Synaptic and circuit functions of multitransmitter neurons in the mammalian brain

Abstract

Neurons in the mammalian brain are not limited to releasing a single neurotransmitter, but often release multiple neurotransmitters onto postsynaptic cells. Here we review recent findings of multitransmitter neurons found throughout the mammalian central nervous system. We highlight recent technological innovations that have made identification of new multitransmitter neurons, and study of their synaptic properties possible. We also focus on mechanisms and molecular constituents required for neurotransmitter corelease at the axon terminal and synaptic vesicle, as well as some possible functions of multitransmitter neurons in diverse brain circuits. We expect these approaches will lead to new insights into mechanism and function of multitransmitter neurons, their role in circuits, and contribution to normal and pathological brain function.

Introduction

A neuron is often defined by the identity of the neurotransmitter it releases (i.e. glutamatergic, GABAergic, cholinergic…) with this label specifying the function of a neuron within a circuit. However, the designation of a molecule as the “principal” neurotransmitter of a neuron obscures the genuine diversity of synaptic signaling and hampers further inquiries into neuronal and circuit function as many, if not all, neurons release more than one neurotransmitter.

The many recent discoveries of multitransmitter neurons in mammalian systems were preceded by a long history of study in other organisms¹ such as the amphibians², mollusks³, and the crustacean somatogastric system⁴. Nevertheless, a review of multitransmitter neurons in the mammalian brain is timely, as exciting new studies exemplify the diverse array of neuronal subtypes in the brain that have the molecular machinery necessary to release more than one neurotransmitter^5–9. Furthermore, new “intersectional” genetic and viral tools permit manipulation of specific neuronal populations suspected of neurotransmitter corelease¹⁰ and enable functional analysis of cotransmitting neurons in vitro and in vivo^11,12.

We expect emerging in vivo studies to lend new insight into how each neurotransmitter released from a multitransmitter neuron(s) contributes to the dynamics and function of interconnected circuits during behavior. As many of the multitransmitter neurons we review here have historically been viewed as releasing a single neurotransmitter, studies in this field have the potential to assign functions previously ascribed to dopamine, acetylcholine, or serotonin to different neurotransmitters released from those same neuronal populations. We conclude that neuroscience is now at a stage where we can confidently identify when and where cotransmission occurs in the brain as well as reveal its contribution to normal and pathological brain function.

Neurotransmitters take many forms, including small molecules (acetylcholine (Ach), glutamate, GABA, and monoamines), peptides (enkephalin, dynorphin, somatostatin), purines (ATP), lipophilic esters (endocannabinoids), and gases (nitric oxide). Here we will focus on activity-dependent neurotransmission of multiple small molecule neurotransmitters (i.e. cotransmitting or multitransmitter neurons) within the mammalian central nervous system (Table 1–2). Furthermore, we will not discuss changes in neurotransmitter release during development, neurotransmitter switching, and corelease of neurotransmitters in other model organisms as recent reviews can be found elsewhere^13–15. We close by speculating on some possible cellular and circuit functions for release of two or more transmitters in different brain regions that we highlight throughout the review.

TABLE 1:

Neurotransmitters and associated vesicular transporters

Neurotransmitter (abbr.)	Vesicular transporter (protein)	Vesicular transporter (gene)
GABA	vGAT/vMAT2	Slc32a1/Slc18a2
Glutamate (Glu)	vGluT1/2/3	Slc17a7/ Slc17a6/ Slc17a8
Acetylcholine (Ach)	vAchT	Slc18a3
Glycine (Gly)	vGAT	Slc32a1
Dopamine (DA)	vMAT2	Slc18a2
Serotonin (5-HT)	vMAT2	Slc18a2

Table 2:

Definitions

Corelease: Process by which two (or more) neurotransmitters are released by a single neuron following an action potential.

Cotransmission: Process by which two (or more) neurotransmitters are synaptically released in response to an action potential and detected by receptors on the postsynaptic cell.

Multitransmitter neuron: A neuron that is capable of releasing more than one neurotransmitter.

Co-packaging: Process by which a coreleasing neuron transports two different neurotransmitters into the same synaptic vesicle.

Mechanisms/determinants for vesicular corelease

Vesicular release of a neurotransmitter requires a neuron to accumulate the signaling molecule in the presynaptic terminal to a concentration that permits it can to be packaged into a vesicle and released. Accumulation can occur by neurotransmitter synthesis and/or reuptake, and packaging is performed by the vesicular transporters. Therefore, concluding that a neuron can only release one or more neurotransmitters (i.e. that it can corelease), requires confirmation that both steps occur. Each of these components can also be independently modulated or modified to restrict or enhance release of a transmitter underscoring their importance to neurotransmission. Finally, knowledge of these components allows precise genetic perturbation of corelease by manipulating each neurotransmitter individually. In this context we will review the examples of multitransmitter neurons from the mammalian central nervous system and identify, if possible, when the molecular requirements for synthesis, accumulation and packaging have been determined. Importantly, neurons can uptake and package neurotransmitters using atypical pathways, making this process sometimes difficult.

Vesicular transporter synergy and antagonism

In multitransmitter neurons, where vesicular packaging of more than one neurotransmitter occurs, it is important to consider how vesicular transporters are differentially sensitive to intra-vesicular factors such as pH or membrane potential¹⁶. This becomes particularly important when considering multitransmitter neurons that exhibit co-packaging of two neurotransmitters within the same vesicle (examples discussed below).

Synaptic vesicles are acidified by vacuolar H+-dependent adenosine triphosphatases (V-ATPases) that pump protons into the vesicle lumen¹⁷. This creates a pH gradient (ΔpH) and a vesicular membrane potential (Δψ) that can both power the uptake of neurotransmitters by vesicular transporters (Table 1)¹⁶. Vesicular transporters for monoamines (vMAT1/2) and Ach (VAChT) transport positively charged neurotransmitters in exchange for 2H+ and are more dependent on ΔpH than Δψ^18,19. Conversely, transport of glutamate by vGluT1/2/3 depends primarily on Δψ. Glutamate is an anion at neutral pH and loading of glutamate into vesicles produces a change in charge and H+ opposite to vMAT and VAChT²⁰. These complementary dependencies may explain the observed synergies between vGluT and VAChT or vGluT and the vesicular zinc transporter (ZnT3) when these are localized to the same synaptic vesicle^5,21,22.

In contrast, the vesicular GABA/glycine transporter (vGAT) transports neutral zwitterions. Studies conclude that vGAT acts as a GABA/H+ antiporter rather than a GABA/Cl- cotransporter and depends on both ΔpH and Δψ^23,24. Therefore, the presence of vGAT on synaptic vesicles may have little impact on the activity of other transporters on the same vesicle with little effect on their function due to its codependence on ΔpH and Δψ and its minimal effect on either gradient. Functional studies on vesicles carrying both vGAT and another vesicular transporter, such as vGluT2, are needed to test if vGAT may alter the co-packaging of other transmitters into the same vesicle^25,26. Below, we take these details into account in multitransmitter neurons where there is evidence for co-packaging and discuss how it may affect signaling.

Multitransmitter neurons with co-packaging

GABA/Glutamate

Typically, neurotransmitter release, at least through actions on ionotropic receptors, is thought to have either an inhibitory or excitatory effect on the postsynaptic cell. Perhaps then, the most confounding example of multitransmitter neurons arises from GABA/glutamate cotransmitting neurons. Recently, using a combination of genetic crosses and in situ hybridization, a screen for GABA/glutamate coreleasing neurons revealed that 30 different brain regions have neurons that express the genes required for synaptic vesicle packaging of both GABA (vGAT; Slc32a1) and glutamate (vGluT2/3; Slc17a6/Slc17a8)⁸. These findings indicated that GABA/glutamate cotransmission may be more widespread than previously thought. However, cotransmission by most of these recently identified cell-types has yet to be physiologically confirmed; therefore, here we will focus on three regions where the existence of GABA/glutamate coreleasing neurons has been most convincingly demonstrated: Supramammilary nucleus (SuM), Ventral tegmental area (VTA), and Entopeduncular nucleus (EP).

Electron microscopy studies of axon terminals in the dentate gyrus (DG) demonstrated anatomical evidence for the existence of GABAergic and glutamatergic vesicular transporters in the same axon²⁷. These studies found that these terminals arose from long-range projections from the SuM and provided evidence that the GABAergic and glutamatergic vesicles segregated into distinct pools and occupied distinct presynaptic terminals²⁸. Furthermore, optogenetic activation of SuM terminals in the DG produces monosynaptic, short latency release of both GABA and glutamate onto postsynaptic granule cells and interneurons (Hashimotodani et al., 2018). Studies using chemogenetic manipulation of projections from the SuM to DG reveal that this circuit is involved in the regulation of arousal²⁹.

Neuronal diversity within the VTA has been studied in depth and excellent detailed reviews exist elsewhere³⁰. One of the VTA neuronal subtypes is distinguished by coexpression of the genes encoding the vesicular transporters for GABA (Slc32a1) and glutamate (Slc17a6)³¹. These neurons are distributed mostly in the medial subregions of the VTA including the interfasicular nucleus, paranigral nucleus, and rostral linear nucleus of the raphe³². They monosynaptically connect to neurons within the lateral habenula (mostly medial subregions) and cotransmit GABA and glutamate onto individual LHb neurons³¹. In vivo optogenetic excitation of VTA input to the LHb (using either Slc17a6-IRES-Cre or Slc32a1-IRES-Cre mice) results in either excitation, inhibition, or both of extracellularly recorded LHb neurons, suggesting that the postsynaptic LHb neuron, presumably by differential insertion of ionotropic GABA and glutamate receptors, may determine whether the input is primarily excitatory or inhibitory. Immunogold electron microscopy indicates that >60% of the axon terminals in LHb from VTA are positive for both vGAT and vGluT2, while smaller subsets label for one or the other transporter. Interestingly, immunogold EM also indicated that single axon terminals make both symmetric (GABAergic) and asymmetric (glutamatergic) synapses (see Figure 5 in³²), suggesting segregation of synaptic vesicles into glutamatergic and GABAergic synapses within the same axon and bouton. Finally, immunogold EM and immunoprecipitation of synaptic vesicles from LHb found that vesicles were largely vGluT2 or vGAT positive, but rarely copositive, suggesting that GABA and glutamate are typically packaged into separate vesicles in the LHb³². Together these studies suggest that even though individual axons from the VTA express both vGluT2 and vGAT, synaptic vesicles and vesicular transporters are largely segregated into distinct pools for release, rather than released together at the same synapse. How a neuron traffics each transporter to a separate vesicle pool and successfully forms and maintains both a symmetric and asymmetric synapse from a single bouton to a common postsynaptic dendrite is unknown and indicates that a great deal remains to be discovered about the molecular processes of synapse specification and maintenance.

The entopeduncular nucleus (globus pallidus internus (GPi) in primates) is a major output nucleus of the basal ganglia, and is typically discussed as a GABAergic region that inhibits targets in the thalamus as part of cortical-basal ganglia-thalamocortical loops³³. However, EP/GPi has diverse neuronal populations and functions, the latter primarily via its dense projections targeting the LHb^34–36. Due to immunohistochemical and behavioral findings, LHb-projecting EP/GPi neurons were first suggested to be glutamatergic, purely excitatory, and signal aversive/negative outcomes^37,38. Later they were found to simultaneously transmit both glutamate and GABA, and be capable of both exciting and inhibiting LHb neurons^26,39 (Figure 1A). GABA/glutamate coreleasing neurons in the EP of mice and GPi in humans were found to be marked by expression of Somatostatin (Sst) facilitating specific targeting of these neurons in functional studies³⁹. Similar to LHb-projecting VTA neurons, LHb-projecting EP neurons have single axon terminals that make both symmetric (GABAergic) and asymmetric (glutamatergic) synapses as determined by EM³².

Figure 1: Synaptic diversity of cotransmitting neurons.

(A) Three examples of cotransmitting neurons that employ vesicular co-packaging of two neurotransmitters at the presynaptic terminal. (B) Ach/GABA cotransmitting neurons found in cortex (VIP/Chat+) release Ach and GABA at different presynaptic terminals and independently package these neurotransmitters into separate vesicle pools. (C) Many midbrain DA neurons release three neurotransmitters, DA and GABA are co-packaged in the same vesicle, whereas glutamate is independently packaged and released at separate presynaptic sites.

Recent studies using electrophysiological recordings and minimal optical stimulation of synapses of EP somatostatin neurons in LHb examined whether GABA and glutamate are co-packaged in the same synaptic vesicle or segregated into independent vesicular pools. Optical quantal analysis determined that the statistical properties of unitary responses resulting from activation of cotransmitting EP release sites were consistent only with co-packaging of GABA and glutamate into the same synaptic vesicle²⁵ (Figure 2). These results support a model in which individual synaptic vesicles express both vGAT and vGluT2 and GABA/glutamate are released simultaneously to activate postsynaptic receptors across the synaptic cleft. The excitatory/inhibitory ratio at each synapse can be adjusted independently by regulating the activity of the vesicular transporters^26,40 or adjusting the composition of postsynaptic receptors. The structure, organization, and regulation of the postsynaptic receptors and related molecular constituents of GABA/glutamate synapses remain unknown, but appear specialized within LHb due to low expression of the glutamatergic postsynaptic terminal associated MAGUK PSD-95²⁵. Further studies are needed to understand the interactions between vGAT and vGluT2 transport into single vesicles, as each transmitter relies on distinct chemical and electrical gradients for vesicular transport which may alter their efficacy when present on the same vesicle^24,41,42.

Figure 2: Methods for investigating individual cotransmitting neurons and synapses.

(A) Methods to evaluate gene expression in single neurons such as single-cell whole transcriptome sequencing (shown) can examine expression levels of many genes in single neurons to determine if the genetic constituents required for neurotransmitter corelease are present (dots circled in purple represent individual Sst+ GABA/Glutamate coreleasing neurons isolated from EP, color represents gene expression level for vGluT2 (left) or VGAT (right)). (B) Fluorescent in situ hybridization (FISH) allows for confirmation of Sc-seq results in tissue without losing spatial patterns of expression. (C) Confocal image of tissue section from the LHb containing axons labeled from Sst+ EP neurons (YFP) and stained for synaptic proteins. Examining protein expression in synaptic terminals using high resolution methods such as array tomography (shown), electron microscopy, and super-resolution imaging is critical for examining the distribution/localization of pre and postsynaptic vesicular transporters, receptors, and synaptic organizers. (D) (Top) Zoomed image of area highlighted in (C) showing overlapping expression of VGAT and VGlut2 in synaptic terminals. (Bottom) Enrichment of each protein within a terminal over scrambled expression patterns demonstrates high concentrations of VGAT and VGluT2 in presynaptic terminals. (E) Diagram of optical components required for stimulation of individual synapses in acute brain slices. (F) Illustration of viral targeting of optogenetic activators to specific genetically defined cotransmitting neurons in the entopeduncular nucleus (EP) and activation of their axons using light guided by a DMD (digital micromirror device) while performing whole cell recordings in LHb. (G) Optical stimulation can be targeted to a grid of many small spots that overlay the recorded neuron and allow of stimulation of single axons. Action potentials are blocked (TTX/4-AP) to restrict spreading of optical axonal stimulation to multiple synapses on the same axon. (H) Careful calibration of optical stimulus parameters is required to enter into a minimal stimulation regime where stimulation of individual synapses is ensured, and quantal analysis can be performed. (I) When the neuron is voltage clamped at an intermediate potential both GABAergic (blue dot) and glutamatergic (red dot) post-synaptic currents and be observed simultaneously on single trials. (J) Scatterplot of the peak amplitudes for all trials shown in (I) to highlight strong correlation between GABAergic and glutamatergic responses, and providing strong evidence for co-packaging of the two neurotransmitters into the same synaptic vesicle (Figure modified from^25,39).

GABA/Glycine

Studies from the 1990s demonstrated cotransmission of GABA and glycine from inhibitory interneurons in the spinal cord⁴³. These studies were some of the first to describe the “co-packaging” phenomenon where two different neurotransmitters were loaded into individual synaptic vesicles. GABA/glycine cotransmission has since been observed in other areas including the auditory brainstem and cerebellum^44–46. By adulthood, both transmitters are inhibitory, activating their cognate chloride permeable ligand gated ion channels, and hyperpolarizing their postsynaptic targets⁴⁷. The kinetics of glycine receptors are faster than those of GABA receptors, resulting in an IPSC with fast and slow components^43,44,48. Notably, GABA and glycine compete for packaging into synaptic vesicles as both are transported by the same vesicular transporter vGAT⁴⁹ (Figure 1A). Therefore, packaging GABA and/or glycine into vesicles is largely dependent on their intracellular (intraaxonal) concentration, which are controlled by either intracellular synthesis (glutamate decarboxylase (GAD) 1/2) or plasma membrane transporters (GlyT1/2), respectively⁵⁰. These results suggest that the potency of a synapse depends not only on the number of postsynaptic receptors, but also on the concentrations of presynaptic transmitter and how much transmitter is loaded into the vesicle. Therefore, assumptions regarding saturation of postsynaptic receptors following single vesicle fusion may not hold at these or other synapses where competition for vesicular transport is high. This may be a general feature of multitransmitter neurons that exhibit co-packaging as it likely puts a lower limit on the amount of each neurotransmitter loaded into a single vesicle when compared to one vesicle with one transmitter. Together these studies frame an interesting role for cotransmission where two transmitters provide hyperpolarization of the postsynaptic cell with different kinetics, allowing the presynaptic neuron tight control over postsynaptic firing rates.

ACh/glutamate

Cholinergic neurons have an essential role in many circuit and cognitive functions within the brain from learning and memory⁵¹, sensory perception⁵², synaptic plasticity⁵³, and arousal⁵⁴. However, many of these neurons also contain the vesicular glutamate transporters (vGluT1/2/3) that are coexpressed with molecular machinery for acetylcholine synthesis, vesicular packaging, and release in multiple brain regions including the striatum, basal forebrain and medial habenula (mHb)^22,55,56. Axon terminals from mHb to the interpeduncular nucleus (IPN) contain both vGluT1/2 and VAChT and cotransmit both glutamate and acetylcholine onto individual neurons in the IPN⁵⁵. Single cell sequencing and ISH studies confirmed that these neurons reside in the ventral and ventrolateral subregion of the mHb and form a genetically distinct subtype^57,58. Although postsynaptic AMPA and NMDA receptors are activated by single action potentials in the presynaptic axon, nicotinic acetylcholine receptors (nAchRs) were only activated slowly following prolonged (50 Hz, 5 s) stimulation of inputs indicating that glutamate and acetylcholine have different transmission modes⁵⁵.

Striatal cholinergic interneurons (CINs) cotransmit Ach and glutamate locally onto spiny projection neurons (SPNs)⁵⁹ and dopaminergic axonal terminals^60–62. Glutamate release onto SPNs activates postsynaptic ionotropic glutamate receptors with distinct AMPA/NMDA ratios than cortico-striatal synapses suggesting distinct plasticity rules from these disparate inputs⁵⁹. Heroic electrophysiological experiments that intracellularly recorded directly from DA axon terminals in striatum unambiguously determined that synaptic release of Ach from CINs onto dopamine axons primarily activate nAchRs and are even capable of locally inducing action potentials (APs)⁶¹. These axonally induced APs are capable of driving DA release in vitro⁶⁰ and have been hypothesized to act as a mechanism for local control of DA release within the striatum^63–67.

Immunogold electron microscopy and immunoprecipitation experiments suggest that vGluT and VAChT co-package glutamate and ACh into the same synaptic vesicle in both mHb to IPN synapses and striatal cholinergic to medium spiny neuron synapses^21,22,55 (Figure 1A). Interestingly, several groups have found that function of vGluT and VAChT are synergistic and that depletion of one transporter or neurotransmitter also depletes loading and release of the other (Frahm et al., 2015; Gras et al., 2008; Nelson et al., 2014). These findings lie in contrast to those described above for neurons that co-package GABA and glycine which show an antagonistic relationship with regards to co-packaging. Together these examples illustrate the importance of vesicular transporter synergism/antagonism for neurons that co-package two neurotransmitters into synaptic vesicles.

5-HT/glutamate

Serotonin neurons from the mid and dorsal raphe nuclei innervate the entire forebrain and act through a wide variety of receptors (14 distinct subtypes) eliciting differential physiological responses^69,70. It then may come as no surprise that a consensus on the primary functions for the serotonin system is lacking, ranging from reinforcement⁷¹, to suppression of locomotion⁷², to promotion of anxiety-like behaviors⁷³. To add to this complexity, it is now clear that a subset of serotonergic neurons in the dorsal and median raphe nuclei express vGluT3 (Slc17a8)^74–76. Those in the dorsal raphe project widely throughout the cortex and increase or decrease their activity to reward and punishment, respectively^75,77. 5-HT/vGluT3+ neurons from the dorsal raphe also target the VTA where they make asymmetric synapses and monosynaptically release glutamate to excite accumbal projecting dopamine neurons⁷⁸. While the membrane serotonin transporter (SERT) is coexpressed at axonal sites with vGluT3, it is unknown if serotonin and glutamate are co-packaged and released from the same presynaptic synapse or if they are packaged into different vesicles and/or released at different sites⁷⁸. In support of co-packaging, synergism has been reported between vGluT3 and vMAT2, by which vGluT3 promotes vesicular loading of 5-HT via vMAT2⁷⁹. In addition, deletion of serotonin synthetic enzyme tryptophan hydroxylase 1 (Tph1) or vGluT3 (Slc18a8) from 5-HT/vGluT3+ neurons both result in increased anxiety-like behaviors suggesting that disrupting signaling from either transmitter has similar consequences^77,79. Compounding the diverse and extensive functions of 5-HT neurons, median raphe 5-HT/vGluT3+ neurons also activate CA1 inhibitory interneurons to disynaptically inhibit CA1 pyramidal neurons⁸⁰. Together these data suggest a complementary or synergistic relationship between glutamate and 5-HT release from 5-HT/vGluT3+ neurons with direct effects on cortical and midbrain structures. The detailed mechanisms by which 5-HT and glutamate release effects circuit or synaptic function in these areas are yet to be revealed.

Multitransmitter neurons with independent release

Ach/GABA

In contrast to neurons that cotransmit Ach and glutamate, separate neuronal subpopulations are capable of cotransmitting Ach and GABA. These two types of Ach-releasing neurons form non-overlapping neuronal populations, reside in different brain regions, and have distinct synaptic mechanisms for corelease of neurotransmitter. Here we review Ach/GABA releasing neurons from the retina, basal forebrain/globus pallidus, and cortex where evidence points to distinct pools of synaptic vesicles containing either Ach or GABA which may be released at separate sites.

Functional and immunohistochemical studies identified starburst amacrine cells (SACs) as the sole source of Ach in the retina^81,82. Further studies suggested these cells also express GABA synthetic enzymes and are capable of releasing GABA^83,84. More recently, an elegant study employing paired recordings from neighboring retinal neurons demonstrated monosynaptic transmission of both Ach and GABA from SACs to direction-sensitive retinal ganglion cells (DSGCs)⁸⁵. Release of GABA from SACs is restricted to the dendrites of the postsynaptic DSGC’s null direction, whereas release of Ach is not spatially restricted. Ach and GABA release are differentially sensitive to calcium concentration and calcium channel antagonists suggesting separate vesicle populations⁸⁵. These studies suggested that Ach and GABA are segregated into different synaptic vesicles and/or axonal boutons to endow single SACs with spatial selectivity over cotransmission of neurotransmitters.

Elsewhere in the CNS, neurons expressing genes for both Ach and GABA synthesis and vesicular transport were detected in the basal forebrain and later in the globus pallidus^86–89. Optogenetic activation of cortically projecting Ach/GABA releasing neurons evoked monosynaptic GABA mediated inhibitory currents and Ach mediated nicotinic excitatory currents in layer 1 interneurons. The GABA mediated IPSC could be ablated by selective genetic deletion of Slc32a1 (vGAT) in Chat-Cre+ neurons while the nicotinic EPSC persisted indicating that the loading of Ach did not depend on GABA vesicular transport⁸⁸. The proportion of postsynaptic layer 1 interneurons receiving an EPSC or an IPSC differed significantly and most received only one type of input. Additional array tomography studies of cortical axons projecting from pallidal Ach/GABA neurons revealed physically separate sites of vGAT and vAChT labeling suggesting distinct vesicular pools for the two neurotransmitters⁸⁹. More recently, olfactory bulb projecting cholinergic neurons from the basal forebrain were also shown to cotransmit GABA and Ach onto deep short axon cells within the internal plexiform layer of the bulb reinforcing the pervasive nature of GABA cotransmission from this population of cholinergic neurons throughout the brain⁹⁰.

Bipolar cortical interneurons expressing vasoactive intestinal polypeptide (VIP) and ChAT were first identified using immunohistochemistry in rodents and account for ~1/3 of all VIP+ cortical neurons^91,92. Transcriptomically similar neuronal subpopulations (VIP/ChAT+) have also been identified with large scale single-cell sequencing of motor and visual cortex in mouse, marmosets, and humans^6,93–95. These studies also confirmed that this cortical subtype expresses all the required genes for synthesis and release of both GABA and Ach. Systematic circuit mapping of the synaptic output of VIP/ChAT+ interneurons was recently undertaken to define how these neurons are integrated into cortical microcircuits^96,97. Optogenetic excitation of VIP/Chat+ neurons in motor and visual cortex revealed GABAergic output primarily to other interneuron subtypes with a bias towards the Sst+ cortical interneurons. Cholinergic EPSCs were sparse, but primarily detected in layer 1 interneurons and other VIP cells⁹⁶. Immunohistochemical analysis also supported independent trafficking of vGAT and vAChT in VIP/Chat+ axon terminals. This suggests, much like the cholinergic neurons in the basal forebrain and globus pallidus, that there are distinct vesicle populations for Ach and GABA release (Figure 1B). Interestingly, studies have revealed significant regional differences in the strength and connectivity of cholinergic output from VIP/ChAT+ neurons, as cholinergic output from VIP/ChAT+ neurons to layer 1 interneurons in mouse and rat medial prefrontal cortex occurs at a higher probability and produces a larger postsynaptic response^96,97. However, these differences may be due to the existence of a unique population of VIP- cholinergic interneurons in this region⁹⁶. Some evidence has begun to point towards a function for VIP/ChAT+ neurons in mPFC as optogenetically inhibiting this population following the cue in a 5-choice serial reaction time task reduced correct responses⁹⁷. Additional studies are needed to dissect how this neuronal population may differentially modulate local circuitry in distinct brain regions.

Glutamate/Glycine

Highlighting the diversity of neurotransmitter combinations and mechanisms of release, a class of vGluT3+ retinal amacrine cells was recently found to release both the inhibitory neurotransmitter glycine and excitatory neurotransmitter glutamate onto distinct populations of retinal ganglion cells (RGCs)^98,99. Glycinergic synapses are highly selective and shape contrast and size selectivity of postsynaptic “Suppressed by contrast” RGCs by depressing tonic firing⁹⁹. Glutamatergic synapses from vGluT3+ retinal amacrine cells are more promiscuous and target OFF alpha ganglion cells and other retinal cell types, but avoid “Suppressed by contrast” RGCs⁹⁸. Therefore, vGluT3+ retinal amacrine cells simultaneously excite cells activated by contrast and inhibit cells suppressed by contrast performing a dual role in retinal circuits. While transmission of glycine (and absence of glutamate) appears specific to a particular postsynaptic cell type, it is unknown if glycine and glutamate are targeted to segregated pools of synaptic vesicles, or if specificity for one neurotransmitter or the other is determined by the presence or absence of postsynaptic receptors. While EM studies have shown amacrine cell processes forming symmetric and asymmetric synapses¹⁰⁰, vGAT is not expressed in vGluT3+ retinal amacrine cells or axon terminals¹⁰¹, leaving the substrate for release of glycine a mystery.

Multitransmitter neurons with mixed co-packaging and independent release

DA/glutamate

Dopamine neurons of the ventral midbrain (DANs) comprise a diverse population of neurons targeting different regions and capable of releasing multiple neurotransmitters in addition to dopamine^30,102,103. Following initial findings of glutamate cotransmission in vitro^104–106, dopamine neurons in the ventral midbrain projecting to ventral and dorsal striatum expressing the vGluT2 were confirmed using optogenetic stimulation in brain slices, to release glutamate onto striatal spiny projection neurons and cholinergic interneurons^107–109. Immunogold EM studies and immunoprecipitation experiments suggest that vGluT2+ and vMAT2+ vesicles segregate into distinct vesicle populations and even separate axonal boutons/microdomains¹¹⁰ raising the possibility that release of these two transmitters may be controlled independently (Figure 1C). Functional studies have also demonstrated differential synaptic release properties and coupling to calcium channels for dopaminergic and glutamatergic transmission from DANs¹¹¹. In contrast, selective deletion of Slc17a6 from DA neurons in vivo abolished glutamate release, decreased DA release in the ventral striatum, reduced locomotor responses to cocaine administration in mice¹¹² and enhanced sucrose and cocaine self-administration¹¹³. These studies also found that vGluT2 was capable of stimulating monoamine uptake into synaptic vesicles by vMAT2 by lowering the intravesicular pH, providing a mechanism for reduced DA release in the absence of vGluT2¹¹². Differences in vesicular colocalization findings in these two studies^110,112 could be explained by the relative abundance of vGluT2/vMAT2 vesicles, compared to vMAT2 and vGluT2 only vesicles. The abundance of the latter would be expected to be in much greater as only a subset of DA terminals contains vGluT2, while the ventral striatum contains many axonal terminals from thalamus with only vGluT2 or from DANs that contain only vMAT2.

DA/GABA

Unlike glutamate release from DANs, which requires vGluT2, striatal GABA release from DANs does not require its typical cognate vesicular transporter vGAT¹¹⁴. Indeed, in DANs, neither GABA synthetic enzymes or vesicular transporters are required for GABA release, instead vMAT2 is required for GABA packaging into vesicles^114,115 (Figure 1C). Thus, the traditional markers used to identify GABAergic neurons cannot be used to determine which DANs release GABA. In support of the function vMAT2 as a bonafide GABA vesicular transporter, it can substitute for vGAT in classical GABAergic neurons to sustain GABA release¹¹⁴. Furthermore, DANs do not synthesize GABA using canonical GABA synthesis (i.e. using GAD65/67), instead they appear to accumulate GABA in two ways, 1) through an aldehyde dehydrogenase 1a1 (Aldha1a)-dependent pathway¹¹⁵, and/or 2) by scavenging extracellular striatal GABA via the membrane GABA transporter GAT1^116,117. However, only the GABA transporter, GAT1, is both necessary and sufficient to sustain GABA release in DANs, as knock out of Aldha1a reduces GABA release from DANs by about 50%^115,117. Furthermore, although Aldha1a was proposed to act within DANs to produce GABA via an atypical biosynthetic pathway¹¹⁵, cell type specific control of Aldha1a expression indicates that its effects on GABA release from DANs occur through a still mysterious action in non-dopaminergic cells¹¹⁷. Therefore, it is likely that all DANs that express GAT1 are capable of GABA release in addition to DA throughout striatum.

Despite being packaged by the same vesicular transporter, net DA and GABA transmission can have differential sensitivity to extracellular calcium concentration and presynaptic modulation by GPCRs. These findings suggest that DA and GABA may be segregated into not fully overlapping and potentially separate vesicle populations despite both depending on vMAT2 for vesicular loading¹¹⁸. Some evidence has begun to point towards a function for GABA corelease from DANs. Decreased GABA release from TH-positive neurons resulting from genetic deletion of E3 ubiquitin ligase Ube3a enhances positive reinforcement leading to increased optogenetic self-stimulation¹¹⁹. Additionally, reducing GABA cotransmission from DANs by genetic deletion of Aldh1a1 increases ethanol intake and preference in mice suggesting that GABA cotransmission from DANs plays a role in reward/ethanol seeking behaviors¹¹⁵. Together, these studies illustrate the complexity of neurotransmitter release from DANs. They have the capability to signal at multiple timescales (see below on hypothesized functions) and spatially separable presynaptic sites depending on the neurotransmitter released (DA, GABA, or glutamate), postsynaptic neuron, or striatal subregion¹²⁰. Teasing out the contribution of each cotransmitted signal to behavioral functions typically ascribed to DA alone would be a major contribution to the fields of motivated behavior, reinforcement learning, and associated diseases.

Circuit functions of cotransmission

Possible cellular and circuit functions for release of two or more transmitters include diverse temporal control over postsynaptic firing, increased ranges of synaptic plasticity, and frequency-dependent regulation of neurotransmission. Here, we highlight several examples of observed and potential cellular or circuit functions of cotransmission.

Multiple levels of temporal control of postsynaptic activity

Real-time monitoring of the kinase activity in behaving animals, indicate that DA release, via regulation of intracellular cAMP production, controls the activity of PKA in SPNs over the time scale of tens of seconds with differential effects on neurons that express type 1 or type 2 dopamine receptors¹²¹. Elegant experiments demonstrating the postsynaptic effects of optogenetically evoked DA on type 1 dopamine receptor expressing SPN firing indicate that DA release has long lasting (mins) effects on spiking output of striatal neurons¹²². DA release affects the postsynaptic cell considerably more slowly (~500 ms delay) than the synaptic currents generated by either cotransmitted GABA or glutamate (2–4 ms)^114,122. These significant differences in the latency of the response to the released transmitter impart each with different functions. GABA or glutamate can precisely activate or inhibit single APs from SPNs (demonstrated in¹¹⁴). In contrast, following a delay, DA could push the neuron into a long-lasting state of increased excitability in which activity of other excitatory synapses (e.g. incoming cortical or thalamic input) could be readily paired with postsynaptic firing and engage mechanisms for spike-timing-dependent plasticity (Figure 3A).

Figure 3: Possible cellular and circuit mechanism of multitransmitter neurons.

(A) GABA and glutamate release from midbrain DANs cause fast excitation and slightly longer inhibition of postsynaptic cells (dSPNs) due to the kinetics of each ionotropic receptor. The physiological effects following activation of type 1 DA receptors is delayed by ~500ms and increases firing rate towards the end of the spike train. Arrowheads mark timing of presynaptic action potential. (B) Ach/glutamate release from mHb to the IPN may be sparse but temporally precise when mHb firing rates are low, with synaptic transmission dominated by glutamate and minimal activation of extrasynaptic Ach receptors. During high frequency activity spillover of Ach reaches and activates extrasynaptic Ach receptors on many postsynaptic cells leading to broad, but temporally imprecise activation of IPN. (C) LHb neurons assigning “value” to each presynaptic input from EP based on good or bad outcomes. Because EP inputs cotransmit GABA and glutamate, LHb can tune each synapse positive or negative by insertion of glutamate or GABA receptors, respectively. This process is the equivalent of perceptron-like learning rules that classify contexts as good or bad in artificial neural networks.

Frequency-dependent regulation of neurotransmitter release

Investigations into Ach/glutamate release from the mHb to the IPN revealed frequency dependent effects of cotransmission on postsynaptic cells. Single optogenetic stimuli of axon terminals from mHb to IPN reliably evoke monosynaptic glutamatergic synaptic currents with no contribution from Ach receptors⁵⁵. However, following prolonged stimulation of axon terminals at higher frequencies (20–50Hz), a slow depolarizing current was observed that was blocked by nAch antagonists⁵⁵. Several groups have found evidence for Ach and glutamate being co-packaged into the same synaptic vesicles at this synapse^21,22,55,123; therefore, glutamate acts as the temporally precise, point-to-point transmitter, while Ach acts as a volume transmitter only accumulating enough to activate postsynaptic receptors at high presynaptic firing rates. Furthermore, whereas glutamate release may only activate a few postsynaptic receptors within the synapse and occasionally elicit a spike, Ach release can activate many more extrasynaptic receptors and result in high frequency action potential firing in multiple neurons within the IPN simultaneously (Figure 3B).

Target specific transmission of distinct neurotransmitters

Ach/GABA release from cortical VIP/Chat+ neurons provides an example of how a single neuron type can have differential effects dependent on the postsynaptic neuronal subtype. Ach is the primary transmitter when the postsynaptic neuron is a layer 1 interneuron or another VIP/Chat+ neuron and the postsynaptic receptors are nAchRs. But GABA is the primary transmitter when the postsynaptic neuron is a Sst+, PV+, or 5HT3aR+ interneuron. The net effect of VIP/Chat+ activity on the cortical microcircuit is disinhibition of excitatory pyramidal neurons, by engaging multiple cell-types across the cortical column^96,97. Accumulating evidence in VIP/Chat+ cells, and other Chat+ neurons throughout the brain, suggests that this is accomplished by specialized presynaptic terminals that target different neurotransmitters to specific synaptic sites allowing synapse specific control of postsynaptic firing^89,96,123.

GABA/glutamate cotransmission as a substrate for learning in a biological neural network

The finding that individual synaptic vesicles from EP somatostatin neurons in the LHb contain both GABA and glutamate opens an interesting possibility for a form of synaptic plasticity whereby the sign and weight of a synapse can be dynamically set in an activity dependent manner²⁵. Unlike other brain areas that separate GABA and glutamate into separate signaling channels, this feature renders this synapse similar to those found in artificial neural networks in which each synapse can take a signed and graded value. Furthermore, GABA/glutamate cotransmitting neurons are rare in the brain, but the LHb receives many projections from such neurons⁸, suggesting a specialized function of cotransmission for the computation performed by this circuit.

LHb neurons increase their activity when adverse outcomes occur or are predicted and provide a disynaptic inhibitory (GABAergic) input to VTA DANs^124,125. Thus, the neurons of LHb, during reward reinforced behaviors, learn to calculate the negative expected value of specific actions or contexts. In contrast, VTA DAN neurons calculate reward-prediction error (RPE) – the difference between the experienced value of an activity and that which had been predicted. As a recent study reported that non-DA neurons within VTA encode experienced value¹²⁶, VTA DA neurons could simply calculate RPE by subtracting the predicted value (provided by LHb) from the experienced value (provided by non-DA VTA neurons).

The activity-dependent insertion and removal of GABA and glutamate ionotropic receptors from postsynaptic terminals opposed to GABA/glutamate coreleasing synapses provides a mechanism for LHb neurons to learn to calculate expected value from information about sensory state, past experience, motor action, and other variables received from diverse brain areas, including EP. Simple activity dependent learning rules such as “insert GABA receptors into synapses that are activated when something good happens, and, conversely, glutamate receptors in those active something bad happens” provide the equivalent of perceptron-like learning rules that classify contexts as good or bad¹²⁷ (Figure 3C). Going further, if the VTA, or another brain structure, provides a graded RPE-like signal to the LHb during learning (RPE is the error signal for value), then a modification of this learning rule performs signed gradient descent to settle on a linear combination of signed and graded synaptic weights that calculate value directly. Such a mechanism is biologically plausible as the strength at many classes of synapses is regulated by the postsynaptic neuron by activity-dependent and dynamic insertion or removal of ionotropic receptors from the postsynaptic terminal. Although RPE-encoding VTA DA release in the LHb is a natural error signal for such plasticity, VTA projections do not appear to release DA in the LHb^128,129, such that the identity of this hypothetical error signal is still unknown. It might be carried by another neuromodulator, such as serotonin or norepinephrine, or by the activity of one of the many other GABA/glutamate coreleasing projections to the LHb.

Caveats and steps forward

Single-cell sequencing is an important new tool for genetically defining cell-types in tissues throughout an organism. Studies performed in the brain have led to discoveries of new neuronal classes, new features of known neuron-types, and perhaps most importantly an initial unbiased framework for further investigation into neuronal diversity in the brain. These studies have also shown the ubiquity of neuron-types that have the genetic constituents required for the release of more than one neurotransmitter^6,7,9,39,130. An important caveat to these findings is that most analyses performed in these studies are performed on populations of neurons. Therefore, it is important to confirm that individual neurons express all the genes necessary for the release of more than one neurotransmitter as analysis of the population alone may lead to incorrect conclusions regarding cotransmission. Careful characterization of mouse lines used to target genetically defined neurons is critical, as off-target Cre expression can easily lead to misinterpretation of cotransmission when in fact, two distinct neuronal populations are labeled. Additionally, in situ transcriptional and electrophysiological analysis of suspected coreleasing neurons are critical confirmatory steps in identifying bona fide coreleasing neurons.

If we are to eventually understand the function that coreleasing neurons play in neural circuits and behavior then the mechanism by which two or more neurotransmitters are released by a single axon or single synapse must be determined. Functionally determining whether both transmitters are simultaneously released by an axon/synapse and whether they are packaged into the same or distinct synaptic vesicles constrains models and hypotheses surrounding the function of cotransmission in a given circuit. To determine these properties at coreleasing axons/synapses minimal stimulation experiments are often necessary to examine the quantal content at an individual release site (Figure 2). Unfortunately, classical minimal stimulation approaches using extracellular stimulation are confounded when axons from many different types of presynaptic neurons are overlapping making it impossible to determine if one axon is being stimulated when looking for two types of postsynaptic response. Fortunately, approaches employing optogenetics permit genetic targeting of stimulation exclusively to a coreleasing neuron class, but careful spatiotemporal calibration of optical stimulation is required to convincingly stimulate individual axons/synapses^11,25. These approaches, combined with superresolution/expansion microscopy and imaging of pre and postsynaptic proteins^131,132 will lead to new insights into the mechanism and function of coreleasing neurons and begin to illuminate mysterious facets of their cell biology. This includes understanding how in cells with multiple classes of synaptic vesicles can target them to different terminals as well as what factors determine if each synaptic vesicle harbors one or multiple classes of neurotransmitter transporters. It is even more mysterious how these properties are maintained through the complex synaptic vesicle life cycle, although a preponderance of kiss-and-run type release events¹³³ might lessen the challenge. Presumably protein domains present on the vesicular transporter, which have previously been implicated in vesicle assembly, sorting or recycling as well as on other vesicular proteins must contribute to establishing the unique biology of multitransmitter neurons^{111,134–136}.

Separately, a combination of experimental and theoretical work is necessary to understand the diversity of functions of multi-transmitter neurons. For example, dopamine, GABA, and glutamate released by DANs may elicit prolonged waves of postsynaptic effects, cascading from the millisecond changes in excitability triggered by ionotropic receptors, to the tens of seconds on biochemical signaling and excitability^{108,114,121,122} (Figure 3A). These temporal waves could be different for each postsynaptic target, which is sensitized to different subsets of transmitters by the complement and type of receptors it expressed^108,137,138. Conversely, GABA and Ach coreleased in cortex may act in parallel on different targets to achieve a single common function, such as placing cortex in a more excitable and pro-plasticity state by simultaneously inhibiting Sst+ interneurons with GABA while using Ach release to excite disinhibitory interneurons and activate muscarinic receptors on pyramidal cells^{89,96,97,139,140}. Lastly, as discussed above, GABA and glutamate corelease might serve to sharpen postsynaptic potentials but might also create signed and graded synapses that provide an ideal substrate for plasticity (Figure 3C). Much remains to be discovered about both the peculiar cell biology of these cells and their function in circuits, and ultimately behavior.