Abstract
The anterolateral motor cortex (ALM) and ventral medial (VM) thalamus are functionally linked to support persistent activity during motor planning. We analyzed the underlying synaptic interconnections using optogenetics and electrophysiology in mice (female/male). In cortex, thalamocortical (TC) axons from VM thalamus excited VM-projecting pyramidal tract (PT) neurons in layer 5B of ALM. These axons also strongly excited layer 2/3 neurons (which strongly excite PT neurons, as previously shown) but not VM-projecting corticothalamic (CT) neurons in layer 6. The strongest connections in the VM → PT circuit were localized to apical tuft dendrites of PT neurons, in layer 1. These tuft inputs were selectively augmented after blocking hyperpolarization-activated cyclic nucleotide-gated (HCN) channels. In thalamus, axons from ALM PT neurons excited ALM-projecting VM neurons, located medially in VM. These axons provided weak input to neurons in mediodorsal nucleus, and little or no input either to neurons in the GABAergic reticular thalamic nucleus or to neurons in VM projecting to primary motor cortex (M1). Conversely, M1 PT axons excited M1- but not ALM-projecting VM neurons. Our findings indicate, first, a set of cell type-specific connections forming an excitatory thalamo-cortico-thalamic loop for ALM ↔ VM communication and a circuit-level substrate for supporting reverberant activity in this system. Second, a key feature of this loop is the prominent involvement of layer 1 synapses onto apical dendrites, a subcellular compartment with distinct signaling properties, including HCN-mediated gain control. Third, the segregation of the ALM ↔ VM loop from M1-related circuits of VM adds cellular-level support for the concept of parallel pathway organization in the motor system.
SIGNIFICANCE STATEMENT Anterolateral motor cortex (ALM), a higher-order motor area in the mouse, and ventromedial (VM) thalamus are anatomically and functionally linked, but their synaptic interconnections at the cellular level are unknown. Our results show that ALM pyramidal tract neurons monosynaptically excite ALM-projecting thalamocortical neurons in a medial subdivision of VM thalamus, and vice versa. The thalamo-cortico-thalamic loop formed by these recurrent connections constitutes a circuit-level substrate for supporting reverberant activity in this system.
Introduction
The mammalian motor cortex comprises multiple subregions, the differentiated functions of which involve interactions with an array of motor-related thalamic nuclei (Strick, 1986; Middleton and Strick, 2000; Jones, 2007; Bosch-Bouju et al., 2013). Anatomically, primary motor cortex (M1) receives particularly prominent projections from the cerebellar-recipient ventral lateral (VL) nucleus (Strick and Sterling, 1974; Jones, 1975; Kuramoto et al., 2009). M1 also receives projections from basal ganglia-recipient nuclei, particularly the ventral anterior nucleus (Strick, 1975; Kuramoto et al., 2009). The ventral medial (VM) nucleus, which is both basal ganglia and cerebellar recipient (Groenewegen and Witter, 2004; Gao et al., 2018), sends projections to multiple motor areas, particularly premotor and other higher-order areas in frontal cortex (Herkenham, 1979; Arbuthnott et al., 1990; Kuramoto et al., 2015). In rodents, the medial subdivision of the posterior nucleus (POm), which receives ascending paralemniscal somatosensory afferents, projects to M1 (in addition to somatosensory cortex; Hooks et al., 2013). The intracortical connections of some of these projections, particularly VL → M1 and POm → M1, have begun to be elucidated in mice (Hooks et al., 2013; Yamawaki et al., 2014; Yamawaki and Shepherd, 2015), as have projections of VM → medial prefrontal cortex (mPFC; Cruikshank et al., 2012; Collins et al., 2018). However, the cellular circuits that link premotor and other higher-order motor cortical areas with VM remain to be characterized.
In the mouse, the anterolateral motor (ALM) cortex has premotor-like properties, particularly in behavioral tasks involving motor planning during a temporal delay (Guo et al., 2014; Inagaki et al., 2018). ALM neurons show persistent activity anticipating specific movements, seconds before movement onset. These studies have implicated strong bidirectional interactions between ALM and the thalamus, including the VM nucleus, as being crucial for these behavioral functions (Guo et al., 2017). The ALM-recipient subdivision of VM receives input from substantia nigra pars reticulata as well as the fastigial nucleus of the cerebellum (Gao et al., 2018). The cellular-level excitatory connections that form thalamo-cortico-thalamic (T-C-T) circuits in this system remain unknown.
The goal of this study was to characterize these connections in the putative T-C-T loop involving ALM and VM, focusing on the identification of connections onto back-projecting neurons in either area, which would form a basis for recurrent excitation in this system. We used previously developed strategies for cell type-specific circuit analysis based on combining optogenetic photostimulation with electrophysiological recordings from retrogradely labeled projection neurons in motor cortex and motor thalamus (Yamawaki and Shepherd, 2015; Yamawaki et al., 2016). Our results delineate a set of cell type-specific excitatory connections constituting the cellular and synaptic underpinnings of a bidirectional ALM ↔ VM T-C-T loop.
Materials and Methods
Animals
Animal studies followed the guidelines of the National Institutes of Health and Society for Neuroscience, and were approved by the Northwestern University Animal Care and Use Committee. The following mouse strains were used in these studies, maintained as in-house breeding colonies. For wild-type mice (WT) we used either C57BL/6 mice or Gad2-mCherry mice (Gad2-T2a-NLS-mCherry, The Jackson Laboratory; RRID:IMSR_JAX:023140). For cell type-specific expression, we used the Cre lines Rbp4-Cre (Rbp4_KL100-Cre, MMRRC; RRID:MMRRC_037128-UCD; Gerfen et al., 2013), Ntsr1-Cre (Ntsr1_GN220-Cre, MMRRC; RRID:MMRRC_030648-UCD; Gong et al., 2007), and Calb1-Cre (Calb1-IRES2-Cre-D, The Jackson Laboratory; RRID:IMSR_JAX:028532), each back-crossed with C57BL/6 mice for at least six generations. A Cre-dependent tdTomato line (Ai14, The Jackson Laboratory; RRID:IMSR_JAX:007908; Madisen et al., 2015) was used as a fluorescent reporter. Female and male mice were used in approximately equal numbers. Animals were housed with a 12 h light/dark cycle, with ad libitum access to water and food. Mice were 1.5–3 months old at the time of the initial surgery and were used for experiments 3–6 weeks later. Animal numbers for each type of experiment are given in the text and figures.
Labeling
Stereotaxic injections were performed as described previously (Yamawaki and Shepherd, 2015). Briefly, mice were deeply anesthetized with isoflurane and placed in a stereotaxic frame. Ophthalmic ointment was applied to protect the eyes during surgery. Thermal support was provided using a feedback-controlled heating pad (Warner). Mice were given preoperative analgesic coverage (buprenorphine 0.3 mg/kg, s.c.). Small craniotomies were opened directly over cortical and/or subcortical targets in the right hemisphere. Injection pipettes were fabricated from glass capillary micropipettes (Wiretrol II, Drummond Scientific Company) by pulling (PP-830, Narishige) to a fine tip and beveling (Micro Grinder EG-400, Narishige) to a sharp edge. Pipettes were loaded with virus or tracer solution by tip filling (i.e., application of negative pressure to the back of the pipette). Pipettes were advanced slowly to their targets, where 40–100 nl of virus was injected, and were kept in place for several minutes before retraction. Animals received postoperative analgesic coverage (meloxicam 1.5 mg/kg, s.c., once every 24 h for 2 d) and were maintained for at least 3 weeks before slice experiments. The experiment-specific details are as follows.
In experiments examining thalamic input to cortical neurons, in most cases adeno-associated virus (AAV)-channelrhodopsin-2 (ChR2)-Venus (AAV1.CAG.ChR2-Venus.WPRE.SV40, AV-1-PV2126; University of Pennsylvania Vector Core) was injected into thalamus of WT mice or Gad2-mCherry mice (which in these experiments were used simply as a source of wild-type mice, not for their labeling pattern). For VM injections, the stereotaxic coordinates were as follows: 1.3–1.7 mm posterior to bregma; 0.8–1.0 mm lateral to midline; and 4.1–4.4 mm below pia. After virus injection, additional injections were made with either latex microspheres (red Retrobeads, LumaFluor) or cholera toxin subunit B conjugated to Alexa Fluor (CTB647, Life Technologies) to retrogradely label projection neurons in ALM. Specifically, tracers were injected into VM (same coordinates as above) to label both pyramidal tract (PT) and layer 6 corticothalamic (CT) neurons, and into pons (3.5–3.7 mm posterior; 0.4–0.6 mm lateral; 5.0–5.8 mm deep) to label PT neurons.
In a subset of the thalamocortical (TC) experiments, AAV-FLEX-ChR2-tdTomato (AAV1.CAGGS.FLEX.ChR2-tdTomato.WPRE.SV40, AV-1-18917P; University of Pennsylvania Vector Core) was injected into VM of Calb1-Cre mice. These mice were used to explore the possibility of facilitating selective labeling of VM, which, characteristic of matrix-type thalamic nuclei, expresses calbindin at relatively high levels (Jones, 2001; Rubio-Garrido et al., 2007). In Calb1-Cre mice, Cre expression is relatively high in VM (transgenic characterizations #574212675, #576527875, and #576528091, Allen Brain Institute; http://connectivity.brain-map.org/transgenic). However, injections targeted to VM resulted in similar labeling patterns in thalamus using either the Cre-dependent or Cre-independent approach, and electrophysiological results were also statistically indistinguishable; the results were therefore pooled.
In experiments examining cortical input to thalamic neurons, we used Rbp4-Cre and Ntsr1-Cre mice to selectively label the PT or CT components, respectively, that together comprise the cortical projection to thalamus. Previous studies have characterized the cellular specificity of Cre expression in cortical neurons in these lines [Gong et al., 2007; Gerfen et al., 2013; Bortone et al., 2014; multiple transgenic characterizations available at http://connectivity.brain-map.org/transgenic (e.g., #167642756 for Rbp4-Cre)] and have used these lines for selectively labeling the PT and CT components of the cortical projection to thalamus (Grant et al., 2016; Jeong et al., 2016). Specifically, AAV-FLEX-ChR2-tdTomato (for details, see above) was injected into the ALM of either Ntsr1-Cre mice, to label layer 6 CT neurons, or Rbp4-Cre mice, to label layer 5 intratelencephalic and PT neurons. Injection of the same Cre-dependent virus into the cortex of WT mice resulted in no labeling (n = 2). The ALM and M1 were also injected with retrograde tracer to label thalamocortical neurons in VM. The ALM coordinates (Komiyama et al., 2010; Guo et al., 2014) were as follows: 2.4–2.6 mm anterior; 1.4–1.6 mm lateral; and 0.1–1.7 mm deep. The M1 coordinates were as follows: 0.1–0.3 mm posterior; 1.5–1.7 mm lateral; and 0.1–1.0 mm deep. To characterize the Cre expression of ALM PT neurons in Rbp4-Cre mice, a Cre-dependent retrograde AAV (Tervo et al., 2016), AAVretro-FLEX-tdTomato (retrograde AAV #28306; Addgene), was injected into VM.
Circuit analysis
Brain slices were prepared as described previously (Yamawaki and Shepherd, 2015). Briefly, mice that had undergone in vivo labeling were killed by anesthetic overdose and decapitation. Brains were rapidly removed and placed in chilled cutting solution (composition, in mm: 110 choline chloride, 11.6 sodium l-ascorbate, 3.1 pyruvic acid, 25 NaHCO3, 25 d-glucose, 2.5 KCl, 7 MgCl2, 0.5 CaCl2, and 1.25 NaH2PO4, aerated with 95% O2/5% CO2). For experiments involving cortical recordings, brains were trimmed using blocking cuts angled to yield off-sagittal slices (0.3 mm) that optimally preserved the apical dendrites of cortical pyramidal neurons. For thalamic recordings, blocking cuts were angled to yield coronal slices (0.25 mm) containing the ventral thalamus. Slices were cut (VT1200S, Leica) in chilled cutting solution, and transferred to artificial CSF (ACSF), composed of the following (in mm): 127 NaCl, 25 NaHCO3, 25 d-glucose, 2.5 KCl, 1 MgCl2, 2 CaCl2, and 1.25 NaH2PO3. Slices were incubated for 30 min at 34°C, and then kept at 22°C for at least an hour before recording.
Electrophysiology and photostimulation.
Brain slices were transferred to the recording chamber of an upright microscope (BX51WI chassis, Olympus) equipped for whole-cell electrophysiology and photostimulation, as previously described (Yamawaki and Shepherd, 2015). The bath solution consisted of ACSF warmed to 34°C by an in-line feedback-controlled heater (TC 324B, Warner). For all cortical recordings, tetrodotoxin (TTX; 1 μm; Tocris Bioscience) and 4-aminopyridine (4AP; 100 μm; Sigma-Aldrich) were added to the ACSF to isolate monosynaptic (i.e., eliminate intracortical polysynaptic) inputs (Petreanu et al., 2009); these reagents were omitted for recordings in thalamus, where intrathalamic polysynaptic excitation was not a concern. For recordings in cortex, 3-(2-carboxypiperazin-4-yl)propyl-1-phosphonic acid (CCP, 5 μm; Tocris Bioscience) and ZD7288 (10 μm; Tocris Bioscience) were included to block NMDA receptors and hyperpolarization-activated cyclic nucleotide-gated (HCN) channels, respectively.
Patch pipettes were fabricated by pulling (P-97; Sutter Instrument) borosilicate glass (inner diameter, 0.86; outer diameter, 1.5 mm; with filament; Warner) to a fine tip (resistance, 2–4 MΩ). Recordings were made in voltage-clamp or current-clamp mode using pipettes filled with cesium- or potassium-based internal solution (composition, in mm: 128 cesium or potassium methanesulfonate, 10 HEPES, 10 phosphocreatine, 4 MgCl2, 4 ATP, 0.4 GTP, 3 ascorbate, 1 EGTA, 1 QX-314, 0.05 Alexa Fluor hydrazide, and 4 mg/ml biocytin, pH 7.25, 290–295 mOsm).
A video camera (Retiga 2000r, Q-Imaging) was used to image slices and cells under wide-field gradient-contrast or epifluorescence optics. Labeling patterns of retrogradely labeled somata and anterogradely labeled axons were visualized using LED illumination (catalog #M470L2, #M530L2, and #M660L3, Thorlabs) and standard filter sets (catalog #U-N41017, #U-N31002, Chroma, #CY5-4040C-OMF, Semrock). Pipettes were advanced under positive pressure to establish >1 GΩ seals onto identified neurons. After membrane rupture to establish whole-cell configuration, intracellular recordings were made with an amplifier (Multiclamp 700B, Molecular Devices). Recordings with series resistance of >40 MΩ were excluded.
For wide-field photostimulation, as previously described (Yamawaki and Shepherd, 2015), a blue LED (catalog #M470L2, Thorlabs) was driven with a TTL (transistor–transistor logic) pulse to generate photostimuli with a duration of 5 ms. The LED intensity controller was set to deliver 1 mW/mm2 at the level of the specimen. For each cell, photostimulation trials were repeated three times at an interstimulus interval of 30 s while recording in voltage-clamp mode with the command potential set to −70 mV.
For subcellular circuit mapping (sCRACM; Petreanu et al., 2009), a blue laser (473 nm, 50 mW; model MLL-FN473, CNI Laser) and scan system (model 6210 galvanometer pair, Cambridge Technology) were used for focal photostimulation and subcellular mapping, as previously described (Petreanu et al., 2009; Suter and Shepherd, 2015). Photostimuli were generated by controlling an electro-optical modulator (350–50, 302 RM, Conoptics) and mechanical shutter (LS2ZM2, VCM-D1, Uniblitz) in the beam path to produce a 1.0-ms-duration pulse of light at each stimulus location. The intensity was in the range of 0.1–2 mW/mm2 at the level of the specimen, adjusted on a cell-by-cell basis to generate a 100–200 pA response to stimulation near the soma. Stimulation grids consisted of at least 26 rows and 10 columns, with uniform 60 μm spacing. The upper edge of the grid (row 1) was aligned to the pia, and the grid was horizontally centered over the soma. The grid parameters (total area, ∼1 mm2; ∼250 sites/mm2) and orientation were chosen to extend from pia to white matter and fully span the dendritic arbors of PT neurons. Each neuron was mapped three times using different pseudorandom sequences, with an interstimulus interval of 0.4 s. Ephus software (Suter et al., 2010) was used to control hardware settings and other parameters for coordinating photostimulation and electrophysiology data acquisition.
Analysis.
To quantify LED-evoked responses, for each cell, the traces from several (generally three) trial repetitions were averaged, and the response amplitude was calculated as the mean amplitude in a poststimulus interval of 50 ms. Data were compared by pooling across slices and animals, as in previous studies (Yamawaki and Shepherd, 2015). Pairwise comparisons were made using the absolute or normalized response amplitudes, as indicated in the text. To quantify sCRACM data, the traces from the three map trials were averaged, mean response amplitudes were determined as described above, and these values were used to construct maps representing the synaptic input at each stimulus location.
Experimental design and statistical analysis
Group comparisons were made using nonparametric tests as indicated in the text, with significance defined as p < 0.05. Unpaired data were compared using the rank-sum test. Paired data were compared using the sign test. For group data, median and median average deviation (m.a.d.) values were calculated as descriptive statistical measures of central tendency and dispersion, except for ratios, for which geometric means and standard factors were calculated. Statistical analyses were conducted using standard Matlab (MathWorks) functions.
Results
Excitatory input from VM thalamocortical axons to VM-projecting ALM neurons is much stronger to layer 5B PT neurons than to layer 6 CT neurons
To dissect the cell type-specific connections at the cortical end of the potential T-C-T loop, we photostimulated ChR2-expressing TC axons from VM in cortical brain slices while recording from cortical projection neurons in ALM. First, we localized ALM-projecting VM neurons (VMALM-proj) by injecting ALM with retrograde tracer and imaging the thalamus (Fig. 1A). Next, in different animals, we targeted this VM region for injection with AAV-ChR2-Venus to label VM projections, and also with a retrograde tracer to label VM-projecting cortical neurons (Fig. 1B). In the same animals, we also injected the pons with retrograde tracer to label PT neurons and thereby identify layer 5B (Qiu et al., 2011). This approach yielded triple-labeled cortical slices (Fig. 1C). Anterogradely labeled axons from TC neurons in VM were densest in layer 1 (Herkenham, 1979; Arbuthnott et al., 1990; Kuramoto et al., 2015). The slices also contained the retrogradely labeled pons-projecting neurons, demarcating layer 5B, and the retrogradely labeled thalamus-projecting neurons in both layers 5B and 6. For clarity and consistency with previous studies, we refer to the thalamus-projecting neurons in layer 5B as PT neurons (i.e., with thalamic branches), and those in layer 6 as CT neurons (Harris and Shepherd, 2015; Yamawaki and Shepherd, 2015).
With this labeling approach, we investigated excitatory connectivity in the VM → ALM pathway, initially focusing on these two types of VM-projecting neurons: layer 5B PT neurons and layer 6 CT neurons (Fig. 1D). Photostimulation of the VM axons (with TTX and 4AP in the bath solution; see Materials and Methods), generated fast, large, monosynaptic excitatory currents in VM-projecting PT neurons that were much stronger than the generally weak or undetectable responses recorded in VM-projecting CT neurons (Fig. 1D–G). Findings were similar with wild-type mice and Calb1-Cre mice (Materials and Methods), and the data were therefore pooled for analysis (Fig. 1D–G). Findings were similar for group comparisons based on individual neurons (Fig. 1E) or animals (Fig. 1F,G).
Specifically, for WT mice (n = 8), injected in VM with AAV-ChR2-Venus, the VM → PT responses were approximately −50 pA [median across n = 14 cells, −56.7 ± 31.7 pA (median ± m.a.d.); median for n = 8 mice, −54.6 ± 32.7 pA] while the VM → CT responses were less than −5 pA (median across n = 13 cells, −1.3 ± 1.1 pA; median for n = 8 mice, −1.0 ± 0.5 pA), a 10-fold difference (unpaired cell-based comparison, p = 0.0002, rank-sum test; pairwise animal-based comparison, p = 0.008, sign test; VM → PT/VM → CT ratio per animal, 19.9 ± 5.9, geometric mean ± geometric standard factor). For Calb1-Cre mice (n = 7) injected in VM with AAV-FLEX-ChR2-tdTomato, the VM → PT responses were approximately −50 pA (median across n = 11 cells, −58.2 ± 12.2 pA; median for n = 7 mice, −55.9 ± 18.3 pA), whereas the VM → CT responses were less than −5 pA (median across n = 13 cells, −3.7 ± 3.1 pA; median for n = 7 mice, −3.8 ± 2.8 pA), a statistically significant difference (unpaired cell-based comparison, p = 0.0001, rank-sum test; pairwise animal-based comparison, p = 0.016, sign test; VM → PT/ VM → CT ratio per animal, 12.9 ± 2.9, geometric mean ± geometric standard factor). For the pooled data (n = 15 mice total, WT and Calb1-Cre), the VM → PT responses were approximately −50 pA (median across n = 25 cells, −58.2 ± 24.1 pA; median for n = 15 mice, −54.7 ± 18.2 pA), while the VM → CT responses were less than −5 pA (median across n = 26 cells, −1.7 ± 1.6 pA; median for n = 15 mice, −3.7 ± 3.1 pA), a statistically significant difference (unpaired cell-based comparison, p = 4 × 10−8, rank-sum test; pairwise animal-based comparison, p = 0.0001, sign test; VM → PT/VM → CT ratio per animal, 16.3 ± 4.3, geometric mean ± geometric standard factor).
These results show that VM projections to ALM excite VM-projecting PT neurons, with little or no input to back-projecting CT neurons (Fig. 1H). An important implication of this result (addressed in experiments presented in a later section) is that if the potential ALM ↔ VM T-C-T loop is tightly “closed” by monosynaptic connections, such connections must involve the PT neurons, rather than the CT neurons, given the prevalence of monosynaptic VM → PT and the paucity of VM → CT connections.
VM axons also strongly excite layer 2/3 neurons in ALM
We also considered the possibility of strong VM inputs to layer 2/3 pyramidal neurons, both because their apical dendrites arborize in layer 1, making such a connection likely by virtue of axodendritic overlap, and because layer 2/3 → PT connections are among the strongest excitatory intracortical connections in anterior motor cortex (Qiu et al., 2011), making such a connection significant by virtue of delineating a parallel disynaptic pathway (i.e., VM → 2/3 → PT). To test this, we used the same approach as above, recording from layer 2/3 pyramidal neurons and PT neurons in ALM slices while photostimulating VM axons (Fig. 2A). Analysis of the photo-evoked EPSCs showed about twice as much input to layer 2/3 neurons as to PT neurons (Fig. 2B–D). Specifically, for WT and Calb1-Cre mice (n = 10 mice total, pooled from 6 WT and 4 Calb1-Cre mice as the results were similar), the VM → PT responses were approximately −50 pA (median across n = 14 cells, −47.1 ± 25.1 pA; median for n = 10 mice, −46.0 ± 11.2 pA) while the VM → L2/3 responses were approximately −100 pA (median across n = 13 cells, −93.7 ± 38.3 pA; median for n = 10 mice, −91.0 ± 39.0 pA), a statistically significant difference (unpaired cell-based comparison, p = 0.014, rank-sum test; pairwise animal-based comparison, p = 0.021, sign test; VM → PT/VM → L2/3 ratio per animal, 0.43 ± 2.3, geometric mean ± geometric standard factor). Thus, layer 2/3 pyramidal neurons also receive strong VM excitation (Fig. 2E).
Localization of VM excitatory synapses to apical tuft dendrites of PT neurons
The high density of VM TC axons in layer 1 (Fig. 1C; Herkenham, 1979; Arbuthnott et al., 1990; Kuramoto et al., 2015) presents the likelihood of axodendritic overlap with the prominent apical tuft dendrites of PT neurons, which suggests that VM input to PT neurons enters via excitatory synapses on apical dendrites in layer 1. However, VM → PT EPSCs could instead or additionally reflect strong perisomatic synapses. To resolve this, we mapped the subcellular locations of inputs on postsynaptic dendrites using sCRACM (Petreanu et al., 2009; Fig. 3A,B; see Materials and Methods). These maps revealed localized hotspots of strong excitatory input to layer 1 apical tufts of PT neurons, as well as a broad region of weaker perisomatic input (Fig. 3B,C).
One implication of these synaptic mapping results is the likelihood that the tuft-targeting VM inputs are shaped by the active properties, particularly H-current (carried by HCN channels), of PT apical dendrites (Lörincz et al., 2002; Sheets et al., 2011; Larkum, 2013; Labarrera et al., 2018). To explore this, we recorded from VM-projecting ALM PT neurons (in current-clamp mode, at the resting potential) and used laser-scanning photostimulation to focally activate VM axons at one site over the apical tuft dendrites and another site over the basal dendrites close to the soma (Fig. 3D). Postsynaptic responses to apical stimulation were markedly enhanced after the application of ZD7288 (10 μm), a selective blocker of HCN channels, while inputs to the basal dendrites were smaller and not significantly enhanced after ZD7288 application (Fig. 3D).
The results of these sCRACM mapping and pharmacology experiments show that VM-projecting PT neurons in ALM receive direct, monosynaptic excitation from VM that preferentially targets apical tuft dendrites in layer 1 (Fig. 3E). They also demonstrate that VM → PT signaling via these distal apical inputs is modulated by H-current, suggesting a candidate mechanism for neuromodulatory regulation of VM → ALM communication.
ALM PT axons excite ALM-projecting VM TC neurons, but not M1-projecting VM TC neurons
Next, we turned our attention to the thalamus, focusing on assessing whether the axons of ALM PT neurons close the potential T-C-T loop by exciting ALM-projecting VM neurons. We used mice from the layer 5-specific Rbp4-Cre line (Materials and Methods), which enabled us to selectively photostimulate ChR2-expressing cortical axons from PT neurons (but not CT neurons) in thalamic slices. Cortical injections of AAV-FLEX-ChR2-tdTomato and retrograde tracers into the ALM cortex of these mice yielded thalamic slices containing anterogradely labeled ChR2-expressing PT axons and retrogradely labeled ALM-projecting VM TC neurons. In addition, because VM axons can have large cortical arbors sometimes spanning multiple motor areas (Kuramoto et al., 2015) and may extend not only to ALM but also M1, we injected another retrograde tracer into M1, allowing us to compare inputs from ALM PT axons to ALM- versus M1-projecting VM TC neurons (Fig. 4A). Relatively strong EPSCs were detected in ALM-projecting VM TC neurons, whereas little or no input was detected in the M1-projecting VM TC neurons (Fig. 4B–E).
Specifically, for these experiments (n = 6 Rbp4-Cre mice), the responses of ALM-projecting VM neurons to ALM PT input (ALM-PT → VMALM-proj responses) were approximately −20 pA (median across n = 16 cells, −19.1 ± 7.7 pA; median for n = 6 mice, −18.3 ± 2.5 pA), while the ALM-PT → M1-projecting VM neuron (VMM1-proj) responses were nearly 0 pA (median across n = 15 cells, −0.7 ± 0.6 pA; median for n = 6 mice, −1.5 ± 0.4 pA), a statistically significant difference (unpaired cell-based comparison, p = 0.00001, rank-sum test; pairwise animal-based comparison, p = 0.03, sign test; ALM-PT → VMALM-proj/ALM-PT → VMM1-proj ratio per animal, 18.7 ± 2.5, geometric mean ± geometric standard factor).
These results show that excitatory PT → VM connections are formed onto back-projecting neurons, thus closing a T-C-T loop at the thalamic end, and that this ALM ↔ VM circuit does not engage other VM neurons that project to M1 (Fig. 4E). Consistent with the electrophysiological findings, the ALM-projecting TC neurons were anatomically localized to a medial part of VM, where ALM PT axons also ramified, whereas M1-projecting TC neurons were found in a more lateral part of VM (Fig. 4F–I). An implication of these findings is that the T-C-T loop is “closed” not only in the sense of recurrent excitatory connectivity between ALM and VM, but also in the sense of not engaging M1-associated T-C-T circuits.
M1 PT axons excite M1-projecting VM TC neurons, but not ALM-projecting VM TC neurons
To further explore these implications, we asked whether the converse connectivity pattern also holds; that is, do M1 PT axons preferentially excite M1-projecting rather than ALM-projecting VM TC neurons. We therefore repeated the experiment above, but injected the AAV-FLEX-ChR2-tdTomato into M1. Recordings in the thalamus from the two types of retrogradely labeled VM TC neurons showed the complementary pattern of excitatory input, with photo-evoked EPSCs detected in M1-projecting VM TC neurons but little or no input to ALM-projecting neurons (Fig. 5A–E). Specifically, for these experiments (n = 6 Rbp4-Cre mice), the M1-PT → VMM1-proj responses were approximately −13 pA (median across n = 16 cells, −13.3 ± 7.6 pA; median for n = 6 mice, −18.9 ± 6.5 pA), while the M1-PT → VMALM-proj responses were nearly 0 pA (median across n = 15 cells, −0.8 ± 0.6 pA; median for n = 6 mice, −1.0 ± 0.7 pA), a statistically significant difference (unpaired cell-based comparison, p = 0.00004, rank-sum test; pairwise animal-based comparison, p = 0.03, sign test; M1-PT → VMM1-proj/M1-PT → VMALM-proj ratio per animal, 17.1 ± 3.2, geometric mean ± geometric standard factor). Consistent with the electrophysiological results, anatomically the M1 PT axons were found to ramify in the laterally located M1-projecting subregion of VM, rather than in the ALM-projecting subregion (Fig. 5F–I). These findings provide further evidence that the T-C-T loops in the motor areas are relatively segregated.
Excitatory input from ALM PT axons is stronger to VM than to mediodorsal
In addition to the dense axonal projection from ALM PT neurons to VM, axons were also observed in the mediodorsal (MD) nucleus (Fig. 6A–D), raising the question of the relative innervation of MD TC neurons versus ALM-projecting VM TC neurons. To resolve this, we injected the ALM of Rbp4 mice with AAV-FLEX-ChR2-tdTomato and retrograde tracer, and recorded in thalamus from MD TC neurons and ALM-projecting VM TC neurons. Although EPSCs were sometimes detected in MD TC neurons, overall these neurons received significantly weaker input compared with ALM-projecting VM TC neurons (Fig. 6E–H). Specifically, for these experiments (n = 9 Rbp4-Cre mice), the PT → VM responses were approximately −20 pA (median across n = 20 cells, −21.4 ± 16.8 pA; median for n = 9 mice, −20.1 ± 7.4 pA), while the PT → MD responses were approximately −13 pA (median across n = 19 cells, −12.5 ± 10.3 pA; median for n = 9 mice, −13.1 ± 1.6 pA), a statistically significant difference (unpaired cell-based comparison, p = 0.006, rank-sum test; pairwise animal-based comparison, p = 0.039, sign test; PT → VM/PT → MD ratio per animal, 2.7 ± 3.3, geometric mean ± geometric standard factor). Thus, these results indicate that ALM-projecting VM TC neurons, rather than MD TC neurons, are the primary thalamic targets of ALM PT axons (Fig. 6I).
ALM CT axons excite both reticular neurons and ALM-projecting VM neurons, while ALM PT axons excite primarily the latter
The preceding experiments focused on the input–output connections of PT neurons with thalamus, because they, rather than CT neurons, were found to form recurrent connections with VM neurons. However, CT neurons are also integral, albeit enigmatic, components of T-C-T circuits, and we therefore performed a limited set of experiments to characterize CT connectivity to thalamic neurons. To assay these connections, we used the CT-specific Ntsr1-Cre line to express ChR2 (using AAV-FLEX-ChR2-tdTomato) in ALM CT axons. Neurons in the reticular nucleus of the thalamus (RTN) and ALM-projecting VM TC neurons both received excitatory input from ALM CT axons (Fig. 7A–F). Specifically, for these experiments (n = 7 Ntsr1-Cre mice), the CT → VM responses (median across n = 14 cells, −14.3 ± 8.6 pA; median for n = 7 mice, −10.8 ± 6.7 pA) and the CT → RTN responses (median across n = 13 cells, −4.2 ± 3.7 pA; median for n = 7 mice, −6.4 ± 5.8 pA) were not significantly different (unpaired cell-based comparison, p = 0.15, rank-sum test; pairwise animal-based comparison: p = 0.13, sign test; CT → VM/CT → RTN ratio per animal, 2.5 ± 3.6, geometric mean ± geometric standard factor). The absolute amplitudes of these ALM CT → RTN connections were notably small compared with those we previously measured for M1 CT → RTN connections (Yamawaki and Shepherd, 2015), likely reflecting the drop-off in Cre expression at the frontal pole in Ntsr1-Cre mice (Fig. 7G,H).
In contrast to ALM CT neurons, ALM PT axons provided little or no input to the RTN neurons (Fig. 7I–N). Specifically, for these experiments (n = 10 Rbp4-Cre mice), the PT → VM responses were approximately −35 pA (median across n = 17 cells, −30.7 ± 18.3 pA; median for n = 10 mice, −41.3 ± 26.1 pA), while the PT → RTN responses were approximately −5 pA (median across n = 17 cells, −3.7 ± 3.4 pA; median for n = 10 mice, −7.0 ± 5.6 pA), a statistically significant difference (unpaired cell-based comparison, p = 0.001, rank-sum test; pairwise animal-based comparison, p = 0.021, sign test; PT → VM/PT → RTN ratio per animal, 6.0 ± 5.2, geometric mean ± geometric standard factor). The Rbp4-Cre line had robust labeling of PT neurons in ALM (Fig. 7O,P). These experiments show that the PT neuron → VM → ALM loop is dominated by excitation, with little feedforward inhibition in the thalamus.
Discussion
Application of optogenetic-electrophysiological tools for cell type-specific analysis of excitatory connections between ALM and VM neurons yielded multiple salient details about circuit architecture in this system (Fig. 8). At the cortical end, TC → PT connections formed part of a recurrent excitatory loop, involving apical tuft dendrites of PT neurons and layer 2/3 pyramidal neurons, but not CT neurons. At the thalamic end, PT → TC connections closed the TC ↔ PT loop between ALM and VM, with weaker engagement of TC neurons in MD. The ALM ↔ VM loop involved VM TC neurons in a medial part of VM, with minimal cross talk between VM TC neurons in a more lateral, M1-associated part of VM. Collectively, these observations provide the outline of a wiring diagram describing how cortical and thalamic neurons in ALM and VM interconnect to form an excitatory T-C-T loop (Fig. 8).
The technical approach used here has many advantages but also limitations for cell type-specific circuit analysis. For example, the combination of ChR2-based wide-field photostimulation and whole-cell recording from tracer-labeled neurons affords high selectivity for stimulating presynaptic axons of interest, high efficiency and sensitivity for detecting connections, and high specificity for sampling postsynaptic neurons of interest, thereby enabling long-range input–output connections to be determined (Luo et al., 2008; Miesenböck, 2009; Schoenenberger et al., 2011; Yamawaki et al., 2016). However, the activation of ChR2 directly at presynaptic terminals by wide-field illumination may alter synaptic release properties, potentially limiting its utility for characterizing dynamic aspects of synaptic transmission (Schoenenberger et al., 2011; Jackman et al., 2014). Laser-based focal stimulation of axons at a distance away from the recorded cell can overcome this limitation (Jackman et al., 2014) but was not attempted in this study, which focused on connectivity rather than dynamics. Moreover, our focus on excitatory connections leaves unexplored the inhibitory circuit mechanisms in this T-C-T loop. In motor cortex, different types of inhibitory interneurons are integrated into local circuits with cell-type and layer specificity (Apicella et al., 2012), and are likely to be involved in shaping responses to thalamic input similar to their roles in VM projections to mPFC (Cruikshank et al., 2012; Delevich et al., 2015; Collins et al., 2018). In motor thalamus, inhibitory inputs from both the RTN and the basal ganglia converge on VM (Jones, 2007; Kase et al., 2015). Our elucidation of excitatory connections provides a starting point for further analysis of the cellular components and biophysical properties of this T-C-T circuit.
ALM-projecting VM neurons receive cortical input from ALM PT neurons and, indeed, appear to be the main target of ALM PT inputs to thalamus, thereby closing the thalamic end of the potential loop. Together, these results demonstrate a set of cell type-specific connections forming a T-C-T loop, in which thalamic branches of PT neurons in ALM mediate recurrent connections with thalamocortical neurons in VM. This recurrent circuit architecture may relate to the physiological functions of the ALM ↔ VM circuit, by supporting persistent activity associated with motor preparation (Li et al., 2016), representing a form of short-term memory.
The axonal projections of PT neurons are diverse and complex, even among PT neurons within the same cortical area (Kita and Kita, 2012; Shepherd, 2014; Rojas-Piloni et al., 2017). Recent results indicate that VM-projecting PT neurons in ALM generate particularly early and sustained preparatory activity in a delayed-response task, contrasting with the late-preparatory and motor-command activity observed in ALM PT neurons that projects instead to medulla (Economo et al., 2018). Further studies will be needed to determine whether these two and potentially additional subtypes of ALM PT neurons also have distinct T-C-T circuit connections.
The architecture shown here for ALM has both shared and distinct features compared with T-C-T circuits of mouse M1. In both M1 and ALM, PT but not CT neurons receive direct, monosynaptic excitation from thalamus (Hooks et al., 2013; Yamawaki and Shepherd, 2015). However, the targeting of thalamic input to apical tuft dendrites is particularly strong for VM → PT connections in ALM (this study), but is relatively moderate for VL → PT connections in forelimb-related M1 (Suter and Shepherd, 2015) and relatively weak for VL → PT connections in vibrissal M1 (Hooks et al., 2013). These differences are partially explained by the anatomical differences of the TC projections: VM projections, characteristic of matrix-type TC projections, ramify densely in layer 1 (Herkenham, 1979; Arbuthnott et al., 1990; Kuramoto et al., 2015), whereas VL projections, which accord more with core-type TC projections, ramify mainly in layers 4 and 5B (Jones, 1998; Clascá et al., 2012; Yamawaki et al., 2014). Perhaps the most prominent difference between the T-C-T circuits of ALM-VM and M1-VL is that the former form a recurrent excitatory loop but the latter do not. Specifically, the finding that ALM PT neurons connect to ALM-projecting VM TC neurons (and vice versa) implies a bidirectional ALM ↔ VM circuit architecture, and contrasts with the markedly weak connections from both PT and CT neurons in M1 to M1-projecting TC neurons in VL (Yamawaki and Shepherd, 2015), which, together with robust VL → PT connections in M1 (Hooks et al., 2013; Suter and Shepherd, 2015; Yamawaki and Shepherd, 2015), implies a predominantly feedforward VL → M1 circuit architecture. These differences in functional anatomy between ALM and M1, situated at the rostral-most and caudal-most regions of the cortical motor system, presumably conform to, and likely mediate, the specialized behavioral functions associated with the two areas. In particular, ALM supports preparatory activity associated with motor planning (discussed above), while an important aspect of M1 function is to integrate cerebellar input into ongoing corticospinal activity associated with motor execution.
Recurrent connections, including direct PT ↔ TC coupling, have also recently been described linking both VM and MD with mPFC (Collins et al., 2018), another frontal area where reverberant activity is implicated in short-term mnemonic functions. In general, numerous findings underscore the many anatomical and functional commonalities across these higher-order T-C-T circuits (Jones, 2007; Cruikshank et al., 2012; Bosch-Bouju et al., 2013; Delevich et al., 2015; Alcaraz et al., 2016, 2018; Acsády, 2017; Collins et al., 2018).
Computational models have emphasized the importance of recurrent excitation to maintain persistent activity underlying short-term memory and motor planning in frontal circuits (Wang, 1999, 2001). It is usually assumed that this recurrent excitation is caused by recurrence in cortical circuits (Goldman-Rakic, 1995). However, cortical circuits are dominated by rapid and strong feedforward and feedback inhibition, which creates challenges for the maintenance of persistent activity. For these reasons biophysical models of persistent activity overcome these challenges by invoking long-duration excitatory currents. Closed T-C-T circuits producing re-entrant feedback excitation may be a key specialization whereby frontal cortical circuits maintain memory-related persistent activity over seconds. In particular, PT neurons evoke little inhibition in the thalamus and the thalamic feedback to ALM is “excitation only,” overcoming the above-mentioned obstacles to the maintenance of persistent activity. This circuit may incorporate additional mechanisms, such as locally recurrent intracortical excitation, NMDA receptors, and synaptic facilitation to promote positive feedback (Wang, 1999; Wang et al., 2006; Mongillo et al., 2008; Kawaguchi, 2017). The localization of VM → PT synapses to apical tuft dendrites suggests that excitation might evoke slow dendritic spikes (Xu et al., 2012; Major et al., 2013) and involve regulation via H-current neuromodulation (Labarrera et al., 2018), which has previously been shown to regulate signaling in motor cortex circuits in a highly PT-specific manner (Sheets et al., 2011). Indeed, pharmacological blockade of HCN channels caused a marked increase in the amplitude of EPSPs generated by activating VM inputs to apical tuft dendrites of PT neurons in layer 1, indicating a candidate neuromodulatory mechanism likely to be involved in regulating activity in this T-C-T loop as well. By identifying key connections in this circuit, our results can facilitate targeted investigation of the mechanisms regulating in vivo activity in this system. Additionally, because aberrant activity in motor-related T-C-T loops—thalamocortical dysrhythmia—has been implicated in movement disorders (Llinás et al., 2005), our results can guide studies of pathophysiological mechanisms in disease models.
The finding that ALM PT input is specific to VMALM-proj rather than VMM1-proj neurons was unexpected, since the wide spread of VM axons across motor cortex observed in single-axon reconstructions suggests high divergence and convergence of VM → cortex pathways (Kuramoto et al., 2015). Although species differences (rat vs mouse) may contribute to the results, technical differences may be a greater factor. In particular, our measurements were of electrophysiological connections, not anatomical projections, and involved sampling from a potentially large population of presynaptic terminals, not single axons. It is also important to consider that the ALM and M1 are located relatively far apart in the frontal cortex, at its rostral-most and caudal-most extents, respectively. The extent to which VM projections to motor areas located between ALM and M1 are anatomically and functionally segregated remains to be determined. Nevertheless, our findings indicate that the thalamic part of the recurrent loop involving ALM is restricted to an ALM-associated medial subdivision of VM, distinct from an M1-associated lateral subdivision of VM. Thus, the connectivity patterns reported here appear broadly consistent with the concept of parallel pathway organization proposed for premotor pathways based on studies in nonhuman primates (Strick, 1986; Middleton and Strick, 2000).
Footnotes
The authors declare no competing financial interests.
This research was supported by National Institutes of Health Grants T32-MH-067564 (K.G.) and R01-NS-061963 (G.M.G.S.). We thank N. Bernstein and C. Maguire for technical assistance. We also thank L. Acsády for discussions and suggestions, and J. Barrett and H. Inagaki for comments on the manuscript.
- Correspondence should be addressed to Gordon M.G. Shepherd, Department of Physiology, Feinberg School of Medicine, Northwestern University, Chicago, IL 60611. g-shepherd{at}northwestern.edu