Neuronal subtypes have been best defined and manipulated by their unique gene expression profiles (Colosimo et al., 2004; Gray et al., 2004; Baumgardt et al., 2007; Flames and Hobert, 2009; Winden et al., 2009). This has partially been made possible through the discovery and creative implementation of site-specific recombination (SSR). SSR affords the ability to selectively target neurons via controlled expression of Cre- and Flp-recombinases (Figure 2A; Dymecki, 1996; Dymecki and Tomasiewicz, 1998; Chai et al., 2000; Awatramani et al., 2003; Branda and Dymecki, 2004; Gong et al., 2007; Anastassiadis et al., 2009; Meinke et al., 2016). Such recombinases can be utilized to excise, invert, or conditionally express genes of interest via the “Lox-Stop-Lox” (LSL), “FRT-Stop-FRT” (Kuhlman and Huang, 2008), or “double-inverted orientation” [DIO; FLEx or FLEx (FRT)] methods (Atasoy et al., 2008; Schwarz et al., 2015). Of these, the DIO approach has proved extremely useful when paired with viral vectors to conditionally express genetic reporters or actuators in a region- or cell-type-specific manner. Additional recombinase/sequence pairs have recently been engineered to expand such approaches, including Dre/rox, VCre/VloxP, and SCre/SloxP (Anastassiadis et al., 2009; Suzuki and Nakayama, 2011). Similar to the Cre/lox system, these pairs can be used to delete, invert, and conditionally express genes of interest depending on the relative orientation of the rox, VloxP, or SloxP sites.

In cases where a neuronal population of interest lacks a single unifying molecular marker, multiple pairs of recombinases and sequences can be combined in an intersectional manner (Figure 2B). The Lox and FRT sites in these systems are oriented such that a gene will only be expressed in cells that express both Flp and Cre (for example, Frt-Stop-Frt followed by a FLEXed gene of interest; Dymecki et al., 2002; Awatramani et al., 2003). To reduce the possibility of leaky reporter expression due to stop codon read-through, an alternative intersectional strategy involves placing Lox and FRT sites at intronic sequences that divide a gene of interest, allowing for tighter control of gene expression. Moreover, intronically engineered sites and their adjacent exons can be oriented to facilitate genetic alterations only in cells that express both recombinases (ex. Cre-On, Flp-On or “COn-FOn”; Figure 2B), only one recombinase but not the other (ex. Cre-On/Flp-Off or Cre-Off/Flp-On), or in all cells except those that express both recombinases (ex. Cre-Off/Flp-Off; Fenno et al., 2014, 2020). Additionally, three-factor-dependent constructs have been generated (ex. Cre-On/Flp-On/VCre-On) to achieve even tighter genetic targeting control (Fenno et al., 2020). Such intersectional strategies also allow for the combination of spatially-restricted and cell type-specific drivers for genetic targeting. For example, using a retrogradely-transported Flp (Tervo et al., 2016) in combination with a cell type-specific Cre driver endows the ability to target cells that express a specific marker and project to a defined anatomical location [as in Rossi et al. (2021)].

While many experiments require the use of transgenic approaches to trace circuits or express actuators, engineering transgenic mice can be time-consuming, costly, and have limited experimental applications. The advent of viral genetics to express conditional genetic constructs (via the DIO or similar method) allows one to perform virtually any experiment in a cell type of interest if a Cre-driver is available (Haggerty et al., 2020; Nectow and Nestler, 2020). Viruses also provide better spatial control when targeting specific brain regions. Adeno-associated viruses (AAVs) are the most commonly used viruses in neuroscience. The simplicity of genomic structure, ease of packaging, and relatively non-pathogenic nature have made engineered AAVs ideal for genetic tracking, activity monitoring, and neuromodulation (Saleeba et al., 2019). Moreover, AAVs have selective tropism endowed by their capsid protein, which can be tailored for tissue of interest. For example, capsids such as AAV DJ/8, 2, and 9 are readily used for brain-specific expression (Haggerty et al., 2020; Nectow and Nestler, 2020). Due to the vast experimental possibilities brought about by viral genetics, teasing apart circuit function has been limited largely by the availability of recombinase-based genetic drivers that specify given cell populations. However, the advent of single-cell RNA sequencing is rapidly expanding to reveal potential gene candidates whose expression may serve as reliable and precise markers of different neuronal populations (Zeisel et al., 2015; Gokce et al., 2016; Chen et al., 2017b; Saunders et al., 2018; Tepe et al., 2018; Rossi et al., 2019; Hashikawa et al., 2020; Wallace et al., 2020). This novel information will provide a much larger repertoire of recombinase-expressing driver mice, a critical tool in the age of viral genetics.

In some instances, single genes – or even sets of genes – are unable to adequately characterize the function of neuronal populations, which may operate as molecularly heterogeneous ensembles. Therefore, it is sometimes desirable to characterize neurons in an activity-dependent rather than a genetically-dependent manner. To this end, elegant methods have recently been developed to target neuronal populations based on their activity (DeNardo and Luo, 2017). These methods co-opt immediate-early genes such as Fos (Greenberg and Ziff, 1984) and Arc (Lyford et al., 1995) to drive recombinase expression in active neuronal populations.

Targeted Recombination in Active Populations (TRAP; Guenthner et al., 2013; DeNardo and Luo, 2017) is one such technology in which a tamoxifen-inducible Cre-ER(T2), has been inserted downstream of a Fos promoter. In the presence of tamoxifen, active cellular populations expressing Fos, and therefore Cre-ER(T2), permit translocation of Cre recombinase to the nucleus to facilitate the expression of a Cre-dependent reporter (Guenthner et al., 2013). Similar models exist that use Arc to drive Cre-ER(T2) expression (Guenthner et al., 2013; Figure 2C). A variation of this technology, termed the Tet-tag system, uses a tetracycline transactivator domain downstream of the Fos promoter (Reijmers et al., 2007). In this approach, tetracycline-controlled transactivator protein (tTA) binds to the tetracycline responsive promoter element (TRE) to drive gene expression. In the presence of tetracycline, tTA cannot bind the TRE, and target gene expression is dampened. To use this system, tetracycline must be continually supplied in chow. Upon cessation of tetracycline provision, active neurons produce tTA, allowing the transcription of downstream reporters in a temporally-precise and activity-dependent manner (Reijmers et al., 2007).

One of the major limitations of both the TRAP and Tet-tag systems is the potential for the temporal lag in accurately capturing real-time neuronal activation. Since TRAP relies on the systemic administration of tamoxifen (or similar analogs), active neurons may be labeled throughout a 24-h time window. Similarly, due to the slow time course of tTA activation following the removal of doxycycline, the Tet-Off system also captures cells over extended times (Reijmers et al., 2007; Guenthner et al., 2013). Thus, more temporally precise labeling methods have been sought-after. One such method called “CANE” (capturing neuronal ensembles) utilizes a transgenic mouse expressing destabilized avian-specific receptor TVA under the control of the Fos promoter (Fos-2A-dsTVA; Sakurai et al., 2016). The dsTVA is fused with a degradation signal so that TVA closely mirrors Fos expression patterns. Upon delivery of pseudotyped lentivirus or rabies virus, neurons actively expressing Fos may be tagged for tracing or manipulation experiments (Sakurai et al., 2016). Other methods that tag activated cells based on calcium signals rather than immediate early gene expression and drug delivery have been shown to be even more temporally constrained. For example, CaMPARI, a photoconvertible genetically encoded calcium indicator, labels neurons only when both calcium levels are high and light is delivered to cells, identifying activated neurons on the millisecond timescale with spatial precision (Fosque et al., 2015). Similarly, FLiCRE (Fast Light and Calcium-Regulated Expression) is a method that labels activated cells at the transcriptional level (Kim et al., 2020; Figure 2D). It not only tags activated cells following a calcium influx but also enables the direct manipulation of the tagged cells via the concomitant light-induced expression of Cre. Collectively, these novel activity-driven targeting approaches allow researchers to screen for neural circuits underlying behaviors of interest. This is useful when a molecular marker for a region/behavior is not known, or when more than one molecular marker may contribute to a particular behavior.

Viruses provide many advantages over conventional tracers. Foremost, viral tracers may be engineered as conditional vectors compatible with diverse genetic markers, permitting cell type-specificity. Second, diverse promoters with varying strengths can be used to modulate expression levels or drive cell type-specificity. Lastly, viruses can be engineered to cross single synapses (Kuypers and Ugolini, 1990; Xu et al., 2020). Delivery of viral vectors using stereotaxic injections affords both temporal and spatial specificity, while also obviating the need to generate transgenic mouse models for genetic labeling experiments, so long as the recombinase driver of choice already exists. The following will give an overview of viruses used for both retrograde and anterograde labeling of neural circuits (Table 1).

Mapping neural circuits using fluorescent reporters combined with confocal microscopy remains a powerful method for navigating the brain. However, to identify connectivity with single-cell resolution, multiplexed mapping and sequencing approaches have evolved (Figure 5). For example, MAPseq (multiplexed analysis of projections by sequencing) and BRICseq (brain-wide individual animal connectome sequencing) are capable of surveying thousands of neurons in a single experiment (Kebschull et al., 2016; Han et al., 2018; Huang et al., 2020). These approaches utilize the Recombinant Sindbis virus, which rapidly achieves very high expression levels, to express unique mRNA barcodes in individual source cells. These barcodes spread throughout each source cell, including its axonal projections. Both the source area and downstream projections are then micro-dissected, dissociated, and sequenced. After sequencing, barcodes from projection areas are matched to the source area to reveal singe-cell projection patterns (Sun et al., 2021; Figures 5A,B). Recently, Gergues et al. (2020) used MAPseq to map projections from the ventral hippocampus (vHPC). They labeled vHPC neurons with a library of random RNA barcodes, examined 7 different target brain regions, and differentiated projection-specific populations of vHPC neurons (Gergues et al., 2020). One caveat with MAPseq/BRICseq is that the precise positions of individual cell bodies are lost due to tissue dissociation. To overcome this limitation, BARseq (barcoded anatomy resolved by sequencing), combines MAPseq with in situ sequencing to maintain cellular spatial localization (Chen et al., 2018, 2019; Figure 5C). BARseq has been applied to map projections from the mouse auditory cortex to 11 areas of the whole brain, confirming the laminar organization of the three top classes of projection neurons in the auditory cortex (Chen et al., 2019).

Other multiplexed mapping methods identify inputs through retrograde transport of barcodes. One such approach is Connect-seq, which determines the molecular identities of individual upstream neurons in a defined circuit by combining retrograde viral tracing and single-cell transcriptomics (Hanchate et al., 2020). Implementing this strategy, Hanchate et al., successfully targeted Cre-dependent PRV infection to corticotropin-releasing hormone neurons within the paraventricular nucleus of the hypothalamus (PVN). Flow cytometry was then used to isolate the upstream inputs, and single-cell RNA-sequencing (scRNA-seq) was used to determine the identities of these inputs (Hanchate et al., 2020). In a similar way, RABID-Seq utilizes pseudotyped rabies virus to label and deliver barcodes to presynaptic inputs in a cell type-specific manner, which may then be identified through scRNA-seq (Clark et al., 2021). Toward this, RABID-Seq has been used to identify axon guidance molecules as mediators of microglia-astrocyte interactions that promote CNS pathologies (Clark et al., 2021; Figure 5D).

In summary, these multiplexed input and output mapping approaches provide unprecedented resolution in the creation of accurate wiring diagrams, providing transcriptomic analyses that are lacking in other contemporary viral tracing approaches.

As neural tracing strategies have evolved, so too have the microscopes and imaging technologies necessary to visualize them. A major obstacle to visualizing deep brain tissue is signal distortion due to light scattering and spherical aberrations. Consequently, imaging long-range projections were effectively impossible. To mitigate these challenges, emerging optical techniques such as CUBIC (Susaki et al., 2014; Murakami et al., 2018; Matsumoto et al., 2019), CLARITY (Chung et al., 2013; Tomer et al., 2014), ScaleS (Hama et al., 2011, 2015), and SeeDB (see deep brain; Ke et al., 2016) “clear” brain tissue. By removing lipids in whole-brain tissue while leaving proteins and nucleic acids intact, tissue effectively becomes transparent, allowing one to image three-dimensional brain architecture or brain-wide neural projections with minimal light scattering. When used in combination with light-sheet microscopy or similar imaging platforms, intricate circuit networks may be visualized in three dimensions, across the entire brain (Newmaster et al., 2021).

Classical electrophysiology centers around the “patch-clamp” technique, which creates a tight seal on a cell’s membrane using a glass electrode or micropipette to record membrane voltage or current (Neher and Sakmann, 1976). Patch clamping has remained the gold standard of electrophysiology for decades, allowing scientists to measure the properties of ion flow across the membrane and investigate the types of ion channels that regulate the electrical activity of cells. However, recent advancements in patch-clamp configuration allow increased flexibility and greater throughput recordings including cell-attached, inside-out, outside-out, perforated patch, and notably, the whole-cell configuration (Hill and Stephens, 2021).

In an increasingly multimodal research world, electrophysiology has adapted to interface with other sophisticated techniques, allowing users to measure multiple dimensions of neural activity. For example, patch-clamp electrophysiology has been combined with RNA sequencing in a technique known as Patch-Seq. Patch-Seq permits the collection of transcriptomic data after performing whole-cell electrophysiology recordings, as patched cells are drawn into a glass capillary and used for single-cell RNA sequencing (Bardy et al., 2016; Cadwell et al., 2016; Chen et al., 2016; Földy et al., 2016; Fuzik et al., 2016; Lipovsek et al., 2021). Using Patch-Seq, both neuronal activity and gene expression patterns are revealed with single-cell resolution. Indeed, this technique has been used to describe the diversity of neuronal subtypes within the mouse motor cortex, as well as to provide insights into the diversity of neuronal populations in the dorsal striatum (Muñoz-Manchado et al., 2018; Gouwens et al., 2020; Scala et al., 2021).

While recording the electrical properties of single cells has been critical for understanding the basic properties of individual neurons, interrogating the activity of interconnected circuits requires recording from multiple cells simultaneously. Toward this, extracellular recordings using wire or silicon-based electrodes measure action potentials in multiple parallel neurons. By using multiple geometrically arranged electrodes, or multielectrode arrays (MEAs), it is now possible to record many neurons simultaneously from one probe and analyze their summed electrical activity, or local field potentials (LFPs; Buzsáki, 2004; Einevoll et al., 2013). The past decade has seen the rise of increasingly powerful MEAs such as Neuropixels, a silicon-based probe with up to thousands of independent channels for recording hundreds of neurons simultaneously (Jun et al., 2017). With the advent of these new high-throughput recording methods, it is now possible to interrogate the electrical and synaptic properties of circuits with unprecedented spatiotemporal resolution (Csicsvari et al., 2003; Shobe et al., 2015; Jun et al., 2017).

Electrophysiology is optimized to measure the activity of neurons with high temporal precision but is limited in its ability to spatially localize the cells being recorded, or to identify their molecular cell type. With the advent of real-time neural activity imaging, activity-dependent fluorescent reporters have been developed to measure circuit dynamics with high spatial resolution in living brain tissue. Genetically Encoded Calcium Indicators (GECIs) are the most widely used activity-dependent fluorescent reporters. GECIs are proteins that bind cytosolic calcium and exhibit increased fluorescence with calcium influx, a proxy for neuronal firing, thereby translating neural activity into spatially resolved optical signals (Bootman and Berridge, 1995; Miyawaki et al., 1997; Oh et al., 2019). Since fluorescent activity reporters are genetically encoded, their expression can be targeted to specific populations through conditional viral or transgenic methods (Pérez Koldenkova and Nagai, 2013). While the temporal dynamics are less precise than electrophysiological techniques, advancements in imaging resolution and GECI sensitivity enable monitoring of diverse firing dynamics at high spatiotemporal resolution. Importantly, such imaging approaches facilitate the investigation of firing and anatomical properties of neuronal circuits from diverse brain structures, from the cortex to deep brain nuclei (Hoogland et al., 2009).

GCaMP reporters are currently one of the most widely used families of GECIs (Table 2). Structurally, GCaMP is a fusion protein between GFP, calmodulin (a calcium-binding protein), and the light chain myosin protein, which was rationally designed to facilitate increased GFP fluorescence with the influx and binding of free calcium (Shimomura et al., 1962; Porumb et al., 1994; Tallini et al., 2006; Akerboom et al., 2009). Additional improvements to GECI technology have been made via large-scale random mutagenesis (GECOs; Table 3), or semi-rational mutation strategies (GCaMP5 family) to identify indicators with stronger fluorescence intensity, varied activity kinetics (GCaMP6s-f), and greater dynamic ranges (Zhao et al., 2011; Akerboom et al., 2012; Helassa et al., 2015). Notably, GCaMP6 and GCaMP8 are widely used for their fast kinetics, stability, and high signal intensity. GCaMP8 has a higher signal intensity than GCaMP6, making it well suited for experiments to monitor population activity signals via imaging of the soma, while GCaMP6 is preferred for visualizing activity at dendritic spines (Ohkura et al., 2012a).

In parallel with the continued development of diverse fluorescent indicator proteins, imaging technologies have also advanced. Classical GECI recordings were often made in superficial brain areas due to the relative ease of access for optical observation. The development of miniaturized fiber optic-based photometry has facilitated studying deep-brain structures in awake, behaving animals in real-time (Helmchen et al., 2001; Grewe et al., 2010; Lütcke et al., 2010). However, fiber photometry measures bulk calcium dynamics, lacking cellular resolution. Now, experiments using implantable gradient refractive index (GRIN) lenses combined with mini-microscopes or 2-photon interfaces permit real-time dynamic recordings from individual cells deep within the brain (Helmchen et al., 2001; Barretto et al., 2009; Ghosh et al., 2011; Zong et al., 2017; Zhang et al., 2019). GECIs represent a powerful tool for monitoring the activity of neural circuits in a cell type-specific manner. This highly optimized and titratable system has revolutionized the field of systems neuroscience, allowing for bulk activity recordings and spike extractions from big data sets such as the Machine Intelligence from Cortical Networks (MICrONS) data set, which contains large scale tracing and functional connectivity data obtained through electron microscopy (Zhou et al., 2020). The combination of highly optimized calcium indicators, genetic targeting technology, sophisticated imaging techniques, and big data analysis through machine learning approaches have opened the door for an in-depth, mechanistic understanding of circuit dynamics with unprecedented precision.

While GECIs remains an invaluable tool to study cell-type-specific circuit dynamics, due to their dependence on calcium activity they represent an indirect measure of cell activity. The development of genetically encoded voltage indicators (GEVIs) directly measures neuronal activity, as indicated by neuronal membrane potential, in a more temporally precise manner than calcium indicators (Table 4). Although the advent of GEVIs predates GECIs until recently, circuit monitoring via GEVIs has suffered from limitations in imaging technologies and available voltage indicators. Advancements in image processing and GEVI development have paved the way for the visualization of voltage dynamics in genetically defined circuits (Chamberland et al., 2017). The first GEVIs were designed as fusion proteins between GFP and voltage-sensing domains (VSDs) like those found in voltage-gated ion channels (Siegel and Isacoff, 1997; Sakai et al., 2001; Dimitrov et al., 2007). Within the class of VSD-based GEVIs, two main families of indicators have been categorized by the number of fluorescent proteins fused to the VSD: FRET GEVIs (also known as ratiometric GEVIs) and single fluorescent protein (monochromatic) GEVIs.

Ratiometric GEVIs use voltage-dependent Förster resonance energy transfer (FRET) to modify the conformation of the VSD, in turn changing the conformation of two fluorescent proteins such that the readout is an anticorrelated switch from fluorescence in one color to another. FRET GEVIs are particularly useful for circuit monitoring applications in vivo, where the anticorrelated color-switching can be parsed out from hemodynamic correlated changes in fluorescents within the same color (Akemann et al., 2012, 2010). Monochromatic GEVIs, on the other hand, contain a single fluorescent protein that emits a single wavelength of light, best suited for cell culture or slice preparation experiments. The advantage of this kind of GEVI is that the optical imaging setup required for measuring its activity is less complex than that of the ratiometric GEVI. However, given that there is only a single wavelength emitted, performing in vivo imaging requires more complex corrective methods to account for hemodynamic changes and/or a multiplexed approach (Borden et al., 2017).

One of the major limitations of VSD-based GEVIs is the weak signal-to-noise ratio. Thus, a second major class of GEVIs was developed based on naturally occurring microbial opsins. These GEVIs take advantage of the native fluorescence of these opsins, but also amplify this signal by binding them to brighter fluorescent proteins, thus reducing the effect of photobleaching that is seen in other GEVIs (Isobe et al., 1983; Adam et al., 2019; Piatkevich et al., 2019). For example, FRET-opsins have been developed that couple rhodopsin to another fluorescent protein, such that the fast dynamics of the opsin can be coupled to the brightness of the additional fluorescent protein for better signal-to-noise ratios. Fluorescent proteins such as yellow mCitrine, the bright green mNeonGreen, or the orange mRuby3 and mScarlet, have been used to generate MacQ-mCitrine, Ace-mNeonGreen, VARNAM, and Ace-mScarlet, respectively (Gong et al., 2014, 2015; Gong, 2015; Kannan et al., 2018; Beck and Gong, 2019; Mollinedo-Gajate et al., 2021). These opsin-based GEVIs have been leveraged in the study of circuit dynamics in cell culture, slice preparations, and in vivo experiments within cerebellar Purkinje cells (Gong, 2015).

Förster resonance energy transfer-opsins represent an advancement in the design of GEVIs to overcome signal-to-noise limitations. However, due to a greater molecular size, they are susceptible to forming intracellular aggregates that impede their localization to the cell plasma membrane (Hochbaum et al., 2014; Kannan et al., 2019). To circumvent this limitation, a new class of chemigenetic hybrid GEVIs has been developed that consists of two components – the first which targets the indicator to a cell type of interest, while the fluorescent protein is separately delivered to the target tissue, often intravenously (Sundukova et al., 2019). Within the class of chemigenetic GEVIs, three main families of indicators exist: Opsin-Dye FRET chemigenetic indicators, fluorescent protein-dye FRET hybrid indicators, and indicators that use photo-induced electron transfer. The two most promising opsin-dye FRET chemigenetic indicators are Voltron and Positron – both of which use a microbial opsin fused to a self-labeling protein tag that allows for the binding of a fluorescent protein (Grimm et al., 2015; Abdelfattah et al., 2019, Abdelfattah et al., 2020). Fluorescent protein-dye FRET indicators use an exogenous chromophore coupled to a fluorescent protein, where the chromophore is used to quench fluorescence based on cell membrane voltage (Chanda et al., 2005; Wang et al., 2010). Lastly, photo-induced electron transfer GEVIs work using a voltage-sensitive fluorescent protein-tagged to the SpyTag protein which covalently binds to the SpyCatcher membrane protein in the cell (Grenier et al., 2019). Voltage sensing is achieved by photo-induced electron transfer through a synthetic wire in the membrane that reversibly quenches the activity of the fluorescent protein (Mollinedo-Gajate et al., 2021). The ability to target GEVI expression with cell-type specificity and improved membrane localization allows for greater reliability and precision in the quest to measure circuit dynamics.

Although fluorescent protein-based GEVIs represent a powerful way to measure neural activity, these indicators are limited in their photostability and potential for toxicity at high expression levels. A class of bioluminescent GEVIs has recently been developed that overcomes these limitations (Inagaki et al., 2019, 2017). Bioluminescent GEVIs provide cells with a substrate that can be converted into a fluorescent reporter based on changes in membrane potential. Traditional bioluminescent GEVIs combine a VSD, luciferase, and a fluorescent protein in a similar configuration to FRET-based GEVIs, whereby voltage changes produce a FRET-based anticorrelated signal between the luciferase and fluorescent reporter. While the anticorrelated design allows for better signal-to-noise ratios, it is limited by the requirement of the exogenous substrate, luciferin. To overcome this limitation, recent advancements have been made such that the bioluminescent GEVI self-generates an energy-rich substrate for light-emitting activity (Srinivasan et al., 2019; Mollinedo-Gajate et al., 2021).

While voltage indicator technology has advanced greatly in recent years, continued improvements to their photostability, signal intensity, and power of imaging platforms are required in order to more readily compete with calcium indicators (Yang and St-Pierre, 2016). However, new and improved GEVIs are continually being engineered to address these concerns, providing direct monitoring of electrical activity in living cells. Despite the high signal intensity provided by GECIs, this tool can only provide binary information on whether a neuron is active, lacking the temporal dynamics required to accurately determine neuronal firing patterns. Additionally, GECIs rely on indirect measures of neuronal activity via calcium. In contrast, GEVIs provide the necessary kinetics to accurately portray complex firing patterns, and even signal propagation throughout an individual neuron, but lack the appropriate signal intensity to generate high-quality data. A paradigm shift in the use of GECIs versus GEVIs may come in the future, but this requires continued advancements in both imaging techniques and protein engineering to improve GEVIs.

Electrophysiology, GECIs, and GEVIs reveal detailed neuronal activity, and with the power of viral and transgenic targeting technologies, interrogate circuit dynamics with cell-type specificity. However, these techniques are limited in their ability to measure cell-cell communication dynamics at the synapse. To study the real-time communications between active neurons, novel protein-based sensors that bind to and report the release of neurotransmitters have emerged, providing an unprecedented view of the dynamics of communication between neurons (Table 5).

Neurotransmitter sensors come in two main classes – periplasmic binding protein (PBP)-based sensors and G-protein coupled receptor (GPCR)-based sensors. PBP-based sensors use microbial PBPs fused to fluorescent proteins to report the presence of neurotransmitters in the synaptic cleft. One of the first PBP-based sensors developed was iGluSnFR, which has been used to detect glutamate in vitro and in vivo in mice, worms, and zebrafish (Marvin et al., 2013). Since then, advanced versions of iGluSnFR with different color and kinetic properties have been developed. Other sensors in the PBP class of neurotransmitter sensors include iGABASnFR, iAChSnFR, and iSeroSnFR, as well as small molecule sensors like iATPSnFR, and iNicSnFR (Lobas et al., 2019; Marvin et al., 2019; Shivange et al., 2019; Borden et al., 2020; Unger et al., 2020). One caveat to using PBP-based sensors is that since they are microbially based, they may interfere with endogenous cellular processes and machinery.

GPCR-based sensors overcome this limitation, as they mimic the properties of endogenous receptors and are quite stable. GPCR sensors bind to neurotransmitters present in the synapse and cause a conformational change that results in a fluorescent signal. Some of the neurotransmitter sensors in this class include the dLight1 and GRAB_DA families of dopaminergic sensors as well as the GRAB_NE, GRAB_5–HT, and GRAB_ACh sensors for norepinephrine, serotonin, and acetylcholine, respectively (Jing et al., 2018; Patriarchi et al., 2018; Sun et al., 2018; Feng et al., 2019; Wan et al., 2021). Furthermore, the advent of neuropeptide GPCR-based sensors such as OxLight1, MTRIA_OT, and GRAB_OT1.₀ for orexin and oxytocin, respectively, permits the visualization of volume transmission from dendrites as well as axon terminals (Ino et al., 2021; Duffet et al., 2022; Qian et al., 2022). However, due to competitive interference from endogenous receptors, the readout of neurotransmitter activity using GPCR sensors will necessarily be based on the relative change in fluorescence rather than the absolute number of neurotransmitter molecules present at the synapse. Additionally, GPCR-based sensors cannot be used for pharmacological studies as they will interfere with the drug-receptor interactions. Thus, depending on the application, either PBP or GPCR sensors may be preferred.

The development of neurotransmitter sensors heralds a new era of neuroscientific discovery, where cell-cell neuronal communication is not only capable of being monitored at the cellular level, but at the synapse itself. In fact, biosensors are even being developed to monitor the activity patterns of second messengers downstream of synaptic transmission (Tewson et al., 2016; Massengill et al., 2021). For example, recent studies using a cAMP biosensor revealed that dopamine-triggered cAMP release in the basal amygdala facilitates learning (Lutas et al., 2022). With the advent of these sensors, neurotransmitter and signaling cascade dynamics may be studied to understand how these molecules play a role in the development and function of circuit architecture within the brain. Additionally, when used in tandem with GECIs or GEVIs, one may monitor both neural activity and neural signaling dynamics at once.

Although electrolytic or chemical lesioning methods remain powerful techniques to interrogate structure and function relationships in the brain (Foster et al., 2003; Lavond and Steinmetz, 2003), these approaches are non-specific, ablating all the cells in a given area. Alternatively, genetically-targeted ablation may be used for cell type-specific lesioning. One such system was adopted from the bacteria Corynebacterium diphtheriae. Its earliest rendition involved the use of transgenic lines that placed a diphtheria toxin (DT) gene under a promoter of interest (Breitman et al., 1987; Saito et al., 2001). Cells that used this promoter would express DT, driving their own death. The principal limitation of this technology was that the timing of cellular ablation could not be controlled. Since ablating cells early in development may result in compensatory plasticity mechanisms, it became desirable to temporally control ablation. In 2001, these limitations were abrogated by controlling cellular expression of the DT receptor (DTR), which is not endogenously expressed in the mouse. In this way, the experimenter could selectively express DTR in specific cell types, and provide exogenous DT at any point in development to selectively ablate cells expressing DTR (Saito et al., 2001). This technology led to the discovery that AgRP/NPY neurons are required for proper feeding behaviors in adult mice, but are not required in neonate mice, perhaps due to compensatory mechanisms (Luquet et al., 2005). One important consideration with cellular ablation is that it is permanent. Other experiments may require reversible manipulation, such as the techniques discussed below.

Neuronal activity can be manipulated by specific chemicals. Commonly, this is due to agonistic or antagonistic effects on neural ion channels (Narahashi et al., 1964; Collingridge et al., 1983; Sweeney et al., 1995; Verderio et al., 1999; Corcoran and Quirk, 2007). By infusing pharmacological substances in particular brain regions, it is readily possible to investigate the effect of activating or inhibiting targeted areas. One common means of drug delivery is through a stereotaxically implanted hollow cannula (Sike et al., 2017). This can be combined with other technologies such as microdialysis or in vivo imaging to dissect the effects of drug treatments on neurotransmitter release, neural activity, and gene expression. Commonly used pharmacologically active chemicals used to manipulate neuronal activity are summarized in Table 6. Importantly, pharmacological manipulations are specific to cells that express a particular receptor but are not selective in the cell types they may target. For reversible and cell type-specific manipulation experiments, opto- or chemogenetics have proven invaluable.

In the late 1970s, Francis Crick remarked that the real challenge of neuroscience would be to precisely control the activity of a desired set of neurons. Nearly three decades later, George Nagel, Peter Hegemann, and colleagues discovered Channelrhodopsin-2 (ChR2), a light-gated cation channel that demonstrated inward cationic current with millisecond time-scale resolution (Nagel et al., 2003). Shortly after this discovery, Boyden, Deisseroth, and colleagues showed that ChR2 could be leveraged to manipulate the activity of neurons using light in a technique known as optogenetics (Boyden et al., 2005). Currently, it remains one of the most temporally and spatially specific methodologies to manipulate neural activity, both ex vivo and in vivo, in which mice are implanted with fiber optics for light delivery (Yizhar et al., 2011). Optogenetics has led to many systems-level discoveries, such as the solidification of arcuate hypothalamic feeding circuitry. In this study, conditional expression of ChR2 and subsequent activation of AgRP cells in the arcuate nucleus of the hypothalamus led to robust acute food consumption, while optogenetic activation of POMC neurons led to decreased food consumption (Aponte et al., 2011).

At the core of optogenetics is the use of opsins. Upon photon absorption, opsins undergo a conformational change that allows for ion transport across the plasma membrane, either depolarizing or hyperpolarizing neurons (Foster and Bellingham, 2002). As described in Table 7, a plethora of opsins has been engineered that allow great flexibility in their experimental use. These opsins have attributes that differ in terms of effects on membrane potential, kinetics, and absorption spectra (Yizhar et al., 2011; Deisseroth, 2015). Furthermore, more than one type of opsin can be expressed by cells. For example, an excitatory opsin activated by blue light (e.g., ChR2) can be co-expressed with an inhibitory opsin activated by yellow light (e.g., halorhodopsin; Han and Boyden, 2007). This allows for either gain- or loss-of-function experimentation within the same animal. Another consideration in the design of an optogenetic experiment is the location of the light source, either at the soma or axon terminals of the opsin-expressing neural population. If placed at the soma, those cells and all of their projection targets will be affected (Figure 6A). If placed at the axon terminals, only that projection site will be affected (Yizhar et al., 2011; Figure 6B). The latter approach allows for greater dissection of the exact circuitry involved in a particular phenotype (Steinberg et al., 2014). Interestingly, stimulating somata versus terminal fields can result in disparate phenotypes. For example, one study used opsins to stimulate or inhibit basolateral amygdala (BLA) projections to the central amygdala, which altered anxiety-related behaviors, while stimulating BLA cell bodies had no effect. This was perhaps due to downstream projections operating in opposition to each other (Tye et al., 2011). Thus, terminal field stimulation is very useful for dissecting circuit function, as cells may have diverse collateralizations.

However, it must be considered that the nature of optogenetics frequently necessitates a level of invasiveness that may be unacceptable. For example, very deep brain regions with functions essential to survival may not be compatible with external light delivery (Chen et al., 2021). Further, the tethering required by most light-delivery systems may compromise the validity of certain behavioral experiments (Won et al., 2020). Finally, light delivery may lead to tissue toxicity, for example, due to heating or ischemia (Chen et al., 2017a). Therefore, at times less invasive means may be needed to manipulate neural circuits transiently, which is possible through chemogenetics.

Chemogenetics refers to a set of techniques whereby genetically engineered receptors expressed on cell surfaces bind to exogenously delivered ligands. Importantly, these ligands are not endogenously produced. Early examples of chemogenetic technologies came in 1998 with Receptors Activated Solely by a Synthetic Ligand (RASSLs; Conklin et al., 2008). Currently, the most commonly used receptors for manipulations are coupled to G proteins known as DREADDs: Designer Receptors Exclusively Activated by Designer Drugs. DREADDs are activated by the ligand Clozapine-N-Oxide (CNO; Whissell et al., 2016; Table 8). Depending on the receptor’s associated downstream signaling cascade, the ligand may be excitatory (Gq, Gs) and increase cell firing, or inhibitory (Gi) and decrease cell firing. While CNO-activated DREADDs are one of the most widely used tools for chemogenetic neural manipulation, this technology has some major limitations. It has been shown that CNO does not cross the blood-brain barrier, but rather its parent metabolite, clozapine (a known antipsychotic drug), exerts effects in the brain (Gomez et al., 2017). Additionally, clozapine has been shown to bind to endogenous receptors throughout the body at high doses and with known effects on animal behavior (Desloovere et al., 2019). However, these findings remain controversial, as other studies report opposing findings in mice (Guettier et al., 2009; Jendryka et al., 2019). To overcome such limitations, studies have shown that at low doses of clozapine, these off-target effects are reduced (Manvich et al., 2018; Matsui et al., 2018). Additionally, novel ligands like olanzapine, JHU37160/152, and deschloroclozapine are being used for DREADD-based manipulations as they have a minimal conversion to other compounds and less off-target effects (Bonaventura et al., 2019; Weston et al., 2019; Nagai et al., 2020).

Some experiments require one to manipulate several neural populations at once or to bidirectionally control the activity of one neural population. Through a recently developed technology labeled DREADD KORD(R), chemogenetics is afforded the possibility of multiplexing. DREADD KORD(R)s are sensitive to a distinct exogenously delivered ligand known as Salvinorin B and completely insensitive to CNO (Vardy et al., 2015). The non-overlapping pharmacologic activation of DREADD KORD(R) and DREADDs provides superior bidirectional manipulation relative to optogenetics, since opsins may absorb photons at overlapping ranges, potentially confounding results. However, the temporal resolution of DREADD action is less precise than optogenetics and operates on the timescale of minutes to hours (Whissell et al., 2016). This is because it takes time for the synthetic ligand (e.g., CNO) to distribute to areas that express the DREADD, and also because GCPRs have slower kinetics than ion channels. This can be beneficial when requiring more chronic forms of manipulation, but less useful for applications where faster or more temporally-precise manipulations are desired.

While chemogenetics is limited by less precise temporal and spatial resolution, newer technologies are emerging. For example, magnetically-sensitive lipid nanoparticles carrying CNO ensure the steady release of CNO to targeted circuits (Rao et al., 2019). Additionally, it is possible to employ both opsins and DREADDs for bidirectional control of neural activity (Han et al., 2017; Siemian et al., 2021). Alternatively, the development of Pharmacologically Selective Activator Modules (PSAMs) which are activated by Pharmacologically Selective Effector Molecules (PSEMs) increases temporal resolution (Shapiro et al., 2012). Unlike DREADDs which alter neuronal activity through the modulation of GPCR-biased signaling cascades, PSAMs affect neuronal activity through modulation of ligand-gated ion channels. Most recent iterations of the technology use the agonist varenicline, an FDA-approved drug. Although varenicline binds to nicotinic receptors (ligand-gated ion channels), scientists have generated a PSAM that binds to varenicline at a low enough concentration that it will not bind to endogenously produced receptors (Magnus et al., 2019).

Both optogenetic and chemogenetic manipulations allow for controlled activation or inhibition of circuitry in a genetically-defined manner. While useful, these technologies are limited by the need to consistently deliver an actuator (e.g., light through fiber optics for optogenetics or ligands for chemogenetics). Thus, using opto- or chemogenetics for studies in which a circuit would need to be activated/inhibited for more than several hours each day (for example, body weight or metabolism studies) is not optimal. For these studies, it is desirable to chronically activate neural activity, as with the overexpression of exogenous ion channels (Johns et al., 1999; Ren et al., 2001; Lin et al., 2010; Patel et al., 2019; Zhu et al., 2020).

Neuronal excitability corresponds to a neuron’s likelihood of triggering an action potential. This likelihood is dependent on ion channel expression and functionality. Given the same stimuli, neurons with greater excitability will fire more frequently than neurons with lesser excitability (Hille, 2001). Scientists can bias excitability by manipulating the types and amounts of ion channels expressed on the neuronal plasma membrane so as to shift resting membrane potential (Johns et al., 1999; Lin et al., 2010; Zhu et al., 2020). Since these methods do not require drug administration or photostimulation, they are less invasive measures for studying neural function. However, these methods lack precise on/off control compared to optogenetics or chemogenetics.

One ion channel used to manipulate neural activity chronically is Kir2.1, an inward rectifying potassium channel that makes action potential generation more difficult (Johns et al., 1999). In 2017, the targeted expression of Kir2.1 helped to elucidate a circuit node responsible for integrating social input and modulating aggressive behavioral output (Watanabe et al., 2017). In 2020, the technology was used to understand the arcuate neuronal contribution to feeding behaviors (Zhu et al., 2020). To increase the temporal flexibility of this technology, methods have been devised to temporally control the expression of the channels using ligands or light, or both. For example, SPARK (Synthetic Photoisomerizable Azobene Regulation K + channels) has a genetically engineered K + channel that is sensitive to an exogenously delivered, light-sensitive ligand (Banghart et al., 2004). In the absence of the ligand, the channel remains open and hyperpolarizes the cell. In the presence of the ligand, and the absence of UV light, the K + channel closes. Finally, in the presence of both the ligand and UV light, the channel opens. This affords the flexibility of choosing when to hyperpolarize the cell or allow resting neuron conditions.

Other exogenous ion channels may be used to increase cell excitability and provide gain-of-function approaches. One such channel is the voltage-gated Sodium Bacterial Channel (NaChBac; Ren et al., 2001; Lin et al., 2010). Most importantly, NaChBac opens at more negative voltages relative to eukaryotic sodium channels. Consequently, subthreshold stimuli open these channels, thereby conferring both a greater depolarizing stimulus and an increased probability of action potential generation. Secondly, upon opening, they enable long-lasting depolarizing stimuli that can drive further action potentials. NaChBacs have been used to understand the roles of both corticotropins-releasing hormone (Zhu et al., 2020) and glutamate (Herman et al., 2016; Patel et al., 2019) in modulating feeding behaviors.

Several experimental approaches combine diverse manipulation methods, such as optogenetics, with viral tracing or circuit monitoring techniques. An elegant intersection between circuit mapping, monitoring, and manipulation is the technique known as Channelrhodopsin Assisted Circuit Mapping (CRACM; Petreanu et al., 2007). This technology is a gold standard in the generation of reliable, functional circuit diagrams. Here, opsins and reporters are expressed on a pre-synaptic cell of interest. Then, by taking whole-cell recordings of a putative post-synaptic cell while stimulating the pre-synapse, one may identify synaptic currents on the post-synaptic cell. The temporal coincidence between light stimulation and post-synaptic currents, combined with the application of pharmacological agents that abolish this response, provides strong evidence for a functional synaptic connection between pre- and post-synaptic cells (Petreanu et al., 2009). While CRACM is useful, depending on the circumstance, it can have drawbacks. For example, depending on the expression profile of the opsin, it can be difficult to determine which side of the pre-or post-synapse is being excited by light when an electrophysiological event is being recorded in the post-synaptic cell. This can be mitigated by conditionally expressing the opsin in a neuronal subtype distinct from the post synaptic neuron. Alternatively, one can use an opsin whose expression is spatially restricted to the soma/proximal dendrites (Baker et al., 2016). One other limitation is the challenge of performing CRACM with peptidergic cell types. This is because neuropeptides often signal via volume transmission, which occurs over large distances and with potentially slower effects compared to fast neurotransmitters (like glutamate and GABA; Fuxe et al., 2013; Nederpelt et al., 2016).

Without sophisticated techniques like Patch-Seq, electrophysiology lacks the ability to identify the exact cells being recorded. Combining electrophysiology with optogenetics allows in vivo targeting of genetically labeled cells for electrophysiological recordings through a “sonar”-like method (Lima et al., 2009; Hangya et al., 2015). The development of small implantable optrodes, or electrode(s) juxtaposed to a fiber optic for light delivery, locates conditional ChR2-expressing neurons via optogenetic stimulation in vivo. Once ChR2-expressing cells of interest are located (as indicated by reliable light-induced single units), extracellular recordings assess cellular electrical activity (Figure 7). The “opto-sonar” technique provides a dependable method to genetically identify and target circuits within a live, freely behaving animal. Indeed, with the integration of optogenetics and implantable MEAs, it is now commonplace to not only monitor the activity of genetically defined circuits but manipulate and assess their activity and temporal dynamics in vivo (Kim C. K. et al., 2017).

Another emergent multiplexed technology is known as holographic imaging. Holograms are light fields that form any simple and complex shape through wavefront interference. By combining imaging technologies with principles of holography, it has become possible to map, monitor, and manipulate populations of neurons in multiple regions of interest (in different focal planes) with an unprecedented combination of spatial and temporal resolution (Quirin et al., 2014; Yang et al., 2015). In a 2018 paper, Yang and colleagues simultaneously photostimulated 80 neurons within the mouse visual cortex while also monitoring the neighboring neuronal activity (Yang et al., 2018). This technology has since been leveraged to understand how neural ensembles within cortical circuitry can drive recall in a complex behavioral task (Carrillo-Reid et al., 2019). While certain technical limitations to the technology exist (e.g., limited field of view, lag in modulating projected holographic patterns, etc.), more profound challenges present themselves in how the output data is analyzed and interpreted (Carrillo-Reid et al., 2017; Yang and Yuste, 2018). To better use this emergent technology, creating methods to analyze these huge datasets must be a major focus moving forward.