with the well-documented differences between SPNs and FSIs.The absence of spines on the dendrites and the presence of fast spiking enabled us to differentiate between FSIs and SPNs without the use of genetically encoded fluorescent markers, such as parvalbumin, which is commonly used to label FSIs (Kawaguchi, 1993;Tepper et al., 2018).

Corticostriatal inputs to SPNs and FSIs reconstructed in 3D using Imaris
The strength of synaptic connectivity from M1 and S1 to striatal SPNs and FSIs may derive from differences in the quantity and distribution of synaptic inputs.Based on our previous work (Lee et al., 2019), we hypothesized that SPNs have  The mediolateral (ML), anteroposterior (AP), and dorsoventral (DV) positions of the recorded neuron within the striatum relative to bregma.
significantly more synaptic inputs from M1 compared to S1, but FSIs have a similar number of synaptic inputs from M1 and S1.To test for this, we quantified M1 and S1 inputs onto SPNs and FSIs from confocal 3D reconstructions using Imaris, similar to previous methods (Fogarty et al., 2013;Kuljis et al., 2019).
In brain slices from mice with M1 and S1 already labeled with sm.FPs, we patched and filled striatal neurons with biocytin for anatomical analysis.Raw fluorescence from single-cell-filled SPNs and FSIs was first converted into 3D "surface" objects using the surface tool in Imaris.Then, we constructed our neurons as "filament" objects because this enabled us to segment the dendritic spines, branch points, and terminals and reduce the inclusion of background fluorescence in the reconstruction (Fig. 3A).The dual labeling of M1 and S1 permitted visualization of both cortical inputs simultaneously, but the distance between the pre and postsynaptic membrane is beyond the limitations of confocal light microscopy (Maidorn et al., 2016).To verify that sm.FPs from M1 and S1 represent putative synaptic inputs, we captured fluorescence from M1 and S1 axons along with a marker for the presynaptic scaffolding protein, bassoon, in a subset of patched neurons.Puncta from sm.FPs that labeled M1 and S1, colocalized with bassoon along the dendrites and spines of filled neurons for both SPNs and FSIs (Fig. 3B,C).
To reconstruct M1 and S1 inputs onto striatal neurons, raw fluorescence from corticostriatal puncta was first masked onto the surface of the reconstructed neuron and then converted into 3D "spot" objects.We quantified the number and distribution of spots whose edge was within ≤0.5 µm from the edge of the filament construction and identified them as putative presynaptic inputs (Fig. 4A).To ensure that our counts were not limited to one side of the Z-stack or confounded by limited light penetration, the Z-positions of the first and last appearance of a dendrite, the soma, as well as the Z-positions of identified M1 and S1 inputs were obtained throughout the stack.We confirmed that there were no inputs counted outside the range of the neuron within the Z-stack and that M1 and S1 fluorescence penetrated equally throughout the stack for both cell types (Fig. 5A).
To quantify an error of detection limit, in a subset of neurons (SPN n = 4, FSI n = 2), we imaged M1 and S1 inputs along with the type 1 vesicular glutamate transporter (Vglut1), a protein expressed in presynaptic corticostriatal synaptic terminals (Deng et al., 2013).We created spots representing Vglut1 in the same manner as we did for M1 and S1.We then used the "colocalize" function in Imaris to identify all of the Vglut1 spots that were within a threshold of 0.5 µm from an identified M1 or S1 input and quantified the error of detection as the mean proportion of M1 or S1 inputs that were Vglut1+.Importantly, we found no significant differences in the detection limit between M1 and S1 inputs, suggesting that there was an equal chance of identifying a putative synapse from M1 and S1 (M1 = 19.95± 2.719%, S1 = 24.04 ± 3.386%, n = 6, p = 0.1787, t = 1.563, df = 5, paired two-tailed t test; Fig. 5D).
The mean ratio of M1 to S1 inputs in FSIs was also similar to SPNs but contained a large amount of variability (SPN M1-S1 = 1.802 ± 0.2797, N = 12 vs FSI M1-S1 = 1.867 ± 0.6133, N = 7, p = 0.4824, Mann-Whitney U = 33; Fig. 5E,F).A homogeneity of variance test was used to compare the variance of M1/S1 ratios, and we found no significant difference between the variance of M1/S1 ratios in SPNs and FSIs (F = 1.944, p = 0.1812, Levene's Test).Therefore, we used a one-sample t test to determine if the ratios were statistically different from a value of 1, which would infer equal innervation by M1 and S1.The results of this test were significant for SPNs, supporting the hypothesis of a consistent bias in M1 innervation.In contrast, we observed no difference for FSIs (SPN M1-S1 = 1.802 ± 0.2797, N = 12, p = 0.0001, one-sample Wilcoxon t test; FSI M1-S1 = 1.867 ± 0.6133, N = 7, p = 0.207, t = 1.414, df = 6, one-sample t test; Fig. 5E,F).
A two-way ANOVA revealed that the origin of cortical input was a significant source of variance in the number of inputs counted (F (1, 17) = 4.840, p = 0.0419), as well as individual subject variability (F (17, 17) = 4.864, p = 0.0011, two-way repeated measures ANOVA).The extent of overlap between the M1 and S1 projection fields can vary along the ML axis of the striatum, and we considered the possibility that the ML position of our recorded cell within the projection field could bias the innervation patterns.However, we found no significant correlation between the ratio of M1 to S1 inputs and the ML position of recorded neurons within the DLS (Fig. 5G), leaving the source of individual variability unknown.
Altogether, although M1 and S1 inputs dually innervate SPNs and FSIs, M1 provides a greater number of synaptic inputs than S1 to SPNs, whereas M1 and S1 provide overall equal input to FSIs.These results complement previous measures of the functional synaptic strength of these inputs (Lee et al., 2019).

M1 and S1 inputs are similarly distributed in the proximal, medial, and distal regions of SPNs and FSIs
The spatial organization of synaptic inputs along the dendritic tree, in addition to the total number of inputs, is an important factor that can influence synaptic integration.To test if there are differences in the spatial distribution of M1 and S1 inputs to SPNs and FSIs, inputs were segmented based on their distance along the dendrites relative to the soma into 10 µm bins.In both SPNs and FSIs, the distribution of M1 and S1 inputs closely followed each other and were found in the proximal (0 < 30 µm), medial (30 < 100 µm), and distal regions (>100 µm) of the neuron.Interestingly, the largest concentration of inputs was found near the soma for both SPNs and FSIs (SPN M1 0-10 µm = 16.17 ± 4.785%, SPN S1 0-10 µm = 15.60 ± 3.998%, N = 12; FSI M1 0-10 µm = 16.92 ± 4.476%, FSI S1 0-10 µm = 17.21 ± 6/972%, N = 7; Fig. 4C,D).This was followed by a sharp dip in the number of inputs synapsing 20-30 µm away from the soma of the SPNs, but this was not as prominent in FSIs.The number of M1 and S1 inputs peaked again at distances 70-80 µm from the soma in SPNs but not in FSIs and then decreased to near zero values at distal dendritic locations >100 µm from the soma in both cell types (Fig. 4C,D).
To estimate the density of inputs in specific regions (soma, dendrites), we divided the number of inputs by the membrane surface of these regions.The overall density of inputs to the soma was far fewer than the density of inputs to dendrites (including spines) for both cell types (SPN M1-Dendrite = 5.021 ± 1.044 vs SPN M1-Soma = 0.4029 ± 0.1431, N = 12, Figure 6.S1 and M1 inputs are found mostly on the spines of SPNs and dendrites of FSIs.A, Mean percent of total S1 and M1 inputs distributed between the spines, dendrites, and soma for SPNs and FSIs.Since FSIs are aspiny, inputs were segmented between the dendrites and soma.B, Mean percent of total S1 and M1 inputs distributed between the spines, dendrites, and soma for inputs that were found in the proximal regions (0 < 30 µm) of SPNs and FSIs.C, Mean density of S1 and M1 inputs represented as contacts per 10 µm.For SPNs, inputs to spines and dendrites were grouped together.p = 0.0011, t = 4.393, df = 11, paired two-tailed t test; SPN S1-Dendrite = 2.855 ± 0.7158 vs SPN S1-Soma = 0.2965 ± 0.06218, N = 12, p = 0.0005, Wilcoxon matched-pair signed-rank test; FSI M1-Dendrite = 3.385 ± 1.772 vs FSI M1-Soma = 0.4092 ± 0.1431, N = 7, p = 0.0156, Wilcoxon matched-pair signed-rank test; FSI S1-Dendrite = 2.497 ± 1.237 vs FSI S1-Soma = 0.3261 ± 0.08445, p = 0.0156, Wilcoxon matched-pair signed-rank test).These results indicate that despite a large number of apparent inputs at the soma, the greatest density of inputs occurs on the dendrites and spines.

Discussion
In this study, using dual anterograde fluorescent tracing, we measured the extent of corticostriatal projections from M1 and S1 to individual neurons in the DLS and identified cell-specific differences in the quantity and distribution of M1 and S1 inputs onto SPNs and FSIs.Our findings indicate that SPNs receive significantly more inputs from M1 compared to S1, while FSIs show no overall bias in the number of M1 or S1 inputs (Figs. 4,5).Individual FSIs, however, show variation in the balance of M1 and S1 inputs.In addition, we found that M1 and S1 inputs have similar distributions across proximal, medial, and distal regions of the dendrites of SPNs and FSIs, but in SPNs, S1 inputs are more likely to be found near an M1 input, forming synaptic clusters (Fig. 7E).These results have implications for how sensory and motor cortical inputs modulate striatal circuitry and behavior.
Cell-specific differences in the number of inputs from M1 and S1 suggest weaker S1 connectivity to SPNs in the DLS The convergence of M1 and S1 inputs to the DLS reflects its important role in sensorimotor integration (Makino et al., 2016;Gritton et al., 2019;Lipton et al., 2019;Martiros et al., 2018).We observed extensive overlap of M1 and S1 projections onto individual neurons in the DLS, in agreement with previous experiments (Figs.1B, 3B,C, 4D, 7A; Hoffer and Alloway, 2001;Hunnicutt et al., 2016;Hooks et al., 2018).
Due to our use of biocytin to label patched neurons, we did not use genetic labels for D1 and D2 SPNs, or parvalbumin for FSIs.However, in our study the proportion of inputs from M1 and S1 to SPNs had low variability (Figs.4B, 5E), suggesting that D1 and D2 SPNs have a similar distribution of M1 and S1 inputs, which we previously demonstrated through optogenetic stimulation of M1 and S1 inputs (Lee et al., 2019).The more notable differences in M1 and S1 innervation were found between SPNs and FSIs, and we distinguished patched FSIs from SPNs by their larger soma size, varicose aspiny dendrites, and high-frequency spiking compared to SPNs, which has been reported in previous studies (Fig. 1C-F , 2;Kawaguchi, 1993Kawaguchi, , 1997;;Tepper et al., 2018).
SPNs and FSIs require many excitatory inputs to generate action potentials due to their hyperpolarized resting membrane voltages seen in Figure 2E (Kawaguchi, 1993(Kawaguchi, , 1997;;Tepper et al., 2018).A greater quantity of inputs leads to increased EPSPs because inputs can spatially summate on dendrites (Magee, 2000).M1 and S1 can form synaptic contacts onto the same SPN or FSI (Figs. 3B, 7A; Ramanathan et al., 2002;Charpier et al., 2020;Johansson and Silberberg, 2020), but DLS SPNs have weak responses to sensory stimuli and strong responses to motor activity (Martiros et al., 2018;Lee et al., 2019, Charpier et al., 2020).In contrast, FSIs respond to M1 and S1 stimulation with EPSPs of similar amplitude (Lee et al., 2019;Johansson and Silberberg, 2020).Directly comparing the number of inputs from M1 or S1 revealed no significant differences between SPNs and FSIs; however, we found that SPNs had significantly fewer inputs from S1 compared to M1 and that the number of inputs between M1 and S1 was overall similar in FSIs (Fig. 5C).The reduced innervation by S1 compared to M1 in SPNs, but equal innervation compared to M1 in FSIs suggest that S1 can promote the feed-forward inhibition of SPNs by preferentially activating FSIs over SPNs (Lee et al 2019, Johansson andSilberberg, 2020).
A large source of variability in the counts stemmed from our group of FSIs (Fig. 5C,E; see ANOVA results).However, this is not due to differences in the penetration of light from our sm.FP constructs during imaging (Fig. 5A) or the total number of inputs counted between SPNs and FSIs (Fig. 5B).Moreover, despite a denser overlap of M1 and S1 in the more lateral aspects of the striatum, we found no major correlation between ML striatal position and the ratio of M1/S1 inputs.This might indicate that the ratio of M1 to S1 inputs to SPN and FSIs is consistent across the DLS (Figs. 1G, 5G) instead of the possibility that the ML position or the location of a patched neuron within the M1/S1 projection field biased innervation patterns towards M1 or S1.Importantly, recent evidence suggests that there are subtypes of parvalbumin-positive FSIs in the striatum with distinct physiological properties and connectivity to cortical brain regions (Muñoz-Manchado et al., 2018;Monteiro et al., 2018;Tokarska and Silberberg, 2022).In addition, due to the lack of spines, which can compartmentalize a synaptic input to specific areas of the membrane, the integration properties of FSIs may be less dependent on the number and location of inputs but rather on the intrinsic properties of the FSI, such as the distribution of ion channels and receptors along the dendrites (Cornford et al., 2019).It is possible that the variability of M1/S1 inputs we observed within the FSI group, which was larger than the variability in SPNs, could reflect different subclasses within the FSI cell group.Future experiments could use further genetic cell typing to determine differences in the ratio of M1 and S1 inputs across FSI cell classes.

The location of M1 and S1 inputs onto SPNs and FSIs suggests integration through spatiotemporal mechanisms
The anatomical measures indicate that the quantity of synapses partly accounts for differences between cell types.The distance between the input from the soma is also important because distal inputs undergo electrotonic decay and result in smaller EPSPs recorded at the soma when compared to proximal inputs (Rall, 1967;Magee, 2000;Straub et al., 2016).We found that M1 and S1 inputs had similar distributions across regions of the neuron for both SPNs and FSIs.A large proportion of apparent inputs were observed closer to the soma for SPNs (Fig. 4C,D).However, our analysis revealed that the majority of inputs on SPNs that we counted were located on dendritic spines (Fig. 6A).Interestingly, reports using VGLUT1, a synaptic marker expressed in corticostriatal terminals, have also observed cortical terminals near the cell body and dendrites with light microscopy but demonstrated with electron microscopy that the majority of Vglut1 synapses form at the dendritic spines of SPNs, with fewer inputs to the shafts and cell body (Reiner et al., 2010;Lei et al., 2013, Deng et al., 2015).Although our detection rate for Vglut1+ M1 and S1 synapses was relatively low, other electron microscopy work has identified corticostriatal synapses onto PV cell somata in addition to dendrites (Ramanathan et al., 2002;Nakano et al., 2018;Zheng et al., 2021).Further studies would be necessary to determine the function of somatic synapses in SPNs and FSIs.
In SPNs, spines are prominent on the primary dendrites distal to the first dendritic ramifications (Figs.1C, 3A, 4D; Wilson and Groves, 1980), likely explaining why we observed a drop in counts in more proximal regions that are aspiny in these neurons (Figs.4C,D, 6B).Interestingly, the number of M1 and S1 inputs peaked again at medial dendritic distances (70-80 µm; Fig. 4C), where SPNs have active calcium conductances that contribute to the generation of "up-states" and increased excitability (Plotkin et al., 2011).Other reports have shown stronger synaptic responses in SPNs and the presence of synaptic clusters at similar dendritic locations (Straub et al., 2016;Hwang et al., 2022).In contrast, "up-states" have not been demonstrated in FSIs, suggesting that they integrate inputs passively (Plenz and Kitai, 1998).
SPNs can perform sublinear, linear, and supralinear integration of excitatory inputs depending on the timing and location of the input (Carter et al., 2007, Straub et al., 2016).The clustering of synapses is optimized for long-term plasticity because the coincident activation of multiple synapses in spines that are close together on a dendrite can produce a selfregenerating dendritic spike, which is stronger and longer lasting than an EPSP (Losonczy and Magee, 2006;Carter et al., 2007;Du et al., 2017, Kastellakis andPoirazi, 2019).Although we observed fewer S1 inputs compared to M1 in SPNs (Figs. 4B, 5C), the majority (>50%) of S1 inputs were in proximity (<5 µm) to a neighboring S1 input or M1 input, and this was not the case for FSIs (Fig. 7A-E).These results support the theory that SPNs can act as coincidence detectors for M1 and S1 and integrate signals from these cortical regions through temporal and spatial summation by clustering sensorimotor inputs.Critically, it has been observed that stimulation of M1 inputs induces an initial membrane depolarization in SPNs that is quickly enhanced by activity generated from S1 (Charpier et al., 2020), and M1 inputs from mice trained on motor tasks form active clusters on the spines of SPNs in the DLS (Hwang et al., 2022).

Implications of findings on sensorimotor integration by SPNs and FSIs in the DLS
In vivo, during sensorimotor learning, SPNs form ensembles that are highly active during the initiation and termination of task-related movements associated with reward (Martiros et al., 2018;Gritton et al., 2019).PV-FSIs have an important contribution to SPN ensemble formation in the DLS (Martiros et al., 2018;Gritton et al., 2019).The cell-specific distribution of M1 and S1 to the DLS that is seen in our results likely contributes to organizing which cells are active during this process.
Synaptic clustering greatly increases the likelihood that activity from M1 and S1 is coincidentally detected on short sections of the dendritic membrane, leading to large depolarizations and changes in plasticity (Carter et al., 2007;Kastellakis and Poirazi, 2019).This increases the chance that these inputs become associated with other cortical and thalamic regions when there is convergent activity (Carter et al., 2007;Huerta-Ocampo et al., 2014, Makino et al., 2016).Therefore, it is likely that while S1 alone does not evoke strong responses in SPNs, SPNs that receive convergent activity from task-related M1 and S1 neurons will overcome local inhibition by FSIs and increase the probability of firing an AP.As a result, they become part of task-related circuits in the DLS that underlie habitual responses to sensory stimuli (Martiros et al., 2018;Lipton et al., 2019).
In conclusion, the organization of M1 and S1 inputs to the DLS suggests that activity from S1 will preferentially excite the striatal FSI population, leading to feed-forward inhibition of SPNs during sensorimotor integration.Our findings have significant implications for how corticostriatal circuits encode learned movements.Future investigations should examine if there are cell-specific requirements for plasticity in corticostriatal inputs to SPNs and FSIs.

Figure 1 .
Figure 1.Viral circuit mapping of M1 and S1 corticostriatal projections to dorsal striatal SPNs and FSIs.A, Viral circuit mapping strategy highlighting spaghetti monster injections into M1 and S1 from a top-down view (left) and the corticostriatal projection pattern onto biocytin-filled striatal cells from a sagittal view (right).B, Top-left: Low magnification view of a coronal striatal section containing the M1 sm.FP injection site and corticostriatal projections to the DLS.CC, corpus callosum; DLS, dorsolateral striatum.Top-right: Low magnification view of a coronal striatal section with S1 corticostriatal projections labeled with sm.FPs.Bottom-left: Low magnification view of merged fluorescence from S1 and M1 corticostriatal projections innervating a striatal SPN filled with biocytin (white box).Bottom-right: High magnification inset bottom-left image with corticostriatal innervation by M1 and S1 onto a biocytin-filled SPN.C, Left: Pseudo-colored representative image of a biocytin-filled SPN with spiny dendrites (inset).Right: Pseudo-colored representative image of a biocytin-filled FSI with aspiny dendrites (inset).D, Left: Average number of primary dendrites (note that FSIs have more primary dendrites extending from the soma compared to SPNs).Right: Average cross-sectional diameter of the soma.E, Comparison of mean IFF and mean HHW at steady state permits differentiation of FSIs from SPNs due to their fast-firing rates.F, Left: Average max firing frequency.Right: Average ISI.G, Schematic representation of the recording location of the SPNs and FSIs in our dataset (note that all cells were recorded in the anterior dorsal striatum where M1 and S1 innervation was dense).

Figure 2 .
Figure 2. Morphological and electrophysiological properties of SPNs and FSIs.A, Mean cross-sectional diameter of the dendritic field.B, Mean depth of the dendritic field along the z-axis.C, Mean HWH amplitude.D, Mean IFF.E, Mean resting membrane potential voltage.F, Mean input resistance.

Figure 3 .
Figure 3. 3D reconstructions of striatal neurons and confirmation of S1 and M1 synaptic puncta.A, 3D reconstruction of biocytin-filled SPNs and FSIs into a "surface" and "filament" object in Imaris.B, Representative raw fluorescence from biocytin, the presynaptic protein bassoon, M1, and S1.C, Merged image from B demonstrating S1 and M1 presynaptic inputs dually innervating an SPN (top) and FSI (bottom).

Figure 4 .
Figure 4. Differential innervation by S1 and M1 to SPNs and FSIs.A, Magnified view of raw fluorescence (left) from S1 and M1 corticostriatal projections onto a biocytin-filled SPN (top) and FSI (bottom) with their associated 3D reconstructions (right).B, Average percent innervation by M1 and S1.C, Mean distribution of S1 and M1 inputs across the neuron when measuring the distance from the soma along the length of a dendrite.D, Representative image of fluorescence from M1 and S1 masked onto a biocytin-filled SPN (left) and FSI (right) that was later used for input analysis.

Figure 5 .
Figure 5. Distribution properties of S1 and M1 inputs to SPNs and FSIs.A, Z-positions of the first and last appearance of a dendrite, and the soma of reconstructed SPNs (left) and FSIs (right) compared with the Z-positions of their associated S1 and M1 inputs.B, Mean total number of combined S1 and M1 inputs counted within 0.5 µm of the filament-constructed SPN or FSI.C, Mean number of inputs counted from S1 and M1 within 0.5 µm of the filament edge.D, Mean percentage of identified M1 and S1 inputs that colocalized with Vglut1.E, Mean ratio of M1 inputs to S1 inputs counted within 0.5 µm of the filament edge.The ratios were statistically compared to a theoretical value of 1 indicated by a red dashed line.F, Cumulative distribution function (CDF) of the M1/S1 ratios.The red dashed line indicates a value of 1, which infers equal innervation by M1 and S1.Gray line = SPNs, black line = FSIs.G, Comparison of M1/S1 ratio with the ML position of the neuron it was recorded from; gray dots = SPNs (N = 12), black dots = FSIs (n = 6).H, Mean distance of an S1 and M1 input from the soma when measured along the length of the dendrite.

Figure 7 .
Figure 7. S1 and M1 input cluster on SPNs but not in FSIs.A, Representative image of fluorescence from M1 (red) and S1 (green) masked onto the dendrites of an SPN (left) and an FSI (right).B, Mean distance between the nearest input from the same cortical region.C, Mean distribution of the shortest distance to the nearest input from the same cortical region.D, Mean distance between inputs from different cortical regions.E, Mean percent of S1 and M1Inputs that are colocalized with an increasing distance threshold for SPNs (right) and FSIs (left).The mean percent was tested against a theoretical value of 50% indicated by the red dashed line using a one-sample t test.F, Mean distance from the soma when measured along the length of the dendrite for S1 and M1 inputs that colocalized within 5 µm of each other.G, Distribution of colocalized inputs from F.