Prefrontal cortical ChAT-VIP interneurons provide local excitation by cholinergic synaptic transmission and control attention

Neocortical choline acetyltransferase (ChAT)-expressing interneurons are a subclass of vasoactive intestinal peptide (ChAT-VIP) neurons of which circuit and behavioural function are unknown. Here, we show that ChAT-VIP neurons directly excite neighbouring neurons in several layers through fast synaptic transmission of acetylcholine (ACh) in rodent medial prefrontal cortex (mPFC). Both interneurons in layers (L)1–3 as well as pyramidal neurons in L2/3 and L6 receive direct inputs from ChAT-VIP neurons mediated by fast cholinergic transmission. A fraction (10–20%) of postsynaptic neurons that received cholinergic input from ChAT-VIP interneurons also received GABAergic input from these neurons. In contrast to regular VIP interneurons, ChAT-VIP neurons did not disinhibit pyramidal neurons. Finally, we show that activity of these neurons is relevant for behaviour and they control attention behaviour distinctly from basal forebrain ACh inputs. Thus, ChAT-VIP neurons are a local source of cortical ACh that directly excite neurons throughout cortical layers and contribute to attention.

T he neurotransmitter acetylcholine (ACh) shapes activity of cortical neurons and supports cognitive functions such as learning, memory and attention [1][2][3] . Rapid ACh concentration changes in rodent medial prefrontal cortex (mPFC) occur during successful stimulus detection in a sustained attention task 4,5 . Traditionally, it is assumed that neocortical ACh is released exclusively from terminals of axonal projections whose cell bodies reside in basal forebrain (BF) nuclei 6,7 . Chemical lesions of cholinergic BF projections impair attention behaviour [8][9][10][11][12] and optogenetic activation of BF cholinergic neurons can mimic ACh concentration changes typically observed during attention behaviour 11 . However, well over 30 years ago, interneurons intrinsic to the neocortex have been described that express the ACh-synthesizing enzyme choline acetyltransferase (ChAT) [13][14][15] . Whether these neurons are a local source of ACh in the prefrontal cortex relevant for attention is not known.
ChAT-expressing interneurons are a sparse population of about 1% of all cortical neurons, which are more abundantly present in superficial cortical layers 2/3 (L2/3) than in deep layers 16 . They have either a bipolar or multipolar morphology and express vasoactive intestinal peptide (VIP) 13,14,17 . In mouse cortex, about 15% of VIP neurons express ChAT 18 , while in the PFC of rats about 30% of VIP neurons express ChAT 16 . Indirect evidence suggested that these ChAT-VIP neurons release ACh in neocortical circuits upon activation, inducing an increase in spontaneous glutamatergic synaptic transmission received by L2/3 pyramidal neurons 19 . Despite molecular, morphological and physiological characterizations, technical limitations thus far prevented a direct demonstration of whether these ChAT-expressing VIP interneurons release ACh. Moreover, basal forebrain cholinergic neurons that project to the neocortex have been shown to form direct point-to-point synapses with several types of neurons in different layers, thereby modulating their activity on a millisecond time scale [20][21][22][23][24] . Activation of ChAT-VIP interneurons can slowly alter local synaptic activity 19 , but it is unknown whether ChAT-VIP interneurons do this via direct cholinergic synaptic transmission, or whether they modulate local neuronal activity more diffusely.
Neocortical circuits contain distinct classes of interneurons with characteristic innervation patterns of local cortical neurons [25][26][27][28] . Fast-spiking (FS), parvalbumin-expressing (PV) interneurons perisomatically innervate pyramidal neurons, while low-threshold spiking (LTS), somatostatin-expressing (SST) genes target more distal regions of dendrites 29 . GABAergic VIP neurons inhibit PV and SST interneurons, thereby disinhibiting pyramidal neurons 28,30,31 . Single cell transcriptomic analysis of cortical neurons has shown that distinct subtypes of VIP interneurons exist with unique gene expression profiles 18,25 . Whether VIP interneuron subtypes are functionally distinct is not known 32 . It is also not known whether ChAT-expressing VIP interneurons show similar innervation patterns, specifically targeting neighbouring PV and SST interneurons, and activating disinhibitory pathways.
Here, we address these issues and focus in particular on elucidating how local neuronal circuitry in various layers of the mPFC are affected by ChAT-VIP neuron activation. We find that ChAT-VIP interneurons directly excite local interneurons and pyramidal neurons in different mPFC layers with fast cholinergic synaptic transmission. In addition, we show that despite their sparseness, activity of ChAT-VIP neurons is involved in sustained attentional performance in freely moving animals in a manner distinct from basal forebrain cholinergic inputs to the prefrontal cortex.

Results
Fast cholinergic synaptic transmission by ChAT-VIP neurons. Previous studies in mice have shown that activation of ChAT-VIP interneurons increases spontaneous excitatory postsynaptic potentials (EPSPs) in layer 5 pyramidal neurons 19 . However, it is unresolved whether ChAT-VIP interneurons directly innervate other neurons in the cortex. To address this, we first expressed channelrhodopsin-2 (ChR2) in ChAT-VIP interneurons in the mPFC of ChAT-cre mice 33 (Fig. 1a) and recorded from L1 interneurons since these neurons are known to reliably express nicotinic acetylcholine receptors (nAChRs) in other neocortical areas [34][35][36] . In mouse mPFC, L1 interneurons received miniature excitatory postsynaptic currents (mEPSCs) mediated by nAChRs ( Supplementary Fig. 1). All brain slice physiology experiments in this study were done in the presence of glutamate receptor blockers (DNQX, 10 µM; AP5, 25 µM). We made simultaneous whole-cell patch-clamp recordings of EYFP-positive ChAT-VIP neurons in L2/3 and nearby L1 interneurons in mouse mPFC (Fig. 1b). EYFP-positive neurons showed similar morphology, ChAT, VIP, CR, GAD expression patterns ( Fig. 1a; Supplementary Figs. 2 and 3) and action potential (AP) profiles (Fig. 1c) as was reported previously 14,16,19 . First, we triggered single APs in presynaptic ChAT-VIP interneurons by short (1 ms) electrical depolarization of the membrane potential (Fig. 1d, n = 11). This resulted in inward currents in postsynaptic L1 interneurons that either were fast and lasted up to 10 ms (Fig. 1d, top traces), or had slower kinetics and lasted over 100 ms (Fig. 1d, bottom traces). The fast postsynaptic currents were fully blocked by methyllycaconitine (MLA, 100 nM), an antagonist of α7-subunit-containing nAChRs (Fig. 1d, grey trace, Fig. 1j). The slow postsynaptic current was fully blocked by DHßE (10 µM), an antagonist of β2subunit-containing nAChRs (Fig. 1d, bottom, grey trace, Fig. 1j). Postsynaptic currents occurred time-locked to presynaptic APs with onset delays of 1.26 ± 0.02 ms (average ± sem) from the peak of the AP (Fig. 1e), suggesting synaptic transmission. Postsynaptic currents showed substantial amplitude fluctuations, in line with synaptic transmission, with an average amplitude of −30 ± 15 pA (Fig. 1f, j). Synaptic transmission between ChAT-VIP and L1 interneuron was quite reliable: the majority of unitary synapses transmitted ACh at each presynaptic AP and average failure rates were 35 ± 10% (Fig. 1f, k). Single cell mRNA expression data from the Allen Institute for Brain Science database on mouse cell types 18,25 shows that ChAT-VIP neurons express the vesicular acetylcholine transporter (VAChT) encoded by the SLC18A3 gene ( Supplementary Fig. 3E). Thus, ACh is likely released by means of vesicles from these neurons.
Next, in the same recordings, we induced APs in presynaptic ChAT-VIP neurons by activating ChR2 with brief blue-light pulses (Fig. 1g). This induced similar inward currents in postsynaptic L1 interneurons with similar current amplitudes (−39 ± 20 pA; Fig. 1g, i, j) and that were blocked by the nicotinic receptor antagonists MLA and DHßE (Fig. 1g, j). The distribution of synaptic onset delays, calculated from the peak of the presynaptic AP, was significantly shorter than with electricallyinduced APs (0.68 ± 0.02 ms; Mann-Whitney test P < 0.00001; Fig. 1h). This lower synaptic onset delay with optogenetic stimulation could result from more rapid activation of additional ChR2-expressing ChAT-VIP neurons in the slice that innervate the same L1 interneuron, or from direct depolarization of presynaptic terminals by blue-light-mediated ChR2 activation, or both. Failure rates of postsynaptic current induction and postsynaptic current amplitude distribution in these recordings were similar with electrical and light-induced presynaptic AP induction ( Fig. 1i-k). To test how abundant ChAT-VIP input to L1 interneurons is in mouse prefrontal cortical slices, we recorded from an additional set of 118 L1 interneurons and activated ChAT-VIP neurons with blue-light pulses. Overall, we found that 36% of L1 interneurons received inputs from ChAT-VIP neurons (n = 42 of 118 neurons, Fig. 1l) and these were all blocked by a cocktail of nicotinic AChR blockers (MLA and DHβE). Thus, in mouse mPFC, ChAT-VIP interneurons provide fast cholinergic synaptic input to L1 interneurons mediated by nicotinic AChRs.
ChAT-VIP neurons are more abundant in the mPFC of rats than in mouse cortex 16,18 . To test whether ChAT-VIP neurons more reliably innervate L1 interneurons in rat neocortex, we expressed ChR2 in ChAT-VIP interneurons in the mPFC of ChAT-cre rats 37 . Following mPFC injections, we did not observe significant retrograde labelling of cells in the basal forebrain ( Supplementary Fig. 4A, B). In rat prefrontal cortex, EYFP-positive L2/3 ChAT-VIP neurons also had a bipolar a b   cholinergic synaptic transmission by ChAT-VIP cells. a EYFP-labelled ChAT-VIP interneuron targeted for recording in an acute mouse mPFC brain slice (DIC-IR microscope). Scale bar 100 μm. b Schematic of simultaneous recordings of presynaptic ChAT-VIP interneurons and postsynaptic L1 interneurons. c Membrane potential responses of a L1 interneuron (top) and L2/3 ChAT-VIP interneuron (bottom) to depolarizing (+200 pA) and hyperpolarizing (−150 pA) somatic current injection. d Example traces of a synaptically-connected ChAT-VIP cell and a L1 interneuron. Top trace: short step depolarization. Second trace: Presynaptic AP in ChAT-VIP interneuron. Third trace: fast postsynaptic inward current of the L1 interneuron (black trace, average of 45 sweeps), blocked by nAChR antagonists (MLA 100 nM, grey trace). Fourth trace: short step depolarization. Fifth trace: Presynaptic AP in ChAT-VIP interneuron. Bottom trace: slow postsynaptic inward current of the L1 interneuron (black trace, average of 57 sweeps), blocked by nAChR antagonists (DHßE 10 µM, grey trace). e Histogram of onset delays of postsynaptic currents relative to the peak amplitude of the electrical stimulationinduced presynaptic AP (from n = 11 connected cell pairs). f Amplitudes of successive unitary synaptic currents in the same ChAT-VIP L1-interneuron cell pair as in (d). g Recordings from the same neuron pair as in (d), but now AP firing was induced by activating ChR2 with a blue-light flash. Traces and graph as in (c). Here and in all other figures, for all example traces of responses to optogenetic stimulations 20 sweeps were averaged. h Histogram of onset delays of postsynaptic currents relative to the peak amplitude of the ChR2-induced presynaptic AP (from n = 11 connected cell pairs). i Amplitudes of successive unitary synaptic currents in the recording of the same ChAT-VIP L1-interneuron cell pair in (f), but with ChR2-induced presynaptic APs. j Summary of postsynaptic current amplitudes. Box plot centre line: median; bounds of box: 25th to 75th percentile; whiskers: min-max data bound. k Summary of failure rates in the same 11 L1 interneurons. l Postsynaptic response incidence in a separate set of 118 L1 interneurons following ChR2-induced ChAT-VIP interneuron activation. morphological appearance (Fig. 2a), as was originally reported 14,16 , confirming their identity. Upon activation of ChR2, ChAT-VIP interneurons fired APs and simultaneously recorded L1 interneurons showed postsynaptic inward currents (Fig. 2b). In all recorded L1 interneurons (n = 119), blue-light activation of ChR2 expressed by ChAT-VIP neurons generated postsynaptic depolarizations and inward currents that were blocked by nAChR blockers (Fig. 2c). These currents were either mono-phasic, consisting of only a fast or slow component, or were biphasic, consisting of a fast and a slow component (Fig. 2d), reminiscent of synaptic fast α7-containing nAChR and slow β2containing nAChR currents observed in Fig. 1d, g and expressed by L1 interneurons in sensory cortical areas [34][35][36] . Indeed, the fast current component was selectively blocked by MLA, while the slow component was selectively blocked by DHßE (Fig. 2d, e).
Since ChAT-VIP interneurons can co-express the acetylcholine (ACh) synthesizing enzyme ChAT and the GABA synthesizing enzyme GAD ( Supplementary Fig. 3) 16,19 , we asked whether these neurons release GABA in addition to ACh. To test this, the membrane potential of rat mPFC L1 interneurons was held at 0 mV in the presence of nAChR blockers (Fig. 2d inset). Blue-light activation of ChR2-expressing ChAT-VIP cells evoked fast outward currents in 11% of the cells (n = 13/119), which were blocked by gabazine (10 µM; Fig. 2d, e). Thus, while all L1 interneurons in rat mPFC received fast cholinergic inputs from ChAT-VIP cells mediated by nAChRs, a minority of L1 neurons also received GABA from ChAT-VIP neurons (Fig. 2f). In all figures below (Figs. 3-6), data from rat mPFC are shown.
Direct excitation, but no disinhibition by ChAT-VIP neurons. VIP interneurons have been shown to disinhibit L2/3 pyramidal neurons by inhibiting activity of fast-spiking, PV-expressing interneurons and low-threshold spiking, SST-expressing interneurons 28,30,31 . To test whether ChAT-VIP interneurons form disinhibitory circuits in L2/3, we made whole-cell patchclamp recordings of L2/3 pyramidal neurons in rat mPFC and triggered activity in ChR2-expressing ChAT-VIP interneurons by applying blue-light pulses. Light-induced activation of ChAT-VIP interneurons did not induce GABAergic synaptic currents in pyramidal neurons, but induced depolarizing inward currents in a minority of pyramidal neurons (n = 3 of 18; Fig. 3a). These inward currents were blocked by nAChR blockers DHßE and MLA. Next, we analysed spontaneous inhibitory postsynaptic currents (sIPSCs) received by L2/3 pyramidal neurons. Lightinduced activation of ChAT-VIP interneurons did not alter the frequency of sIPSCs received by L2/3 pyramidal neurons (Fig. 3b, c), indicating that activity of ChAT-VIP neurons did not change inhibition received by L2/3 pyramidal neurons. To test whether ChAT-VIP neurons target and inhibit other local interneuron types, we recorded from rat mPFC FS (Fig. 3d) and LTS ( Fig. 3f) interneurons, as defined by their AP firing profiles, while triggering activity in ChR2-expressing ChAT-VIP interneurons by applying blue-light pulses (Fig. 3e, g). We did not observe any GABA-mediated inhibitory postsynaptic currents at 0 mV membrane potential in both interneuron types following activation of ChAT-VIP interneurons (Fig. 3e, g). However, a subgroup of FS (n = 4/6) as well as LTS (n = 5/8) interneurons showed small inward currents at −70 mV that were mediated by fast α7containing or slow β2-containing nAChRs and were blocked after application of the nAChR antagonists DHßE and MLA (Fig. 3e, g). Thus, we find no evidence that ChAT-VIP neurons form disinhibitory circuits in L2/3, as has been reported for other VIP interneurons, but we do find evidence that ChAT-VIP interneurons directly excite subgroups of local interneurons as well as a minority of L2/3 pyramidal neurons.
Cholinergic synaptic inputs to L6 pyramidal neurons. Previous studies have shown that a majority of layer 6 pyramidal neurons express nAChRs 38,39 and these neurons can be activated by cholinergic inputs from the BF 21,22,40 . EYFP-positive fibres of ChAT-VIP cells show abundant arborization in deep layers of rat mPFC ( Fig. 4a; Supplementary Fig. 2). We therefore asked whether L6 pyramidal neurons receive direct inputs from ChAT-VIP interneurons. To test this, we made whole-cell patch-clamp recordings from rat mPFC L6 pyramidal neurons combined with activation of ChR2-expressing ChAT-VIP interneurons by applying blue-light pulses (Fig. 4b, c). Seventy-one percent (n = 20/28) of recorded L6 pyramidal neurons showed nAChR antagonist sensitive inward currents with slow kinetics (Fig. 4b,  d). Although the amplitude of these currents was on average about 5 pA, the currents lasted about 1 s and ChAT-VIP activation resulted in a significant depolarization of the membrane potential (Fig. 4b, c) due to the relatively high membrane resistance of these cells 22,34 . Six of the L6 pyramidal cells showed an additional gabazine-sensitive fast outward current at 0 mV in the presence of nAChR blockers (Fig. 4c, d). These findings show that 20 out of 28 L6 pyramidal neurons receive direct cholinergic inputs from local ChAT-VIP interneurons, and 6 out of 28 L6 pyramidal neurons received both ACh and GABA (Fig. 4e).
Consequences of co-transmission of ACh and GABA. ChAT-VIP cells activate postsynaptic nAChRs that depolarize pyramidal neurons and interneurons in L1, L2/3 and L6. It is somewhat surprising that some of these postsynaptic neurons also receive GABA and show inhibitory GABAR currents. It is not known what the consequences for the excitability of postsynaptic cells is of co-transmission of GABA and ACh. Typically, GABAR and nAChR currents had distinct kinetics (Fig. 5a, b). The majority of excitatory nAChR-mediated synaptic responses had slow kinetics with rise times of 155.5 ± 26.5 ms (n = 9/13 postsynaptic L1 cells with combined GABAR and nAChR currents; Fig. 5a, c, d). GABAergic currents had much faster kinetics that decayed back to baseline in about 30 ms (Fig. 5b, d). In only four postsynaptic L1 interneurons that showed a combined nAChR and GABARmediated currents we found both the fast MLA-sensitive nAChR current that would match the activation kinetics of the fast GABAR current (n = 4/13). Thus, in the majority of combined GABA and ACh responses, fast GABAR currents were followed by slower nAChR currents.
Activation of GABAR currents could lead to shunting inhibition, preventing or postponing subsequent AP firing. Alternatively, hyperpolarizing GABAergic inputs can give rise to rebound excitation by deinactivation of intrinsic voltage-gated conductances [41][42][43] . The excitation induced by slow inward nAChR currents may thus theoretically be amplified by rebound excitation following the GABAergic hyperpolarization. To test whether this occurs, we recorded from L1 interneurons and monitored AP timing in response to ramp depolarizations with and without blue-light activation of ChR2-expressing ChAT-VIP interneurons in rat mPFC. First, to test the effect of the cholinergic component of ChAT-VIP input, only recordings with nAChR-mediated postsynaptic currents without GABAR currents were included (Fig. 5e). Activation by blue-light pulses of ChR2-expressing ChAT-VIP interneurons advanced the timing of the first AP during ramp depolarizations (Fig. 5f), reducing the AP onset delay (Fig. 5i). Next, we investigated whether cotransmission of GABA advances or postpones the first AP. Now, only recordings with combined nAChR/GABAR-mediated postsynaptic currents were included (Fig. 5g). Blocking GABAergic inhibition with the GABA A receptor antagonist gabazine resulted in a shortening of the delay to the first AP in L1 interneurons during ramp depolarizations (Fig. 5h, i), suggesting that the postsynaptic GABAR currents provided shunting inhibition that postponed AP firing. Gabazine did not alter the input resistance ( Fig. 5i inset), nor excitability of L1 interneurons and did not advance spiking in L1 neurons that did not show co-transmission of GABA ( Supplementary Fig. 5F). In line with these findings, at near-AP threshold membrane potentials in L1 interneurons, bluelight activation of ChR2-expressing ChAT-VIP interneurons augmented AP firing probability much more when GABARs were blocked by gabazine ( Supplementary Fig. 5A-E). Taken together, these results show that postsynaptic nAChR currents induced by ChAT-VIP interneurons directly excited L1 interneurons, increasing AP firing probability and shortening delays to first AP firing. Co-transmission of GABA provided shunting inhibition, postponing AP firing, rather than facilitating rebound excitation.
ChAT-VIP neurons are required for attention. Is activity of ChAT-VIP interneurons relevant for mPFC function during behaviour? To test this, we optogenetically inhibited ChAT-VIP cells during a well-validated task for assessing attention behaviour in rodents, the 5 choice serial reaction time task (5CSRTT) 44 ( Fig. 6; Supplementary Fig 6). In this task, animals observe a curved wall with 5 holes in an operant chamber. After an intertrial interval, typically 5 s, a cue light will briefly turn on for 1 s, in a randomly assigned hole. Following cue light presentation, animals have to respond within 2 s with a nose poke in the illuminated hole, to obtain a food reward (Fig. 6b). During a daily session, animals complete 100 trials. Since ChAT-VIP cells release ACh in the mPFC, similar to basal forebrain (BF) cholinergic inputs, we also tested whether inhibiting ChAT-VIP interneurons affected attention behaviour distinct from inhibiting BF cholinergic inputs to mPFC. Therefore, ChAT::cre rats received AAV5::DIO-EYFP-ARCH3.0 (or AAV5::DIO-EYFP in controls) injections either in the mPFC or the BF and optic fibres were placed over the prelimbic (PrL) mPFC in all groups ( Fig. 6a;  Supplementary Figs. 4 and 6). By randomly assigning half of the trials to green laser light ON and the other half to laser OFF (50 trials each, randomly mixed ON/OFF), ChAT-VIP cells or BF-to-mPFC projections were either free to fire APs or were inhibited in the same animals for five seconds during the pre-cue period when rats show preparatory attention for the upcoming stimulus presentation (Fig. 6b). For each animal, behavioural performance during laser-ON trials was compared with its own behavioural performance during laser-OFF trials. Inhibiting ChAT-VIP cells or BF-to-mPFC projections impaired response accuracy (Fig. 6c), and both inhibition of BF cholinergic neurons as well as inhibition of ChAT-VIP interneurons reduced correct responses and increased errors in each animal (Fig. 6d, e). Interestingly, no changes in any of the other Interference with the cholinergic system can produce fluctuations in attentive engagement in 5CSRTT 45,46 . Analysis of attention performance in distinct temporal phases of 5CSRTT sessions showed that inhibiting cholinergic BF-to-mPFC projections reduced accuracy of responding only during the first half of the session (early trials 0-50), but not the second half of the session (late trials 51-100) (Fig. 6h). In contrast, inhibiting mPFC 10 mV ChAT-VIP interneurons significantly reduced attention performance in the second half of the session (Fig. 6h). These results indicate that BF ChAT neurons and mPFC ChAT-VIP interneurons affect attention performance distinctly: BF cholinergic neurons support early phases of attention performance, while activity of mPFC ChAT-VIP interneurons is required to sustain attention during the late phase of the session.

Discussion
In this study, we asked how cortical ChAT-VIP interneurons affect local circuitry in the mPFC, whether they function similar to other cortical VIP cells and whether they are involved in attention behaviour. We found that ChAT-VIP interneurons release ACh locally in both mouse and rat mPFC and directly excite interneurons and pyramidal neurons in different layers via fast synaptic transmission. In contrast to regular VIP interneurons, this ChAT-expressing subtype of VIP interneurons does not inhibit neighbouring fast-spiking and lowthreshold spiking interneurons. Our experiments revealed that activity of ChAT-VIP interneurons contributes to attention behaviour in a distinct manner from activity of basal forebrain ACh inputs to mPFC: ChAT-VIP neurons support sustained attention. These findings challenge the classical view that behaviourally relevant cholinergic modulation of neocortical circuits originates solely from BF cholinergic projections in rodent brain 47 .
Various reports over the last 30 years identified neocortical ChAT-expressing VIP interneurons and these were suggested as a local source of ACh in the cortex [13][14][15][16][17][18][19]25 . Simultaneous recordings from cortical ChAT-VIP interneurons and pyramidal neurons showed an AChR-dependent increase of excitatory inputs received by pyramidal neurons following high frequency stimulation of ChAT-VIP interneurons 19 . However, no evidence was found for direct cholinergic synaptic transmission between ChAT-VIP and other neurons in cortical L2/3 48 . We took a different approach from previous studies by recording from neuron populations that have strong nAChR expression 36,39,49 , i.e. L1 interneurons and L6 pyramidal neurons in both mouse and rat mPFC, as well as using ChR2-mediated activation of ChAT-VIP neurons. Our results suggest that ChAT-VIP interneurons form fast cholinergic synapses onto local neurons, since in unitary synaptic recordings the delay between presynaptic AP and postsynaptic response was 1.26 ms, suggesting mono-synaptic connections. Therefore, it is unlikely that ChAT-VIP interneurons triggered poly-synaptic events, exciting terminals of BF neurons and triggering ACh release from these terminals. Fast cholinergic inputs from ChAT-VIP neurons are more abundant in rat mPFC L1 interneurons than in mouse mPFC, in line with a larger percentage of VIP cells expressing ChAT in rat cortex 16 . Von Engelhardt et al. 19 did not observe nAChR currents activated by mouse ChAT-VIP cells in other L3 interneurons, which we did find in rat mPFC. This may be due to species differences or brain region difference in the two studies. Nevertheless, our findings show that in addition to cholinergic fibres from the BF, ChAT-VIP interneurons act as a local source of ACh modulating neuronal activity in mPFC.
Regular cortical VIP interneurons disinhibit local pyramidal neurons by selectively inhibiting somatostatin (SST) and parvalbumin (PV)-expressing interneurons 26,28,30,31,50 . In contrast, we did not find evidence that ChAT-VIP neurons form disinhibitory circuits. Low-threshold spiking and fast-spiking interneurons receive exclusively cholinergic excitatory inputs and no GABAergic inhibitory input from ChAT-VIP interneurons. Prefrontal cortical ChAT-VIP neurons also did not indirectly disinhibit L2/3 pyramidal neurons through excitation of L1 interneurons 34 . In mouse auditory cortex, fear-induced activation of L1 interneurons by cholinergic inputs from the BF results in feed-forward inhibition of L2/3 FS interneurons and disinhibition of L2/3 pyramidal neurons 34 . ChAT-VIP interneurons in the mPFC might in principle play a similar role in exciting L1 interneurons as BF cholinergic inputs do in mouse auditory cortex. However, in our experiments we did not find evidence that ChR2-mediated activation of ChAT-VIP neurons altered ongoing inhibition and spontaneous inhibitory inputs to L2/3 pyramidal neurons. In contrast, we found that ChAT-VIP interneurons directly targeted a subgroup of L2/3 pyramidal neurons and provided direct excitation to these pyramidal neurons.
Recent anatomical and functional evidence shows that VIP interneurons in rodent brain are morphologically and functionally diverse and that prefrontal cortical VIP cells can directly target pyramidal neurons [51][52][53] . Both multipolar and bipolar VIP cells form synapses on apical and basal dendrites of pyramidal neurons in superficial and deep layers and VIP neurons directly inhibit pyramidal neuron firing [51][52][53] . Frontal cortical VIP cells rapidly and directly inhibit pyramidal neurons, while they can also indirectly excite these pyramidal neurons via parallel disinhibition. These findings suggest that not all VIP cell subtypes adhere to targeting only other types of interneurons, and regulating cortical activity through disinhibition only. VIP interneurons represent about 15% of all cortical interneurons in mouse brain, and recent single cell RNA sequencing profiling identified 12 different molecular VIP-positive subtypes, of which 3 types express ChAT 18,25 . Our findings show that ChAT-VIP interneurons project to both interneurons as well as pyramidal neurons and directly excite them, in contrast to most regular VIP interneurons. Thereby, activation of ChAT-VIP interneurons in L2/3 of the mPFC can lead to increased excitability of inhibitory as well as excitatory neurons.
In mouse brain, cholinergic fibres from BF neurons can cotransmit the excitatory neurotransmitter ACh with the inhibitory neurotransmitter GABA in the cortex 41,42,54 . We find here that in rat mPFC, a minority of L1 interneurons (11%) and L6 pyramidal neurons (21%) receive co-transmission of GABA and ACh from the ChAT-VIP interneuron population. Whether both neurotransmitters are actually released from the same ChAT-VIP neuron awaits direct demonstration, but may be likely given simultaneous GAD and ChAT expression by ChAT-VIP neurons. How these two neurotransmitters interact with each other, and what the effect on postsynaptic neurons is, was under debate 42,55 . Nicotinic AChRs show a range of activation kinetics. Heteromeric β2-subunit-containing nAChR currents have relatively slow activation kinetics with 20-80% rise time of 150 ms, while Arroyo et al. 20 showed that homomeric α7-subunit-containing nAChR currents activate rapidly with time constants of 2.6 ms and decay time constants of 4.9 ms in neocortical L1 interneurons 20 , comparable to kinetics of synaptic GABAergic currents. This suggests that when the fast α7-subunit-mediated nAChR currents are induced in L1 interneurons by activation of ChAT-VIP cells, the additional GABAR currents that have similar kinetics will shunt the cholinergic depolarization. In case L1 interneurons express only the slower β2-subunit-containing nAChR currents, cotransmission of GABA could augment the excitatory action of ChAT-VIP neurons by rebound excitation. However, depolarizing ramps or near-threshold AP firing probabilities revealed that GABA acted inhibitory in both cases, decreasing spiking probability. Therefore, co-transmission of GABA in addition to ACh postpones AP firing in postsynaptic neurons compared with synaptic transmission of only ACh, forcing a temporal window of inhibition followed by excitation.
This scheme of postsynaptic GABAR current and AChR current interaction will depend on the physical mode of release, whether these neurotransmitters are released from the same ChAT-VIP cell and the same terminals or not 42,54 . In our experiments using wide-field illumination to activate ChR2 on multiple ChAT-VIP neurons simultaneously, we could not distinguish whether ACh and GABA were released from the same nerve terminals or even from the same ChAT-VIP neuron. It is also not known whether GABA and ACh are packaged in the same vesicles or separately. As such, it is not clear whether cotransmission of GABA and ACh occurs from single ChAT-VIP neurons. However, it is unlikely that ChAT-VIP neurons release only GABA in the mPFC, since we never observed isolated postsynaptic responses mediated only by GABARs, whereas all postsynaptic responses in rat L1 interneurons following ChAT-VIP neuron activation consist of only AChR currents. Thus, regardless of the mode of co-transmitter release, ChAT-VIP activity results in excitation and increased spiking probability throughout the mPFC layers. In addition to depolarization by nAChR activation, calcium influx through these receptors 56 could potentially affect postsynaptic cell function.
Cholinergic signalling in the mPFC controls attention and task-related cue detection 4,11,48,57 . In contrast to the general view that ACh is solely released in the mPFC from cholinergic projections from neurons located in the BF, we present evidence that there is a second source of ACh that supports attentional performance. The different temporal requirements of activity of BF-mPFC projections and ChAT-VIP interneurons in attention suggests that the two sources of cortical ACh interact in shaping cortical network activity during attentional processing. Our findings indicate that activity of cholinergic projections from the BF is required for early phases of attention performance. In contrast, activity of ChAT-VIP interneurons supports later phases of the attention task. Given the sparseness of these neurons, only 15-30% of VIP interneurons express ChAT 16,18 , it is surprising that inhibition of this small population in a single brain region has an effect on brain function and behaviour. Even though activation of archaerhodopsin expressed by ChAT-VIP cells or axons may lead to increased activity 58 or suppression of activity, our experiments do show that specific manipulation of these cell populations affect attention. How synaptic connectivity in local PFC circuitry and activity of these neurons actually gives rise to or contributes to attention behaviour in a mechanistic sense is beyond our understanding at this moment.
Recent findings indicate that BF cholinergic neurons are preferentially activated by reward and punishment, rather than attention 59 . Hangya et al. 59 suggested that the cholinergic basal forebrain may provide the cortex with reinforcement signals for fast cortical activation, preparing the cortex to perform a complex cognitive task in the context of reward. Still, rapid transient changes in ACh levels in the mPFC may support cognitive operations 5 and may mediate shifts from a state of monitoring for cues, to generation of a cue-directed response 11,57 . Since we find that activity of ChAT-VIP neurons is required during sustained attention, it remains to be determined whether ACh release from local ChAT-VIP interneurons is responsible for or contributes to the generation of cue-directed responses.

Methods
Animals. All experimental procedures were in accordance with European and Dutch law and approved by the animal ethical care committees of the VU University and VU University Medical Center, Amsterdam. Mice: experiments were done on acute brain tissue of both female and male ChAT-IRES-Cre mice (JAX laboratory, mouse line B6;129S6-Chattm2(cre)Lowl/J 33 ). Average age at time of injection was 9 weeks; average age at time of sacrifice was 16 weeks. Rats: male ChAT-cre rats (kindly provided by the Deisseroth lab 37 ) were bred in our facility, individually housed on a reversed 12 h light-dark cycle (lights OFF: 7 a.m.) and were 12-13 weeks old at experiment start. Only when assigned to behavioural experiments, rats were food deprived (start 1 week before operant training, 85-90% of the free-feeding body weight). Water was provided ad libitum.
For blocking nAChRs the following antagonists were bath applied: DHßE (10 µM) and Methyllycaconitine (MLA, 100 nM). GABAA receptor mediated responses were blocked by bath application of the antagonist gabazine (10 µM). For whole-cell recordings of EYFP-positive ChAT-VIP interneurons and other L2/3 interneurons a potassium-based internal solution was used containing (in mM): Kgluconate 135, NaCl 4, Hepes 10, Mg-ATP 2, Phosphocreatine 10, GTP (sodium salt) 0.3, EGTA 0.2. During recordings, ChAT-VIP interneurons were kept at a membrane potential of −70 mV. Whole-cell recordings of L1 interneurons and pyramidal neurons were made using a caesium gluconate-based intracellular solution containing in mM: Cs gluconate 120, CsCl 10, NaCl 8, MgATP 2, Phosphocreatine 10, GTP (sodium salt) 0.3, EGTA 0.2, HEPES 10. For paired recordings between ChAT-VIP interneurons and L1 interneurons, potassiumbased internal solution was used for both cells. Interneurons and pyramidal neurons were identified by their morphology under IR-DIC, the distance of the soma to the pia and their spiking profile. Membrane potentials were kept at −70 or 0 mV to investigate nAChR or GABAR currents.
Opsins were activated by green (530 nm, eARCH3.0) or blue light (470 nm, ChR2). Light pulses with the specific wavelengths were applied to the slices by using a Fluorescence lamp (X-Cite Series 120q, Lumen Dynamics) or a DC4100 4channel LED-driver (Thorlabs, Newton, NJ) as light source. During recordings from brain slices from animals injected with eARCH3.0, 20 sweeps, each 10 s apart were applied. One sweep consists of a 1-s long light pulse. The intensity of the light source was adjusted to 1.7, 3, 7 and 17 W.
Immunohistochemistry. For processing of brain slices with biocytin-filled neurons ( Supplementary Fig. 2) brains from AAV5.EF1a.DIO.EYFP-injected ChAT-cre rats were sectioned in 30-µm-thick slices 22,37 . BF and mPFC slices were stored in PBS overnight and subsequently incubated in citrate buffer pH 6.0 for 3 × 10 min. Thereafter sections were incubated with heated citrated buffer with 0.05% Tween-20 at 90 o C for 15 min, left to cool down, and subsequently, rinsed with 0.05 M TBS. Next, sections were incubated overnight in 0.05 M TBS with 0.5% triton (Tx) containing all five primary antibodies as a cocktail at room temperature. After rinsing slices with TBS (3 × 10 min), sections were incubated for 2 h with secondary antibodies in TBS-Tx. Finally, slices were rinsed in Tris-HCL and mounted on glass slides in 0.2% gelatin, dried, mounted with Mowiol (hecht assistant 1.5 H coverslips). As controls, adjacent sections were included for all five labels.
Cell counts in basal forebrain. To quantify potential retrograde labelling by AAV5 from the mPFC to the BF ( Supplementary Fig. 4), rats were injected with AAV5-DIO::eYFP either in the mPFC or the BF at the coordinates used for behavioural and physiological experiments. 50 µm slices of the brains were cut using a vibratome (Leica, 1200T, Germany). Slices were stained for eYFP and mounted on glass slides covered by 2% Mowiol, anti-fading mounting agent and cover slip. Images were acquired using a confocal laser scanning microscope (CLSM; Zeiss LSM 510 Meta) with an excitation wavelength of 514 nm (bandpass 530-600 nm). Cell counting was performed using the cell count function of ImageJ.
Attention behaviour. After 1 week of recovery from surgery and 1 week of habituation to the reversed light/dark cycle, rats started training in the 5CSRTT in operant cages (Med Associates Inc., St. Albans, VT, USA) ( Supplementary Fig. 6). In short, following the initial training phase, progression was based on individual performance of each rat, and was reduced from 16 to 1 s 60 . Criteria to move to a shortened stimulus duration were the percentage of accuracy (>80%) and omitted trials (<20%). When the criterion of 1 s stimulus duration was reached animals were moved to the pretesting phase. In the pretesting phase, a green custom-made LED replaced the normal house-light of the operant cages (<1 mW intensity) to mask reflections by the laser light used for the experiments.
After three consecutive sessions during which rats performed according to criteria with the LED on in the operant cage, three additional baseline sessions were conducted. During these sessions, rats were connected to the patch-cable (Doric Lenses, Quebec city, Canada) used to deliver the light into the brain. In this condition, percentage accuracy was above 80%. However, rats often did not show <20% omissions within sessions. This was most likely due to the fact that the animals were connected to the optic fibre patch-cable and therefore less free to move in combination with the short time window for the animal to respond (i.e. within 2 s after the cue light went off). Therefore, in line with previous work 60 , the omission criterion was increased to <40% omissions.
Following acquisition of baseline performance, rats were assigned to the testing phase where the task comprised 100 consecutive trials with a random assignment of laser-ON or laser-OFF trials. For the testing phase, the following parameters were acquired and analysed through a box-computer interface (Med-PC, USA) and custom-written MATLAB scripts (Mathworks): accuracy on responding to cues (ratio between the number of correct responses per session over the sum between correct and incorrect hits, expressed as percentage); absolute and percentage of correct, incorrect responses and errors of omission; correct or incorrect response latency; latency to collect reward; number of premature and perseverative responses. Percent of correct, incorrect and omissions were calculated based on the number of started trials 61 to allow a more sensitive evaluation of the parameters.
Optical inhibition during behaviour. To light-activate the opsins in vivo, we used a diode-pumped laser (532 nm, Shanghai Laser & Optics Century Co, China) directly connected to the rat optic glass fibre implant. Light was delivered at 7-8 mW from the fibre tip for experiments carried out with eARCH3.0. These stimulation regimens are able to produce a theoretical irradiance which ranges between 7.59 and 8.68 mW mm −2 (http://web.stanford.edu/group/dlab/cgi-bin/ graph/chart.php). Light was delivered according to scheduled epochs by a stimulator (Master 9, AMPI Jerusalem, Israel) connected to the computer interface, which semi-randomly assigned the different trials to laser-OFF or laser-ON conditions (50% of each). In the laser-ON condition, light was delivered during the whole preparatory period (5 s) that precedes stimulus presentation. Optical inhibition sessions were repeated two times per rat with a baseline session in between to control for potential carry-over effects.
Moreover, reported data for the majority of rats refer to the first two optical inhibition sessions after establishment of stable baseline performance. Power analysis based on the effect size determined the minimal sample size to detect a statistical significance (7 or more) with a power of β = 0.9.
Histological verification. After behavioural testing, brains were checked for fibre placement and viral expression. For this, rats were anesthetized with isoflurane and a mix of ketamine (200 mg kg −1 i.p.) and dormitol (100 mg kg −1 i.p.) and then transcardially perfused (50-100 mL NaCl and 200-400 mL PFA 4%). Brains were removed and maintained in 4% PFA for at least 24 h. After that, brains were sliced with a vibratome (Leica Biosystem, Germany) into 50-100 µm coronal sections and mPFC slices were mounted on glass slides covered by 2% Mowiol, anti-fading agent and cover slipped. Images were taken with a CLSM (LSM 510 Meta; Zeiss, Germany) with excitation wavelength of 514 nm bandpass filtered between 530 and 600 nm, and further analysed using ImageJ (NIH, USA).
Quantification and statisitical analysis. To evaluate behavioural performance between the ARCH3.0 groups and EYFP control group, two-way ANOVAs for repeated measures were performed. Corrected values for multiple comparison with Sidak's test were used when the interaction between light and virus was significant. In all cases, the ANOVAs were preceded by the Kolmogorov-Smirnov (KS) test for normal distribution. In cases when the KS p-value was >0.05, factorial analysis was performed on the raw data per parameter. In other cases, raw data were first transformed with square-root or arcsin transformation. Analysis of other parameters were performed with student's t test, Wilcoxon test and always preceded by KS test to check for normal distribution of the sample. Data were analysed by MATLAB 2016a (Mathworks), Microsoft Excel (Office) and graphs were plotted by GraphPad Prism. In all cases the significance level was p < 0.05.
To statistically evaluate the results between nAChR blockers and aCSF conditions in acute slice experiments, two-tailed paired Student's t-test was employed. To evaluate differences with GABAR blockers two-way ANOVA for repeated measures was used. If the amplitude of the response was lower than four times the standard deviation of the baseline preceding the stimulation, the event was counted as a failure. The average time for onset delay from AP peak was calculated by fitting a lognormal function to the histogram of onset delays. To quantify the spike delay time and probability two-tailed paired student's t-test was used. Significance level was set to p < 0.05.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.