Mesodiencephalic junction GABAergic inputs are processed separately from motor cortical inputs in the basilar pons

Summary The basilar pontine nuclei (bPN) are known to receive excitatory input from the entire neocortex and constitute the main source of mossy fibers to the cerebellum. Various potential inhibitory afferents have been described, but their origin, synaptic plasticity, and network function have remained elusive. Here we identify the mesodiencephalic junction (MDJ) as a prominent source of monosynaptic GABAergic inputs to the bPN. We found no evidence that these inputs converge with motor cortex (M1) inputs at the single neuron or at the local network level. Tracing the inputs to GABAergic MDJ neurons revealed inputs to these neurons from neocortical areas. Additionally, we observed little short-term synaptic facilitation or depression in afferents from the MDJ, enabling MDJ inputs to carry sign-inversed neocortical inputs. Thus, our results show a prominent source of GABAergic inhibition to the bPN that could enrich input to the cerebellar granule cell layer.


INTRODUCTION
Motor control relies on brain-wide networks. Motor cortex directs voluntary movements (Guo et al., 2015) and the cerebellum coordinates movements (Manto et al., 2012). Reciprocal connections between these structures are necessary for proper motor control. Indeed, the cerebellum projects to the motor cortex via the thalamus (Sawyer et al., 1994;Aumann 2002;Gornati et al., 2018), while the motor cortex projects to the cerebellum via the pontine nuclei (Schwarz and Mock 2001;Kratochwil et al., 2017). This closedloop connectivity is proposed to enable forward and inverse models for motor control (Wolpert et al., 1998;Shadmehr and Krakauer 2008). Interestingly, other parts of the neocortex and cerebellum are also connected (Kelly and Strick 2003;Henschke and Pakan 2020;Pisano et al., 2021), potentially enabling similar computational mechanisms for cognitive processes (Ito 2008).
This places the pontine nuclei at the nexus of information transfer between neocortex and cerebellum. Indeed, afferents from the basilar pontine nuclei (bPN) constitute the principal source of mossy fibers in the cerebellum (Kratochwil et al., 2017). The bPN also receives inputs from numerous non-neocortical regions of the brain (Burne et al., 1981;Wiesendanger and Wiesendanger 1982;Kosinski et al., 1986;Mihailoff et al., 1988Mihailoff et al., , 1989. These afferents generally terminate in topographically organized zones in the bPN (Leergaard and Bjaalie 2007;Proville et al., 2014;Kratochwil et al., 2017). Similarly, mossy fibers originating from the bPN project to specific zones in the cerebellum (Pä ä llysaho et al., 1991;Mihailoff 1993;Odeh et al., 2005;Huang et al., 2013;Kratochwil et al., 2017). Consequently, the bPN is often considered to be a relay for information destined for the cerebellum rather than having a role in active processing.
Still, synaptic plasticity of inputs to the bPN has been described, suggesting a potential way of input processing  and shaping spiking activity in the bPN Mö ck et al., 2006;Guo et al., 2021). Furthermore, various sources of GABAergic input to bPN neurons have been suggested Mihailoff and Border 1990;Mö ck et al., 1999), but these inputs have not been physiologically confirmed or characterized, precluding conclusions about their function and integration in the cerebro-cerebellar circuit.
Here we identify the mesodiencephalic junction (MDJ) as the main source of GABAergic signaling to the bPN. This inhibition does not seem to interact with afferents from the motor cortex at the single neuron

Monosynaptic inputs from M1 to the basilar pontine nuclei display marked synaptic depression
To characterize the short-term plasticity of cortico-pontine synapses, we expressed Chronos in the motor cortex (M1) to enable the visualization and stimulation of M1 afferents to the bPN. Whole-cell voltage-clamp recordings from bPN neurons at À70 mV were made, and M1 axons were stimulated with short pulses of blue light (Figure 2A and 2B). All inputs evoked from M1 had a short delay to onset (2.4 G 0.06 ms), a fast-rising (B) Examples of retrobead-labeled neurons in the inferior olivary nucleus (IO), lateral reticular nucleus (LRN), external cuneate nucleus (eCU), basilar pontine nuclei (bPN), and the absence of labeled neurons in the primary sensory cortex (S1), and primary motor cortex (S1). Images were produced by acquiring monochromatic photos in the red spectrum and then inversing these images. As a result, retrobeads are depicted in black. Scale bar represents 50 mm. (C) Same as for A, but for retrobead injections in the bPN (N = 2 animals). (D) Same as for B, but for retrograde labeling after injections in bPN. Example sections are shown with labeling concentrated in the neocortex, sparse signal of retrogradely labeled neurons in midbrain, and dense staining in the injection site. iScience Article phase and decay (10-90%: 1.0 G 0.22 ms and 90-10%: 20 G 12 ms, respectively), and were reduced to 4.4 G 0.8% of the original response by DNQX application (ACSF: 20 G 5 pA; DNQX: 1.0 G 0.3 pA; p < 0.001; n = 4 neurons; Figure 2C), confirming that M1 provides glutamatergic inputs to bPN. During train stimulation, we observed prominent short-term synaptic depression of M1 inputs across all tested frequencies, with more pronounced depression at higher frequencies and later in the stimulus train ( Figure 2E, n = 4 neurons in N = 4 mice). To check for possible opsin-specific influences on these inputs, we repeated these experiments with ChrimsonR. ChrimsonR-evoked responses were more depressed at higher stimulation frequencies (Figure S2A), which is likely owing to incomplete recovery of the ion channel (Klapoetke et al., 2014). Therefore, we assessed all short-term plasticity with the stimulation of Chronos. In mice expressing Chronos in M1, synaptic responses recorded in the bPN were depressed at 50 and 100 Hz after a 20-pulse train stimulus to 0.7 G 0% and 0 G 0% of initial amplitude, respectively (steady-state, average of last five responses). After train stimuli at 10 and 20 Hz, responses were depressed to 41 G 3% and 22 G 2%, respectively. To confirm that M1 inputs to the bPN are monosynaptic, we applied tetrodotoxin (TTX) to block AP-generated neurotransmitter release. In this situation, stimulated axons can only depolarize during the optogenetic stimulation, after which they are quickly repolarized, preventing invasion of positive charge into synaptic boutons. As expected, optogenetically evoked responses were virtually absent in the presence of TTX (2.2 G 0.2% of the original response; ACSF: 50 G 18 pA; TTX: 1.0 G 0.7 pA; n = 3). Subsequent co-application of 4-aminopyridine (4-AP), which prolongs optogenetically evoked depolarization and therefore increases the likelihood of stimulation of boutons, even with remote axonal stimulation, recovered the synaptic responses (130 G 56% of amplitude in ACSF; TTX + 4-AP: 50 G 28 pA, n = 3; Figure 2D; Repeated measures ANOVA on measurements normalized to ACSF condition: F(2,2) = 11.21, p = 0.02; Post-hoc Bonferroni corrected t-tests ACSF vs TTX p < 0.001; TTX vs TTX+4AP p = 0.22; ACSF vs TTX+4AP p = 0.98). These results show that M1 provides prominent, but strongly depressing monosynaptic glutamatergic inputs to bPN neurons.
The mesodiencephalic junction sends prominent monosynaptic GABAergic inputs to the basilar pontine nuclei Our retrograde tracing experiments ( Figure 1) showed that in addition to neocortical regions, some subcortical brain regions also provide inputs to the bPN. These regions might provide GABAergic inputs as has been suggested before Mihailoff et al., 1988Mihailoff et al., , 1989Mihailoff and Border 1990;Mihailoff 1995). To investigate possible GABAergic signaling in the bPN, we stained sections of mouse brain for the enzyme Glutamate decarboxylate 67 (GAD67) to identify GABAergic neurons. We never observed GAD67+ somata in the bPN of these mice, but we did observe prominent and numerous GAD67+ boutons (Figures S2B and S2C; N = 6 mice). We further confirmed these observations in GAD-GFP mice (Chattopadhyaya et al., 2004) (Figure S2D; N = 4 mice). This indicates that there is a prominent extrinsic source of GABA in the bPN. Closer investigation of the afferent areas to bPN revealed that the majority of inputs from midbrain arose from the MDJ (3% of all projections to the bPN, Figure 1H), an iScience Article area intimately involved with the cerebellar circuit (Ruigrok 2004). Glutamatergic MDJ neurons that project to the inferior olive have been described previously (de Zeeuw et al., 1989;Ruigrok and Voogd 1995) and these neurons are positioned intermixed with neurons that contain other neurotransmitters (De Zeeuw and Ruigrok 1994).
To confirm that the MDJ is a source of GABAergic inputs to the bPN, we injected AAVs to express Chronos in this region. In acute slices, we performed whole-cell recordings in regions of the bPN that also receive inputs from M1. We observed outward currents in neurons clamped at 0 mV, with a short rise, and long decay (2.1 G 0.36 ms and 140 G 46 ms, respectively) when stimulating with light ( Figures 2F and 2G). These inputs were reduced to 6 G 7% in the presence of Gabazine (ACSF: 20 pA G 11 pA; Gabazine: 0.5 G 0.41 pA; n = 10 neurons; p < 0.001; Figure 2H). Furthermore, we did not observe a change in holding current (ACSF: 140 G 28 pA vs Gabazine: 160 G 33 pA, n = 10 neurons, p = 0.27), indicating that inhibition from MDJ to bPN neurons is predominantly phasic Contrary to glutamatergic M1 inputs, GABAergic MDJ inputs showed remarkably little short-term synaptic plasticity at intervals >20 ms, even after a 20 pulse stimulation-train, we observed 108 G 5% and 105 G 7% of the initial amplitude for 10 and 20 Hz stimulation trains, respectively ( Figure 2J, n = 11 neurons in N = 9 mice). The amplitude of responses was only depressed toward the end of a pulse train at frequencies S50 Hz (to 55 G 7% and 14 G 7% for 50 and 100 Hz stimulus trains, respectively). Similar to the observed ChrimsonR effects on M1 inputs, MDJ afferents expressing ChrimsonR showed enhanced short-term synaptic depression ( Figure S2A). To confirm that inputs from the MDJ are monosynaptic, we applied TTX followed by the combined application of TTX and 4-AP. Inputs from the MDJ are blocked upon TTX application (5 G 3.9% of the response in ACSF; ACSF: 23 G 8 pA vs TTX: 1.5 G 0.61 pA; n = 5 neurons) and subsequently rescued after co-application with 4-AP (to 400 G 760% of ACSF response; TTX +4-AP: 50 G 23pA; n = 5 neurons Figure 2I; Repeated measures ANOVA on measurements normalized to ASCF condition: F(2,4) = 27.77, p < 0.001; Post-hoc Bonferroni corrected t-tests ACSF vs TTX p < 0.001; TTX vs TTX+4AP p = 0.05; ACSF vs TTX+4AP p = 1.00).

Segregated information streams to the basilar pontine nuclei
Our results thus far indicate that neurons in the bPN receive depressing excitatory input from M1 and inhibitory input from the MDJ that undergoes very little short-term plasticity. A possible role of the bPN is modulating incoming cortical inputs, for example via inhibitory inputs from MDJ. However, this can only be achieved if these inputs interact in a network. To investigate whether single bPN neurons receive inputs from both M1 and from MDJ, we analyzed data from long full-field optical stimulation of all neurons that responded to either M1 or MDJ axon stimulation (see method details). The success rate of evoking opsin-induced currents in bPN neurons was generally low, while we could detect spontaneous excitatory and inhibitory events in many neurons. This indicates that potentially many other sources of input to bPN neurons exist. Neurons were clamped at À70 mV and subsequently at 0 mV to enable the detection of EPSCs and IPSCs, respectively ( Figures 3A-3C). Of all bPN neurons that responded to optogenetic stimulation, 60% (33 out of 53) of neurons only received inputs from MDJ and 40% (20 out of 53) only received inputs from M1 Figure 3D). We did not observe any neurons that received both M1 and MDJ inputs, suggesting that these afferents target different neurons within the bPN. To investigate whether these neurons might represent different classes, we compared several passive electrical properties between the two groups. However, we found no statistically significant differences in membrane resistance (M1: 320 G 49 MU; MDJ: 220 G 25 MU; p = 0.08; n = 52), membrane capacitance (M1: 100 G 15 pF; MDJ: 108 G 8.2; p = 0.86; n = 52) or membrane decay time constant (M1: 1.18 G 0.08 ms; MDJ: 1.2 G 0.10 ms; p = 0.96; n = 52) between these two groups, providing no indication that these neurons represent separate classes ( Figure S3). Thus, our results show that convergence of inputs from M1 and MDJ in the bPN is rare if not absent, making it unlikely that MDJ inputs directly modulate M1 inputs in bPN.
Nonetheless, it is possible that inputs from M1 and MDJ indirectly interact in the bPN via a local network.
One study reports no short-range interactions between bPN neurons, but this dataset only comprised twenty tested pairs and strictly probed proximate connections with a maximal distance well below 100 mm (Mö ck et al., 2006). To confirm and expand on this finding, we probed a total of 250 unidirectional connections spaced up to 500 mm apart, in slices cut in both the coronal (n = 168) and sagittal (n = 82) orientation to avoid confounding effects of slice orientation (Shinoda et al., 1992) ( Figure 3E). We did not ll OPEN ACCESS iScience 25, 104641, July 15, 2022 5 iScience Article detect evidence of any synaptic contacts between neurons. Therefore, it is also unlikely that M1 and MDJ inputs interact via a local circuit, but rather that M1 and MDJ inputs are processed separately by the bPN. iScience Article This could explain a recent report that the firing rates of distinct pontine neuron populations differentially change during a voluntary reaching and grabbing task (Guo et al., 2021). If bPN-projecting MDJ neurons are indeed recruited during movement, we predict that they receive prominent inputs from the neocortex.

Basilar pontine nuclei-projecting mesodiencephalic junction neurons receive input from neocortex
To investigate whether the neocortex projects to bPN-projecting MDJ neurons we used monosynaptic rabies tracing (Wickersham et al., 2006). We first checked whether we could trace connections from neocortex, through bPN to cerebellum. We injected a retrograde AAV (Tervo et al., 2016) into cerebellum to express cre in bPN neurons, followed by AAVs to express TVA and rabies glycoprotein in bPN and subsequent glycoprotein-deleted EnvA rabies virus after one week (see method details). With this approach, we could visualize rabies-infected neurons in neocortex, and GAD+ rabies-infected neurons in MDJ (Figure S4). In other mice, we injected retrograde AAV in the bPN to express Cre in all afferent areas to bPN. Subsequent injections with AAVs to express TVA and optimized G protein were made into the MDJ, followed by pseudotyped rabies virus. In these experiments ( Figure 4A, n = 3 mice) we observed widespread labeling of rabies virus throughout the brain (Figures 4B and 4F). As expected, starter neurons in the MDJ were GAD+ (Figure 4C), and we could observe many GAD+ axon terminals in bPN from these neurons (Figure 4D). This confirmed that MDJ GAD+ neurons, indeed, make contacts in the bPN. In the neocortex of these mice, we observed rabies-virus labeled pyramidal neurons in the deep layers of neocortex iScience Article ( Figures 4B and 4E). These results show that GABAergic neurons in the MDJ that project to the bPN receive inputs from neocortex.

DISCUSSION
We show that the bPN receives prominent inputs from neocortex and from MDJ. Using whole-cell recordings and optogenetic stimulation we show that the bPN receive synaptically depressing glutamatergic inputs from M1 and GABAergic inputs from MDJ that show remarkably little short-term plasticity. Furthermore, we did not observe convergence of M1 and MDJ inputs onto single bPN neurons, and our paired recording data show that convergence via locally connected bPN neurons is exceedingly rare at best. This suggests that M1 and MDJ represent separate streams of information through the bPN. Finally, using Rabies-virus tracing we show that MDJ neurons that project to the bPN receive prominent input from neocortical output neurons. Thus, our results show and characterize a previously unknown source of GABA to bPN from the MDJ, which could provide sign-inversed inputs from neocortex to cerebellar granule cells.
It has long been unclear whether inhibitory inputs to the bPN exist and from where their afferents arise. Several sources of GABAergic inputs have been suggested, including the zona incerta, anterior pretectal nucleus, and cerebellar nuclei . Local interneurons in the bPN have also been suggested to provide GABAergic inhibition Mihailoff 1985, 1990;Brodal et al., 1988). We did not find a pronounced number of afferents from the sources suggested previously. Instead, retrograde tracing from the bPN produced labeling in the caudal part of the MDJ, which we confirmed to be a source of GABAergic inputs to bPN neurons. The MDJ is a region located in the tegmentum that receives prominent inputs from neocortex and richly innervates the inferior olive with glutamatergic afferents (De Zeeuw et al., 1998;Kubo et al., 2018;Wang et al., 2022). In a functional study, it was suggested that the inhibition of bPN neurons could be induced via a polysynaptic pathway from motor cortex (Guo et al., 2021), and inhibition in bPN neurons has been observed in vitro after the stimulation of the cerebral peduncle and the tegmentum . GABAergic inhibition to bPN neurons, therefore, seems to be completely extrinsic. However, there may still be sources other than the MDJ that provide inhibition to the bPN. For example, in this study, we did not consider possible glycinergic afferents (Aas and Brodal 1990). At the same time, several midbrain regions have been suggested to also provide inputs to bPN neurons, though it remains unclear which neurotransmitters are involved (Mihailoff et al., 1989). Interestingly, we did not observe excitatory inputs during the optical stimulation of MDJ afferents in our experiments, suggesting that MDJ inputs to bPN are exclusively GABAergic.
The bPN are thought to integrate incoming motor and sensory information from the neocortex at the single-cell level (Potter et al., 1978). Indeed, some neurons in the bPN respond only to movement, whereas others are responsive to multiple modalities such as movement and cue (Guo et al., 2021), though this might also reflect earlier integration in the cortex. The precise extent of convergent streams in the bPN remains an important unanswered question. Based on anatomical tracing data, it is suggested that excitatory afferents from different regions could converge onto single bPN neurons despite the general topographical organization Lee and Mihailoff 1990;Schwarz and Thier 1999;Leergaard 2003;Leergaard and Bjaalie 2007). It is, therefore, striking that we did not find any convergence of excitatory M1 and inhibitory MDJ inputs in the bPN. Furthermore, we have found no evidence of synaptic connectivity between bPN neurons. Although we cannot unequivocally rule out the presence of synaptically connected bPN neurons, based on our results and a similar earlier report we expect that such connections would be too sparse to be functionally relevant (see also Mö ck et al., 2006). Thus, information carried by MDJ and M1 inputs very likely remains segregated in the output of the bPN.
In addition to targeting different populations, we show that GABAergic MDJ and glutamatergic M1 inputs are also markedly different in their short-term plasticity. M1 inputs show clear synaptic depression across all tested frequencies, which is particularly prominent during stimulation at relatively high frequencies.
Conversely, MDJ inputs undergo little synaptic plasticity except for slight depression towards the end of a pulse train at higher stimulation frequencies. These differences are important, as synaptic plasticity plays an important role in shaping the activity of neurons (Silver 2010). Layer 5 neurons provide the output from neocortex to bPN (Tervo et al., 2016), and respond with changes in firing rate up to 50 Hz during movement (Park et al., 2021;Guo et al., 2021 iScience Article (Guo et al., 2021). We took care to estimate short-term synaptic plasticity by stimulating over axons with short pulses of light, while avoiding stimulating over boutons or somata . However, even these estimates of short-term plasticity can be confounded by the choice of opsins, generating artificial depression. Indeed, ChrimsonR, a channelrhodopsin with relatively slow kinetics, showed more pronounced depression than Chronos, a faster variant (Klapoetke et al., 2014), indicating that synaptic plasticity estimates are also depending on the speed of the opsin. Still, stimulation with ChrimsonR and Chronos yielded comparable results for frequencies up to 20 Hz, indicating that our estimates of synaptic plasticity probably are not an artifact from optogenetic stimulation in this frequency range. However, we do notice subtle synaptic depression of MDJ inputs in the 50-100 Hz range, which is possibly an optogenetic stimulation artifact. Thus, we cannot exclude the possibility that the short-term plasticity of GABA signaling from the MDJ is frequency dependent.
The mechanism by which inhibition in the bPN contributes to voluntary motor control remains an important question to be addressed. Although our findings do not decisively point to one single mechanism, we are able to rule out several hypotheses. Our rabies tracings suggest that GABAergic MDJ afferents to the bPN could be recruited by cortical activation. This possibly explains why optogenetic stimulation of the motor cortex induces diverse changes in the firing rate of bPN neurons (Guo et al., 2021). Although the effects of inhibition were previously studied by optogenetically silencing the entire bPN (Wagner et al., 2019), our data show that GABAergic MDJ inputs specifically target bPN neurons that likely receive excitatory inputs from cortical areas other than M1. Therefore, we consider it unlikely that feedforward inhibition from the MDJ serves as a gating mechanism (Crowley et al., 2009;Geborek et al., 2013). Furthermore, the phasic nature of GABA signaling in the bPN suggests a timing-dependent mechanism rather than gain adjustment (Silver 2010). It is, therefore, more likely that the MDJ specifically provides the bPN with a negative signal based on neocortex inputs to MDJ. This is further supported by the fact that we only observe purely GABAergic inputs coming from the MDJ. In this arrangement, the bPN would transmit one direct positive signal based on corticopontine inputs, and one negative signal based on cortico-MDJ-pontine inputs. This would greatly enrich the inputs that are provided to the input layer of the cerebellar cortex, which would support cerebellar learning (Chabrol et al., 2015;Cayco-Gajic et al., 2017;Straub et al., 2020).
Thus, we propose that the bPN are more than a passive relay for information destined for the cerebellum. The seemingly targeted projection of GABAergic MDJ afferents suggests that inhibition to the bPN fulfills a specific role that is likely timed with cortical activation. An important remaining question is how the activation of the MDJ impacts voluntary motor behavior and whether its timing differentially affects the outcome of a planned movement. Further investigations of inhibition in the bPN should focus on performance during a learned voluntary behavioral task in order to address these questions. Given its crucial position within the cerebro-cerebellar circuit, expanding our knowledge of bPN functionality will likely aid in further understanding the mechanisms underlying voluntary motor control and cerebellar learning.

Limitations of the study
In the present study, we used viral tracing and optogenetics as tools to study M1 and MDJ inputs to the bPN, as well as their synaptic plasticity and possible convergence onto single neurons. Even though channelrhodopsin is a widely used light-gated ion channel, the kinetics of channelrhodopsin is not shared across variants, which may differentially affect estimations of synaptic plasticity. We based our conclusions on input data recorded using Chronos. Despite having the fastest on-off kinetics among all available variants, we here confirm that Chronos-induced axonal stimulation is still only reliable up to firing frequencies of around 50 Hz. This precludes us from accurately describing synaptic plasticity of M1 and MDJ inputs during high frequency (i.e. >50 Hz) stimulation. Still, owing to the known properties of Chronos and the comparable results between Chronos and ChrimsonR, the estimates of synaptic plasticity for both M1 and MDJ inputs in the lower frequency range (10-20 Hz) are likely to be accurate.
To investigate whether M1 and MDJ inputs converge at the single-cell level we analyzed responses to fullfield light stimulation. Although in none of the neurons recorded we observed responses to both M1 and MDJ afferent stimulation, it is possible that such neurons do exist and can be found through sampling a much larger number of neurons.

OPEN ACCESS
iScience 25, 104641, July 15, 2022 9 iScience Article We hypothesized that the inhibition of M1 inputs in the bPN could have important functional implications. However, if we missed convergence of M1 and MDJ afferents onto single bPN neurons as a result of undersampling, we still expect that such convergence is unlikely to hold substantial functional significance owing to its rarity.
We then investigated whether there are local networks in the bPN by simultaneously patching pairs of neurons. We performed these experiments in brain slices in vitro. It is important to note that many dendrites and axons are cut during the slicing procedure, particularly at this slice thickness (i.e. 250 mm), thus lowering the chance of finding synaptically connected neurons. We tried to minimize this issue by sampling from pairs in two slicing orientations. However, owing to this limitation, we cannot rule out the possibility of a few local connections in the bPN. Ideally, the presence of local networks in the bPN is studied in vivo where the complete bPN network remains intact.
Finally, the use of rabies viral tracing has taken a flight over the past few years, but several shortcomings have been reported, such as an undersampling of presynaptic partners, selective tropism, and heavy dependence on helper plasmid and injection timing (Ginger et al., 2013). In our current study, the use of glycoprotein-deleted rabies tracing could have biased our results to a particular neuronal population, potentially missing another population of neurons presynaptic to bPN neurons. At the same time, the viral tracing results might underestimate the total number of neurons impinging on bPN neurons.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

AUTHOR CONTRIBUTIONS
L.W. designed the study. A.K. performed and analyzed electrophysiological experiments. L.W. performed and analyzed anatomical experiments. All authors checked the data analysis. A.K. and L.W. wrote the article. All authors critically revised the article.

Paired whole-cell recordings
Sagittal or coronal slices were prepared for paired recordings. Up to three neurons were recorded at the same time, and potential connections between neurons were probed by evoking spike trains successively in each neuron. Ten action potentials were evoked presynaptically using current injections of 2 nA at 50 Hz, followed by a single current injection after 500 ms. Cells were kept at or around resting membrane potential throughout recording to detect EPCSs. Current clamp recordings were acquired at a 50.0 kHz sample rate with a 10 kHz low pass filter. We did not compensate for the liquid junction potential.
Cells from paired whole-cell patch clamp experiments were analyzed when: (1) stimulation evoked action potentials (APs); (2) cells did not have a negative leak current exceeding 500 pA; (3) recordings had a stable resting membrane potential; (4) at least fifteen sweeps were collected. To detect connections, we looked for EPSCs in the average postsynaptic response in the first 18 ms after the AP to accommodate for monoand disynaptic connections. Then, the postsynaptic response was determined by calculating the average amplitude over a 1-ms time window of the peak amplitude.

Quantification of labeled neurons
Retrobead tracing was analyzed by first marking all labeled neurons by hand via cellcounter in matlab (https://github.molgen.mpg.de/MPIBR/CellCounter) and then aligning the sections with labeled neurons using the wholebrain tool in R (Fü rth et al., 2018) to the Allen Brain Atlas between bregma +3.0 and À5.6. For rabies tracing, sections were visualized under a Zeiss Axio Observer Z1, photographed and counted by hand.

QUANTIFICATION AND STATISTICAL ANALYSIS
All statistical analyses were performed in Igor Pro version 7. To compare the effects of pharmacological blockers on synaptic input, one sample t-tests were performed on data normalized to ACSF. For TTX and TTX+4AP experiments, repeated measures ANOVA was used on measurements normalized to the ACSF condition. Post-hoc testing to establish different groups was done using the paired t-test. p values less than 0.05 were considered statistically significant.