The animal study was approved by the Montreal General Hospital Facility Animal Care Committee and adhered to the guidelines of the Canadian Council on Animal Care. To drive expression of Channelrhodopsin-2 (ChR2) and enhanced yellow fluorescent protein (EYFP) in VIP INs, we crossed homozygous VIP^TM1(cre)Zjh/J mice (JAX strain 010908) (Taniguchi et al., 2011) with homozygous B6.Cg-Gt(ROSA)26Sor^{TM32(CAG–COP4*H134R/EYFP)Hze}/J mice (also known as Ai32, JAX strain 024109) to obtain VIP^Cre/+;Ai32^flox/+ mice, henceforth referred to as VIP-ChR2 mice. Experiments were carried out in male and female postnatal day (P)21-P40 VIP-ChR2 mice. Animals were anesthetized with isoflurane and sacrificed once the hind-limb withdrawal reflex was lost.

To optimize slice quality obtained from these relatively mature animals, we relied on a sucrose-based cutting solution containing (in mM) 200 sucrose, 2.5 KCl, 1.0 NH₂PO₄, 2.5 CaCl₂, 1.3 MgCl₂, 47 D-glucose and 26.2 NaHCO₃. The solution was bubbled with 95% O₂/5% CO₂ for 10 min and cooled on ice to ∼4°C. Osmolality was adjusted to 338 mOsm with glucose, measured using Model 3300 or Osmo1 osmometers (Advanced Instruments Inc., Norwood, MA, USA).

After decapitation, the brain was removed and placed in ice-cold sucrose cutting solution. Coronal 300-μm-thick acute brain slices were prepared using a Campden Instruments 5000 mz-2 vibratome (Campden Instruments, Loughborough, UK) and ceramic blades (Lafayette Instrument, Lafayette, IN, USA). Brain slices were kept at ∼33°C in oxygenated artificial cerebrospinal fluid (ACSF), containing (in mM) 125 NaCl, 2.5 KCl, 1 MgCl₂, 1.25 NaH₂PO₄, 2 CaCl₂, 26 NaHCO₃ and 25 glucose, bubbled with 95% O₂/5% CO₂, for ∼10 min and then allowed to cool at room temperature for at least 1 h before starting the recordings. Osmolality of the ACSF was adjusted to 338 mOsm with glucose. We carried out experiments with ACSF heated to 32–34°C with a resistive inline heater (Scientifica Ltd, Uckfield, UK), with temperature recorded and verified offline. Recordings were truncated or not used if outside this range.

An internal solution was prepared containing (in mM) 1 or 5 KCl, 115 K-Gluconate, 10 K-HEPES, 4 Mg-ATP, 0.3 Na-GTP, 10 Na₂-Phosphocreatine and 0.1% biocytin. KOH was added to reach a pH of 7.2 to 7.4 and sucrose was added to reach the target osmolality of 310 mOsm. To visualize patched cells, 20 μM of Alexa 594 Hydrazide dye (Life Technologies, Eugene, OR, USA) was added to the internal solution. Patch pipettes were pulled using the P-1000 puller (Sutter Instruments, Novato, CA, USA). The pipette resistances varied between 4 and 7 MΩ.

We obtained whole-cell recordings using BVC-700A amplifiers (Dagan Corporation, Minneapolis, MN, USA) in current-clamp configuration. Amplified signals were low-pass filtered at 5 kHz and acquired at 40 kHz using PCI-6229 boards (NI, Austin, TX, USA). All data was acquired in Igor Pro 8 or 9 (WaveMetrics Inc., Lake Oswego, OR, USA) using custom software (Sjöström et al., 2001; Sjöström et al., 2003).¹ We monitored input resistance, series resistance, perfusion temperature, and resting membrane potential during experiments and performed further analyses offline. We did not compensate for series resistance, nor did we account for the liquid junction potential (10 mV).

Cells were patched with a LUMPlanFL N 40 × /0.80 objective (Olympus, Olympus, Melville, NY, USA) using infrared video Dodt contrast on a custom-modified Scientifica SliceScope as previously described (Buchanan et al., 2012). A Chameleon ULTRA II (Coherent, Santa Clara, CA, USA) titanium-sapphire laser tuned to 920 or 820 nm was used to excite EYFP and Alexa 594 fluorophores, respectively. VIP INs were targeted based on EYFP expression visualized with 2-photon (2P) microscopy at 920 nm. L5 BCs and MCs were targeted based on their small round-shaped soma which were distinctly different from L5 pyramidal cells (PCs) which have a triangular-shaped soma and prominent apical dendrite. Cell identity was verified post hoc using electrophysiological and morphological properties (see below and Figure 1, Supplementary Figures 5, 7, and Supplementary Tables 1, 2). Briefly, BCs were characterized by their typical fast-spiking physiology, narrow action potential half width, and high rheobase as well as their densely branching axons and dendrites. MCs were characterized by their accommodating firing pattern and lower rheobase as well as their ascending axon and dangling dendrites (Silberberg and Markram, 2007; Buchanan et al., 2012; Sippy and Yuste, 2013; Tremblay et al., 2016).

To explore long-term plasticity at VIP IN outputs, BCs and MCs were targeted for whole-cell recording in acute slices from P21-P40 VIP-ChR2 mice. L2/3 VIP INs were visualized using 2P microscopy at 920 nm. To activate ChR2-expressing L2/3 VIP INs, a blue laser (1-W 445-nm Blue Laser Diode Module, Item Id: 131542738201, Laserland, eBay.ca) was guided into the same light path as the 2P beam using a dichroic (FF665-Di02, Semrock Inc., Rochester, NY, USA) and controlled with a pair of 6215H 3-mm galvanometric mirrors (Cambridge Technologies, Bedford, MA, USA). The blue laser was gated by the MultiPatch software described above, thus enabling synchronization with electrophysiology acquisition. Blue laser pulses aimed onto fluorescent L2/3 VIP INs had a power of 20 mW and a duration of either 2 ms or 5 ms. L5 BCs and MCs that showed inhibitory postsynaptic potentials (IPSPs) in response to ChR2 activation were used for experiments.

To explore long-term plasticity at VIP IN inputs, L2/3 VIP INs were targeted for whole-cell recording using 2P microscopy at 920 nm. An extracellular stimulating pipette filled with ACSF was brought into the slice ∼100–200 μm from the patched cell and was used to activate VIP IN inputs. Extracellular stimulation was performed using a Biphasic Stimulation Isolator BSI-950 (Dagan Corporation, Minneapolis, MN, USA) that was manipulated via the MultiPatch software described above. Extracellular stimulation pulses were 100 μs in duration. Excitatory postsynaptic potential (EPSP) responses in patched VIP INs were inspected to ensure they were due to the activation of VIP IN inputs rather than direct stimulation of the patched VIP IN itself. A depolarization onset that emerged directly from the stimulation artifact was indicative of direct stimulation, whereas a depolarization onset that occurred 1–2 ms after the stimulation artifact was indicative of indirect stimulation. An input-output curve was used to measure the response amplitude to incremental increases in extracellular stimulation strength in the patched cell. The stimulation strength that yielded EPSPs at least 1 mV in amplitude and below the spiking threshold was used for the experiment.

For long-term plasticity experiments at VIP IN inputs and outputs, an initial pre-induction baseline consisted of two laser or extracellular stimulation pulses followed by two current pulses, both delivered at 30 Hz and offset by 700 ms, repeated 60 times over a period of 10 min. The induction protocol consisted of five laser or extracellular stimulation pulses and five current pulses delivered at 50 Hz and offset by ± 10 or +25 ms or delivered at 20 Hz and offset by ± 25 ms. The induction protocol was repeated 15 times for 2.5 min. The post-pairing baseline — which was contents-wise identical to the initial baseline — was repeated for up to 1 h. Control experiments had only presynaptic (pre only) or only postsynaptic spiking (post only) during the induction period. The time window for quantifying post-induction synaptic response amplitude started 10 min after the end of the induction and continued until the end of each individual experiment. This was compared to or normalized to the synaptic responses acquired during the pre-induction baseline period.

The paired-pulse ratio (PPR) was calculated as IPSP₂/IPSP₁ or EPSP₂/EPSP₁ for the pre-pairing and post-pairing periods. The change in PPR (ΔPPR) was calculated as PPR_after - PPR_before. Coefficient of variation (CV) analysis was performed as previously described (Brock et al., 2020). Briefly, the mean and CV of IPSP₁ or EPSP₁ were calculated for the pre-pairing period, followed by normalizing mean and CV^–2 to the post-pairing period. The angle (θ) was defined by the diagonal unity line and the line formed by linking the starting coordinate (1, 1) and CV analysis endpoint. θ < 0 (i.e., clockwise from diagonal) indicated a postsynaptic locus of plasticity expression, while θ > 0 indicated a presynaptic locus (Brock et al., 2020).

The motor cortex was targeted based on the location of the corpus callosum white matter tract. L1 and white matter were identified based on their lack of cell bodies. For electrophysiological experiments, we differentiated between L2/3, L5 and, L6 based on PC morphology. In L2/3, PC somata are relatively small, whereas in L5, PCs have large somata and a thick apical dendrite. L6 PCs have rounded somata and a thin apical dendrite.

Layer boundaries for immunohistochemistry and biocytin histology were informed by NeuN cell counts and in situ hybridization (ISH) data from the Allen Institute Mouse Brain Atlas (Lein et al., 2007). We selected ISH images stained for Stard8, Rorb, Bend5, and Ighm expression, which was restricted to L2/3, L4, L5, and L6, respectively. Using Fiji/ImageJ (Schindelin et al., 2012), we selected a ∼800-μm-wide linear region of interest spanning the motor cortex from pial surface to white matter and measured the intensity profile across the cortical thickness. We thus overlayed the intensity profile for each gene marker (Supplementary Figure 1). Each density profile was baseline-subtracted, peak-normalized, and box-smoothed (setting 10) in Igor Pro 9. The point of intersection between pixel intensity profiles were then used to define layer boundaries. The percentage distance normalized from pial surface to white matter was used to inform layer boundaries in individual slices.

We were concerned that fixed tissue samples might be altered, e.g., due to fixation, cover slipping, or other histology steps, which might distort the apparent cortical thickness. We therefore verified the cortical thickness in fixed tissue by comparing to the acute slice in electrophysiology experiments, which revealed a good match, thus validating the accuracy of fixed slice samples.

P21-P40 VIP-ChR2 mice were anesthetized with isofluorane and transcardially perfused with 0.1 M phosphate buffered saline (PBS) followed by 4% paraformaldehyde. Brains were incubated in 4% paraformaldehyde overnight, and then stored for two additional days in a 30% (w/v) sucrose solution. Next, brains were mounted in plastic cubes containing Optimal Cutting Temperature embedding medium (25608-930, VWR, Radnor, PA, USA) and then frozen using a bath of 100% EtOH and dry ice. Fixed brains were sectioned with a cryostat at 50 μm thickness through the motor cortex and sections were placed in a 0.01 M PBS solution. Sections underwent a 20-min wash in 0.01 M PBS with 1% Triton-X followed by a 90-min wash in 0.01 M PBS with 0.3% Triton-X and 10% normal donkey serum (NDS; 017-000-121 Jackson ImmunoResearch, West Grove, PA, USA). All antibody incubations were performed in 0.01 M PBS with 0.3% Triton-X and 1% NDS.

Sections were incubated overnight at 4°C in the following primary antibodies: 1:500 rabbit anti-VIP (20077, ImmunoStar, Dietzenbach, Germany), 1:100 rat anti-somatostatin (ab30788, Abcam, Boston, MA, USA), 1:1000 chicken anti-GFP (ab13970, Abcam), 1:500 mouse anti-parvalbumin (p3088, Sigma, St. Louis, MO, USA), and 1:1000 mouse anti-NeuN (ab104224, Abcam). Twenty-four hours later, tissue underwent three 15-min washes in 0.01 M PBS with 0.3% Triton-X and 1% NDS, followed by a 90-min incubation in the following Alexa Fluor secondary antibodies at 1:1000: donkey anti-rabbit 647 (711-605-152, Jackson ImmunoResearch), donkey anti-rat 594 (712-585-150, Jackson ImmunoResearch), donkey anti-chicken 488 (703-545-155, Jackson ImmunoResearch), goat anti-mouse 647 (A21240, ThermoFisher Scientific, Waltham, MA, USA), and donkey anti-mouse 568 (SAB4600075, Sigma). Next, the tissue underwent three 20-min washes in 0.01 M PBS with 0.3% Triton-X and 1% NDS. Following this procedure, coronal slices were mounted using coverslips with a 40 μl bolus of ProLong Gold Antifade Mountant (ThermoFisher Scientific).

Sections were imaged using a Fluoview FV1000 confocal laser scanning microscope and Fluoview software (Olympus Canada, Richmond Hill, ON, Canada) or a Zeiss LSM780 confocal laser scanning microscope and ZEN software (Zeiss, Oberkochen, Germany). Image analysis and quantification were performed using Fiji/ImageJ (Schindelin et al., 2012) and Igor Pro 9 (Wavemetrics). Cell counts for neurons expressing VIP, EYFP, somatostatin (SST) and parvalbumin (PV) were carried out across all six cortical layers and in both hemispheres.

Patched VIP INs, MCs, and BCs used in long-term plasticity experiments were saved for neuronal reconstruction. Once the experiment was completed, the patch pipette was removed slowly while applying light positive pressure. Sections were then incubated in 4% paraformaldehyde overnight and were stored in 0.01 M PBS solution for up to 3 weeks before staining.

Sections underwent four 10-min washes in 0.01 M Tris-buffered saline (TBS) solution with 0.3% Triton-X followed by a 1-h wash in 0.01 M TBS with 0.3% Triton-X and 10% NDS. Sections were incubated overnight at 4°C in 0.01 M TBS with 0.3% Triton-X and 1% NDS, supplemented with 1:200 Alexa Fluor 647- or Alexa fluor 488-conjugated Streptavidin (ThermoFisher Scientific). Twenty-four hours later, tissue underwent four 10-min washes in 0.01 M TBS. Following this procedure, sections were mounted using coverslips with a 40 μl bolus of ProLong Gold Antifade Mountant (ThermoFisher Scientific). 3D image stacks were acquired using a Zeiss LSM780 confocal laser scanning microscope and ZEN software (Zeiss) and used for morphological reconstructions.

3D confocal image stacks were contrast adjusted and converted to 8 bits or 16 bits in Fiji (Schindelin et al., 2012) and then imported into Neuromantic V1.7.5 (Myatt et al., 2012) for manual tracing. Morphometry was performed in Igor Pro 9 (Wavemetrics) using the qMorph in-house custom software as previously described (Buchanan et al., 2012; Zhou et al., 2021).²

Unless otherwise noted, results are reported as the mean ± standard error of the mean (SEM). Significance levels are denoted using asterisks (*p < 0.05, **p < 0.01, ***p < 0.001). Pairwise comparisons were carried out using a two-tailed Student’s t-test for equal means. If an equality of variances F test gave p < 0.05, we employed the unequal variances t-test. Wilcoxon-Mann-Whitney’s non-parametric test was always used in parallel to the t-test, yielding similar outcomes. Statistical tests were performed in Igor Pro 9 (Wavemetrics).

We created a VIP-ChR2 mouse line by crossing a VIP-Cre driver line with the Ai32 ChR2/EYFP reporter line (Methods). We validated our VIP-ChR2 mice by exploring the degree of overlap between the genetic EYFP tag and VIP expression. To do so, we relied on immunohistochemistry (Methods). This revealed that ∼80% of EYFP-positive cells were also positive for VIP and that ∼88% of VIP-positive cells were also positive for EYFP (Supplementary Figure 2), demonstrating that our VIP-ChR2 mouse line was highly specific for VIP INs, in agreement with the prior literature (Taniguchi et al., 2011).

Next, we looked at the spatial distribution of VIP INs across the cortical layers. Similar to what has been shown in the barrel cortex (Prönneke et al., 2015; Almási et al., 2019) and visual cortex (Gonchar et al., 2007), we found in the motor cortex that most VIP INs were located in L2/3 (Supplementary Figure 3). We also explored whether VIP INs expressed SST or PV — molecular markers of MCs and BCs, respectively (Blackman et al., 2013) — but found that they did not (Supplementary Figure 4), in agreement with the prior literature (Prönneke et al., 2015).

We compared the electrophysiological properties of a total of 46 patched VIP INs from L2/3, L5, and L6. We found no detectable differences across layers in basic electrophysiological properties such as resting membrane potential, firing threshold, action potential height, action potential half width, rheobase, membrane time constant, and input resistance (Supplementary Table 1). We also found that VIP INs had varying spike patterns. Following the Petilla convention (Ascoli et al., 2008), we found that VIP INs exhibited three different action potential firing patterns: adapting, burst firing, and irregular firing (Supplementary Figure 5 and Supplementary Table 1).

We investigated the morphology of VIP INs in the mouse motor cortex. We used biocytin histology and confocal imaging to 3D reconstruct patched VIP INs in L2/3, L5 and L6. Consistent with previous findings in the barrel cortex (Prönneke et al., 2015), we found that VIP INs have dendrites that project toward the pial surface and axons that extend into deeper cortical layers (Figures 1A–C). Sholl analysis (Sholl, 1953) additionally revealed that VIP IN dendrites and axons were most densely branched around 100 μm away from the soma (Figure 1D). L2/3 VIP IN dendrites were mostly localized to superficial layers, whereas their axons extended into deeper cortical layers. L5 and L6 VIP INs dendrites and axons were mostly localized to deeper cortical layers (Figure 1E).

Synapse-type-specific plasticity has been reported throughout the brain (Larsen and Sjöström, 2015) and allows synapses to adapt differentially depending on factors such as target cell (Blackman et al., 2013) or functional role in the microcircuit (McFarlan et al., 2023). In our study, we found that plasticity outcomes at VIP IN outputs depend on synapse type. This synapse-type-specific plasticity has been reported at I→I synapses in L2/3 visual cortex, where presynaptic tetanic stimulation induced I→I LTP at PV IN→PV IN synapses but not at SST IN→PV IN synapses (Sarihi et al., 2012). Similarly, increased PC activity in L2/3 prefrontal cortex selectively potentiated SST IN→PC synapses but not PV IN→PC synapses (Chiu et al., 2018). Additionally, I→E synapses in the developing cortex exhibited different plasticity rules depending on the location of the postsynaptic cell. For example, auditory cortex L4 IN→L5 PC synapses exhibited causal LTP (Field et al., 2020), whereas visual cortex L4 BC→L4 PC synapses exhibited causal LTD that later switched to LTP after critical period sensory experience (Vickers et al., 2018). The switch in plasticity rules at L4 BC→L4 PC synapses suggested that these I→E synapses may be important for promoting plasticity, whereas L4 IN→L5 PC synapses may provide stability. Overall, these studies highlight the many factors that contribute to synapse-type-specific plasticity which include cell type, cell location, and synapse type.

In our study, plasticity was induced at VIP IN→MC synapses but not VIP IN→BC synapses. Considering that VIP INs weakly synapse with BCs compared to MCs (Pfeffer et al., 2013; Kepecs and Fishell, 2014; Tremblay et al., 2016; Apicella and Marchionni, 2022; Campagnola et al., 2022), it is possible that VIP INs and BCs do not have enough consistent coincident activity to induce STDP at VIP IN→BC synapses. Induction of plasticity at VIP IN→BC synapses may instead depend on pre- or postsynaptic activity alone. High frequency stimulation of PV INs, for example, resulted in E→I LTP and I→I LTP in the visual cortex (Sarihi et al., 2012; Chistiakova et al., 2019). Moreover, low frequency theta-burst stimulation at excitatory inputs onto hippocampal BCs resulted in E→I LTP, whereas high frequency theta-burst stimulation resulted in E→I LTD (Camiré and Topolnik, 2014). Other factors such as neuromodulators may be required for VIP IN→BC plasticity. It was demonstrated that the release of the neuropeptide enkephalin by hippocampal VIP INs long-term disinhibited CA2 PCs via I→E LTD at PV IN outputs. This VIP IN-mediated disinhibition allowed for enhanced information transfer between hippocampal CA3 and CA2 PCs, which was important for social memory storage (Leroy et al., 2022). Thus, this form of plasticity did not require the coincident firing of pre and postsynaptic cells, but rather a neuromodulator. Together, these results highlight the need to further explore different induction protocols to elucidate the plasticity rules that govern VIP IN→BC synapses.

Though plasticity at a given synapse type depends on several factors including rate, timing, depolarization, and higher-order spiking statistics (Sjöström et al., 2001; Froemke and Dan, 2002; Froemke et al., 2006; Pfister and Gerstner, 2006), we focused here on how plasticity at VIP IN inputs and outputs depends on the relative timing of pre- and postsynaptic spiking. We found that LTD induction at VIP IN→MC synapses was sensitive to timing (Figure 2). At E→VIP IN synapses, however, LTP did not depend on the sign of relative timing, yet pre- or postsynaptic firing alone yielded no plasticity (Figure 4). This demonstrated that this form of LTP was still dependent on spike timing, a defining feature of STDP (Markram et al., 2012). Previous studies have additionally shown that STDP can be symmetric around the origin (Egger et al., 1999; Abbott and Nelson, 2000; Lu et al., 2007). Taken together, our findings highlight how timing is a key determinant of VIP IN STDP.

The locus of expression of long-term plasticity is important because it carries with it computational implications (Costa et al., 2015; Costa et al., 2017; Mizusaki et al., 2022). Postsynaptically expressed plasticity generally alters only synaptic strength while leaving short-term dynamics unaffected, although there are exceptions to this rule (Poncer and Malinow, 2001). Presynaptically expressed long-term plasticity, however, typically alters both short-term plasticity and synaptic gain (Markram and Tsodyks, 1996; Sjöström et al., 2003; Sjöström et al., 2007). This is because during high-frequency spike trains, the readily releasable pool of vesicles is depleted, which causes short-term synaptic depression (Zucker and Regehr, 2002), although at some synapse, short-term facilitation dominates for other mechanistic reasons (Blackman et al., 2013). Either way, these short-term synaptic dynamics are generally strongly affected by pre- but not by postsynaptically expressed long-term plasticity. Short-term plasticity is functionally important as it filters information transferred by a synapse (Fortune and Rose, 2001; Fuhrmann et al., 2002). In this view, short-term facilitating connections act as high-pass filtering burst detectors (Maass and Zador, 1999; Matveev and Wang, 2000), while short-term depressing connections low-pass filter information for correlation detection and gain-control (Abbott et al., 1997; Rosenbaum et al., 2012). For instance, presynaptic LTP would be expected to increase the release probability, thereby depleting the readily-releasable pool of vesicles faster during high-frequency bursts. The ensuing increase in short-term depression thus biases the synapse toward correlation detection at the expense of burst detection (Blackman et al., 2013; Costa et al., 2017).

In our study, we analyzed PPR and CV to assess the locus of expression. For VIP IN→MC LTD, we found no change in PPR, suggesting a postsynaptic locus of expression (Figure 2E), whereas CV analysis suggested that plasticity was expressed presynaptically (Figure 2F; Brock et al., 2020). This apparent discrepancy could be explained by the paired-pulse stimulation frequency not being potent enough to sufficiently deplete the readily-releasable pool of vesicles, leading it to be inconclusive. With this interpretation, LTD at VIP IN→MC was presynaptically expressed. In this view, this form of LTD would thus be expected to alter the information filtering of VIP IN→MC synapses in addition to weakening them (Blackman et al., 2013; Costa et al., 2017).

Mechanistically, presynaptically expressed plasticity at VIP IN→MC synapses could be mediated by retrograde messengers like endocannabinoids (Castillo et al., 2012; Piette et al., 2020) or BDNF (Inagaki et al., 2008; Vickers et al., 2018). Additionally, given that the postsynaptic MC needs to be active in order to coincidentally fire with the presynaptic VIP IN, postsynaptic NMDA receptor signaling may be involved. Indeed, NMDA receptor signaling has been shown to control GABA_A receptor stability at inhibitory synapses (Muir et al., 2010).

For inputs to VIP INs, on the other hand, ΔPPR and CV analysis agreed that E→VIP IN LTP had a postsynaptic locus of expression (Figure 4). The expression of postsynaptic E→E LTP is typically due to the insertion of postsynaptic AMPA receptors (Malinow and Malenka, 2002; Bredt and Nicoll, 2003; Kessels and Malinow, 2009). The NMDA receptor, which is particularly well suited for STDP given its capacity for coincidence detection, is also highly implicated in E→E LTP (Wong et al., 2021). Thus, AMPA receptor insertion and NMDA receptor signaling are strong candidates for the postsynaptic expression of E→I LTP at VIP IN inputs.

To understand the consequences of long-term plasticity at VIP IN inputs and outputs, it is important to consider its impact at the circuit level (McFarlan et al., 2023). LTP of excitatory inputs to VIP INs promotes VIP IN activity, which by inhibiting BCs and MCs is expected to increase circuit activity overall. Consequently, it has been proposed that attention operates through cortical disinhibitory circuits, e.g., by neuromodulation or by top-down control (Kullander and Topolnik, 2021; Speed and Haider, 2021). It has been argued that cholinergic modulation specifically of VIP INs is required for attention (Obermayer et al., 2019), although some disagree and propose that VIP IN activation is orthogonal to attention (Myers-Joseph et al., 2023). Regardless, we show here how activity-driven VIP IN plasticity can potentially contribute to such attentional effects.

The plasticity of VIP INs may also more generally regulate learning. For instance, VIP IN-mediated suppression of SST INs has been correlated with improved motor learning in the motor cortex (Adler et al., 2019) as well as enhanced adult plasticity in the visual cortex (Fu et al., 2014; Fu et al., 2015). Thus, E→I LTP at VIP IN inputs may help to boost learning and plasticity in cortical circuits.

In addition, VIP IN→MC LTD is expected to increase inhibition specifically of PC dendrites (Wang et al., 2004). Although E→VIP IN LTP and VIP IN→MC LTD may superficially seem to oppose each other if triggered simultaneously, they would additionally redistribute inhibition across the somato-dendritic axis of pyramidal cells (Pouille and Scanziani, 2004; Blackman et al., 2013). According to influential theoretical frameworks on cortical associative learning (Larkum, 2013), such dendritic inhibition is expected to regulate neocortical information storage.

As VIP INs are able to mediate disinhibition (Artinian and Lacaille, 2018; Kullander and Topolnik, 2021), they are ideally positioned as key regulators of activity in local circuits, with implications for disease states such as epilepsy (Cunha-Reis and Caulino-Rocha, 2020). For example, it has been reported that, whereas inhibition of SST INs prolongs seizures, inhibition of VIP INs reduces seizure propensity (Khoshkhoo et al., 2017). Consequently, the protective role of VIP IN→MC LTD might be possible to harness as a therapy for epilepsy.

One caveat in our study is the discrepancy between ΔPPR and CV analysis for determining the locus of expression of VIP IN→MC LTD. CV analysis suggested that LTD was expressed presynaptically, whereas PPR analysis suggested it was expressed postsynaptically. This discrepancy between CV and PPR analyses could occur if the 30-Hz stimulation was not sufficiently high to deplete the readily releasable pool. It is possible that depleting VIP IN→MC synapses better would reveal a change in PPR.

Another potential caveat with our experimental paradigm comes from the use of ChR2 to activate presynaptic VIP INs. Since ChR2 fluxes calcium (Nagel et al., 2003), it may directly trigger release at synaptic boutons. ChR2 may also depolarize synaptic terminals, again contributing to release. This artifact, if present, would be expected to elevate release and increase short-term depression. Consequently, this may be particularly important when using PPR as a measure for determining the locus of plasticity expression at VIP IN outputs, whereas CV analysis may be less affected. However, given the many hundreds of micrometer distance between presynaptic laser stimulation and postsynaptic cells where VIP IN synapses form (Figures 2A, 3A), this artifact seemed unlikely.

Furthermore, the lack of single-cell resolution with respect to presynaptic cell activation — we likely optogenetically stimulated more than one presynaptic L2/3 VIP IN at a time. An alternative approach for unitary synapse resolution would be paired recordings (Lalanne et al., 2016). However, given appreciable distance between L2/3 and L5, the sparsity of VIP INs in the cortex (Supplementary Figure 3), and the overall low connectivity of VIP IN outputs (Walker et al., 2016), paired recordings would be an impractical tool. Another alternative approach for unitary synapse resolution is by using 2P optogenetic activation (Chou et al., 2023), which in the future should enable the study of long-term plasticity at multiple VIP IN→MC/BC synapses in parallel.

To further our understanding of how disinhibitory plasticity impacts the healthy brain as well as neuropathologies, future research will need to explore the plasticity at the many different I→I and I→I→E synapses, to contribute to the plasticitome of cortical INs (McFarlan et al., 2023). Because it is challenging to explore many different synapse types, this effort will likely require new high-throughput plasticity-mapping approaches (Sjöström, 2021).

We relied here on the STDP experimental paradigm, which is widely believed to be biologically plausible (Markram et al., 2011; Feldman, 2012; Markram et al., 2012). However, some have disagreed and instead argued that STDP is of limited biological relevance (Lisman and Spruston, 2005; Lisman and Spruston, 2010). On a related note, due to the experimental challenges associated with targeting these specific synapse types, we could only explore a limited plasticity induction parameter space in this study. Induction protocols that e.g., rely on local dendritic spikes (Holthoff et al., 2004; Remy and Spruston, 2007; Bittner et al., 2017) may thus reveal plasticity at VIP IN→BC synapses. We also did not explore neuromodulation, which may additionally gate plasticity (Pawlak et al., 2010), even synapse specifically for the same cell type (Edelmann et al., 2017). Our study is thus not a final verdict on VIP IN plasticity, but rather a starting point.