Abstract
Inhibition of GABAergic interneurons has been found to critically fine-tune the excitation–inhibition balance of the cortex. Inhibition is mediated by many connectivity motifs formed by GABAergic neurons. One such motif is the inhibition of somatostatin (SST)-expressing neurons by vasoactive intestinal polypeptide (VIP)-expressing neurons. We studied the synaptic properties of layer (L) 2/3 VIP cells onto L4 SST cells in somatosensory (S1) and visual (V1) cortices of mice of either sex using paired whole-cell patch-clamp recordings, followed by morphological reconstructions. We identified strong differences in the morphological features of L4 SST cells, wherein cells in S1 fell into the non-Martinotti cell (nMC) subclass, while in V1 presented with Martinotti cell (MC)-like features. Approximately 40–45% of tested SST cells were inhibited by VIP cells in both cortices. While unitary connectivity properties of the VIP-to-nMC and VIP-to-MC motifs were comparable, we observed stark differences in short-term plasticity. During high-frequency stimulation of both motifs, some connections showed short-term facilitation while others showed a stable response, with a fraction of VIP-to-nMC connections showing short-term depression. We thus provide evidence that VIP cells target morphological subclasses of SST cells differentially, forming cell-type–specific inhibitory motifs.
- Martinotti cell
- non-Martinotti cell
- plasticity
- short-term depression
- short-term facilitation
- somatosensory cortex
- somatostatin
- vasoactive intestinal polypeptide
- visual cortex
Significance Statement
Inhibitory circuits are involved in a wide variety of cortical computations. In particular, the inhibition of somatostatin-expressing (SST) neurons by vasoactive intestinal polypeptide-expressing (VIP) neurons has been well-documented in layer (L) 2/3 of sensory cortices. It was recently identified that L4 SST neurons of somatosensory (S1) and visual (V1) exhibit two different morphological subtypes, namely, non-Martinotti cells (nMCs) in S1 and Martinotti cells (MCs) in V1. We show that L2/3 VIP neurons inhibit both SST subtypes in L4 with similar dynamics. However, we also find that under high-frequency stimulations, the VIP-to-nMC motif exhibits strong short-term depression, but this was not observed in the VIP-to-MC motif. Therefore, we identified morphologically distinct, inhibitory cell-type–specific motifs in the sensory cortices of mice.
Introduction
An ever-increasing number of cortical cell types with cell-type–specific connectivity is the basis of higher cognitive functions like sensory perception and goal-directed motor behavior (Harris and Mrsic-Flogel, 2013; Staiger and Petersen, 2021). Neuronal connectivity utilizes chemical synaptic transmission across dynamic time scales (Hooks and Chen, 2020; Chéreau et al., 2022) for the consolidation of external sensory information into relevant behavioral and cognitive tasks (Jurjut et al., 2017; Wood et al., 2017; Ramos-Prats et al., 2022). Chemical transmission can occur in a variety of plasticity modalities (facilitating to depressing) and can occur between glutamatergic and GABAergic neurons (Galarreta and Hestrin, 1998; Jackman and Regehr, 2017; Martinetti et al., 2022), thus increasing the range of sensory processing and coding (Fuhrmann et al., 2002; Tong et al., 2020).
Inhibitory interneurons are vastly diverse in morphology, electrophysiological properties, expression of molecular markers, and connectivity patterns (Ascoli et al., 2008; Cadwell et al., 2016; Tremblay, 2016; Gouwens et al., 2020; Yuste et al., 2020; Staiger and Petersen, 2021). Single-cell RNA sequencing and Patch-seq analysis have classified GABAergic neurons into six major subpopulations: parvalbumin- (PV), somatostatin- (SST), vasoactive intestinal polypeptide- (VIP), Lamp5-, Sncg-, and Serpinf-1-expressing neurons (Tasic et al., 2016, 2018), which contain a considerable number of different cell types (Scala et al., 2019; Gouwens et al., 2020). The interplay of these inhibitory neurons, their recruitment via intracortical, long-range, or neuromodulatory inputs, and their subsequent targeting of excitatory cells (Silberberg and Markram, 2007; Karnani et al., 2014; Wall et al., 2016; Urban-Ciecko et al., 2018; Gasselin et al., 2021; Hafner et al., 2021; Naskar et al., 2021) is key to maintaining an appropriate moment-to-moment excitation–inhibition balance of the neocortex (Isaacson and Scanziani, 2011).
An inhibitory microcircuit, which has been well-described in the sensory and prefrontal cortices of mice, mostly in layer (L) 2/3, is the VIP-to-SST motif (Lee et al., 2013; Pfeffer et al., 2013; Pi et al., 2013; Walker et al., 2016; Garcia-Junco-Clemente et al., 2017). This motif has been attributed to a downstream suppression of SST-to-excitatory cell inhibition (Gentet et al., 2012; Muñoz et al., 2017; Yu et al., 2019), leading to the VIP-to-SST motif being dubbed the “disinhibitory” motif (for reviews, see Roelfsema and Holtmaat, 2018; Kullander and Topolnik, 2021). Notably, all SST cells shown to receive VIP cell inhibition thus far are within L2/3 and are supposed to be Martinotti cells (MCs), which are the predominant morphological subtype of SST cells. MCs are characterized by a L1-projecting axonal arbor, with dense horizontal ramifications (Kawaguchi and Kubota, 1996; Wang et al., 2004; Ma et al., 2006).
However, SST cells have been shown to exhibit varied axonal tree morphologies, depending on the cortical layer and the sensory cortex in which the soma is housed. For instance, L4 of the primary somatosensory cortex (S1) is primarily populated by a different morphological subclass of SST cells, known as non-Martinotti cells (nMCs; Ma et al., 2006; Xu et al., 2013; Scala et al., 2019). nMCs differ from MCs by axonal prevalence in L4 and dense innervation of the home barrel. In contrast, SST cells in L4 of the primary visual cortex (V1) were shown to be MCs (Scala et al., 2019). MCs are also the predominant type of SST cell found in L5 of S1 (Wang et al., 2004; Ma et al., 2006; Yavorska and Wehr, 2016). Indeed, within S1, the differential axonal architecture of MCs and nMCs across L2–5 has been implicated in their complementary contribution to network dynamics (Muñoz et al., 2017; Nigro, 2018; Naka et al., 2019; Hage et al., 2022).
While SST cells occur more often in deeper layers, a large proportion of VIP cells is localized in L2/3 (Prönneke et al., 2015; Almási et al., 2019). The axon is laterally restricted to a single column while spanning nearly all cortical layers, with a majority of it densely occupying L2/3 or L5, although a proportion of synaptic boutons are attributed to L4 as well (Prönneke et al., 2015; Gouwens et al., 2020). While the VIP-to-SST motif has been well studied within L2/3 across cortical areas, it has not been characterized in a translaminar setting, until very recently (Jiang et al., 2015; Campagnola et al., 2022). Moreover, the effect of VIP inputs on different postsynaptic subtypes of neurons belonging to the same molecular class is still unknown.
Using single and paired whole-cell patch-clamp recordings in a novel reporter mouse model, we studied the L2/3 VIP-to-L4 SST microcircuit in the S1 and V1 cortex. Besides revealing VIP-to-nMC and VIP-to-MC motifs and studying their unitary properties, we also examined short-term plasticity (STP) across a broad frequency spectrum. We identified abundant targeting of VIP cells onto SST cells, with comparable unitary connectivity properties, regardless of the morphological subtype. We also uncovered diverse STP types within and across VIP-to-nMC and VIP-to-MC motifs. While short-term facilitation prevailed overall, we observed short-term depression exclusively in S1, in the VIP-to-nMC motif. Altogether, these results provide insights into how cortical microcircuits differ across areas and suggest that early sensory processing might be modulated by a feedback translaminar disinhibitory circuit motif.
Materials and Methods
Animals
All animals used for breeding the experimental mouse lines except the VIP-EGFP line were obtained from The Jackson Laboratory. The VIP-EGFP line was obtained from the University of California Davis Mutant Mouse Regional Resource Centers. All animals were housed in standard cage conditions, with access to food and water ad libitum. Mice were housed on a 12 h light/dark cycle. VIP-ires-FLP (He et al., 2016; Viptm2.1(flpo)Zjh/J) mice were crossed with homozygous Ai193 (B6;129S6-Igs7tm193(CAG-EGFP,CAG-tdTomato)Tasic/J) mice, and the resulting VIP-FLP::Ai193 progeny were crossed with SST-Cre (Taniguchi et al., 2011; Ssttm2.1(cre)Zjh/J) mice to obtain a triple transgenic VIP-FLP::SST-Cre::Ai193 mouse line. Using these lines, VIP-expressing and SST-expressing cells could be identified by their specific fluorescent label (VIP cells: tdTomato fluorescence, SST cells: EGFP fluorescence). A subset of experiments was performed with SST-Cre::Ai9::VIP-EGFP animals. For this mouse line, SST-Cre (Ssttm2.1(cre)Zjh/J) was crossed with Ai9 reporter mouse (Madisen et al., 2010; B6.Cg-Gt(ROSA)26Sortm9(CAG-tdTomato)Tasic/J) and the VIP-EGFP transgenic line (Tg(Vip-EGFP)37Gsat/Mmcd). This mouse line resulted in VIP cells expressing EGFP and SST cells expressing tdTomato fluorescence. All animal experiments were performed in accordance with the German laws on animal research. The stock numbers for all utilized mouse lines are as follows:
VIP-ires-FLP: Jax Strain #028578
SST-ires-Cre: Jax Strain #013044
Ai193 reporter: Jax Strain #034111
Ai9 reporter: Jax Strain #:007909
VIP-EGFP: 031009-UCD-HEMI-F
Slice preparation
To obtain acute brain slices of somatosensory and visual cortices, mice of either sex (postnatal days 28–50; median, 34) were deeply anesthetized with isoflurane and decapitated. The brain was extracted, the hemispheres were separated, and the brain was kept at 4°C in an oxygenated (95% O2, 5% CO2) cutting solution containing the following (in mmol): 75 sucrose, 87 NaCl, 2.5 KCl, 0.5 CaCl2, 7.0 MgCl2, 26 NaHCO3, 1.25 NaH2PO4, and 10 glucose (pH 7.4). Thalamocortical slices were obtained for S1 (Porter et al., 2001) while coronal slices were preferred for V1 to ensure minimal truncation of axonal and dendritic processes. Slices were 300 µm thick and were cut using a vibratome (Leica VT1200S). Slices were then incubated in artificial cerebrospinal fluid (ACSF) containing (in mmol): 125 NaCl, 2.5 KCl, 2 CaCl2, 1 MgCl2, 26 NaHCO3, 1.25 NaH2PO4, and 25 glucose (pH 7.4, equilibrated with 95% O2, 5% CO2) at 34°C for 30–45 min and then allowed to cool to room temperature.
Electrophysiology and data acquisition
Slices were transferred to a recording chamber which was continuously perfused with ACSF at the rate of 2 ml/min. The recording chamber was fitted with an upright microscope (Axio Examiner A1, Zeiss) with a low magnification 2.5× objective (EC Plan-Neofluar 2.5×; NA, 0.075; working space, 9.5 mm; Zeiss) to enable slice visualization under brightfield illumination, and 40× water-immersion objective (40×/0.75 W; NA, 0.80; working distance, 3.3 mm; Olympus) to enable cell visualization. Epifluorescence was enabled with a mercury arc lamp (HXP 120 C, Zeiss), filtered by a dichroic mirror. Filamented borosilicate capillaries (Science Products) of 5–8 MΩ resistance were filled with standard potassium gluconate solution containing (in mmol) 135 K-gluconate, 5 KCl, 0.5 EGTA, 10 HEPES, 4 Mg-ATP, 0.3 Na-GTP, and 10 Na-phosphocreatine phosphate (pH 7.4), and whole-cell patch-clamp electrophysiology was performed in VIP cells and a subset of SST cells under current-clamp (CC) conditions. To assess synaptic connectivity, SST cells in paired recordings were patched with cesium methylsulfonate intracellular solution containing (in mmol) 135 CsMeSO4, 5 CsCl, 2 MgCl2, 0.5 EGTA, 10 HEPES, 4 Mg-ATP, 0.3 Na-GTP, and 10 Na-phosphocreatine phosphate (pH 7.4) and held at 0 mV under voltage-clamp (VC) conditions. Both intracellular solutions contained 0.3–0.5% biocytin to ensure post hoc morphology visualization and analysis. Care was taken to ensure that the somas of the VIP and SST cells were as vertically aligned as possible, keeping with the columnar orientation of the VIP axon. Membrane potentials were not corrected for liquid junction potentials.
Membrane potentials and currents were recorded using an SEC 05 L amplifier (NPI Electronics) in discontinuous current-clamp (CC) mode or voltage-clamp (VC) mode, with a switching frequency of 50 kHz. The signals were filtered at 3 kHz and digitized at 10–25 kHz using a CED Power1401 interface (CED Limited). Electrode capacitance was compensated immediately after break-in using visual control. Compensation was performed for current-clamp recordings as necessary throughout the experiment. For voltage-clamp recordings, access resistance was calculated post hoc using a −5 mV prepulse during paired recording data collection and was found to be 23.97 ± 10.53 MΩ. Passive and active membrane properties were measured immediately after break-in, using rectangular hyperpolarizing and depolarizing current pulses of 1 s, in steps of 10 pA. All characterization was carried out while the cells were held at −65 mV.
To investigate synaptic connectivity, SST cells patched with cesium methylsulfonate intracellular solution were held at 0 mV in VC. This ensured that the driving force of chloride currents was maximized, to facilitate the detection of inhibitory postsynaptic currents (IPSCs). Recordings were collected and analyzed using Signal 5 software (CED Limited).
Analysis of electrophysiological data
All electrophysiological data were analyzed using custom-made scripts for Signal 5 software. All passive properties were analyzed using averages of 20 responses to 1 s–10 pA rectangular current pulse. The membrane time constant (τ) was determined by fitting an exponential [f(x) = a0e−x/a1 + a2, where a0 is the amplitude (mV), a1 is τ (s), and a2 is offset (mV)] to the average membrane response. Input resistance (Ri) was calculated according to Ohm's law, using the maximum change in amplitude of the membrane potential response. Sag index was calculated based on the following formula: ((1/steady state Ri) − (1/highest deflection Ri)/1/steady state Ri) × 100 (denoted in percentage). Membrane capacity was calculated as C = τ/Ri. Firing threshold, spike amplitude, spike width, spike time to peak, and slope were all determined from action potentials evoked at rheobase. The firing threshold was calculated as the membrane potential at the point in time where the instantaneous slope was <10% of the average slope of the upstroke, in a 1 ms window preceding the spike peak. The amplitude of the afterhyperpolarization (AHP) was measured by calculating the difference in voltage from the firing threshold to the maximum deflection of repolarization. The time to peak of the AHP was determined as the time point between the repolarization of the action potential crossing the firing threshold and the maximum amplitude of the AHP. Frequency plots were generated based on protocols established in Prönneke et al. (2015). Briefly, we calculated the interspike intervals (ISIs) for each neuron, and the reciprocal of ISIs was determined to be the instantaneous firing frequency (IFF). The IFF for each neuron was plotted as a percentage in a cumulative manner. The plots were binned in steps of 10 Hz, from 10 to 400 Hz. A direct comparison of frequency plots between S1 and V1 was done by plotting the noncumulative percentile distribution of IFF.
Paired recordings and analysis
Paired recordings were performed by holding the presynaptic VIP cells in CC at −65 mV, while postsynaptic SST cells were held in VC at 0 mV. Unitary connectivity properties were assessed by injecting suprathreshold current pulses for 5 ms (≥20 iterations, 10 s interstimulus interval) into the VIP cell, to ensure single action potential firing. All measures of synaptic connectivity were carried out on averages of 20–30 individual sweeps. Only sweeps with a detectable IPSC (>baseline + 3 × SD) were selected for further analysis. The following unitary connectivity properties were analyzed: amplitude (difference from baseline to peak), latency (time from spike peak to onset of response), normalized slope of the ascending phase (ms−1), 10–90% rise time (time between the 10 and 90% of the IPSC amplitude), and synaptic success rate. The normalized slope of the IPSCs was used to avoid the influence of response amplitudes on slope determination. Short-term synaptic plasticity was assessed by injecting trains of suprathreshold currents (5 ms) into the VIP cells to evoke five action potentials at 1, 10, and 50 Hz frequencies (≥20 iterations, 10 s intertrain interval). Of the 52 connected pairs from which short-term plasticity data were recorded (25 in S1 and 27 in V1), 48 pairs (∼92%) were recorded in the order of 50, 1, and 10 Hz, 3 pairs followed a different order, although all three frequencies were still recorded, and 1 pair in V1 was only recorded with 50 Hz recording data. Amplitude ratios for each evoked IPSC were derived relative to the first IPSC (nth IPSC/first IPSC). To measure amplitudes of overlapping IPSCs, an exponential was fitted to the decaying phase of the preceding IPSC and extrapolated to the baseline level. Amplitude was then measured as the difference between the peak of the IPSC and the fit at that point in time.
Connections were determined to be short-term facilitating, short-term depressing, or stable based on a sliding window method and statistical testing. A rank sum test was performed to test for statistical significance between the 50 Hz recording data from S1 and V1. Since no significant difference was observed between corresponding IPSC response ratios in S1 and V1, 50 Hz data between both cortices were pooled to determine the upper and lower limits of stable responses. We first calculated the mean of the IPSC ratio 1 to ratio 5 for each connected pair. We then sorted the calculated mean responses and the IPSC ratio 5 from highest to lowest and selected data where the scatter of both these values was <10%. IPSC ratio 5 values that fulfilled this criterion were tested against a hypothetical stable response of 1. The selected values were sequentially replaced by higher and lower values to determine the upper and lower limits, respectively. This method of sliding window and statistical testing against the value of 1 was continued until a statistically significant result was reached. We performed a parametric t test in the case of normally distributed data and nonparametric Mann–Whitney U test in the case of non-normal data distribution. Mean ± 3 × SD of the last set of values that were not significantly different from 1 was used to calculate the upper and lower limits of the stable response data set. For our data, the upper and lower limits of stable responses were calculated to be 1.2 and 0.8, respectively. All connected pairs with a fifth IPSC ratio ≥1.2 were classified as short-term facilitating, while responses ≤0.8 were categorized as short-term depressing. All connected pairs with a fifth IPSC ratio between 0.8 and 1.2 were classified as stable responses. All frequency data were recorded for at least 20 trials (at most 40). All sweeps were aligned to the presynaptic VIP spike before averaging, to prevent disturbances of the IPSC waveform due to jitter.
Staining
After all electrophysiological measures were recorded, the pipette was detached from the membrane by slowly retracting the patch pipette while simultaneously moving it back and forth on the y-axis. This ensured reclosure of the membrane after recordings, which aided the optimum morphological recovery, with minimal biocytin background noise. Cells were stained with the appropriate antibodies to visualize GFP- and RFP-expressing cells. Staining of biocytin-filled cells was done as described previously (Staiger, 2004; Staiger et al., 2015). In brief, slices were fixed in 4% paraformaldehyde solution with 15% picric acid at 4°C for at least 24 h. Slices were then washed repeatedly with phosphate-buffered saline (PBS; 0.1 M, pH 7.4) solution until the picric acid residues were visibly reduced and then washed twice for 15 min each with PB, TB, TBS, and TBST buffers. Subsequently, slices were blocked in bovine serum albumin (BSA)-containing solution (0.25% BSA, 10% normal donkey serum, and 0.5% Triton X-100, pH 7.6 in PBS) for 90 min and later incubated in primary antibodies (goat anti-GFP, 1:2,000 Abcam; rabbit anti-RFP, 1:500 Rockland) for 48–72 h at 4°C. Slices were then rinsed in TBST (four times, 15 min each) and incubated in secondary antibodies overnight at 4°C. The secondary antibodies used were donkey anti-goat AF488 and donkey anti-rabbit AF546 (both 1:500, Molecular Probes) and streptavidin-conjugated AF633 (1:300). After rinsing in TBST buffer and staining with DAPI (4′,6-diamidino-2-phenylindole, 1:5,000, Molecular Probes), slices were mounted in Aqua-Poly/Mount solution (Polysciences) and covered with a coverslip (24 × 50 mm, 0.08–0.12 mm thick, Menzel).
FISH and data analysis
Sections of VIP-FLP::SST-Cre::Ai193 mice were stained with RNA probes against Vip and Sst. Probes were generated using plasmids (vector backbone: pGEM-T Easy, Promega) containing cDNA inserts of VIP and SST genes. Each section was 40 µm thick, and we analysed Vip probe in 11 sections and Sst probe in 12 sections from two animals. Primers for the inserts were as follows: Vip (nested 367 bp, Allen Brain Atlas Riboprobe ID: RP_080916_03_C04): forward primer (FP) nested, CTGTTCTCTCAGTCGCTGGC; reverse primer (RP) nested, GCTTTCTGAGGCGGGTGTAG; Sst (514 bp, Allen Brain Atlas Riboprobe ID: RP_081204_01_A03): FP, ACGCTACCGAAGCCGTC; RP, GGGGCCAGGAGTTAAGGA. The staining procedure was carried out as described previously (Hafner et al., 2021). The tissue was amplified for RFP and GFP signal using the immunohistochemistry procedure described above, without utilizing Triton X-100. Vip probe-hybridized sections were imaged using an upright epifluorescence microscope (Axio Observer, Zeiss) while Sst probe-hybridized sections were imaged using a confocal inverted microscope (LSM 880, Zeiss). We utilized a 25× water-immersion objective for both conditions. Brain regions were identified based on cytoarchitectonic landmarks visible under DAPI-generated nuclear staining. Somas expressing RFP or GFP fluorescence, as well as colocalization with Vip or Sst probes, were manually counted using Neurolucida software (MBF Bioscience). Layers were identified based on nuclear density, as well as previously established relative thickness criteria (Prönneke et al., 2015). Cell counts were exported to Neurolucida Explorer software (MBF Bioscience) and then subsequently to MS Excel for data extraction.
Imaging and reconstruction of biocytin-filled cells
Overview images of the slices from pia to white matter were obtained with an upright epifluorescence microscope (Axio Observer, Zeiss), with 10× objective controlled by ZEN software (Zeiss). Slices were imaged for DAPI and AF 633 to (1) visualize the layer localization of stained cells for post hoc confirmation and (2) perform a quality control check for morphological recovery of patched neurons. If dendrites and axon were reasonably well-recovered, the slice was subsequently imaged as a stack with a confocal inverted microscope (LSM 880, Zeiss) with either a 40× water- or 63× water- or oil-immersion objective controlled by ZEN Black software (Zeiss). A subset of images was also obtained using the Fast Airyscan mode with the LSM 880 for a higher signal-to-noise ratio (Huff, 2015). Images were stitched, and maximum-intensity projections were obtained using ZEN Black software for image representation purposes. Stitched images were then converted into jpx format for morphological reconstruction. Single or pairs of neurons were reconstructed using Neurolucida software. Only well-preserved neurons with intense staining of dendrites and axons as well as no truncations close to the soma were selected for further reconstruction. Dendrites were differentiated from axons by their fine structures, diameter, and branching pattern, as described previously (Prönneke et al., 2015). Keeping with Allen Brain Atlas conventions, L5 of S1 was subdivided into L5a and L5b using cytoarchitectonic markers during reconstruction, but neurite features were grouped together for comparison with V1 L5.
Analysis of morphological data
Parameters of the reconstructed neurons were extracted using NeuroExplorer software. Data were not corrected for tissue shrinkage. Superimposition of dendrites and axons was done by aligning the processes on a common cortical layer border delineation [average of layers determined using DAPI staining (Prönneke et al., 2015)]. Binary images were generated by superimposing neurites from multiple neurons, and a Gaussian filter was applied with a comparable radial sigma (20 at 300 dpi) for all structures. A look-up table ranging from cold (blue and green for white to light gray) to warm colors (yellow and red for dark gray to black) was applied to the resulting image. This was then superimposed onto the original grayscale image and merged with the original black and white image to get heat maps visualizing areas of the highest dendritic and axonal densities.
Statistical tests
Statistical tests and graphs were generated using either SigmaPlot Software (Grafiti, 14.0) or Prism program (Graphpad Software, 8.0). We tested all data for normality using the Shapiro–Wilk test. If the normality test was passed, the data were subjected to a two-tailed Student's t test; otherwise, we used a Mann–Whitney U test. To avoid familywise errors, p-values were adjusted using Bonferroni’s correction. Pearson’s product–moment correlation tests were used to identify any correlations between values. We also used ANOVA on rank analysis to test for statistical differences between IPSC ratios. Fischer's exact test was used to analyze the statistical significance of connection probabilities. Values for morphological data are given as mean ± SD if normally distributed, or otherwise as median. Values for electrophysiological data are given as mean ± SEM if normally distributed, or otherwise as median. All values for FISH are given as mean ± SD. Adobe Illustrator and InDesign software (Adobe) were used for figure generation.
Results
Layer 4 SST cells present with unique morphological features in S1 and V1
We utilized a novel triple transgenic VIP-FLP::SST-Cre::Ai193 mouse line (Ben-Simon et al., 2024) that ensured effective targeting of cells of interest through Cre::FLP recombinase (Figs. 1A,B, 2). This mouse line allowed us to obtain single and dual recordings of VIP and SST cells in the desired cortical areas and layers, recover the morphology of recorded cells (Fig. 1C,E), and reconstruct their morphology (Fig. 1D,F).
A Cre- and FLP-dependent dual transgenic Ai193 reporter mouse line permits targeted simultaneous whole-cell recordings from L2/3 VIP and L4 SST neurons. A, B, Coronal sections through the primary somatosensory cortex (S1, A) and primary visual cortex (V1, B) of VIP-FLP::SST-Cre::Ai193 mice, with immunohistochemically amplified dual recombinase-mediated fluorescence in VIP and SST neurons. VIP neurons are labeled with RFP, SST neurons with GFP, and blue is DAPI (used to identify cortical areas and layers). C, E, Biocytin labeling of VIP and SST neurons through paired whole-cell patch-clamp recordings. Depicted here are two synaptically connected pairs of cells in S1 (C) and V1 (E), after post hoc morphologically recovery. Maximum-intensity projection is shown in white (pseudocolored), and layer borders are depicted with dashed horizontal lines. D, F, Neurolucida reconstructions of synaptically connected pairs seen in (C, E). Somatodendritic compartment of VIP is depicted in red, while axon is displayed in orange. S1 SST neuron soma and dendrites are displayed in deep blue (D), and axon is depicted in light blue, while V1 SST neuron is depicted with soma and dendrites in deep green and axon in light green (F). Scale bar, 200 µm.
FISH analysis of a novel VIP-FLP::SST-Cre::Ai193 dual transgenic reporter mouse line. A–A″, B–B″, 40-µm-thick sections from S1 and V1 were subjected to fluorescent in situ hybridization (FISH) using probes hybridizing with RNA corresponding to the Vip gene in S1 and V1. VIP neurons expressing RFP fluorescence in L2/3 are seen on the left, VIP probe in the middle, and merged images on the right. Cell bodies colocalizing transgenic and mRNA probe fluorescence are marked by arrowheads. C, D, Results of population analysis data for FISH analysis of VIP cells (2 animals, 11 sections) are plotted in a summarized manner as cells per cubic millimeter of the cortex. Red bars depict cells that displayed only RFP fluorescence, gray bars depict cells only displaying Vip probe fluorescence, and yellow bars depict cells with colocalization of RFP and Vip probe fluorescence. Percentages above the colocalization bar depict the fraction of RFP-expressing cells colocalized with Vip probe (error bars are SD). E–E″, F–F″, Same as A–A″ but for SST neurons in L4 of S1 and V1, respectively. One to two cell bodies expressing GFP transgenic fluorescence but not colocalizing with the Sst mRNA probe are seen only in V1 images and are marked by arrows. Scale bar, 20 µm. G, H, Results of population analysis data for FISH analysis of SST cells (2 animals, 12 sections) are plotted in a summarized manner as cells per cubic millimeter of the cortex. Green bars depict cells that displayed only GFP fluorescence, gray bars depict cells only displaying Sst probe fluorescence, and yellow bars depict cells with colocalization of GFP and Sst probe fluorescence. Percentages above the colocalization bar depict the fraction of GFP-expressing cells colocalized with Sst probe (error bars are SD).
Whole-cell patch-clamp electrophysiology, combined with biocytin-mediated morphological recovery, was used to characterize the electrophysiological and morphological features of SST cells. After passing stringent quality control criteria (see Materials and Methods), we reconstructed the morphology of 12 and 10 SST cells in S1 and V1, respectively (Fig. 3). Neurite distribution and densities of individual dendrites and axons are shown superimposed, and as heat maps, to visualize density patterns across cortices (Fig. 4, see Materials and Methods).
Morphological reconstructions of L4 SST neurons of primary somatosensory and visual cortices. A, Reconstructions of L4 SST neurons are shown here in a schematized primary somatosensory cortex (S1). L4 barrel outlines are depicted in dotted outlines, while dashed horizontal lines represent the layer borders. Somatodendritic compartment is depicted in deep blue, while axonal arbor is depicted in light blue. Neurons are organized from left to right as most nMC-like to least nMC-like. The neuron on the extreme right is an MC neuron. It can be appreciated that although the penultimate neuron on the right (i.e., least nMC-like) displays numerous axonal branches in L2/3, it exhibits minimal L1 axon. B, Same as A, but L4 SST neurons are shown here in a schematized primary visual cortex (V1). Somatodendritic compartment is depicted in deep green, while axonal arbor is depicted in light green. Dashed horizontal lines represent the layer borders. Most neurons depicted fall under the classification of MCs, which can be identified by their L1-reaching axon and dense horizontal ramifications upon reaching this layer. MCs are organized from left to right on a scale of most MC-like to least MC-like. The neuron on the extreme right is a nMC neuron, with the soma localized to the L3/L4 border. Cortical layers are indicated to the left of both images. Scale bar, 200 µm. C, C′, High-resolution confocal image of an S1 L4 nMC on the left and corresponding reconstruction on the right. Axon can be seen originating directly from the soma. D, D′, Similar to C and C′, but depicting a V1 L4 MC. Dendrite-originating axon can be seen in light green. Scale bar: D′ (for C, D′) 20 µm.
Somatodendritic and axonal distribution of nMCs and MCs across the cortical depth. A, Somatodendritic compartments derived from reconstructions of S1 nMCs and V1 MCs are superimposed, and a heat map is applied on the resulting structure. Areas surrounding the somas show the highest dendritic density, as can be seen from the heatmap. B, Dendritic length distribution of S1 nMCs and V1 MCs across the cortical depth. Neither cell type extended its dendrites to L1 or L6. Dendritic lengths in L2/3, L4, and L5 were found to be comparable between nMCs and MCs. Results are given as mean ± SD. C, Axonal arbors of S1 nMCs and V1 MCs are superimposed, and a heat map is applied to the resulting structure. nMCs exhibit the largest density of neurites within L4 and, to a small extent, L3. On the other hand, MCs show strong neurite density in L1, L2/3, and L4. In both A and C, the positions of somas are vertically aligned and depicted by open black circles. The density of the neurites is visualized by heat maps ranging from red (most dense) to blue (least dense). Scale bar, 200 µm. D, Axonal length distribution of S1 nMCs and V1 MCs across the cortical depth. The axonal length in L1, L4, and L6 was significantly different between nMCs and MCs, with MCs displaying a denser axonal compartment in L1, while nMCs exhibited higher axonal arbor lengths in L4 and L6, underscoring the differences in axon distribution that classically highlight the two SST cell types. Results are given as mean ± SD. E, Comparative boxplots show total dendritic and axonal length as well as the number of synaptic boutons found per 100 µm of axons of nMCs and MCs. No significant difference was found in the total dendritic length between both cell types. Despite strong differences in the density and distribution of axonal arbor, the total axonal length and bouton density were invariant across nMCs and MCs. F, A representative confocal image showing the axonal branches of a V1 MC (top) and corresponding reconstruction (bottom) can be seen. Two boutons are depicted by the white arrowheads, while the corresponding marker is shown by the black arrowheads. The red arrowhead indicates an axonal segment from the connected L2/3 VIP cell, which is not seen in the SST reconstruction. Scale bars, 20 µm. G, Same as B and D, but depicting the synaptic bouton distribution of S1 nMCs and V1 MCs across the cortical depth. The bouton numbers in L1 and L6 were significantly different between nMCs and MCs, with MCs displaying a higher bouton count in L1, while nMCs exhibited slightly higher bouton numbers in L6. Results are given as mean ± SD.
SST cells were found to have distinct morphological features within L4 of S1 and V1. Most (11/12) L4 SST cells of S1 that we observed could be classified into the morphological category of non-Martinotti cells (nMCs; Fig. 3A) which have been previously described in L4 S1 of rodents (Wang et al., 2004; Ma et al., 2006; Yavorska and Wehr, 2016). Reconstructed cells presented with typical nMC-like features such as a predominant restriction of dendritic arbor to L4 (80.3%, dendritic length in L4 median 1,992.2 µm vs non-L4 dendritic length median 256.5 µm, p = 0.0002; Figs. 3A, 4A,B], often not extending beyond the home barrel, as well as an abundance of axonal arborization within L4 (62%, axonal length in L4 median 16,947.1 µm vs non-L4 axonal length median 4,617.9 µm, p = 0.0336; Figs. 3A, 4C,D). Crucially, nMCs also presented with either a complete absence or very minimal axonal presence within L1 (0.68%, axonal length in L1 median 0 µm vs non-L1 axonal length median 24,222.1 µm, p < 0.0001; Figs. 3A, 4C,D). Interestingly, nMCs also displayed extensive vertically oriented axonal branches (between three and five for each nMC) projecting to infragranular layers, with very few branch sites. SST cell somas in our sample were often localized within the barrel region. We only found one SST soma in S1 that had a septal localization (not reconstructed).
In stark contrast, L4 SST cells of V1 largely (9/10) presented with features typical for MCs (Fig. 3B), wherein a larger proportion of the axon is housed within L1 in comparison with L4 nMCs (17.7%; axonal length in L1, 3,780.71 ± 2,168.2 µm; Figs. 3B, 4C,D). Dendrites were not restricted to L4 (60.1%; dendritic length in L4 1,665.7 ± 649.8 µm vs non-L4 dendritic length 1,102.1 ± 898.2 µm, p = 0.1698; Figs. 3B, 4A,B), but were extended to L2/3 (30%; dendritic length in L2/3, 872.9 ± 778.7 µm; Figs. 3B, 4A,B) and sometimes even L5 (6.4%; dendritic length in L5, 227.9 ± 163.1; Figs. 3B, 4A,B).
When morphological features of SST cells were compared between the two cortices, axonal length in L1 was significantly higher in V1 SST cells in comparison with S1 (S1 median, 0; V1 median, 2,931.9; p = 0.0050; Fig. 4D) while the opposite was true for axonal length in L4 (S1, 15,524.3 ± 6,482.4; V1, 6,816.2 ± 3,957.2; p = 0.0091; Fig. 4D) and L6 (S1 median, 534.1; V1 median, 0; p = 0.0095; Fig. 4D). However, overall axonal length (S1, 25,026.69 ± 6,721.9; V1, 21,271.33 ± 3,708.9; p = 0.1719; Fig. 4E), and consequently the number of axonal boutons (S1, 10,503.45 ± 3,475.5; V1, 9,681.7 ± 3,014.2; p = 0.6030; Fig. 4E), was not significantly different. The density of boutons per 100 µm axon was also comparable between the two cortices (S1, 41.48 ± 7.2; V1, 45.6 ± 12.1; p = 0.3841; Fig. 4E). However, the number of axonal boutons in L1 was significantly higher in V1 SST cells in comparison with S1 (S1 median, 0; V1 median, 1,411; p = 0.0003; Fig. 4G) while the opposite was true for number of boutons in L6 (S1 median, 255; V1 median, 0; p = 0.01; Fig. 4G). An example of a high-resolution image of MC synaptic boutons and corresponding reconstruction can be seen in Figure 4F. While V1 MCs showed a preference for the axon to originate from the dendrite (77%, 7/9; Fig. 3D,D′), S1 nMCs showed an equal tendency for the axon to originate either from the dendrite (55%, 6/11) or directly from the soma (45%, 5/11; Fig. 3C,C′).
Some (S1, 7/12; V1, 3/10) of the SST neurons described above were synaptically connected to an L2/3 VIP cell, which were also reconstructed post hoc and are depicted in Figure 5.
Morphology of synaptically connected VIP and SST pairs in S1 and V1 of mouse cortex. A, Reconstructions of L2/3 VIP cells synaptically connected to L4 SST cells in S1 are depicted here. VIP soma and dendrites are depicted in red, while the axon is shown in orange. SST cell somatodendritic compartment is shown in deep blue, while axon is depicted in light blue. Layer borders and barrels in L4 are depicted in dotted lines, while the borders of the cortex are depicted in solid lines. All SST cells visualized are identified to be nMCs, with the exception of the SST cell on the extreme right, which is classified as an MC. B, Same as A but for connected L2/3 VIP and L4 SST cell pairs in V1. While the VIP cells are depicted with the same color scheme in A, SST cells are shown with somatodendritic compartment in deep green and axon in light green. All SST neurons are identified as MCs except the cell on the extreme right which was identified as an nMC. Scale bar: A, B, 200 µm.
The dendritic length of V1 MCs was not significantly different from S1 nMCs (S1, 2,521.6 ± 588.3 µm; V1, 2,767.9 ± 372.4 µm; p = 0.3149; Fig. 4E). While dendrites of SST cells from either cortical area did not extend to L1 or L6, the dendritic length of L4 SST cells in V1 and S1 was comparable (S1, 2,026.5 ± 631.1; V1, 1,665.7 ± 649.8; p = 0.2253; Fig. 4B). Interestingly, in all but two S1 nMCs, all dendrites projected to one neighboring barrel but not to both, underscoring a tendency toward low horizontal dendritic spread and preference for home barrel localization.
In contrast to axonal and dendritic features, somatic features of SST cells were invariant across both cortices. Most SST cells showed a multipolar soma (roundness S1 0.5 ± 0.1. vs roundness V1 0.6 ± 0.1, p = 0.0755). The surface area of the soma (S1 median 764.2 µm2 vs V1 median 417.1 µm2, p = 0.1506) was also comparable across cortices. The number of dendrites arising from the soma was slightly more variable in S1 nMCs than in V1 MCs (S1, 3–8; V1, 5–6). A detailed account of the morphological properties can be found in Extended Data Tables 4-1 and 4-2.
Table 4-1
Comparison of somato-dendritic morphological properties of L4 SST cells in S1 and V1 SD = Standard deviation, nMC = non-Martinotti cell, MC = Martinotti cell. Feret max and feret min refer to the largest and smallest dimensions of the soma contour and are considered a proxy for the long and the short diameter of the cell body. If data passed a normality test (Shapiro-Wilk), t-test was performed, otherwise Mann Whitney U test was performed. P values were corrected post-hoc for alpha inflation error by Bonferroni method. Download Table 4-1, DOCX file.
Table 4-2
Comparison of axonal properties of L4 SST cells in S1 and V1 Statistically significant differences are marked in grey. If data passed a normality test (Shapiro-Wilk), t-test was performed, otherwise Mann Whitney U test was performed. P values were corrected post-hoc for alpha inflation error by Bonferroni method. Download Table 4-2, DOCX file.
In summary, L4 SST cells of S1 and V1 were significantly different in the layer-specific distribution of their axon. While non-Martinotti cells of S1 showed a strong preference to ramify within the home layer L4, Martinotti cells of V1 largely targeted their axon to L2/3 and L1. Due to the differential morphological characteristics exhibited, we will refer to L4 SST cells of S1 as nMCs and those of V1 as MCs henceforth, for the entirety of this manuscript.
Electrophysiological properties of SST cells are heterogeneous within and across cortices
We then determined the intrinsic electrophysiological properties of nMCs and MCs from L4 of S1 and V1 (nMCs n = 36, MCs n = 47, Fig. 6). Since we observed strong morphological differences depending on which cortical area the SST cell is housed in, we asked whether differences could also be found at the level of electrophysiological characteristics.
Electrophysiological properties of L4 SST neurons in S1 and V1 cortices. A, Example action potential traces of nMC neurons from S1 where individual traces are depicted as i–iii. Rheobase recordings are shown in the top traces (values in picoampere), afterhyperpolarizations are shown as inserts within dotted boxes, and hyperpolarizing traces at −100 pA are shown directly below. The bottom traces depict the firing pattern of the neuron depicted at 2× or 3× rheobase (values in picoampere) represented above the firing pattern trace. All recordings were taken while the cell was held in a current clamp (CC) at −65 mV. B, Distribution of firing pattern types among nMC neurons of S1. More than three-quarters (∼78%) of cells displayed adapting firing pattern, while the rest (∼22%) exhibited stuttering firing patterns. C, Same as A but for MC neurons of V1. D, Distribution of firing pattern types among MC neurons of V1. Both stuttering and adapting firing patterns were distributed in comparative proportions across the population. E, F, Box plots depicting passive (E) and active (F) electrophysiological properties of L4 SST neurons from S1 (n = 36) and V1 (n = 47). Significant differences were found only in the τ values.
Single action potentials displayed a variety of afterhyperpolarization (AHP) waveforms. In S1, nMCs exhibited a fast AHP (fAHP) during rheobase recordings, with either a completely absent or very small slower medium AHP (mAHP) component (50%, 18/36, Fig. 6Ai). A subset of cells from S1 also exhibited a biphasic AHP waveform (Ma et al., 2006) with pronounced fAHP and mAHP (50%, 18/36; Fig. 6Aii,Aiii). MCs of V1 showed a similar sequence of fAHP and mAHP (60%, 28/47; Fig. 6Ci,Cii), with some cells only showing fAHP features (40%, 19/47, Fig. 6Ciii).
In terms of firing patterns, SST cells of both cortical areas present with a variety of characteristics. For instance, 78% of recorded cells in S1 nMCs (28/36; Fig. 6Ai,Aii,B) and 47% in V1 MCs (22/47; Fig. 6Ci,D) showed a continuous adapting form of firing, when observed at double or triple rheobase stimulation. Close to a quarter of S1 nMCs (22%, 8/36; Fig. 6Aiii,B) and more than half the V1 MCs (53%, 25/47; Fig. 6Cii,Ciii,D) showed stuttering (i.e., highly irregular) high-frequency firing, interspersed by periods of quiescence. The nomenclature of firing patterns used in this study is in line with the Petilla group conventions (Ascoli et al., 2008). We also compared the frequency spectra of S1 nMCs and V1 MCs by plotting the instantaneous firing frequencies (IFFs) of all responses (see Materials and Methods). In both cortices, the upper limit of firing of SST cells was ∼180 Hz during current injection of 1 s. The proportion of frequencies among binned frequency domains was also found to be comparable between S1 and V1 (Fig. 7).
Comparison of frequency spectra between nMCs and MCs of S1 and V1. A, Frequency spectra of nMC (in light blue) and MC (in light green) neurons of S1 and V1 are shown. The averaged distribution can be seen in deep blue for nMCs and deep green for MCs. All recorded instantaneous firing frequencies (IFFs) were pooled, and frequency proportions were plotted in 10 Hz bins. Ninety-eight percent of interspike intervals (ISIs) were found between 10 and 180 Hz for both nMCs and MCs (depicted by the gray vertical line). B, Fraction of IFFs plotted against binned frequencies for nMCs (in light blue) and MCs (in light green) is shown here. Average fraction per bin is depicted in either deep blue or deep green for nMCs and MCs, respectively. The fraction of IFFs increases in the frequency range of 10–50 Hz for nMCs, while MCs show a dip in the fraction between 10 and 20 Hz, followed by an increase in the range of 20–50 Hz.
SST cells from both cortical areas displayed largely similar properties, except in certain passive membrane features (Fig. 6E). Membrane time constant (τ) was found to be significantly different between both cortical areas, with MCs in V1 exhibiting higher values than their S1 counterparts, nMCs (nMC median 8.2 ms vs MC median 12.14 ms, p = 0.0008). The input resistance of the membrane was not found to be significantly different (nMC median 126.55 MΩ vs MC median 153.31 MΩ, p = 0.0644). In response to hyperpolarizing stimuli, SST cells showed a high degree of variability in membrane voltage sag within the same cortical area (nMC median 11.69% vs MC median 8%, p = 0.0812). In none of the cases was the rebound depolarization amplitude large enough to elicit spiking during offset of current simulation (Extended Data Table 6-1).
Table 6-1
Comparison of electrophysiological properties of L4 SST cells in S1 and V1 Statistically significant differences are marked in grey. If data passed a normality test (Shapiro-Wilk), t-test was performed, otherwise Mann Whitney U test was performed. P values were corrected post-hoc for alpha inflation error by Bonferroni method. Download Table 6-1, DOCX file.
In contrast, no action potential property was found to be significantly different between both cortices (Fig. 6F). Action potential waveform features, such as threshold of firing (nMC median −37.08 mV vs MC median −37.4 mV, p = 0.6183), amplitude (nMC 54.96 ± 1.35 mV vs MC 55.89 ± 0.94 mV, p = 0.5647), and time to peak (nMC median 0.33 ms vs MC median 0.34 ms, p = 0.7456), were also invariant between the two cortical areas. Interestingly, the width of the action potential was also not significantly different between S1 and V1 (nMC median 0.35 vs MC median 0.35 ms, p = 0.6183), although L4 SST cells in S1 have been described as “quasifast spiking” cells (Ma et al., 2006) due to their spike width being more closely aligned with fast-spiking (putative PV-positive) cells, rather than other inhibitory or excitatory neurons. However, our results are in agreement with a recent study which also did not find any significant differences in spike width between nMCs and MCs (Scala et al., 2019). A detailed account of the electrophysiological properties can be found in Extended Data Table 6-1.
We also compared passive and active membrane properties for a subset of VIP cells (S1, 19; V1, 18; Fig. 8). Active properties like rheobase, firing threshold, and spike amplitude were invariant across the cortices. Two passive membrane properties, however, were significantly different: input resistance (S1 median 227.38 MΩ vs V1 median 335.75 MΩ, p = 0.0216) and membrane capacitance (S1 median 52.73 pF vs V1 median 42.01 pF, p = 0.0216, Fig. 8A, Extended Data Table 8-1).
Electrophysiological properties of L2/3 VIP neurons in S1 and V1 cortices. Passive (A) and active (B) electrophysiological properties and (C) afterhyperpolarization (AHP) properties are shown. Significant differences are marked with the asterisk. If data passed a normality test (Shapiro–Wilk), t test was performed; otherwise, the Mann–Whitney U test was performed. p-values were corrected post hoc for alpha inflation error by Bonferroni’s method.
Table 8-1
Comparison of electrophysiological properties of L2/3 VIP cells in S1 and V1 Statistically significant differences are marked in grey. If data passed a normality test (Shapiro-Wilk), t-test was performed, otherwise Mann Whitney U test was performed. P values were corrected post-hoc for alpha inflation error by Bonferroni method. Download Table 8-1, DOCX file.
Taken together, the largest difference in electrophysiological properties between L4 SST cells of S1 and V1 arises from passive membrane properties, and not from action potential features. However, electrophysiological properties do not indicate how neurons might be embedded into the cortical circuits that they are a part of. Therefore, it is imperative to study the L2/3 VIP-to- L4 SST circuit to identify the connectivity properties of nMCs and MCs.
Unitary synaptic connectivity properties between presynaptic L2/3 VIP cells and postsynaptic L4 SST cells are similar between cortices
Intralaminar connectivity between L4 SST cells and excitatory cells of the primary somatosensory and primary visual cortices has been recently studied (Scala et al., 2019). However, translaminar inhibitory sources of input to L4 SST cells, and thus potential disinhibitory downstream circuitry, have not been fully addressed. Particularly, within S1, translaminar VIP input to L4 SST cells has not been shown previously. For L2/3 VIP cells, 13% of their boutons exist in the granular layer of the home column (Prönneke et al., 2015), and their putative targets have not yet been identified. By using the novel dual transgenic VIP-FLP::SST-Cre::Ai193 reporter mouse line, we were able to effectively perform targeted paired whole-cell patch-clamp electrophysiology, to probe the L2/3 VIP connectivity onto L4 SST cells in S1 and V1 cortices.
VIP cells were patched with standard potassium gluconate solution and induced to fire single action potentials by current injection. VIP cells have previously been suggested to target downstream cells on intermediate dendritic processes (Zhou et al., 2017), leading to an attenuated response being recorded at the soma. To reduce the space-clamp error (Williams and Mitchell, 2008), we resorted to using cesium methylsulfonate. Furthermore, to maximize the likelihood of capturing small IPSCs mediated by VIP cells, the SST cell was clamped at 0 mV to increase the driving force.
We identified robust targeting of L4 SST cells by L2/3 VIP cells (Fig. 9A). The connection probability was comparable in S1 (∼47%, 28/60 connections) and in V1 (∼36%, 27/74 connections, Fig. 9B). The success rate of synaptic transmission was similar in both cortices (S1, 57.5%; V1, 66%; Fig. 9C). The amplitude of evoked unitary IPSCs was similar between cortices (S1 median 8.35 pA vs V1 median 8.3 pA, p = 0.5308, Fig. 9C). Notably in V1, a few unitary responses were observed to be approximately fourfold larger than the population average, which we did not observe in S1 (Fig. 9A). Evoked IPSCs in S1 and V1 also showed similar latencies (S1 median 1.14 ms vs V1 median 1.02 ms, p = 0.2178), 10–90% rise times (S1 median 2.04 ms vs V1 median 2.12 ms, p = 0.9534), and normalized slopes (S1 0.29 ± 0.03 ms−1 vs V1 0.26 ± 0.02 ms−1, p = 0.4665, Fig. 9C). Furthermore, in a subset of pairs (S1: connected pairs 18/28, unconnected pairs 20/32; V1: connected pairs 16/27, unconnected pairs 23/47), we measured the depth of the SST soma from the pial surface to identify a correlation between SST cell depth and probability of connectivity. We did not observe any significant differences between the location of connected and unconnected SST cells within L4 in either cortex, indicating no correlation between the depth of the postsynaptic cell and connection probability.
Unitary synaptic connectivity properties of the VIP-to-nMC and VIP-to-MC motifs. A, Grand averages of nMC (in deep blue, n = 28) and MC neurons (in deep green, n = 24) held at 0 mV voltage clamp (VC) in response to firing of VIP neurons, evoked through current injection to elicit single spikes. Gray traces represent averages of 20–30 individual traces per synaptically connected pair. B, Probability of connectivity of the VIP-to-nMC and VIP-to-MC motifs are depicted here, which were comparable between S1 and V1 cortices (deep blue, S1 ∼47%, 28/60 connections; deep green, V1 ∼41%, 24/59 connections). C, Parameters of unitary connectivity such as synaptic success rate, amplitude, latency, 10–90% rise time, and normalized slope of the IPSC were quantified (S1, deep blue; V1, deep green). Tested unitary connectivity parameters were invariant across the two motifs.
We also probed the reciprocal connectivity of L4 SST cells onto L2/3 VIP cells (Fig. 10). SST cells were evoked to spike once, and VIP responses were recorded at a holding potential of −45 mV. Interestingly, SST cells were more often bidirectionally connected to VIP cells (S1, ∼29%; 13/45; V1, ∼23%; 14/62; Fig. 10B), while unidirectional SST-to-VIP connectivity was found in 9% of cases in S1 (4/45) and 17% in V1 (11/62). Due to the loading of SST cells with cesium methysulfonate leading to changes in action potential waveform, we did not further analyze the unitary properties of the SST-to-VIP motif. We did not identify any statistical differences between the unidirectional and bidirectional connection probabilities in S1 and V1 using Fischer's exact test. Furthermore, we also checked the rates of VIP-to-SST connectivity within the dataset of 45 and 62 pairs in S1 and V1, respectively, in which we tested reciprocal connectivity. We did not find any higher probability of the SST cell being connected to the VIP, if the VIP is connected to the SST or vice versa.
Unitary synaptic connectivity of the reciprocal nMC-to-VIP and MC-to-VIP motifs. A, nMC and MC neurons (in deep blue and deep green, respectively) are evoked through current injection to fire single action potentials while under the influence of cesium methylsulfonate IC. Grand averages of responses of a VIP neuron, held at −50 mV current clamp (CC), to nMC (in deep blue, n = 17) and MC neuron (in deep green, n = 25) stimulation. Gray traces represent averages of 20–30 individual traces per synaptically connected pair. B, Probability of connectivity of the nMC-to-VIP and MC-to-VIP motifs are depicted here, which were comparable between S1 and V1 cortices (deep blue, S1 ∼38%, 17/45 connections; deep green, V1 ∼40%, 25/62 connections). Bidirectional connectivity was more prevalent than unidirectional connectivity in both cortices (bidirectional S1 ∼29%, 13/45; unidirectional S1∼9%, 4/45; bidirectional V1 ∼23%, 14/62; unidirectional V1 ∼18%, 11/62).
We therefore have been able to show that L2/3 VIP cells target nMCs as well as MCs of L4 with similar connectivity properties. The VIP-to-SST circuit between L2/3 and L4 seems to be conserved across sensory cortices of mice, regardless of the morphological subclass to which the postsynaptic SST cell belongs. Initial evidence suggests a substantial rate of reciprocal connectivity in the MC-to-VIP and nMC-to-VIP circuits.
High-frequency stimulation of a proportion of L2/3 VIP cells induces short-term depression in nMCs but not in MCs
Cortical information transfer relies on dynamic neuronal firing over short time scales. Previously, we have detected short-term plasticity differences that were dependent on the molecularly defined subclass of GABAergic neurons (Walker et al., 2016). To study the short-term plasticity (STP) of the VIP-to-nMC and VIP-to-MC motifs, we evoked trains of presynaptic VIP firing over a range of frequencies. We observed different forms of STP within both cortical areas, with differences seen most clearly at high frequencies (Figs. 11, 12).
Short-term plasticity dynamics exhibited by the VIP-to-nMC motif in a range of frequencies in S1. IPSC responses of SST neurons evoked by stimulation of VIP neurons at 1 Hz (n = 25; A), 10 Hz (C), and 50 Hz (E). Grand average traces are depicted in deep blue, while averages of 20–30 iterations from each individual connected pair are in gray. Short-term plasticity forms were exhibited only under 50 Hz stimulation frequency. VIP-to-nMC motif either exhibited short-term depression (n = 7/25) or short-term facilitation (n = 9/25) forms of short-term plasticity. A proportion of connections (n = 9/25) also did not display any appreciable form of plasticity (depicted here as stable responses). The amplitude ratio (nth response/first response) of 1 Hz (B), 10 Hz (D), and 50 Hz (F) traces for each IPSC is segregated based on the type of short-term plasticity exhibited by the motif at 50 Hz. Note that no appreciable form of short-term plasticity is observed for 1 and 10 Hz stimulations. G, Schematic representation of the proportion of each short-term plasticity type exhibited by the VIP-to-nMC motif.
Short-term plasticity dynamics exhibited by the VIP-to-MC motif in a range of frequencies in V1. IPSC responses of SST neurons evoked by stimulation of VIP neurons at 1 Hz (n = 26; A), 10 Hz (n = 26; C), and 50 Hz (n = 27; E). Grand average traces are depicted in deep blue, while averages of 20–30 iterations taken from each individual connected pair are seen in gray. Short-term plasticity forms were exhibited only under 50 Hz stimulation frequency. VIP-to-MC motif exhibited short-term facilitation in the majority of tested connections (n = 14/27) forms of short-term plasticity. Nearly half of the connections (n = 13/27) also did not display any appreciable form of plasticity (depicted here as stable responses). The amplitude ratio (nth response/first response) of 1 Hz (B), 10 Hz (D), and 50 Hz (F) traces for each IPSC is segregated based on the type of short-term plasticity exhibited by the motif at 50 Hz. Note that no appreciable form of short-term plasticity is observed for 1 and 10 Hz stimulations. G, Schematic representation of the proportion of each short-term plasticity type exhibited by the VIP-to-MC motif.
We recorded IPSCs evoked by the firing of five presynaptic VIP cell spikes at 1, 10, and 50 Hz, covering different physiologically relevant frequency ranges. All recorded IPSC amplitudes were normalized to the first IPSC, and the ratio of the fifth IPSC at 50 Hz stimulation was used to classify the type of STP exhibited (see Materials and Methods). The first IPSC response for all tested connections was found to be comparable in amplitude within and across cortices.
The VIP-to-nMC motif in S1 (n = 25) exhibited three types of STP during high frequency (50 Hz) firing of presynaptic VIP cells (Fig. 11): short-term facilitation (Fig. 11E,F; n = 9/25), stable responses (Fig. 11E,F; n = 9/25), and short-term depression (Fig. 11E,F; n = 7/25). Connections showing short-term facilitation at 50 Hz did not facilitate at 1 or 10 Hz frequencies (Fig. 11B,D). Similarly, connections showing stable responses at 50 Hz also did not exhibit any STP at 1 or 10 Hz frequencies (Fig. 11B,D). In contrast, connections showing short-term depression did so during all tested frequencies (Fig. 11B,D). All three STP types presented with a near-equal occurrence probability within tested connections (Fig. 11G).
The VIP-to-MC motif in V1 (n = 27) also exhibited short-term facilitation (Fig. 12E,F; n = 14/27) and stable responses (Fig. 12E,F; n = 13/27) at high-frequency stimulation. However, in contrast to the VIP-to-nMC motif, we did not observe any connections that presented with short-term depression at any tested frequency. The short-term facilitation and stable responses were similar in magnitude to S1 (Fig. 12B,D). Both types observed were also found in almost equal proportions within V1 (Fig. 12G). In most cases, the IPSC ratio evoked in response to the second presynaptic spike at high frequency was a good predictor for the responses to the third, fourth, and fifth spikes.
We used ANOVA on ranks to identify statistical differences between amplitude ratios. In S1, we identified a significant difference between response ratios at IPSC 1 and IPSC 3, 4, and 5 for short-term facilitating responses. For short-term depression, we identified a significant difference in the response ratio of IPSC 1 and IPSC 5. In V1, all tested response ratios were significantly different from IPSC 1. We did not identify any differences in the stable responses in either S1 or V1.
We also analyzed the correlation between the amplitude of the first response and the STP dynamic that is exhibited, through Pearson’s product–moment correlation. We did not identify any significant relationship between the amplitude of the first response and the type of STP that is exhibited. We also did not find any correlation between the synaptic kinetic features studied (rise time, rise 10–90%, normalized slope, decay, or synaptic success rate) and the STP dynamics exhibited. Furthermore, we also did not find a correlation between the depth at which the SST cell resides and the type of STP that is exhibited in either S1 or V1.
Therefore, here we show motif-specific, as well as frequency-specific, differences in the short-term plasticity of the VIP-to-nMC and VIP-to-MC motifs, a feature that points to sensory system-specific properties of a translaminar feedback disinhibitory circuit.
Discussion
In this study, we used a dual transgenic mouse line to characterize the morphological and electrophysiological differences between L4 SST cells of the mouse primary somatosensory and visual cortices. In both cortices, we discovered a translaminar L2/3 VIP-to-L4 SST motif, with a remarkable rate of occurrence. While the unitary synaptic properties of L2/3 VIP-to-L4 S1 nMC and V1 MC circuits were similar, differences in short-term plasticity dynamics were observed within and across cortices. A subset of L2/3 VIP-to-L4 SST connections within S1, but not in V1, showed short-term depression. In addition, we observed short-term facilitation and stable responses within both S1 and V1. This offers a potential gatekeeping mechanism for the main sensory input layer with cortical area-adjusted temporal dynamics that might be responsible for recently observed in vivo contextual modulation (Keller et al., 2020; Millman et al., 2020; Szadai et al., 2022).
L4 SST of S1 and V1 differ in morphological, but not electrophysiological, properties
First, we performed an in-depth characterization of the morphological features of L4 SST cells within both cortices. MCs and nMCs have long been distinguished as distinct morphological subtypes of SST cells solely based on their qualitative presence or absence of axonal ramifications in L1 (Kawaguchi and Kubota, 1996; Wang et al., 2004; Ma et al., 2006) with little quantitative analysis of axons or dendrites, as well as electrophysiological properties. Several passive electrophysiological properties of the cell largely depend on the somatodendritic structure. Therefore, characterizing the dendrite and electrophysiological properties may have important implications in identifying the potential impact of incoming IPSCs.
We observed stark differences in the distribution of S1 nMC and V1 MC dendrites. Nearly the entire dendritic length of S1 nMC was confined within L4, in contrast to V1 MC, which tended to reach toward and branch within L2/3 and L5. This could indicate that while S1 nMCs receive bottom-up sensory inputs projecting to the cortical input layer (L4), MCs are morphologically primed to receive more input from the higher-order association layer (L2/3) and the output layer (L5). Indeed, SST cells of S1 have been shown to receive very weak top-down recruitment from higher-order areas (Lee et al., 2013; Naskar et al., 2021; Shen et al., 2022), while SST cells of V1 are strongly recruited via various means of top-down inputs (Zhang et al., 2014; Shen et al., 2022). This opens a possibility for MCs and nMCs to be integrated differently within the microcircuit dynamics of both sensory areas.
The site of axon origination influences the efficiency with which action potentials are evoked (Thome et al., 2014). Approximately 90% of the SST cells we studied in V1 had dendrite-originating axons, in comparison with ∼60% of the SST cells in S1. These dendrite-originating axons of both S1 nMCs and V1 MCs arose from dendrites oriented toward the pial surface. Therefore, presynaptic input to these dendrites could have a larger impact on the axonal discharge of SST cells by circumventing somatic integration (Hodapp et al., 2022). Indeed, input onto axon-carrying dendrites in pyramidal cells has been shown to more likely lead to dendritic spikes and increased efficacy of action potential generation (Thome et al., 2014). Dendrite-originating axons were also implicated in boosting distal EPSPs onto the oriens–alveus interneurons of the hippocampus (Martina et al., 2000). Therefore, since a larger proportion of V1 MCs exhibit dendrite-originating axons, we hypothesize that the top-down recruitment of V1 MCs is even more efficient than for S1 nMCs.
The pia-orientation selectivity of V1 MC axon origin parallels the upward projecting apical dendrites of excitatory cells, the main targets of MCs (Kapfer et al., 2007; Silberberg and Markram, 2007; Tremblay, 2016). The main targets of L4 nMCs, however, are L4 PV cells (Xu et al., 2013), with dendrites that project to and branch out in L3 (Feldmeyer et al., 2013; Scala et al., 2019). This could explain the upward projection of nMC axons, although a recent study has reported strong targeting of nMCs onto spiny stellates as well (Nigro, 2018). In some cases, we could observe a turning point of the nMC axonal arbor close to or within cortical L1, reinforcing the nMC identity of the cell, as has been previously shown for SST cells within the X94 line (Ma et al., 2006).
Electrophysiological properties of SST cells were heterogeneous and in alignment with what has been previously reported (Nigro, 2018; Scala et al., 2019). Except for τ, no other passive or active membrane property of S1 nMCs and V1 MCs showed any significant difference. When electrophysiological parameters of MCs and nMCs within S1 were compared, several properties were found to be significantly different (Ma et al., 2006). The difference between their study and ours could be explained by differences in age range and mouse lines, as well as the analysis methods that were used. It remains to be seen whether MCs of L4 V1 are a class of inhibitory neurons that are electrophysiologically different from MCs of supra- and infragranular layers of S1.
Action potential properties were also compared between S1 nMCs and V1 MCs. We did not observe a significant difference in any of the measured spike parameters. Interestingly, the spike width was also found to be very comparable between S1 nMC and V1 MC in contrast to previous reports (Ma et al., 2006), which identify S1 nMC as “quasifast spiking” cells, due to their narrow spikes. We did, however, observe a proportion of S1 nMCs that presented with fAHP and no or very small mAHP components that are in alignment with previous descriptions of S1 nMCs (Ma et al., 2006).
The L2/3 VIP-to-L4 SST motif is a conserved feature of sensory cortical networks
An important finding of our study is that L2/3 VIP cells target SST cells of L4 in the S1 and V1 cortices. We also observed that both morphological subtypes of SST cells, S1 nMCs and V1 MCs, were targeted with a similar high probability. Furthermore, we identified that unitary properties of the VIP-to-nMC and VIP-to-MC motifs were also comparable. This is indicative of the previously underestimated fact that VIP cells comprehensively target SST cells, regardless of the morphological subclass to which the SST belongs.
We describe the unitary connectivity properties and synaptic dynamics of the L2/3 VIP-to-L4 SST motif in S1. In V1, the L2/3 VIP-to-L4 SST motif has been recently identified in a large-scale dataset (Campagnola et al., 2022). However, we report approximately threefold higher connection probability than what has been previously reported. This discrepancy could be attributed to the differences in methods employed in both studies. In our study, we used Cs–methylsulfonate intracellular solution in the postsynaptic cell, which was held at 0 mV. This ensured optimized recording conditions by maximizing the amount of driving force for chloride-mediated inputs, as well as space-clamp error reduction. On the other hand, Campagnola and colleagues utilized a K-gluconate intracellular solution to record IPSPs and IPSCs from postsynaptic cells. Consequently, inhibitory responses were measured at a holding current of −55 mV, resulting in smaller driving forces and, therefore, smaller responses. Furthermore, membrane fluctuations of inhibitory cells clamped at −55 mV can be quite disruptive, leading to potential masking of some smaller amplitude responses. L2/3 VIP cells have been previously described to synapse onto dendrites of GABA-positive downstream cells (Zhou et al., 2017), and somatic recordings could lead to an underestimation of the probability of connection due to attenuated responses. We hypothesize that the difference in recording conditions could have led to the discrepancy in connectivity rates between the two studies.
The amplitude of the VIP-to-nMC and VIP-to-MC response was in line with intralaminar L2/3 connections (Walker et al., 2016), although we observed a few responses in the VIP-to-MC motif that were many folds larger. Such an occurrence could be explained by VIP inputs targeting the perisomatic compartment of the MC, rather than the dendritic processes. Due to the lack of morphological recovery of these particular VIP axons, we cannot confirm this assumption. The short latencies of these connections, however, do point toward the presence of perisomatic VIP synapses. Unitary properties of other connections are more in agreement with distal dendritic targeting.
We furthermore report that L4 SST cells also inhibited L2/3 VIP cells in ∼38% of cases in both cortices. A high level of reciprocal connectivity between VIP and SST cells has been reported recently within the mouse cortex (Campagnola et al., 2022; Kannan et al., 2022). Each inhibitory motif has been hypothesized to be recruited during different “switch states,” which may be evoked by frequency-dependent changes in the oscillation of the microcircuit (Hahn et al., 2022).
Synaptic dynamics
Synaptic responses are highly dynamic, and activity-dependent changes in synaptic transmission on a scale of milliseconds to minutes (i.e., short-term plasticity) have been implicated in the information processing of the cortex (Fuhrmann et al., 2002; Abbott and Regehr, 2004; Tong et al., 2020). Hence, rate-dependent synaptic transmission lends physiological relevance to a connectivity motif. Therefore, we also studied the response of S1 nMCs and V1 MCs to 1, 10, and 50 Hz firing of VIP cells. Here, we show that a subset of S1 nMCs, but not V1 MCs, respond to 50 Hz stimulation with prominent short-term depression. This type of STP has been previously reported in the L2/3 VIP-to-excitatory cell motif in S1 (Karnani et al., 2016b). VIP cells frequently fire at frequencies approximately and above 50 Hz (Prönneke et al., 2015; Sachidhanandam et al., 2016), and it could be hypothesized that the short-term depression might be recruited during a behavior which is specific to the somatosensory cortex, such as explorative whisking (Van Der Bourg et al., 2017).
Both the VIP-to-nMC and the VIP-to-MC motifs exhibited short-term facilitation and stable synaptic state during high-frequency stimulation. While short-term facilitation has been described for the VIP-to-excitatory, as well as VIP-to-SST motif within L2/3 before (Walker et al., 2016; Campagnola et al., 2022), the stable synaptic state has only been described for PV-to-excitatory and excitatory-to-SST cell connections (Pala and Petersen, 2015; Campagnola et al., 2022).
Each type of short-term dynamics that we identified could be observed in the entire range of experimental animal age, ruling out age-related effects (Extended Data Table 11-1). L2/3 VIP cells have been reported to broadly exhibit three different firing patterns—continuous adapting (CA), irregular spiking (IS), and burst spiking (BS; Prönneke et al., 2015, 2019). We did not observe specific firing patterns to result in particular STP types (Extended Data Table 11-1). While we observed IS to be more frequent than CA, we also observed very few cases (only once in each cortex) of BS neurons targeting L4 SST cells. The reasons for this could be twofold: (1) BS neurons have been reported to be localized in upper L2/3. Neurons are more likely to be synaptically connected if their intersomatic distances are within the range of 100–200 µm (Karnani et al., 2016a). (2) BS might be more prominent in VIP-CCK-expressing neurons (He et al., 2016). Using electron microscopy, some VIP cells have been shown to target excitatory neurons (Zhou et al., 2017), a finding which has not yet been replicated using paired recording techniques. Therefore, it could be hypothesized that burst-spiking VIP cells are a distinct subclass of neurons that preferentially target excitatory cells rather than SST cells.
Table 11-1
Depiction of age ranges of mice used, short-term plasticity observed and firing pattern of connected presynaptic VIP neuron Short-term facilitation, stable responses and short-term depression were recorded in the entire spectrum of experimental mice age. Similarly, all types of VIP neuron firing pattern (continuous adapting, irregular spiking and burst spiking) could be seen across age ranges, and the type of plasticity exhibited by the SST neuron was found to be independent of the firing pattern exhibited by the presynaptic VIP neuron. Download Table 11-1, DOCX file.
Possible functional relevance
A mixture of STP effects within the same cortical motif has been observed before. Differential STP between excitatory cells of S1 has been reported in both intralaminar and translaminar conditions (Lefort and Petersen, 2017). In V1, connections from interneurons onto different postsynaptic neurons have been recently reported to exhibit mixed STP dynamics, from strong facilitation to depression (Campagnola et al., 2022). Within L6, the occurrence of facilitation and depression between different subpopulations of excitatory neurons was attributed to the differential recruitment of these cells by thalamic afferents (Beierlein and Connors, 2002). L4 of S1 and V1 has been reported to consist of microcolumns which make up specific areas of anatomical and functional connectivity (Bruno et al., 2003; Egger, 2008; Kondo et al., 2016). A recent anatomical study reported horizontally compartmentalized occurrence of excitatory subtypes within the C2 barrel (Hua et al., 2022). One could hypothesize that VIP-to-SST motifs might exert temporally different disinhibition onto anatomical subdomains, based on different STP patterns.
SST cells of V1 have been reported to be recruited via various means of top-down inputs (Zhang et al., 2014; Shen et al., 2022). However, SST cells of S1 receive very weak top-down recruitment from higher-order areas (Lee et al., 2013; Naskar et al., 2021; Shen et al., 2022). Since top-down inputs converge heavily onto the supragranular layers of both S1 and V1 (Lee et al., 2013; Shen et al., 2022), we could hypothesize that the lack of recruitment of S1 nMCs could be attributed to the preferential localization of dendrites within the granular layer. This lower level of excitation received by SST cells of S1 introduces the need to reduce the amount of inhibition they receive, to maintain the delicate excitation–inhibition balance of the cortex. Therefore, we hypothesize that short-term depression exerted by a subset of VIP cells onto SST cells only within S1 is a direct consequence of the need for reduced inhibitory recruitment of S1 nMCs.
Footnotes
We thank Patricia Sprysch, Sandra Heinzl, and Pavel Truschow for their excellent technical assistance and Sophia Heidenreich, Sabrina Hübner, Ima Mansori, Leander Matthes, Paul Molis, Marvin Schmidt, and Nicolas Zdun for the morphological reconstructions. We also thank Dafna Ljubotina for contributing data toward this project. This work was supported by grants from the Deutsche Forschungsgemeinschaft (STA 431/14-1, 21-1; WI5636/2-1).
The authors declare no competing financial interests.
- Correspondence should be addressed to Jochen F. Staiger at jochen.staiger{at}med.uni-goettingen.de.
