Synaptic co-transmission of acetylcholine and GABA regulates hippocampal states

The basal forebrain cholinergic system is widely assumed to control cortical functions via non-synaptic transmission of a single neurotransmitter, acetylcholine. Yet, using immune-electron tomographic, molecular anatomical, optogenetic and physiological techniques, we find that mouse hippocampal cholinergic terminals invariably establish synapses and their vesicles dock at synapses only. We demonstrate that these synapses do not co-release but co-transmit GABA and acetylcholine via different vesicles, whose release is triggered by distinct calcium channels. This co-transmission evokes fast composite postsynaptic potentials, which are mutually cross-regulated by presynaptic auto-receptors and display different short-term plasticity. The GABAergic component alone effectively suppresses hippocampal sharp wave-ripples and epileptiform activity. The synaptic nature of the forebrain cholinergic system with differentially regulated, fast, GABAergic and cholinergic co-transmission suggests a hitherto unrecognized level of synaptic control over cortical states. This novel model of hippocampal cholinergic neurotransmission could form the basis for alternative pharmacotherapies after cholinergic deinnervation seen in neurodegenerative disorders. Supplementary materials are attached after the main text.

Contemporary models of the basal forebrain cholinergic system and efforts to develop pro-cholinergic treatments have been based largely on the assumption that cholinergic cells release only a single transmitter and it is released non-synaptically (8)(9)(10)(11)(12)(13). The seemingly rare synapses on cholinergic fibers (see Supplemental Information) supported the concept of nonsynaptic transmission. However, highly precise cholinergic transmission during reward and punishment (14), recordings of phasic release (10,15,16), and the dependence of hippocampal synaptic plasticity on the millisecond-scale timing of the cholinergic input (17) challenge this textbook model of non-synaptic transmission by cholinergic fibers.
Therefore, we hypothesized that all cholinergic terminals establish synapses. After immunolabeling, we analyzed the real incidence of synapses, localized vesicle pools using STORM super-resolution imaging and we also localized membrane-docked neurotransmitter vesicles using electron tomography. Because previous data suggested the co-localization of acetylcholine and GABA in retina and other brain areas (18)(19)(20)(21)(22)(23), we also hypothesized that hippocampal cholinergic synapses may be GABAergic as well. Using immunolabeling and optogenetics combined with in vitro electrophysiology, we investigated the possible presence and subcellular regulation of hippocampal co-transmission of acetylcholine and GABA, and the role of its GABAergic component in controlling hippocampal network activity.
Challenging a decades-old model, we show that all hippocampal cholinergic terminals established synapses and vesicles dock only at synapses. We also show that hippocampal cholinergic synapses are GABAergic as well and evoke fast composite (hyperpolarizing and depolarizing) postsynaptic potentials. The two transmitters showed different short-term plasticity, were modulated by distinct calcium channels, and were mutually regulated by presynaptic auto-receptors. We demonstrate here that synaptic release of GABA from cholinergic terminals alone can suppress hippocampal sharp wave-ripples effectively and it can attenuate hippocampal epileptiform activity as well.
Our data urge the reinterpretation of previous studies about the basal forebrain cholinergic system and offer a new explanation for the emergence of hippocampal epileptiform activity associated with Alzheimer's disease related loss of cholinergic innervation.

All hippocampal cholinergic axon terminals form synapses, with GABA as a co-transmitter
Previous studies concluded that few cholinergic terminals form synapses (see Discussion and Supplemental Information). We hypothesized that all cholinergic terminals form synapses.
Having verified that all hippocampal cholinergic terminals originate from basal forebrain cholinergic cells (for controls see Supplemental Information and Figure S2), we found that all hippocampal cholinergic terminals examined established one or more NL2 positive synapses (Figure 1, S1; Table S4, Supplemental Information).
Cholinergic synapses innervated pyramidal dendritic shafts (63%), spine-necks (27%), and interneuron dendrites (5%). Data from the somatosensory cortex was similar (Figure 1 O, Table S4; Supplemental Information). The linear density of synapses along cholinergic axons was similar to that found along GABAergic axons that are known to establish synapses abundantly ( Figure S1, Table S4, and Supplemental Information).
Next, we tested whether these synapses are GABAergic as well. First, we localized the elements of postsynaptic GABAergic signaling machinery in these cholinergic contacts. We localized gephyrin first because it is known to interact with both GABA A receptors and NL2. We found, that more than 81% of hippocampal cholinergic synapses contained gephyrin postsynaptically, on dendrites and spine-necks ( Figure 1, Supplemental Information, Table S3).
In addition, we found that more than 80% of the cholinergic synapses showed GABA A receptor gamma2 subunit labeling that was also readily detected on both dendrites and spine-necks ( Figure 1, S1, Table S3, and Supplemental Information).
Then, we localized the elements of presynaptic GABAergic and cholinergic signaling machinery in these terminals. By crossing a zsGreen fluorescent reporter-mouse-line with a vesicular GABA transporter (vGAT)-Cre mouse line and labeling the medial septum for ChAT, we found that all septo-hippocampal cholinergic cells are also vGAT positive (Figure 2A We recorded the membrane potential of inhibitory neurons in CA1 str. lacunosummoleculare, because they are known to display cholinergic responses (25). Cells were recorded using whole-cell patch-clamp in responses to 1ms optical stimulation ( Figure 3A, Supplemental Experimental Procedures) that resulted in a composite membrane potential response: a quick GABA A receptor-dependent hyperpolarization (peak @ 13.8 ms), and a slightly delayed (peak @ 92 ms) acetylcholine receptor-dependent depolarization ( Figure 3C-E, S3A-C). Although these synaptically released transmitters may act on non-synaptic receptors as well, both responses had short onset latency ( Figure S3A-C), which together with our anatomical data suggests synaptic release and action of GABA. Synaptic spill-over of GABA and acetylcholine may act extrasynaptically as well.
Next, we blocked both nicotinic and muscarinic AChRs and recorded inhibitory postsynaptic currents (IPSC) on pyramidal cells (PCs) and interneurons (INs) after optical stimulation of cholinergic fibers ( Figure 3F-I). Single IPSC kinetics and short-term plasticity in PCs and INs were tested using five short light pulses at physiologically relevant firing rates (at 2, 5, 10 and 20 Hz) measured in vivo (26,27). The amplitude of the evoked inhibitory currents (calculated for the first stimulus) was larger on INs than on PCs, but their rise-time (20-80%) was not significantly different. IPSCs evoked in INs had a shorter decay time ( Figure 3H). The series of light pulses revealed strong short-term depression (STD) of inhibitory currents onto both PCs and INs ( Figure 3G and I). GABAergic STD was observed in every tested neuron and was accompanied by a decrease in transmission probability during the stimulation sequence ( Figure   S3), suggesting a presynaptic mechanism for STD.
Calcium influx through ChR2 expressed on axon terminals can alter intrinsic short-term plasticity of the synapses (28). To exclude this possibility, we used a digital micromirror device (DMD) inserted into the optical path of the microscope, to illuminate cholinergic axons running towards the recorded cell, but not the terminals ( Figure S3D). This resulted in STD similar to that recorded with optic fiber illumination ( Figure S3E-F), excluding the possibility of ChR2 mediated Ca 2+ influx, as the reason for the observed STD. A series of stimuli could also decrease the driving force of chloride through GABA A Rs, resulting in apparent STD (29,30). In this case, the putative site of STD would be postsynaptic instead of presynaptic. When we used a high Clintracellular solution to prevent shifts in Clreversal, a series of stimuli resulted in STD similar to that shown above ( Figure S3G and H).

Cholinergic and GABAergic vesicle docking is restricted to synaptic active zones in cholinergic fibers
Our physiological and anatomical data showed that cholinergic terminals establish effective cholinergic-GABAergic synapses. Non-synaptic acetylcholine release, however, might still be possible if cholinergic vesicles could be docked and fused outside of the synaptic active zones as well. Using advanced electron tomography tools, we were able to reconstruct cholinergic fiber segments, to localize vesicles with 1 nm resolution and to analyze the precise distribution of synaptic vesicles in hippocampal cholinergic terminals. We identified terminals using vAChT immunogold labeling and reconstructed them using electron tomography ( Figure   4A-D). Three-dimensional reconstruction revealed that synaptic vesicles clustered close to the active zones ( Figure 4D). We measured the distances of vesicles to the closest (synaptic or nonsynaptic) cell membranes, and compared their density at different distance intervals from the membranes ( Figure 4F). The density of the vesicles within 60 nm from the membranes was 6.5 times higher in the synaptic active zone than extrasynaptically ( Figure 4D, F). We did not find any docked (less than 5 nm from the membrane) or fused (undergoing exocytotic fusion) vesicles non-synaptically, but we detected both of these at synapses ( Figure 4B, C, F). The distribution of vesicles in cholinergic terminals was similar to that found at other glutamatergic or GABAergic terminals, arguing against non-synaptic vesicular docking or release in cholinergic terminals.
Vesicular volume can correlate with its transmitter content (31). Therefore, using electron tomography, we compared the vesicular morphology between cholinergic and GABAergic terminals ( Figure 4E). Vesicles of cholinergic terminals were significantly (p<0.001), about 60% larger than those in GABAergic terminals ( Figure 4E, Supplemental Information) and their volumes were significantly more variable as well (p<0.001). Smaller (putatively GABAergic) vesicles of cholinergic terminals were similar to those in purely GABAergic terminals ( Figure 4E, see Supplemental Information), suggesting an even larger difference between the two types of vesicles.

Acetylcholine and GABA are released at the same synapses, but from different vesicles in cholinergic terminals
To further confirm that the two transmitter systems use the same active zones, we used confocal fluorescent imaging in mouse hippocampus and found that vAChT-positive vesicles were concentrated at gephyrin-labeled, putative GABAergic-cholinergic synapses ( Figure 4G, H, Supplemental Information). Scale-free analysis confirmed that the likelihood of vAChT labeling was the highest at the gephyrin-labeled synaptic active zones ( Figure 4I, Supplemental Information).
Using super-resolution STORM imaging of vGAT and vAChT immunolabeling, combined with correlated fluorescent confocal laser scanning microscopy (CLSM) of cholinergic fibers, we demonstrated that cholinergic-GABAergic vesicle pools were mixed and were confined to a small volume of the AAV-eYFP labeled septo-hippocampal terminals ( Figure 4J, for details, see Supplemental Information).
Using isolated mouse cortical synaptic vesicles, we found that acetylcholine and GABA are packed into different vesicles ( Figure S4A, Supplemental Information, and Supplemental Experimental Procedures). Both flow cytometry ( Figure S4B) and electron microscopy data ( Figure S4C, E) confirmed the purity of our sample (see also Supplemental Information and Experimental Procedures). All vesicles were labeled with synaptophysin (SYP). After quadrupleimmunolabeling of the isolated synaptic vesicles, 29% were only SYP-positive, 45% were doublelabeled for vesicular glutamate transporter 1 and SYP, 14% were double-labeled for vGAT and SYP, 11% were double-labeled for vAChT and SYP. Only a negligible amount of vesicles (0.9%) were triple labeled with any combinations and only 0.14% of all vesicles were co-labeled for cAChT and vGAT, suggesting that vesicular transporters for GABA and acetylcholine are expressed by distinct vesicle populations in cortical samples ( Figure S4H; Supplemental Information). Acetylcholine and GABA release shows different short-term plasticity, modulated by distinct voltage-dependent calcium channels, and are mutually regulated by different presynaptic

auto-receptors
Our anatomical data demonstrated that acetylcholine and GABA are released from different vesicles. However, distinct vesicle pools of the same terminals can couple to different release mechanisms with different short-term plasticity (20). Because GABAergic transmission always show strong STD in cholinergic terminals, we examined, whether cholinergic responses display similar short-term dynamics. We recorded inhibitory neurons and light-stimulated ChR2 expressing cholinergic axons ( Figure 5G). Their net depolarization for a series of five stimuli did not decrease but increased slightly with frequency ( Figure 5H). We found that cholinergic dynamics did not show short-term depression present in GABAergic responses, in which case net depolarization should have decreased. The observed slight increase in the net depolarization with increasing stimulation frequency indicates a form of facilitation in cholinergic responses. This differential dynamics of GABA and acetylcholine release confirmed that the two transmitters are released from distinct vesicular pools.
Dynamics of vesicle release likely depends on the type of voltage-dependent calcium channels (VDCCs), which mediate Ca 2+ triggered vesicle release (32). However, different VDCCs in the same cell (20) can be associated with distinct types of vesicles. Indeed, we found that selective blockade of P/Q-type calcium channels decreased GABAergic IPSPs significantly, but caused no change in the cholinergic components ( Figure 5E, Supplemental Information).
Conversely, the cholinergic component was robustly decreased after the selective blockade of N-type calcium channels, while GABAergic IPSPs showed no change ( Figure 5F, Supplemental Information). Besides confirming the presence of different vesicular pools, this revealed different molecular pathways for their regulation during co-transmission.
We also investigated presynaptic modulation of this vesicular release. Blocking AChRs increased the amplitude of GABAergic hyperpolarization ( Figure 3D), for which presynaptic muscarinic receptors may be responsible (25). We held CA1 interneurons at -50 mV, to record GABAergic hyperpolarization and cholinergic depolarization concurrently ( Figure 5A). Muscarinic receptor blocker atropine (10 µM) significantly increased the amplitude of both GABAergic IPSPs and cholinergic EPSPs ( Figure 5B). By blocking M2-type AChRs, reported abundant in hippocampal cholinergic terminals (25,33), we reproduced PSP increases evoked by atropine ( Figure 5C). This confirms the role of M2-type AChRs in regulating both acetylcholine and GABA release from cholinergic terminals.
Previously, we described the presence of presynaptic GABA B autoreceptors in septal cholinergic cells (34). To test their role in regulating synaptic co-transmission, we blocked GABA B receptors. This led to a significant increase in the amplitude of both GABAergic IPSPs and cholinergic EPSPs ( Figure 5D). However, we did not see GABA B receptor-dependent postsynaptic hyperpolarization in response to cholinergic fiber stimulation in our recordings. Overall, these experiments revealed that GABA and acetylcholine cross-and auto-regulate their cotransmission from cholinergic-GABAergic terminals, presynaptically.

GABA release from septal cholinergic cells is essential for shaping healthy and pathological cortical network activities
Septal cholinergic neurons control hippocampal activity states (35,36) and suppress in vivo sharp wave-ripples (SWR) in the hippocampus (37). However, the effect of their hippocampal GABA release has not been investigated. We recorded spontaneous SWRs in hippocampal slices ( Figure 6, Supplemental Experimental Procedures). Illumination of control slices without ChR2 expression caused no change in SWR activity. In AAV-ChR2 infected ChAT-Cre mice, we found that without blocking cholinergic transmission, optical stimulation of cholinergic fibers decreased SWR rate significantly ( Figure 6C-D), followed by a transient increase in SWR rate after the cessation of optical stimuli. However, when cholinergic transmission was blocked, the same optical stimulation caused a similar decrease in SWR rate ( Figure 6E-F), with no rebound after stimulation. Although cholinergic fibers have been reported to be responsible for inhibiting SWRs (37), here we demonstrated that GABA release from cholinergic synapses alone is sufficient to downregulate the rate of SWRs.
The GABA release from septo-hippocampal cholinergic fibers may also affect the occurrence of hippocampal epileptiform activity. Therefore, we examined the effect of GABA release from cholinergic axons in the in vitro high potassium epilepsy model ( Figure 6G-K, 38,39). Optical stimulation of cholinergic fibers reduced the rate of epileptiform events ( Figure 6H-I) and a similar reduction was present, when AChRs were blocked ( Figure 6J-K). These results demonstrated that synaptic co-transmission of GABA from cholinergic fibers alone can effectively control hippocampal epileptiform activity.

Discussion
Our results challenge the model of non-synaptic single transmitter release from cholinergic fibers in the hippocampus and demonstrate a synaptic communication using two neurotransmitters. Our results also show non-synaptic action of these synaptically released transmitters. We showed that hippocampal cholinergic synapses were effective GABAergic synapses as well. Cholinergic fibers are among the densest and most influential subcortical pathways in hippocampus; therefore, our findings suggest a fundamental change in our view of the regulation of hippocampal states. We also found that acetylcholine and GABA were not co-released but co-transmitted from the same synapse. We confirmed this by showing that (i) isolated cortical transmitter vesicles do not co-express vGAT and vAChT, (ii) GABAergic and cholinergic components show different short-term plasticity in these fibers, (iii) they are preferentially modulated by distinct voltage-dependent calcium channels, and (iv) electron tomography suggested differences in the volume of GABA and acetylcholine containing vesicles. This co-transmission (similar to that in retina 20) may require nanoscale subsynaptic organization of presynaptic molecules, as proposed before (40). Emphasizing the functional role of GABA release, we also demonstrated that synaptic GABA release from cholinergic terminals alone can effectively suppress hippocampal sharp wave-ripples and per se can attenuate hippocampal epileptiform activity. These confirmed the functional importance of this GABAergic-cholinergic co-transmission in healthy and pathological states and led to a novel model of the septo-hippocampal cholinergic neurotransmission.

Cholinergic terminals establish synapses
For decades, the predominant form of cholinergic communication was thought to be a form of "non-synaptic transmission" (8-10, 12, 41), which was seemingly supported by studies showing cholinergic fibers with few synapses (for details see Supplemental Information).
Originally, the presence of extrasynaptic acetylcholine receptors and micro-dialysis experiments seemed to support a non-synaptic transmission hypothesis; however, later, sensitive microelectrodes showed faster changes in extracellular acetylcholine levels (Supplemental Information). In addition, basal forebrain cholinergic neurons were also shown to respond to reward and punishment with extremely high speed and precision (14), and recent data suggested that cholinergic cells may regulate cortical information processing with a remarkable, millisecond-scale temporal precision (15,17,42,43). These data urged the re-examination of whether acetylcholine release is synaptic (44). Using electron tomography and direct labeling of cholinergic synapses using neuroligin 2, we showed that, all terminals of the dense meshwork of hippocampal cholinergic fibers established (one or more) synapses and no docked or fused vesicles were detectable non-synaptically. Most of these synapses were missed previously, probably, because of their weak membrane thickening and narrower synaptic gap. Furthermore, we found that vAChT labeled vesicle pools associated only with synaptic active zones. Although synaptically released acetylcholine or GABA may act on non-synaptic receptors as well, in a "volume transmission" fashion, none of our data supports non-synaptic release. Therefore, the most likely resolution for the mismatch between this and the classic non-synaptic volume transmission hypothesis is a strong "spill-over" of transmitters from these frequent synapses that would allow transmitters to reach extrasynaptic receptors as well. In fact, the much larger cholinergic synaptic vesicles that we found could contain more acetylcholine to counteract its effective extrasynaptic removal by extracellular acetylcholine esterase (AChE).

The tight regulation of synaptic GABAergic co-transmission from basal forebrain cholinergic fibers and their role in hippocampal synaptic plasticity
Co-transmission can significantly increase the efficacy of information transfer (45)(46)(47).
Using purified vesicle preparation, we showed that GABA and acetylcholine are not "coreleased" from the same vesicles, but "co-transmitted" from separate vesicles, sequentially inhibiting and exciting hippocampal neurons. To unlock the full potential of co-transmission, these coexisting vesicle pools need complex regulation. Indeed, we showed that the release of GABAergic vesicles are preferentially regulated by P/Q-type calcium channels, while the cholinergic ones are more affected by N-type calcium channels in the same synapses, similar to that of retinal co-transmission of GABA and acetylcholine (20). Differential calcium channel regulation may be achieved by sub-synaptic nanoscale organization of presynaptic molecules (40) providing partly different access for the distinct vesicle pools. Negative feedback via presynaptic autoreceptors could provide a further level of control. Previously, we demonstrated that septo-hippocampal cholinergic neurons express GABA B -receptors (34), while the expression of muscarinic M2 receptors was also described in these cells (33). Indeed, we confirmed that blocking either M2-or GABA B -receptors increased the release of both acetylcholine and GABA, suggesting the presence of tonically active autoreceptor regulation of these transmitters, presynaptically. In addition, our results, verified by a series of control experiments, suggested that GABAergic STD is an intrinsic property of synapses established by cholinergic fibers in the mouse hippocampus.
The timing of this co-transmission seems crucially important in hippocampal Schaffer collateral (SC) to CA1 synaptic plasticity. If the cholinergic input was activated 100 or 10 ms before SC stimulation, it resulted in ionotropic AChR-dependent long-term potentiation (LTP) or short-term depression, respectively, while delaying cholinergic stimulation until 10 ms after SC stimulation resulted in metabotropic AChR-dependent LTP (17). Here, we found that the GABAergic and cholinergic components of this co-transmission reach their IPSP and EPSP peak about 13.8 and 92 ms after stimulation, respectively, which suggest that synaptic plasticity may depend on the coincidence of SC stimulation with the GABAergic or cholinergic component of PSPs from these basal forebrain fibers. Nevertheless, the GABAergic component seems to have its own inhibitory role in the target network, making basal forebrain cholinergic fibers an unorthodox but effective source of GABAergic inhibitory control of the hippocampus, as suggested by our demonstration of its effect on network dynamics.

Role of basal forebrain cholinergic system in regulating hippocampal network states
Basal forebrain cholinergic cells play a pivotal role in transforming activity states in the hippocampus (35). High cholinergic cell activity is associated with theta oscillation and the fast, yet unstable storage of external information in the hippocampus, while low cholinergic activity accompanies sharp wave-ripple (SWR) activity (48), which seems important for the consolidation and relocation of unstable memory traces from the hippocampus to the neocortex (49,50). Classic theories of cholinergic modulation presume that diffusely released acetylcholine would slowly retune cortical network activity, enabling the appearance of distinct network dynamics (8,51,52). Indeed, a recent study demonstrated that medial septal cholinergic cells suppress SWR activity in vivo (37). The authors suspected an M2 cholinergic receptor-mediated suppression of GABAergic interneurons. We re-examined that hypothesis and confirmed that cholinergic fiber activation indeed suppressed SWRs; however, we also showed that GABA (without acetylcholine) release from cholinergic fibers alone was able to achieve that. This GABAergic inhibition could lower the probability of concurrently active pyramidal cells in a given time-window, which is required for stochastic SWR initiation (53).
Blocking all GABA receptors would have eliminated SWRs (54), therefore the effect of acetylcholine release alone, without the contribution of GABA co-transmission could not be addressed. The presence of a transient increase in SWR rate after the cessation of optical stimuli, might reflect longer-lasting changes in network or cell excitability, mediated by "synaptic spill-over" of acetylcholine, while the effect of GABA release from smaller vesicles is likely confined to synaptic (and presynaptic) receptors. However, this effect was not reported in vivo (37), therefore, it could have been the result of a relatively low AChE activity in our dualsuperfusion system.
Degeneration of the cholinergic system is a typical characteristic of Alzheimer's disease pathology (7), and these patients often develop epileptic seizures (55). Indeed, selective lesion of medial septal cholinergic projection increases seizure incidence in the hippocampus (56)(57)(58), while REM-sleep, which is associated with enhanced cholinergic cell activity (27), has a suppressing effect on epileptic seizures (59,60). The mechanisms, however, were unclear, because several cholinergic agonists were shown to trigger epileptic discharges (61). Therefore, we hypothesize that degeneration of basal forebrain cholinergic cells leads to epileptiform activity primarily, because it deprives the hippocampus of a massive GABAergic input. Indeed, our results supported this idea, because GABA release from basal forebrain cholinergic fibers alone was sufficient to decrease the occurrence of epileptiform activity in the hippocampus.
Our results, showing a tightly regulated, effective, synaptic GABA co-transmission from hippocampal cholinergic fibers urge the re-interpretation of previous models and can lead to alternative pharmacotherapies to treat Alzheimer's disease-related loss of cholinergic innervation.

Figures Figure 1. All cholinergic axon terminals establish synapses, and also express GABAergic markers postsynaptically, while innervating pyramidal cells and interneurons
A: Three-dimensional EM reconstructions show that hippocampal cholinergic fibers form synapses (arrows) frequently. Axon A is labeled for ChAT in a WT mouse. Axon B is an AAV-eYFP virus-labeled septo-hippocampal fiber in a ChAT-Cre mouse. Insets show two typical terminals with synapses (blue).
The plasma membrane was made partially transparent to reveal mitochondria (mito, green). Gold labelings of NL2 (axon A, synapse 1-4) or gephyrin (axon B, synapse 5-7) were used to recognize synapses (black dots and arrows in the insets). EM images show terminal boutons (b) of the reconstructed axonal segments establishing synapses 1-7 (arrowheads, indicated by the same numbers on the left) on dendrites (d) and a spine (s). Next to synapse 1, a ChAT-negative, putative GABAergic terminal bouton   blocks the remaining depolarizing response (magenta). D: Conversely, first blocking the AChRs, resulted in a putative GABAergic IPSP (magenta), which was blocked by GABA A Rs inhibitor gabazine (black). The increase of IPSP amplitude for AChR block is addressed in Figure 5.

E: Magnified cholinergic EPSP (black) from panel C, and GABAergic IPSP from panel D (magenta)
demonstrate their short latency (see also Figure S3).  E: Scatterplot shows the relationship of vesicle size (volume) and elongation factor in the sampled vesicles. Vesicles in the cholinergic terminals had a broader size-distribution than vesicles of purely GABAergic terminals, which were smaller and more elongated (Supplemental Information). Some vesicles (black markers) could be identified as vAChT-positive because immunogold particles clearly labelled them.
F: The density of the vesicles within a 60-nm-wide band from the membrane is plotted as a function of distance from the terminal membrane at synaptic (green) and extrasynaptic (black) areas. Vesicle density was 6.5 times higher (in the first 60 nm from the membrane) in the synaptic active zone than extrasynaptically (n=183 vesicles from 8 terminals in 2 mice). Docked (closer to the membrane than 5 nm) or fused (undergoing exocytotic fusion) vesicles were absent extrasynaptically, but were present in the active zones.    showed that GABA release from cholinergic fibers alone could decrease SWR rate even more effectively (F). There was no SWR rate increase after the cessation of optical stimulation (for data see Supplemental Information). G: Epileptiform activity was evoked by elevating K + concentration in the ACSF to 8.5mM in slices. LFP (bottom) and its wavelet transform (top) shows the signature of a pharmacologically evoked epileptiform events. Note the robust differences in amplitude, length and frequency components compared to physiological SWR activity ( Figure 6B) that is consistent with literature data (39).  We reconstructed randomly selected, long axonal segments [6 -33 µm, average: 21 µm, n= 17, labeled either with anti-choline acetyltransferase (ChAT) antibody in wild type (WT) mice or with eYFP-adeno-associated viruses (AAV) injected in ChAT-Cre mice, Figure 1, S1, Table S4, for controls see Suppl. Experimental Procedures] and identified their synapses with NL2 or gephyrin immunogold labeling. All of them established synapses abundantly (Figure 1, S1). The average density of synapses was 42 synapses/100 µm. Some of these contact sites would not have been considered synapses earlier, because of their weak membrane thickening and narrower intercellular synaptic gap (e.g. Figure 1A, synapse 2-3) (1); however, NL2 and gephyrin labeling clearly identified their active zones. For comparison, we have also reconstructed GABAergic axonal segments (labeled for cannabinoid receptor type 1, CB 1 ; n=2, 18 and 29 µm, Figure S1, Table S4), which are known to establish synapses abundantly. We found that practically all cholinergic terminals formed one or sometimes two synapses (Figure 1 and Figure   S1). As a consequence, the linear density of synapses along cholinergic axons was similar to that along GABAergic axons ( Figure S1, Table S4, number of synapses per 100 µm cholinergic axons: 42 in CA1, 40 in S1, number of synapses per 100 µm CB 1 -positive axons: 51).

Postsynaptic targets of cholinergic axons
Using electron microscopy in hippocampal CA1, we found that NL2 and gephyrin positive cholinergic synapses (n=107, collected from 4 mice) predominantly innervated pyramidal dendritic shafts (63%) and spine-necks (27%), and they also innervated interneuron dendrites (5%), while some (6%) postsynaptic targets could not be classified (Figure 1 O, Table S4). All innervated spines received another, putatively glutamatergic asymmetric, type I input from an unlabeled terminal, suggesting that, contrary to previous suggestion (2), cholinergic synapses alone do not induce spine formation. These data suggest that about 90% of cholinergic synapses target pyramidal cells in CA1, whereas they also innervate interneurons (at least 5%), which ratio is close to the neuronal ratios in CA1.

Cholinergic synapses express GABA A receptors and its scaffolding protein postsynaptically
Previously, we demonstrated that cholinergic synapses (in CA1, S1 somatosensory cortex, prefrontal cortex, basolateral amygdala and centrolateral thalamic nucleus) expressed We found that at least 81% of 188 synapses collected in the CA1 area of two WT and two ChAT-Cre mice contained gephyrin postsynaptically (Table S3). Immunogold particles in these synapses were associated with the cytoplasmic side of the postsynaptic membrane of the innervated pyramidal cell dendrites (Figure 1B-D, G-I) and spines ( Figure S1B) or interneuron dendrites ( Figure 1J). The antibody used for GABA A receptor-labeling was directed against an extracellular epitope of the γ2 subunit (7,8) and labeled cholinergic synaptic clefts accordingly.
In the CA1 of three WT and two ChAT-Cre mice, at least 80% of the cholinergic synapses (out of 172) showed GABA A receptor γ2 subunit labeling (Table S3). Cholinergic fibers established GABA A receptor containing synapses on both dendrites ( Figure 1E,F,K,L,N; Figure S1 C, D) and spine-necks ( Figure 1M). In addition, we found that at least 83% of synapses of vAChT-positive terminals in the somatosensory cortex S1 (n=36, Figure S1 E-F) contained these GABA A receptor γ2 subunits. Because reliable nicotinic receptor antibodies are not available, they could not be localized directly.

Cholinergic terminals possess the molecular machinery for GABA-release
By crossing a zsGreen fluorescent reporter-mouse-line with a vesicular GABA transporter (vGAT)-Cre mouse line, we created mice, in which all GABAergic cells are zsGreen labeled (vGAT-zsGreen mice). After co-labeling medial septum sections for ChAT, we found that all cholinergic cells were also positive for zsGreen (n=243 cells in 2 mice, Figure 2A To confirm vGAT protein expression in cholinergic terminals we performed vGAT-eYFP and vAChT-eYFP multiple labelings, and found that vGAT was present in the majority of the eYFP positive cholinergic terminals (at least 82.9%, n=311 in 3 mice Figure 2D). We also quantified vAChT expression, which was found to be present in at least 63.6% of the eYFP positive cholinergic terminals (n=364 in 3 mice).
Background level was estimated measuring gold particles on vAChT-negative, putative glutamatergic terminals that formed asymmetric synapses in the vicinity of the examined vAChT-positive terminals (n=2 mice, 34 terminals). Data from two mice were not different statistically; therefore, they were pooled. We found a significantly higher level of immunogold labeling for GABA in vAChT-positive terminals than in glutamatergic terminals (3.3 times higher density; Figure 2F), suggesting the presence of GABA in these terminals.

Synaptic vesicles of cholinergic terminals are highly heterogeneous and relatively large
We analyzed the volume and elongation of vesicles (9,10) in cholinergic terminals along with similar data from purely GABAergic terminals ( Figure 4E). As expected, GABAergic vesicles were small and elongated (median volume: 13730 nm 3 , 11174-16434 nm 3 interquartile range; median of elongation factor: 2.90, 2.52-3.41 interquartile range; n=54 vesicles from 2 mice).
However, the volume of cholinergic vesicles were significantly larger (Mann-Whitney test, p<0.001). The volume of the vesicles in cholinergic terminals showed a significantly higher variability as well (F-test, p<0.001), ranging from the very large and round ones to the small and elongated vesicles (median volume: 23267 nm 3 , 19160-27539 nm 3 interquartile range; median elongation factor: 1.80, 1.57-2.04 interquartile range; n=140 vesicles from two mice, Figure 4E).
The small and elongated vesicles in the cholinergic terminals were similar to the purely GABAergic ones from GABAergic interneuron terminals ( Figure 4E). These data suggest that cholinergic terminals contain both smaller, more elongated GABAergic vesicles and larger, rounder cholinergic vesicles, both of which are released from the same synaptic active zone.
Interestingly, some vesicles that were directly labeled with vAChT-immunogold particles were rather large and round ( Figure 4E). We also observed a vAChT-labeled vesicle, fused to the synaptic membrane, likely releasing its transmitter into the synaptic cleft ( Figure 4C). In subsequent experiments, we collected further evidence that acetylcholine and GABA are filled into different vesicles allowing their separately-regulated co-transmission.

Acetylcholine and GABA are released at the same active zone in cholinergic terminals
We also tested, whether the two transmitter systems use the same or distinct active zones. We performed multiple immunofluorescent labeling experiments on virus labeled (eYFP) cholinergic fibers in ChAT-Cre mice for gephyrin, vAChT and eYFP, followed by confocal fluorescent imaging. We observed gephyrin-labeled puncta opposed to eYFP-positive terminals ( Figure 4G, H), identifying the postsynaptic active zones of these fibers. vAChT-labeling was clearly concentrated opposite to the gephyrin-puncta, proving a tight association to the synaptic active zones (Figure 4G, H). Scale-free analysis confirmed that the likelihood of vAChT labeling was the highest at the synaptic active zones ( Figure 4I; n=32 synapses from two mice). To directly examine the existence of a mixed cholinergic/GABAergic vesicle pool, we labeled brain slices for vGAT, vAChT and eYFP, and performed correlated fluorescent confocal laser scanning microscopy (CLSM) and superresolution STORM imaging ( Figure 4J). The superresolution images confirmed that vAChT-and vGAT-labeled vesicle pools overlap, and were localized to the same small, confined portions of the eYFP septo-hippocampal terminals.

Acetylcholine and GABA are released from different vesicles in cholinergic terminals
Although a previous study in rat has suggested that GABA-containing synaptic vesicles do not contain acetylcholine (11), using a highly specific method, we confirmed that GABA and acetylcholine vesicular transporters are localized on different vesicles in mouse cortical axon terminals. We used isolated synaptic vesicles to test whether acetylcholine and GABA are packed into the same vesicles. Isolation from neocortex and hippocampus was performed according to Mutch et al. (Figure S4A, 12). vesicles; Figure S4D). When PSFs in different channels colocalized, their centroids were never farther away from each other than 0.130 µm (median: 0.03 µm, 10-50 µm interquartile range, 0-0.13 min-max, n=92 vesicles; Figure S4D). These experiments confirmed that co-localizing PSFs correspond to a single vesicle. Next, we analyzed co-localizations of the PSFs in different channels ( Figure S4G) and found that 29.2% of vesicles were labeled only for SYP, 44.5% were double-labeled for VG1 and SYP, 14.3% were double-labeled for vGAT and SYP, 11.  Figure S4I). These data suggest that vesicular transporters for glutamate, GABA and acetylcholine are expressed by distinct vesicle populations in cortical samples ( Figure S4H; n=353 vesicles). Therefore, acetylcholine and GABA may be released at the same active zones, but from different vesicles.

Identification of basal forebrain cholinergic fibers in the hippocampus: control experiments
We either used immunolabeling against the vesicular acetylcholine transporter (vAChT), or performed anti-eYFP staining on sections from ChAT-Cre mice, the medial septal areas of which have previously been injected with Cre-dependent eYFP-adeno associated virus (AAV).
Both of these methods had to be verified for selectivity and specificity, thus we completed a comprehensive set of control experiments. The cholinergic innervation of the hippocampus is reported to originate exclusively from the basal forebrain. Although the presence of a local cholinergic cell population in the mouse hippocampus was reported to be an artifact (14) we also tested for it. We injected Cre-dependent eYFP-AAV into the hippocampi of ChAT-Cre mice ( Figure S2C, inset), and stained hippocampal sections for eYFP, vAChT and vGAT ( Figure S2E, F).
We found a few eYFP positive cells in the hippocampus. They were extremely rare and resembled dentate gyrus granule cells and CA3 pyramidal cells. We also found some sparsely distributed eYFP positive fibers originating from them, but vAChT or vGAT immunoreactivity was never found in these eYFP positive terminals (0 out of 323 terminals, from 2 mice, Figure S2F).
We also tested vAChT positive terminals in the same samples, and never found any eYFPpositivity in them (0 out of 3673 from 2 mice, Figure S2F). Thus, we confirmed that there are no cholinergic cells in the hippocampus, only some extremely rare ectopic expression of the Cre enzyme. These results also confirmed that we can reliably label the septo-hippocampal cholinergic fibers with vAChT labeling.
To verify the other approach, we injected eYFP-AAV into the medial septal areas of ChAT-Cre mice ( Figure S2C), and peformed PV/ChAT/eYFP triple labelings ( Figure S2D). 97.6% of all tested eYFP positive cells in the MS were also positive for ChAT (the few % of false negative cells are likely due to not perfectly efficient antibody penetration), but none of them were positive for PV (n=212 in 2 mice). We also tested the fibers of these cells in the hippocampus, and performed a PV/vAChT/eYFP triple labeling ( Figure S2A, B). We found that eYFP positive terminals colocalized with vAChT-labeling, but were never positive for PV (n=252 terminals from 2 mice). These results confirmed that eYFP positive fibers in these animals originate exclusively from cholinergic cells.               (23,24), rat hippocampus (25,26), mouse hippocampus (27)], while some papers reported more frequent synapses [44-67%: in macaque prefrontal cortex (28), human temporal lobe (29), rat parietal cortex (30)]. Although acetylcholine esterase (AChE) was known to be highly effective in terminating extracellular cholinergic signal, the presence of certain extrasynaptic acetylcholine receptors ("receptor mismatch", 27, 31) suggested that extracellular diffusion of acetylcholine occurs. Micro-dialysis experiments (32) and the localization of AChE, distant from cholinergic terminals, also seemed to support a non-synaptic "volume" transmission hypothesis (15). However, later, highly sensitive microelectrodes showed faster, phasic changes in extracellular acetylcholine levels that facilitated cue detection and cortical information processing (21,(33)(34)(35)(36). Basal forebrain cholinergic neurons were also shown to respond to reward and punishment with extremely high speed and precision (37), and recent data suggested that cholinergic cells may regulate cortical information processing with a remarkable, millisecond-scale temporal precision (35,(38)(39)(40). However, such a delicate temporal precision is hard to imagine without synapses and it remains inconclusive, whether the mode of acetylcholine signaling is synaptic "wired" transmission or "non-synaptic volume" transmission by ambient acetylcholine (27).

GABAergic markers in cholinergic cells
Basal forebrain cholinergic cells share a common developmental origin with different populations of cortical, striatal and basal forebrain GABAergic neurons (41)(42)(43)(44). Previous studies have suggested that less than 2% of cholinergic cells express GABAergic markers (45,46), while about 8% of ChAT-positive boutons in the cat striate cortex was shown to contain GABA (47).
The recognition of GABAergic signaling in the BF cholinergic system may have been hampered by its lack of GAD67 (48) and GABA transporter 1 (49). Although the associations of cholinergic terminals with gephyrin (50) and NL2 (1) suggested the capability of GABAergic signaling from these terminals. While, for example, GABA is released together with glutamate or aspartate in the hippocampus, or with dopamine in periglomerular cells (51,52), and acetylcholine is released with glutamate in striatum (53); GABA and acetylcholine were also shown to be released together in retina and frontal cortex (46,48,54,55). However, the precise architecture, the mechanism of the dual cholinergic/GABAergic transmission and their hippocampal synaptic physiological and network effects have not yet been investigated. Boehringer Ingelheim, Germany) intraperitoneally to support recovery, and we placed them into separate cages to survive for at least three weeks before decapitations or perfusions.

Slice preparation and recording conditions.
After 3-6 weeks following injection (to reach an appropriate expression level in longrange projecting axons) horizontal slices were prepared. In all cases, mice were decapitated under deep isoflurane anesthesia. The brain was removed into ice-cold cutting solution, which had been bubbled with 95% O 2 -5% CO 2 (carbogen gas) for at least 15 min before use. The cutting solution contained the following (in mM): 205 sucrose, 2.5 KCl, 26 NaHCO 3 , 0.5 CaCl 2 , 5 MgCl 2 , 1.25 NaH 2 PO 4 , 10 glucose, saturated with 95% O 2 -5% CO 2 . Horizontal hippocampal slices of 300 µm or 450 µm thicknesses (for whole-cell and LFP recordings, respectively) were cut using a vibratome (Leica VT1000S). After acute slice preparation the slices were placed into an interface-type holding chamber for recovery. This chamber contained standard artificial cerebrospinal fluid (ACSF) at 35°C that gradually cooled down to room temperature. The ACSF had the following composition (in mM): 126 NaCl, 2.5 KCl, 26 NaHCO 3 , 2 CaCl 2 , 2 MgCl 2 , 1.25 NaH 2 PO 4 , 10 glucose, saturated with 95% O 2 -5% CO 2 . After incubation for at least 1.5 h, slices were transferred individually into a submerged-style recording chamber equipped with a single superfusion system. In case of LFP recording of hippocampal network activities, a double superfusion system was used for improved slices maintenance conditions (57,58). In the latter design, the slices were placed on a metal mesh and two separate fluid inlets allowed ACSF to flow both above and below the slices with a rate of 3-5 ml/min for each flow channel at 30-32°C (Supertech Instruments; www.super-tech.eu). The composition of the modified ACSF (mACSF) used in experiment presented in Figure 6G-K was the following (in mM): 126 NaCl, 3

Anatomical identification
The recorded cells were filled with biocytin. After the recording, the slices were fixed in

Optogenetic illumination
For illumination, we used a blue laser diode (447 nm, Roithner LaserTechnik GmbH) attached to a single optic fiber (Thorlabs) positioned above the hippocampal slice. In all cases, 1 ms pulse length was used. In case of structured illumination ( Figure S3), the same laser was used with a digital micromirror device (Polygon400, Mightex Systems, Toronto, Canada) integrated into the optical path of the microscope. We have used Polygon400 for the illumination of the slice at a remote location (~5-600 µm), to avoid direct illumination of ChR2expressing terminals on the recorded cell.

Electrophysiological recordings and analysis
Both whole-cell and LFP recordings were performed with a Multiclamp 700A or 700B amplifier (Molecular Devices) and were low-pass filtered at 3 kHz using the built-in Bessel filter of the amplifier. Data were digitized at 10 kHz with a PCI-6042E board (National Instruments) and recorded with EVAN 1. Transmission probability was defined, as the probability of detectable IPSC in response to the optical stimulation. For characterizing short-term plasticity, each stimulation frequency was repeated 20 times with an interval of 3 seconds. To record in parallel the changes in cholinergic EPSP and GABAergic IPSPs (for the drug applications presented on Figure 5) cells were measured at -50 mV in current clamp mode. Since the amplitude of both PSPs increased, qualitative changes in amplitude could be measured, however in this way, we may have underestimated the changes in amplitude due to conductances overlapping in time and acting in the opposite directions. In the case of VDCC blocker application, we used five pulses to evoke larger cholinergic responses. We quantified EPSP area changes for ω-agatoxin and ω-conotoxin application after low-pass filtering (at 2 Hz) the response (e.g. subtracting GABAergic component).  (59). LFP signals were filtered with a two way RC filter to preserve phase.
SWRs were pre-detected on 30 Hz low-pass-filtered field recordings using a threshold value of four times the SD of the signal. The pre-detected SWRs were then redetected using a program that detected SWR peaks and eliminated recording artifacts similar to SWRs. Epileptic event peaks were detected similarly. Event rates were calculated with equal time periods for control, stimulation and recovery phases. All automatic detection steps were manually supervised.

Perfusion
For perfusion, mice were deeply anesthetized as above. Mice for electron microscopical reconstruction of axonal segments and immunofluorescent labelings were perfused transcardially first with 0.9 % NaCl in 0.1M phosphate buffer (PBS) solution for 2 min followed by 4% paraformaldehyde (PFA) in 0.1 M phosphate buffer (pH=7.4; PB) for 40 min and finally with PBS for 10 min. In the case of one WT mouse for reconstruction the fixative also contained 0.5% glutaraldehyde. Two WT mice for postembedding GABA-labeling and electron tomography were perfused first with PBS for 2 min followed by a fixative containing 2% PFA and 2% and then embedded in 2% agarose for sectioning. Coronal sections were cut on a Leica VT1200S vibratome at 50 µm. All series of sections were rinsed in PB, cryoprotected sequentially in PB containing 10% and 30% sucrose, frozen in liquid nitrogen and stored at -70°C until further processing.

Antibodies
We have summarized the primary antibodies used, their concentrations and information on their specificity in Table S1. The secondary antibodies (see Table S2) were extensively tested for possible cross-reactivity with the other secondary or primary antibodies, and possible tissue labeling without primary antibodies was also tested to exclude autofluorescence or specific background labeling by the secondary antibodies. No specific-like staining was observed under these control conditions.

Single and double-labeling preembedding immunoelectron microscopy
Sections were freeze-thawed two times over liquid nitrogen and washed in PB. Sections for gephyrin immunolabeling were treated with 1% sodium borohydride in PB for 10 min. For detection of GABA A receptors, the sections were incubated in 0.2 M HCl solution containing 2 mg/ml pepsin (Dako) at 37°C for 2-4 min. After extensive washes in PB and 0.05 M Tris-buffered saline (pH 7.4; TBS) sections were blocked in 1 % human serum albumin (HSA; Sigma-Aldrich) in TBS. Then, they were incubated in a mixture of primary antibodies (Table S1) diluted in TBS containing 0.05% sodium azide for 2-3 days. After repeated washes in TBS, the sections were  Figure 2E); therefore, only every second electron microscope grid was reacted for GABA and the analysis of vACHT-positive terminals were carried out using the so called mirror technique. Sections on copper grids adjacent to the first sections on nickel grids were systemically scanned for preembedding immunogold-labeled vAChT-positive terminals. These terminals were identified in the next, GABA-labeled section, followed in consecutive serial sections and digital images were taken at 50,000 times magnification in each serial section. Postembedding immunogold particles were counted on these GABA-labeled terminals and the examined surface area was measured using the Reconstruct software (63). For comparison, we have also measured the immunogold densities of putative glutamatergic terminals using the same method, in the same series of images.
Terminals forming type I (asymmetric) synapses were considered glutamatergic. Although periodic acid-etching of the sections is known to remove preembedding antibodies together with immunogold and silver precipitation, we confirmed this again in a control experiment because, here, both the preembedding labeling and the postembedding GABA staining was performed using rabbit antibodies. In these control experiments, the postembedding GABA immunostaining was carried out without the primary anti-GABA antibody, in which case practically no postembedding gold particles were detected, confirming the specificity of the method.

Electron microscopy analysis of cholinergic synapses
For evaluation of the gephyrin and GABA A receptor content at synapses of cholinergic axons, we performed preembedding double labeling. VAChT (in wild type mice) or eYFP (in ChAT-Cre mice, see above) was labeled with DAB, while gephyrin or GABA A γ2 subunit was labeled with immunogold (see above, Table S1). Electron microscopic serial sections were systemically scanned for synapses of DAB-labeled terminals. Parallel appositions between the membranes of a DAB-containing terminal and a putative postsynaptic target were regarded as a synapse if they displayed widening of the extracellular space at the presumptive synaptic cleft and clustered synaptic vesicles in the terminal. Synapses found were followed and photographed at 30,000 magnification in every section, where they were present throughout the series: thus they were fully reconstructed. Since the background labeling (measured in putative glutamatergic synapses in the same series of sections) was negligible, synapses containing at least one gold particle were regarded as positive.

3D reconstruction of axonal segments
Cholinergic axonal segments (n=17) were reconstructed using consecutive serial 70 nm- length, Blender software (www.blender.org) was used. Postsynaptic targets of cholinergic axons were classified as described earlier (64). Briefly, spines were recognized by their small size and specific morphology. In CA1 str. oriens and radiatum, dendrites that have spines and do not receive type I (asymmetric) inputs on their shafts are known to be pyramidal cells (65), whereas dendrites receiving type I synapses on their shafts are interneurons (66). The robustness of this classification method was reconfirmed recently (64). Because in str. lacunosum-moleculare the shafts of pyramidal dendrites may receive type 1 inputs (65) they were not distinguished from interneuron dendritic shafts. Also, the dendritic shafts in somatosensory cortex S1 could not be identified based on morphological features.

Immunofluorescent labeling and confocal laser-scanning microscopy
Before the immunofluorescent staining, the 50 µm thick sections were washed in PB and Tris-buffered saline (TBS). This was followed by blocking for 1 hour in 1% human serum albumin (HSA) and 0.1% Triton X-100 dissolved in TBS. After this, sections were incubated in mixtures of primary antibodies overnight at room temperature. After incubation, sections were washed in TBS, and were incubated overnight at 4°C in the mixture of, all diluted in TBS. Secondary antibody incubation was followed by washes in TBS, PB, the sections were mounted on glass slides, and coverslipped with Aqua-Poly/Mount (Polysciences). Immunofluorescence was analyzed using a Nikon Eclipse Ti-E inverted microscope (Nikon Instruments Europe B.V., Amsterdam, The Netherlands), with a CFI Plan Apochromat VC 60XH oil immersion objective (numerical aperture: 1.4) and an A1R laser confocal system. We used 405, 488, 561 and 647 nm lasers (CVI Melles Griot), and scanning was done in line serial mode. Image stacks were obtained with NIS-Elements AR software, and deconvolved using Huygens Professional software (www.svi.nl).

STORM super-resolution microscopy
Before the immunofluorescent staining for the super-resolution experiments, the 20 µm thick sections were washed in PB and TBS. This was followed by blocking for 1 hour in 1% HSA dissolved in TBS. After this, sections were incubated in mixtures of primary antibodies; overnight at room temperature. After incubation, sections were washed in TBS, and sections were incubated overnight at 4°C in the mixtures of secondary antibodies. Secondary antibody incubation was followed by washing in TBS, PB; then hippocampi were cut out with scalpels in buffer, sections were dried on clean coverslips, and stored for maximum 3 weeks at room temperature in a dry environment before imaging. 3D direct-STORM (direct Stochastic Optical Reconstruction Microscopy) acquisition protocol was used as described before (67  illumination. The acquisition in the two channels was done in sequential mode. Confocal stacks were deconvolved (Huygens Professional), and -together with the STORM molecule listsprocessed for further analysis in VividSTORM software (see "Analysis"). Imaging was performed with identical parameters (depth in the section, laser intensities, etc.) for all samples.

Electron tomography
For the electron tomographic investigation we used 150 nm thick sections from the hippocampal CA1 region from the anti-vAChT immunogold stained material (see: "Preembedding immunoelectron-microscopy"). Before electron tomography, serial sections on single-slot copper grids were photographed with a Hitachi H-7100 electron microscope and a Veleta CCD camera. Serial sections were examined at lower magnification, and vAChT-positive synaptic terminals from the CA1 area selected. For each terminal, the section containing the largest synaptic cross section was chosen for electron tomography. After this, grids were put on drops of 10% HSA in TBS for 10 minutes, dipped in distilled water (DW), put on drops of 10 nm gold conjugated Protein-A in DW (1:3), and washed in DW. Finally, we deposited 5 and 5 nm thick layers of carbon on both sides of the grids. Electron tomography was performed using a Tecnai T12 BioTwin electron microscope equipped with a computer-controlled precision stage (CompuStage, FEI). Acquisition was controlled via the Xplore3D software (FEI). Regions of interest were pre-illuminated for 4-6 minutes to prevent further shrinkage. Dual-axis tilt series were collected at 2 degree increment steps between -65 and +65 degrees at 120 kV acceleration voltage and 23000x magnification with -1.6 --2 µm objective lens defocus. Reconstruction was performed using the IMOD software package (68). Isotropic voxel size was 0.49 nm in the reconstructed volumes. After combining the reconstructed tomograms from the two axes, the nonlinear anisotropic diffusion (NAD) filtering algorithm was applied to the volumes.
Segmentation of the terminals has been performed on the virtual sections using the 3Dmod software, and measurements were done on the scaled 3D models.

Isolation of synaptic vesicles
Synaptic vesicles were isolated according to (12), with some modifications. Briefly, cortices and hippocampi of ten C57Bl6 mice (80 days old, mixed gender) were pulverized mechanically after freezing in liquid nitrogen. The resulting fine powder was then homogenized

Immunostaining and flow cytometry analysis of isolated synaptic vesicles
The vesicles were dialyzed overnight, at 4 o C, in phosphate buffered saline (PBS; 800ml). After resin polymerization, small pieces were cut and removed from the coverslip surface, glued onto plastic blocks, and 40 nm thick sections were prepared using an ultramicrotome (Leica EM UC6) and picked up on Formvar-coated single-slot grids. Ultrathin sections were examined in a Hitachi 7100 electron microscope equipped with a Veleta CCD camera (Olympus Soft Imaging Solutions, Germany). For CLSM analysis coverslips were washed in PB and Tris-buffered saline (TBS). This was followed by blocking for 1 hour in 1% human serum albumin (HSA) dissolved in TBS. After this, sections were incubated in mixtures of primary antibodies for 1 hour at room temperature. After incubation, sections were washed in TBS, and were incubated for 1 hour at room temperature in the mixture of secondary antibodies, all diluted in TBS. Secondary antibody incubation was followed by washes in TBS, PB, DW and the coverslips were mounted on glass slides with Aqua-Poly/Mount (Polysciences). Immunofluorescence was analyzed using a Nikon Eclipse Ti-E inverted microscope (Nikon Instruments Europe B.V., Amsterdam, The Netherlands), with a CFI Plan Apochromat VC 60XH oil immersion objective (numerical aperture: 1.4) and an A1R laser confocal system. The possible influence of chromatic aberration on PSFposition was controlled using fluorescent TetraSpeck™ Microspheres (0.1 µm diameter, blue/green/orange/dark red, ThermoFisher Scientific). We used 405, 488, 561 and 647 nm lasers (CVI Melles Griot), and scanning was done in line serial mode. Image stacks were obtained with NIS-Elements AR software, and deconvolved using Huygens Professional software (www.svi.nl).

Statistical analysis
During segmentation of tomographic volumes, z-scaling was calculated from the thickness difference of the reconstructed volume and the original section thickness, and applied to the 3D models. Mesh surface areas and volumes inside meshed objects were measured with the "imodinfo" program.
In STORM imaging "molecule lists" were exported from NIS in txt format, and the 3 image planes of the ics-ids file pairs from the deconvolved confocal stacks matching the STORMvolume were converted to the ome-tiff format using Fiji software. Confocal and corresponding STORM images were fitted in VividSTORM (69).
When data populations in this work had a Gaussian distribution according to the Shapiro-Wilks W test, we reported parametric statistical features (mean ± SD), otherwise we reported non-parametric statistical features (median, interquartile range). For the presentation of electrophysiological data we used median and interquartile range, because the data did not show Gaussian distribution. Two non-parametric groups were compared using the Mann-Whitney U test. Wilcoxon signed-rank test was used for calculating significance between dependent non-parametric data groups. F-test was used to compare the variability of data groups. The types of statistical tests used in different investigations are indicated in the text. We   : We did not identify the cell types that establised the dendritic shafts in CA1 lacunosum-moleculare and S1 (see Supplemental Experimental Procedures).