Correlated Somatosensory Input in Parvalbumin/Pyramidal Cells in Mouse Motor Cortex

Abstract In mammalian cortex, feedforward excitatory connections recruit feedforward inhibition. This is often carried by parvalbumin (PV+) interneurons, which may densely connect to local pyramidal (Pyr) neurons. Whether this inhibition affects all local excitatory cells indiscriminately or is targeted to specific subnetworks is unknown. Here, we test how feedforward inhibition is recruited by using two-channel circuit mapping to excite cortical and thalamic inputs to PV+ interneurons and Pyr neurons to mouse primary vibrissal motor cortex (M1). Single Pyr and PV+ neurons receive input from both cortex and thalamus. Connected pairs of PV+ interneurons and excitatory Pyr neurons receive correlated cortical and thalamic inputs. While PV+ interneurons are more likely to form local connections to Pyr neurons, Pyr neurons are much more likely to form reciprocal connections with PV+ interneurons that inhibit them. This suggests that Pyr and PV ensembles may be organized based on their local and long-range connections, an organization that supports the idea of local subnetworks for signal transduction and processing. Excitatory inputs to M1 can thus target inhibitory networks in a specific pattern which permits recruitment of feedforward inhibition to specific subnetworks within the cortical column.

How inhibitory interneurons integrate in subnetworks may differ from Pyr neurons. Interneurons may connect nonspecifically to Pyr cells in nearby local, intralaminar circuits where they pool inputs from excitatory cells with different response properties (Fino and Yuste, 2011;Packer and Yuste, 2011; for review, see Mountcastle, 2003;Sohya et al., 2007;Fino et al., 2013).Thus, inhibitory interneurons have broader tuning curves for stimuli orientation and spatial frequency in visual cortex (Sohya et al., 2007;Niell and Stryker, 2008;Bock et al., 2011;, although inhibitory interneurons show selectivity in some species Wiesel, 1962, 1963;Ohki et al., 2005; but see Hirsch et al., 2003;Cardin et al., 2007;Ma et al., 2010;Runyan et al., 2010;Moore and Wehr, 2013;Ringach et al., 2016;Wilson et al., 2017). How to reconcile specific, connected subnetworks of excitatory cells with the nonspecific targeting of Pyr cells by interneurons? One possibility is that connectivity is dense, with high probability of connection, but synapse strength is weighted higher within subnetworks but weaker to outside networks (Znamenskiy et al., 2018). Another nonmutually exclusive possibility is that long-range inputs and outputs are organized into subnetworks during brain development through Hebbian plasticity (Tezuka et al., 2022; for review, see Katz and Shatz, 1996).
Here, we examined the functional organization of longrange thalamic (PO) and cortical (vS1) inputs to PV1 and Pyr cells in M1. Here, we refer to these sensory inputs as recruiting feedforward inhibition in M1, although thalamic input to cortex may also be referred to as feedback in other contexts. Using two-channel optogenetic stimulation with paired whole-cell patch-clamp recording, we show that thalamic and somatosensory inputs were more correlated in connected pairs compared with nonconnected pairs. Thus, recruitment of feedforward inhibition by thalamic and somatosensory inputs to motor cortex in mice is subnetwork specific and may depend on functional connections between excitatory and inhibitory cells. Specific differences in connectivity in circuits promoting different movements might assist in motor control.

Animals
Animal protocols were approved by the Institutional Animal Care and Use Committee at University of Pittsburgh.

Stereotactic injections
Animals were anesthetized using isoflurane and placed in a custom stereotactic apparatus. Mice at P14-P40 were injected with AAV expressing excitatory opsins. Injections were made with glass pipettes (Drummond) using a custom-made injector (Narashige). The injection apparatus was a positive displacement pump allowing slow injection of nanoliter volumes. Injection coordinates (Table 1; Extended Data Fig. 1-2) on the anterior/ posterior (A-P) axis are reported relative to bregma (positive values anterior to bregma); medial/lateral (M-L) axis coordinates are reported relative to the midline; and dorsal/ventral (D-V) axis coordinates are reported as depth from pia. Injections were made at two depths in cortex. For posterior thalamic injections, we used two adjacent sites, covering the elongated shape (in the A-P axis) of the PO nucleus. The more anterior set of those thalamic anterior injections was done in different mice (n = 15) for 11 connected and 12 nonconnected PV1 and Pyr pairs. Injections in both sites resulted in similar axon patterns in M1 (Hooks et al., 2013;for review, see Castro-Alamancos and Connors, 1997) and were pooled. As in our previous studies, we examined the injection site in thalamus during sectioning to confirm injection targeting to PO. We also confirmed the axonal projection pattern in cortex arborized in layer (L)1 and the L2/3-5A border, as is typical of PO injections (Petreanu et al., 2009;Hooks et al., 2013).

Electrophysiology and photostimulation
Whole cell recordings were performed at 22°C in oxygenated ACSF with borosilicate pipettes (3-6 MV; Warner Instruments) containing potassium gluconate-based internal solution (in mM: 128 potassium gluconate, 4 MgCl 2 , 10 HEPES, 1 EGTA, 4 Na 2 ATP, 0.4 Na 2 GTP, 10 sodium phosphocreatine, three sodium L-ascorbate; pH 7.27; 287 mOsm). Data were acquired at 10 kHz using an Axopatch 700B (Molecular Devices) and Ephus software (www. ephus.org; Suter et al., 2010) on a custom-built laser scanning photostimulation microscope with inversion recovery differential interference (Shepherd et al., 2003) using a Retiga 2000R camera (QICAM; QImaging). Slices were visualized with 4Â, 0.16 numerical aperture, UPlanSApo; Olympus power objective. Individual neurons were visualized with a 60Â, 1 numerical aperture Olympus Fluor LUMPlanFL water-immersion objective. Series resistance errors were minimized with bridge balance in currentclamp mode. Current-clamp recording was performed to confirm stable conditions. To measure excitability, 500-ms current steps were applied starting from À150 to 700 pA in 50-pA steps. Connections between pairs of cells were tested in current-clamp with a train of five 3-nA pulses of 0.5-ms duration repeated 40 times (20 for 5 Hz, 20 for 40 Hz) while the other cell was held in voltage-clamp mode. Each sweep length was 2 s with a 5-s delay between the sweeps. While testing connections from PV1 interneurons to Pyr cells, the Pyr cell was held at À50 mV (0 mV in some cases, N = 5 Pyr cells) to detect IPSCs, while the reciprocal connection was tested with PV1 interneurons held at À70 mV to detect EPSCs.
Photostimulation was done as previously described (Hooks et al., 2015) using 590-and 470-nm LEDs (OptoLED, Cairn). Photostimuli in single-channel experiments were ;2 mW/mm 2 . Photon flux was matched for 590-nm and 470nm stimuli in the same experiment. Light ,585 nm from the 590-nm LED was blocked using a bandpass filter (D607/45, Chroma). Voltage-clamp experiments with LED photostimulation were performed at À70 mV for Channelrhodopsin-induced EPSCs and at 0 mV for Channelrhodopsin-induced IPSCs recruited through feedforward inhibition. Under these Anterior/posterior (A-P) axis coordinates are reported relative to bregma (positive values anterior to bregma). Medial/lateral (M-L) axis coordinates are reported relative to the midline. Dorsal/ventral (D-V) axis coordinates are reported as depth from pia. Injections were made at two depths in cortex. Distances in millimeters (Extended Data Fig. 1-2), (Paxinos and Franklin, 2004;Hooks et al., 2013).
conditions, 590-nm LED pulses of 50, 100, 250, and 500 ms were followed by 50-ms pulses of 470-nm LED. Depolarized ReaChR-expressing and ChR2-expressing axons, respectively, triggered the local release of glutamate. Sweeps were repeated four times with a 20-s gap. The four different delay protocols (50-500 ms) were used as a control for the effects of activation with two different wavelengths (590 and 470 nm) on the glutamate release caused by activation of red-shifted ReaChR channels first and the blue shifted ChR2 second. If the first stimulus (590 nm) does not affect the features of the second response, then we are confident that our paradigm is independently stimulating the two pathways. In some experiments, biocytin was added to the intracellular solution (3 mg/ml biocytin or neurobiotin). The experimental recording sequence started with series resistance test of five pulses in voltage-clamp, followed by current-clamp to test action potential (AP) firing and confirm the passive and active electrophysiological properties. This was followed by the connectivity test between the pairs and finally, optogenetic stimulation at different holding potentials in voltage-clamp. Cells with stable access were used for quantification of passive electrophysiological properties.

Histologic preparations and image analysis
Some of the slices were processed for biocytin recovery. Samples were postfixed overnight in 4% PFA. Sections were processed as free floating and stained with Hoechst reagent. Blocking was done in TBS containing 10% normal goat serum or normal donkey serum (MilliporeSigma) and 0.5% Triton X-100 (MilliporeSigma) for 2 h at room temperature. Tissue was washed three times in TBS with 2% normal goat serum and 0.4% Triton X-100 (washing solution), followed by incubation with streptavidin conjugated Alexa-647 (1:200) overnight to 24 h at 4°C in the washing solution. After overnight incubation the tissue was stained with nuclear staining Hoecsht (10 mg/ml in water, 1:3000 dilution; Invitrogen) for 10min and washed four times for 10min. Tissue was mounted on Fisherbrand ColorFrost Plus microscope slides submerged in Fluoromount G (ThermoFisher).
Images were acquired with a Nikon A1R confocal microscope with 20Â or 60Â oil-immersion objectives. All images were processed in the ImageJ-Fiji package. Image processing for publication was done in Fiji and Corel Draw Graphics Suite X8 (Corel) or Adobe Illustrator.

Experimental design and statistical analysis
Data analysis was performed with custom routines written in MATLAB. Electrophysiology data were low pass filtered (1 kHz) with an 8-pole low-pass Bessel filter. EPSCs were detected with a threshold of .2Â SD from baseline. All data measurements were kept in Microsoft Excel (Microsoft) and in Origin (OriginLab). Statistical analysis of the data was done in SPSS v.24-v.28 (IBM). For large samples, one-way ANOVA with Tukey's post hoc correction was used. When the samples had nonhomogeneous variance (significant Levene's test for equality of variance), Welch's test with Games-Howell post hoc correction was used. For small samples from different observations, independent-samples two-tailed Student's t test was used, and depending on Levene's test significance, the t statistics for equal or unequal variance are reported. For measurements coming from the same neurons before and after treatment, paired-samples two-tailed Student's t test was used. For non-normally distributed data, the nonparametric Wilcoxon signed-rank or Kolmogorov-Smirnov tests were used. All data are shown as arithmetic average 6 SEM or 695% confidence intervals, unless otherwise specified.

Code accessibility
Data analysis was performed with custom routines written in MATLAB. The acquired data as well as the data acquisition and analysis software (M-files in MATLAB format) are available on request.

Results
Subnetworks by definition contain cells connected to each other more frequently than to cells in the outside networks. We hypothesized that long-range synaptic inputs may be different to cells in one subnetwork compared with a different subnetwork. To understand how long-range projections to primary vibrissal motor cortex (vM1) excite specific networks of interneurons, we recorded from connected and nonconnected pairs of parvalbumin positive inhibitory interneurons (PV1) and Pyr excitatory neurons (Pyr) to explore differences in their circuit connectivity. We used the two-channel Channelrhodopsin-assisted circuit mapping (2CRACM) approach developed by our lab (Petreanu et al., 2007(Petreanu et al., , 2009Hooks et al., 2015). We injected viral vectors containing Channelrhodopsin-2 (ChR2-mCherry) into posterior thalamus (PO) and the red-shifted Channelrhodopsin variant ReaChR (ReaChR-mCitrine) into vS1. We activated these opsins by sequential 590-nm and 470-nm stimulation ( Fig.  1A,B,J,K). The intracranial injections were done with a custom-made positive displacement system in PV1-Cre 1/1 ; lsl-tdTomato (ai14) 1/1 mice. To allow for opsin expression, recording started two weeks after injections. Pairs of adjacent (,120 mm) PV1 and Pyr neurons were recorded in whole-cell current-clamp and voltage-clamp configuration. Passive membrane properties and series resistance were measured at À70 mV. Current injections of 500 ms in 50-pA steps characterized active membrane properties including AP firing (  Table 1-1). Connectivity was tested in each direction (PV1 $ Pyr), holding one neuron in currentclamp and applying 3 nA, 0.5-ms current steps while the other neuron was held in voltage-clamp (Fig. 1F,I). Some of the slices were processed for biocytin recovery at the end of the experiments, which allowed confirmation of cell type and laminar position (Fig. 1D,G). Input was quantified in voltage-clamp using the 2CRACM approach, ReaChR-mCitrine was stimulated with 590-nm LED light (50-to 500ms pulses), followed by stimulation of ChR2-mCherry with 470-nm blue LED light (50-ms pulses immediately following), with additional 500-ms 590-nm only LED stimulation to have ReaChR trace only for subtraction (Fig. 1J,K).
The Channelrhodopsin-induced EPSCs kinetic properties are shown in Figure 2. These include the EPSC onset delay, the rise time, normalized amplitude, and decay . Vibrissal M1 (vM1) receiving long-range projection inputs from vS1 (green lines, ReaChR-mcitrine expressing), and from posterior thalamus (red lines, ChR2-mcherry expressing). Paired whole-cell patch-clamp recording targeted to a pyramidal excitatory neuron (Pyr, empty triangular shaped) and parvalbumin positive inhibitory interneuron (PV1, oval shaped, red) receiving inputs from vS1 (stimulated with orange LED, 590 nm), and from posterior thalamus (stimulated with blue LED, 470 nm). B, Illustration of stimulation paradigm in brain slices. ReaChR expressing axons (vS1, green) were stimulated first with 590 nm, orange LED (50-500 ms) immediately followed by stimulation of ChR2 containing axons (PO) with 470-nm blue LED 50 ms with equal light intensity (;2 mW/mm 2 ). Example traces are shown in J-K containing axons (PO) with 470-nm blue LED 50 ms with equal light intensity (;2 mW/mm 2 ). Example traces are shown in J-K. C, Illustration of the scientific inquiry question and color coding. D, Example reconstruction of a recorded biocytin filled PV1 cell. Scale bar above. E, Example current-clamp traces showing responses of a PV1 inhibitory fast-spiking cell recorded in those experiments with current steps in between and the scale bar to the right. For electrophysiological properties, please see Extended Data Figure 1-1. Extended Data Table 1-1. F, Example traces of a connectivity test between a PV1 cell in current-clamp mode with 3 nA 0.5-ms current step to elicit single AP and the voltage-clamp IPSC response of the Pyr cell held at 0 mV. G, Example reconstructed biocytin filled Pyr cell. Scale bar above. H, Example current-clamp recording of Pyr cell with current steps as labeled. Scale bar to the right. I, Example of connectivity test, with the same protocol as in E, Pyr in current-clamp, EPSC in PV1 cell held at À70 mV. J, K, Examples of two connected PV1 and Pyr cells in voltage-clamp showing responses to LED stimulation. First column 590-nm and 470-nm LED stimulation (colored bars indicate time of LED on/off). Middle traces are 500-ms 590-nm alone. Third column is a subtraction of middle traces from the first column to reveal 470-nm response. time course. EPSC onset times may be slightly slower for vS1 inputs compared with PO inputs, because of the slower kinetics of ReaChR (Lin et al., 2013). However, these do not vary significantly across delay times, suggesting that activating vS1 inputs first does not affect responses from PO afferents.
Somatosensory cortical excitation is stronger than thalamic input to PV1 neurons in layer 2/3 of vM1 To compare the difference in recruitment of feedforward inhibition mediated by PV1 cells and excitation of Pyr cells, we recorded opsin-mediated EPSCs in pairs of PV1 and Pyr cells in the whole-cell voltage-clamp configuration. Cortical laminae were defined based on visible boundaries formed by differential cell densities in the brightfield image of the slice (Weiler et al., 2008; and reported as the normalized distance of the cells between pia and white matter. L5A is the pale band in the brightfield image above the more heavily myelinated L5B . L2/3 cells were within 8-38% and L5A cells were within 20-58% of the slice thickness, depending on the curvature and anterior-posterior position of the slice (Fig. 3C). Our prior data, using subcellular Channelrhodopsin-assisted circuit mapping (sCRACM), had shown that both vS1 input and PO input similarly excited vM1 L5A Pyr neurons most strongly and L2/3 Pyr neurons with ;70-90% of this strength (Mao et al., 2011;Hooks et al., 2013). But for PV1 neurons, this pattern shifted and vS1 excited L2/3 PV1 neurons more strongly than L5A, while PO excited L5A PV1 neurons more strongly than L2/3 (Okoro et al., 2022). Based on this difference in connection strength, EPSCs amplitudes in PV1 cells divided by the amplitudes of EPSCs in Pyr cells should result in larger ratios from vS1 stimulation compared with PO stimulation in L2/3 in our paired recordings. Consistent results confirm that the wide-field LED stimulation in the 2CRACM approach produces similar results to those predicted from earlier circuit mapping approaches (which differs by the use of tetrodotoxin Figure 2. 2CRACM EPSCs kinetics. A, Example response traces of PV1 (red) and Pyr (black) to 590 nm (vS1, orange) LED stimulation, following by 470 nm (PO, blue) LED stimulation. Description of kinetics that were measured and compared, as indicated by arrows in the panel. Onset was measured from start of the stimulation to 10% of the EPSC peak, the rise time was measured from 10% to 90% of the EPSC peak; and the decay time was measured from 90% to 50% of the EPSC peak. Peaks are shown by pink arrowheads. Because vS1 responses onset before PO responses, onset and rise kinetics are averaged across all delay protocols. For the uniformity purposes only, 100-ms and 250-ms delay protocols kinetics are shown to constrain to the data analyzed and presented in other figures. B, vS1 EPSC onset averaged across all delay protocols (5-500 ms; PV1 n = 123-125; Pyr n = 114-115). C, PO EPSC onset (PV1 n = 84-108; Pyr n = 85-109). D, vS1 EPSC averaged rise time (PV1 n = 123-125; Pyr n = 114-115). E, PO EPSC rise time (PV1 n = 84-108; Pyr n = 85-109). F, vS1 EPSC normalized to its own peak at 250-ms delay protocol response (PV1 n = 123-125; Pyr n = 114-115). G, PO EPSC normalized to its own peak at 250-ms delay protocol response (PV1 n = 89-110; Pyr n = 118). H, vS1 EPSC decay time (PV1 n = 59-110; Pyr n = 47-102). I, PO EPSC decay time (PV1 n = 76-98; Pyr n = 81-100). Means are shown with 95% confidence intervals. Figure 3. vS1 input is larger than PO input to vM1 layer 2/3 PV1 interneurons. A, Illustration of the vM1 slice preparation receiving long-range projections inputs from vS1 (green lines, ReaChR-mcitrine) and from PO (red lines, ChR2-mcherry) during paired recording. B, Example voltage-clamp traces in cells held at À70 mV from a pair of PV1 inhibitory interneuron and Pyr excitatory neuron showing responses to 590-nm LED stimulation of vS1 axons in vM1 immediately followed by 470-nm LED stimulation of PO axons. Middle panel is the response to 590-nm LED stimulation alone and the last panel is a subtraction of the middle traces from the first traces to isolate 470 nm-induced EPSCs. C, Example of an off-coronal 300-mm brain slice of vM1 with two cells in patch-clamp. Left panel shows the location of the cells relative to pia and white matter. Approximate layer boundaries indicated. Middle panel shows vS1 axonal fluorescence in green (arrowhead). Right panel is a confocal image showing PO axonal fluorescence in red (arrowheads) in L1 and the lower L2/3 and L5A together with PV1 interneurons. A biocytin filled example pair (green) of cells is shown. Scale bars are 500 mm. D, Cumulative distribution of EPSC amplitudes comparing vS1 and PO inputs to PV1 (left) and Pyr (right) neurons in layers 2/3 (top) and 5A (bottom). EPSCs in L2/3 PV1 neurons were significantly larger from vS1 than PO (Wilcoxon signed-rank test = 2131, p = 5E-6, effect size h 2 = Z 2 /(nÀ1), h 2 = 0.296). Differences between vS1 and PO EPSCs in L2/3 Pyr neurons were significant (Wilcoxon signed-rank test = 620, p = 1.63E-4, effect size, h 2 = 0.203; however, the effect size Hedges' g and confidence interval show no statistical significance, Extended Data  EPSC amplitudes showed that vS1 inputs to PV1 cells are stronger compared with PO inputs in vM1 L2/3 (Wilcoxon signed-rank p = 5E-6, effect size h 2 = 0.296, [95.0%CI 0.351, 0.85]; Fig. 3D), while comparisons in L5A showed similar synaptic strength (Fig. 3D,G).

Figure 4. Characterization of connectivity between PV1 and Pyr neurons. A, Example of biocytin filled 3 cells (green) and currentclamp traces show characteristic PV1 (red) and
Pyr neurons subthreshold responses and suprathreshold AP firing in response to depolarizing current steps (PV1 red, Pyr gray) with ChR2 positive axons (cyan). One PV1 filled with biocytin is shown in yellow, because of overlap of tdtomato labeling. B, Example of connection test traces for reciprocally connected (top), connected in one-way from PV1 to Pyr (second row), connected in one-way from Pyr to PV1 (3rd row), and nonconnected (fourth row). C, Percentage of connected pairs (Fisher's exact test 20.81, p = 1.6E-5; Extended Data Table 4-1). D, Normalized distance to pia, the distance from pia to white matter is 100%, means (circles) and medians (squares) are to the left with whiskers showing 695% confidence intervals. Because of nonhomogenous variance, Levene's test p , 0.001, Welch test used for comparison of the means did not show any significant difference, F (2,104.52) = 1.05, p = 0.36 with Games-Howell post hoc ns., between the reciprocally connected pairs (pink), unidirectionally connected pairs (powder blue) and nonconnected pairs (infant blue). E, Euclidian distance between all the cell pairs were recorded within 120 mm of each other and show no difference, although the tendency of connected pairs having a smaller Euclidian intersomatic distance is showing (average Euclidian distance between reciprocally connected pairs = 23.65 mm, median = 21.29 mm, n = 18; between unidirectionally connected pairs avg = 26.96 mm, median = 24.22 mm, n = 58; between nonconnected pairs avg = 31.11 mm, median = 27.6 mm, n = 119, one-way ANOVA F = 2.52, p = 0.08 ns.; Extended Data Table 4-1). Scale bar for A is 50 mm.
Research Article: New Research Figure 5. Normalized vS1 and PO inputs to connected and not pairs of PV1 and Pyr neurons show different trends. All inputs are from the 100-ms delay protocol normalized to the maximum slice peak at 250-ms delay protocol. A, C, E, vS1 (orange) inputs for connected PV1 (red) and Pyr (black) pairs. A, Original figure. C is A bubble plotted to PO (blue) inputs to PV1 neurons in the same pairs. E is A bubble plotted to PO inputs to Pyr in the same pairs. B, D, F, vS1 inputs for nonconnected PV1 (magenta) and Pyr

Connectivity between PV1 and Pyr neuron pairs
Paired recordings allowed identification of Pyr and PV1 neuron pairs that were connected or unconnected, providing comparison between neurons in the same local network. Connectivity was tested bidirectionally between 197 PV1 and Pyr pairs (Fig. 4A-C). Of these, ;39% were connected (N = 77/197), either unidirectionally or bidirectionally.

Somatosensory cortical and thalamic excitation in connected versus nonconnected PV1 and Pyr neurons
To test whether excitation from somatosensory cortex and thalamus is organized differently in connected versus nonconnected pairs of PV1 and Pyr cells, we compared whole-cell voltage-clamp responses to vS1 and PO excitation in simultaneously recorded pairs of neurons. The similar distribution of ChR2-induced EPSC amplitude with ReaChR-induced EPSC amplitude (Fig. 3D) following the same amount of viral volume injected into PO and vS1 suggests both pathways excite vM1 with roughly similar strength. To compare the responses across different slices and animals and account for the difference in ChR2 and ReaChR expression levels between individual animals, opsin-mediated EPSCs amplitudes were normalized to the maximum response in the slice during the 250ms delay sweep (Fig. 5). Only the slices with at least two cells were included. To visualize both somatosensory cortical and thalamic inputs in the same pairs in the same graphs, we used scatter bubble plots, that allows to show the third dimension by controlling the size of the data points. Thus, plots comparing vS1 input to PV1 and Pyr neurons could also show the strength of PO input with the size of the marker. Visual inspection shows a difference in normalized EPSCs amplitudes between connected and nonconnected pairs of PV1 and Pyr cells, which is emphasized when the thalamic inputs to PV1 or Pyr cells are chosen as a third dimension/variable (controlling the size of each data point in the pair receiving input from somatosensory cortex; Fig. 5). Specifically, PV1 neurons receive stronger vS1 inputs than nearby Pyr neurons (Fig. 5A,B), resulting in most points falling below the unity line (gray). Furthermore, the scatter of these points is reduced in connected versus nonconnected pairs (Fig. 5A-F), resulting in fewer points scattered above the line, suggesting less variance. This trend also seems to hold for PO inputs to PV1 and Pyr neurons (Fig. 5G-L), although less pronounced. The normalized input strength is comparable for PV1 and Pyr in both groups, yet the distribution is shifted (skewness for the VS1 normalized inputs in nonconnected PV1 = 0.454, in nonconnected Pyr = 0.329; in connected PV1 skewness = À0.353, in connected Pyr = 1.284; kurtosis in nonconnected PV1 = 0.089, in nonconnected Pyr = À1.097, in connected PV1 kurtosis = À0.916, in connected Pyr kurtosis = 1.798; skewness for the PO normalized inputs in nonconnected PV1 = À0.013, in nonconnected Pyr = 0.890, in connected PV1 skewness = À0.370, in connected Pyr = 0.474; kurtosis for the PO normalized inputs in nonconnected PV1 = continued (gray) pairs. B, Original figure. D is B bubble plotted to PO inputs to PV1 neurons in the same pairs. F is B bubble plotted to PO inputs to Pyr in the same pairs. G, I, K, PO inputs for connected PV1 and Pyr pairs. G, Original figure. I is G bubble plotted to vS1 inputs to PV1 neurons in the same pairs. K is G bubble plotted to vS1 inputs to Pyr neurons in the same pairs. H, J, L, PO inputs for nonconnected PV1 and Pyr pairs. H, Original figure. J is H bubble plotted to vS1 inputs to PV1 neurons in the same pairs. L is H bubble plotted to vS1 inputs to Pyr neurons in the same pairs. Bubble plots scale is from 0.1 to 1.6 with D of 0.5. M, Schematics of scientific inquiry question, methods to study it, bubble plots color coding and size for vS1 inputs. N, Schematics of scientific inquiry question, methods to study it, bubble plots color coding and size for PO inputs. O is A and B shown as a box plots. P is G and H shown as a box plots. Means are shown by circles, medians by squares and whiskers represent 95% confidence intervals. À1.182, in nonconnected Pyr = 0.501, in connected PV1 kurtosis = À1.377, in connected Pyr = À1.012; Fig. 5O,P).
We then sought to test whether the connected cell pairs had correlated inputs. We speculated that interconnected subnetworks of neurons performing similar computations might get correlated input, such as both PV1 and Pyr neurons receiving strong or weak input. The alternative is that long-range input strength would be random with respect to whether pairs were connected. We plotted the EPSC amplitudes for connected and nonconnected pairs (Fig. 6B,C,F,G). We then fit these with a linear regression, finding that EPSCs from both vS1 and PO inputs before normalization showed higher correlation in connected (vS1 inputs in connected pairs Spearman r = 0.46, p = 2.39E-3, confidence interval of [95.  Fig. 6; Extended Data Figs. 6-1, 6-2, 6-3, the data were resampled 1000 times bootstrapped and produced correlation coefficients that were compared between those groups, p , E-10) suggesting that co-targeting of long-range projections is dependent on whether they contact connected or nonconnected pairs.

Excitation-to-inhibition ratio of vS1 and PO inputs
We wanted to test how the recruitment of PV1 cells by long-range projections is correlated with the ReaChR and ChR2-induced IPSCs in both PV1 and Pyr cells. As before, we thought that interconnected subnetworks of Figure 6. Increased correlation of the long-range inputs to connected pairs. A, Example of voltage-clamp traces from PV1 (red) and Pyr pair with 590-nm stimulation of vS1 axons and 470-nm stimulation of PO (blue) axons middle traces are 590nm stimulation alone and the right traces show the result of subtraction of middle traces from the first traces. Connected pairs are B-E. B, Scatter plot of vS1 in connected pairs showing a higher correlation compared with nonconnected pairs in F, also see Extended Data Figures 6-2 and 6-3 for layer-specific information. The data in B and F were resampled using 1000 samples bootstrap and produced Spearman's r correlation coefficients, which were squared and compared with nonparametric independent samples Mann-Whitney test, U = 208,790.0, p , 1E-10 (Extended Data Fig. 6-1). The estimation of confidence interval for the Spearman's r correlation coefficient was also done with 10,000 Bootstrap resampling, [95.0%CI 0.324, 0.760;Ho et al. (2019)]. C, Scatter plot of PO EPSCs in connected pairs showing a higher correlation compared with nonconnected pairs in G. The data in C and G were resampled using 1000 samples bootstrap and produced Spearman's r correlation coefficients, which were squared and compared with nonparametric independent samples Mann-Whitney test, U = 279,207.0, p , 1E-10 (Extended Data Fig. 6-1). The estimation of confidence interval for the Spearman's r correlation coefficient was also  . C, vS1 and PO EPSCs (e's) at À70 mV holding potential multiplied by À1 for the convenience of presentation, and IPSCs (I's) at 0 mV holding potential; EPSCs in L2/3 PV1 neurons were significantly larger from vS1 than PO (Wilcoxon signed-rank test = 2131, p = 5E-6, effect size h 2 = Z 2 /(nÀ1), h 2 = 0.296; Fig. 3D). vS1 Is were significantly larger in L2/3 than L5A Pyr [independent samples Mann-Whitney (M-W) U = 39, p = 3.08E-2, h 2 = 0.180]. D, Averaged layer-specific excitation-to-inhibition ratio where the amplitude of EPSCs at À70 mV is divided by the amplitude of IPSCs at 0 mV for each cell. E/I ratio for vS1 inputs was significantly larger in PV1 than in Pyr in L2 Comparison of excitatory to inhibitory response ratio within the same cells between L2/3 and L5A showed that for Pyr cells the vS1 excitatory drive was larger in L5A compared with L2/3 (E/I ratio, Mann-Whitney, p = 4.89E-3, h 2 = 0.254; Fig. 7C,D) confirming the results from a previous study (Mao et al., 2011). For both L2/3 and L5A, the vS1 excitatory drive was larger for PV1 cells than for Pyr cells, while PO excitatory drive was significantly larger only in L5A PV1 cells compared with L5A Pyr cells (L2/3 vS1 inputs E/I ratio Mann-Whitney, p = 2.2E-5, h 2 = 0.409; L5A vS1 inputs E/I ratio Mann-Whitney, p = 1.83E-2, h 2 = 0.197; L5A PO inputs E/I ratio Mann-Whitney, p = 4.0E-4, h 2 = 0.344; Fig. 7C,D). Consistent with larger vS1 inputs in L2/3 PV1 cells the vS1 inhibitory responses were larger in L2/3 Pyr cells compared with L5A Pyr cells (Mann-Whitney, p = 3.08E-2, h 2 = 0.180; Fig. 7B-D). This suggests that long-range inputs excite more PV1 cells eliciting stronger feedforward inhibition in a layer-specific manner.

Discussion
Here, we tested whether locally connected subnetworks of Pyr and PV1 neurons exist in mouse M1 and whether these local networks differed in input from two major sources of long-range excitation. The data show that connected Pyr and PV1 neuron pairs indeed share correlated long-range input. Furthermore, Pyr neuron connectivity is elevated to PV1 neurons that inhibit them. This data collectively suggests that inhibitory neurons in motor cortex are specifically connected in local subnetworks.
Motor cortex is somatotopically organized. Thus, M1 might contain distinct subnetworks for different motor functions, such as different movements of a given limb. Thus, we tested the local and long-range connectivity of PV1 and Pyr neurons to assess the existence of specific subnetworks. Connected pairs represented withinnetwork neurons, and unconnected pairs were chosen to represent different networks. Here, we show that connected pairs of PV1 and Pyr cells in M1 have stronger correlation of long-range somatosensory (vS1) and thalamic (PO) projections than nonconnected pairs (Fig. 6). Thus, subnetworks with strong vS1 input exist and more strongly excite connected PV1 and Pyr neurons. and IPSCs (y-axis) for connected (left panels) PV1 (red) and Pyr (black) and nonconnected (right panels, magenta for PV1 and gray for Pyr). The estimation of confidence interval for the Spearman's r correlation coefficient was also done with 10,000 bootstrap resampling, for VS1 inputs to PV1 connected (A,  upper

Integration of interneurons into local subnetworks
It is controversial the degree to which interneurons participate in such subnetworks. Imaging interneuron response properties in visual cortex suggests that these cells are more broadly tuned than excitatory neurons (Kerlin et al., 2010), while other studies suggest PV1 cells may be selective to orientation and direction (Runyan et al., 2010). Further, in direction-selective or orientation-selective inhibitory interneurons (52%) and their clusters, 75% of clustered Pyr cells shared direction tuning with their corresponding inhibitory neuron (Palagina et al., 2019). Relatively dense local connectivity of PV1 neurons has been proposed, with both nonspecific but frequent output to local Pyr neurons (Packer and Yuste, 2011;Fino et al., 2013) and pooled excitatory input from Pyr cells with different properties (Sohya et al., 2007).
How to reconcile a role for selective inhibitory connectivity with nonselective all-to-all inhibition remains to be answered. One possibility is that connections are common, but synaptic strength selectively varies within or across subnetworks (Znamenskiy et al., 2018). One statistical/structural explanation for subnetworks is the targeting by feedforward and feedback long-range projections, which may increase the information propagation and processing in cortical circuits by selecting clustered cells that have higher probability of connecting to each other (Nigam et al., 2016;Rost et al., 2018;Faber et al., 2019;Palagina et al., 2019;Peron et al., 2020). Thus, synaptic inputs from long-range projections may be correlated in connected clusters.
In our data, selectively connected PV1 interneurons share correlated input strength with in-network pyramidal neurons. Furthermore, the contribution of interneurons to subnetworks is not simply correlated connection strength, but also enhanced connection probability. While PV1 neurons make frequent local connections to Pyr neurons (N = 71/197 pairs, 36.0%), Pyr neurons make sparser connections to PV1 cells (N = 24/197, 12.2%). However, Pyr neurons are much more likely to excite PV1 cells that reciprocally connect to them, with connection rates as high as those for PV1 output (N = 18/53, 34.0%). Thus, connected pairs were reciprocally connected at much higher than random rate (Fig. 4). This arrangement is consistent with a given Pyr neuron activating its own feedback inhibition, providing a negative feedback mechanism to stabilize excitability. The overall connection probability is potentially underestimated, as some connections may be severed in the brain slices. But it is unlikely that slice preparation differentially severs connections between neurons that lack correlated excitatory input.

Thalamocortical and corticocortical connection strength in M1
These data show that, consistent with earlier work (Hooks et al., 2015), single Pyr and PV1 neurons receive input from both cortical (vS1) and thalamic (PO) sources. In assessing strength of connections, vS1 inputs were stronger in L2/3 PV1 cells than PO inputs (Fig. 3). Furthermore, both classes of excitatory inputs were much stronger in PV1 cells than to Pyr cells (Figs. 3, 5). This is in line with previous studies showing increased thalamocortical inputs to PV1 compared with Pyr cells in somatosensory cortex (Gibson et al., 1999;Gabernet et al., 2005;Cruikshank et al., 2007Cruikshank et al., , 2010. There may be differences in measured EPSC strength because of differences between the cell types. PV1 cells may be more electrotonically compact, making larger inputs easier to measure, and differences in intrinsic excitability may favor higher firing rates in PV1 neurons (Extended Data Fig. 1-1). These differences amplify the effectiveness of inputs in PV1 neurons. Thus, in general, the E/I ratio of PV1 cells is higher than the E/I ratio of Pyr cells (Fig. 7). Further, comparing input strength across layers (Fig. 7D), E/I ratio for vS1 input to Pyr cells increases from L2/3 to L5A implicating greater recruitment of feedforward inhibition in L2/3, presumed to originate from PV1 cells. In contrast, E/I ratio for thalamic input to Pyr cells goes down from L2/3 to L5A, implicating greater recruitment of feedforward inhibition in L5A (Okoro et al., 2022).

Technical notes
This current work uses long optogenetic stimulations (50-500 ms), which may result in polysynaptic, recurrent activity in the vM1 circuits. Some opsin evoked EPSCs do show multiple peaks (Fig. 2). However, comparing sweeps with a range of stimulus durations (50, 100, 250, and 500 ms; Fig. 1J,K) suggests that longer duration stimuli do not activate increased polysynaptic responses, as the initial EPSC is of similar amplitude for different durations of 590-nm stimulation. There is a small possibility that the stimuli may activate the same small set of recurrent synapses in all durations protocols, in which case these recurrent synaptic inputs still would be considered to originate within the subnetwork by definition (clusters of connected cells), and at the same time it will need to produce input amplitudes bigger than the opsin evoked EPSCs to contribute increased variance to the analysis of current results. Further, previous work in our lab using subcellular CRACM, with application of TTX to block non-Channelrhodopsin evoked activity can produce similar multiple peak responses following short 1 ms, localized laser stimulation (Okoro et al., 2022). Comparison of opsin-evoked EPSCs onset latencies between sCRACM from our previous work and 2CRACM in current work also suggest minimal contribution of recurrent synaptic activity despite longer stimuli and EPSCs poly-peaks recordings (data not shown). It is also worth noting that a similar laminar pattern of input strength occurs following LED stimulation (this work) as is measured in TTX following laser stimulation (Okoro et al., 2022). Coupled with the low spontaneous AP firing rate in the cortex (for review, see Barth and Poulet, 2012), polysynaptic responses on wide-field LED stimulation are not expected to be major contributors to these results.
In conclusion, this work proposes a role for long-range projections as part of the neural circuit organization that differentiates the primary whisker motor cortex into subnetworks, to some degree as in visual and somatosensory cortices. Differences in cortical and thalamic input to different local subnetworks can result in local circuit elements specialized for processing different streams of information. However, these experiments are done ex vivo in mouse cortical slices, which may underestimate connectivity. Whether similar results can be obtained in vivo remains to be tested.