Viral vectors for neuronal cell type-specific visualization and manipulations

Abstract

Characterizing neuronal cell types demands efficient strategies for specific labeling and manipulation of individual subtypes to dissect their connectivity and functions. Recombinant viral technology offers a powerful toolbox for targeted transgene expression in specific neuronal populations. In order to achieve cell type-specific targeting, exciting progress has been made to: alter viral tropisms, design rational delivery strategies, and drive selective expression patterns with engineered DNA sequences in viral genomes. For the latter case, emerging single-cell genomic analyses provide rich databases. In this review, we will summarize current status, and point out challenges, of using viral vectors for neuronal cell type-specific visualization and manipulations. With concerted efforts, progress will continue to be made towards developing viral vectors for the vast array of neuronal subtypes in the mammalian nervous system.

Introduction

The ability to examine and manipulate the function of specific brain nuclei and neuronal types is critical for understanding how brain circuitry works. This is evident from the widespread applications of transgenic mouse driver lines, which enable specific transgene expression in a few selected cell types or brain regions [1]. In light of the complex nature of the nervous system, more versatile tools of this kind would be extremely desirable. This is an even more pressing issue with current advances in single cell analysis methods, such as single-cell RNA sequencing (scRNA-seq), that are characterizing ever-growing numbers of neuronal cell types in an increasing number of mammalian species (including mice and humans). In principle, if a neuronal type is uniquely defined by the combinatorial expression of a set of genes, one could target such neurons specifically using sophisticated genetics. For example, multiple transgenic animals can be generated in which each marker gene is used to drive either a different DNA recombinase (e.g. Cre, Flp, Dre), or a reporter line, so that a desired transgene (e.g. XFP, opsins, activity indicators) is only expressed when all recombinases and the reporter are expressed in the same neuronal cells. While the advancement of CRISPR technology has made it possible to generate gene-targeted animals efficiently, crossing these different lines to achieve cell-type specific expression is still a lengthy and expensive process.

On the other hand, recombinant viral vectors with cross-species targeting and minimal cytotoxic effects offer unprecedented delivery systems to selectively express different desired transgenes in specific neuronal populations. In addition to the natural tropisms of individual viral types [2,3], their transduction specificity can, in principle, be refined by altering their capsid or envelope proteins (to dictate which cells to bind and enter) and/or the promoter/enhancer in the viral genome (to determine whether or when to express the target gene). Combining these techniques to alter viral tropism with various delivery methods would allow recombinant viral vectors to serve as a collection of region-selective, neuron type-specific, and activity-dependent tools that can be readily used in multiple species. Here we review four main strategies for, and recent progresses toward, designing and optimizing viral vectors for neuronal type-specific expression, namely: (1) axonal projection-dependent; (2) viral serotype/tropism-based; (3) gene regulatory element (promoter/enhancer/microRNA)-driven; and (4) delivery-controlled strategies for targeting of distinct neuronal cell types (Figure 1).

Figure 1: Strategies for neuronal cell type-specific targeting with viral vectors.

A) Axonal projection-dependent retrograde labeling: viruses with axonal tropism (rabies, HiRet lentivirus, CAV2, and retroAAV) mediate retrograde targeting for neurons of interest in both the brain and spinal cord. B) Viral serotype/capsid determines tissue, and in some cases cell-type specific, tropism. Here, AAV color indicates various capsids that correlate with the specific tissue or cell-type that the AAV is able to infect. Specifically: PHP.eB (red) is preferentially expressed in the CNS, PHP.S (green) in the PNS, AAV2 (purple) in RGCs and amacrine cells, ShH10 (yellow) in muller glia, and AAV2.7m8 (blue) in photoreceptors. Grey indicates cells that are not targeted. C) Sequencing and -omics approaches can be harnessed to identify novel gene regulatory elements (e.g. promoters, enhancers, microRNA) to label specific neuronal and non-neuronal subtypes. D) Finally, targeting of distinct neuronal cell types can be aided by unique delivery methods (stereotaxic, intrathecal, tail-vein, intravitreal, and sub-retinal). [Portions of this figure used and/or altered artwork from https://smart.servier.com/; Purkinje neuron from Ramón y Cajal].

Viral vectors for axonal projection-dependent targeting of specific neurons

Neurons residing within one brain region expressing similar genetic markers can have distinct projection targets and can thereby be classified based on their efferent axonal innervation patterns [4–6]. Thus, one can use recombinant viruses capable of infecting neurons from their axon terminals to target a more specific cell type within a larger neuronal population through retrograde viral infection (Figure 1A). Many types of viral vectors have axonal tropism (Table 1), and four different types of retrograde viral vectors have been developed to this end.

Table 1.

Summary of features of recombinant viruses commonly used in neuroscience

Family	Genome	Neurotropism	Neuronal transport		Features	Limitations
			Direction	Cross-synapse
Adeno-associated virus (AAV)	Single strand DNA	Engineered capsid: e.g. PhP.eB (CNS), PhP.S (PNS) Enhancer/promotor specific	Anterograde; Engineered for retrograde: rAAV2-retro	No Exception: AAV1or 9-Cre, Anterograde	Do not integrate; High and stable expression; Selective neurotropism; Capable of recombinase-based modification	Low packaging capacity
Lentivirus	Single strand (+) RNA	No pseudotyping: immune cells Pseudotyping (VSV-G): neuron and glia	Anterograde; Engineered for retrograde: FuG-B, C & E	No	Viral genome integrates into host genome; Selective targeting by pseudotyping (EnV-TVA); Capable of recombinase-based modification	Expression level low
Rabies virus	Single strand (−) RNA	Neurotrophic	Retrograde	Retro-Exception: Sensory neurons, Anterograde	Engineered for non-crossing retrograde (RV-dG) or mono-synaptic (RV-dG with G) transmission; Selective targeting by pseudotyping (EnV-TVA)	Highly cyto-toxic; Incompatible with Cre/Flp
α-herpes viruses: Pseudorabies virus (PRV) and HSV1	Double strand DNA	Neurotrophic	Bi-directional	Strain dependent: PRV: Retrograde HSV1: Retro-/Anterograde	Poly-synaptic, bi-directional transmission; Engineered for mono-synaptic transmission (HSV: H129-dTK)	Difficult for genome engineering; Highly cyto-toxic

Recombinant rabies viruses:

It has long been known that rabies viruses infect mammalian neurons from axon terminals [7]. However, the use of rabies viruses for manipulating or labeling specific neuronal populations is limited because they are neurotoxic and uncontrollably propagate between hierarchically-connected neurons. Within the five genes of rabies viral genome, only G encodes the glycoprotein to mediate retrograde, trans-synaptic viral transmission. Thus, when the rabies envelope glycoprotein (RG) was replaced by fluorescent proteins (e.g. GFP), the recombinant rabies virus is used for non-transsynaptic retrograde neuronal labeling [8]. To further reduce its neurotoxicity, Wickersham’s group developed a double deletion (L and G proteins) recombinant rabies virus [9]. This new version enables long-term anatomical or physiological study of retrogradely-targeted projection neurons.

Retrograde lentiviruses:

Considering the aforementioned limitations of the rabies viruses, a promising alternative candidate is the notably less toxic lentivirus genus. Lentiviruses are single stranded RNA viruses that stably integrate into the host genome of both pre- and post-mitotic cells [2]. In light of previous findings that RG pseudotyped equine infectious anemia virus (EIAV) is capable of retrograde tracing [10], Kobayashi’s group decided to pseudotype lentiviruses using fusion proteins incorporating RG domains. They discovered that packaging lentiviruses with a fusion envelope composed of the extracellular domain of RG and the transmembrane and intracellular domain of the vesicular stomatitis virus (VSV) glycoprotein resulted in retrograde lentiviral vectors (called HiRet-LV, also called RG-LV) that were highly effective in infecting axon terminals of long-distance projection neurons in in the brain and spinal cord of both rodents and non-human primates (NHPs) [11–13]. Retrograde lentiviruses have since been successfully used by multiple labs for axonal projection-dependent labeling and manipulation of neuronal cell types [14–17].

CAV-2:

Canine adenovirus type 2 (CAV-2) displays potent retrograde axonal transport ability, through its selective interaction with the coxsackievirus and adenovirus receptor (CAR) at axon terminals [18]. CAV-2 can drive stable, long-term transgene expression with minimal cytotoxicity and has been effectively used as a retrograde vector to study many circuits [19,20]. However, CAV-2 has limited tropism since many neurons do not express the CAR receptor. This can be overcome by overexpression of CAR in candidate projection neurons [21].

rAAV2-retro:

Using in vivo screening for adeno-associated virus (AAV) capsids that would enable the packaged AAVs to be efficiently taken up by axonal terminals, recombinant AAV2-retro (rAAV2-retro) capsid was discovered to exhibit high efficiency in retrograde labeling of long-projecting neurons such as corticospinal neurons [22]. rAAV2-retro vector has broad, but not pan-neuronal, tropism for retrograde infection and can drive high-level gene expression like other AAVs. Thus, it is a very versatile retrograde vector, but which needs its tropism for desired neurons tested before use.

Applications of retrograde vectors:

The development of retrograde viral tools enables virus-based intersectional approaches to precisely manipulate target projection neurons that are embedded in a mixture of different neuronal populations. To achieve this, the retrograde viral vectors encoding recombinases (e.g. Cre or Flp) are often injected at specific sites where the axons of the target projection neurons innervate. These viral particles are selectively absorbed by axon terminals and then undergo retrograde gene transfer. Next, AAVs encoding Cre- or Flp-dependent genes (e.g. XFP, DTR, ChR2 or GCamp6) are injected into areas where the somas of the targeted projection neurons are located (Figure 2). Such a multi-step, axonal projection-dependent intersectional approach has been successfully applied in sophisticated neuronal circuit mapping and functional characterization in both rodent and NHP models [4,16,17,23–25]. More recently, combining this approach with brain clearing and 3D-imaging reconstruction, the Janelia MouseLight project has reconstructed the entire cell bodies and neurites of >1000 cortical projection neurons in the mouse brain, revealing previously unknown subtypes of cortical projection neurons [26] (MouseLight Neuron Browser: http://ml-neuronbrowser.janelia.org).

Figure 2: Viral-based intersectional targeting approach.

A) Schematic drawing of viral-based intersectional targeting approach. B) Images of a coronal brain section showing corticospinal neurons targeted by an engineered lenti-viral vector (HiRet-GFP, green), a traditional neuronal tracer (BDA, red) and their merge. The cartoon drawing demonstrated that almost all corticospinal neurons randomly labeled by BDA are co-labeled with HiRet-GFP and thus indicated the high efficiency of viral mediated retrograde labeling. In addition, compared to BDA, viral based retrograde labeling allows further functional observation and manipulations of targeted neurons.

AAV serotype/capsids-based cell type-specific neuronal targeting

Recombinant AAVs (rAAVs) are comprised of a single-stranded DNA genome with a capsid outer shell and are unable to integrate into host genomes, unlike naturally occurring ones that indeed integrate their genome into a preferred locus of the host genome [27]. Because of minimal cytotoxic effects and different tropisms associated with individual serotypes, AAV vectors have been increasingly used for mapping brain connectivity and for monitoring and manipulating neuronal functions in many vertebrate species, including mouse, rat, songbird, and NHPs. In addition, AAV vectors are also used clinically for gene therapy [28]. In most cases, the serotypes of rAAVs are determined by variations in their capsid proteins [29], with AAV1-9 being the most commonly used. Most of these serotypes are capable of infecting many neuronal cells, although with various efficacies in targeting different neuronal and glial cell types [30–32]. For example, intravitreally-injected AAV2 is highly effective to generally target retinal ganglion cells [33,34].

The existence of many serotypes raises the possibility of altering/screening capsid proteins to achieve cell type-specific infection, and has inspired many such efforts. In fact, the above mentioned rAAV-retro capsid was developed based on similar ideas. Briefly, directed evolution, which involves mutagenesis of the Cap gene followed by iterative rounds of phenotypic screening, has been widely used to identify a desired viral capsid. For example, most natural AAVs have limited ability to cross the blood brain barrier (BBB). By using a Cre-dependent selection system, Gradinaru’s group discovered novel capsids named PHP.B that can cross the BBB to infect the entire nervous system [35]. In a follow-up study, they further developed engineered capsids named PHP.eB and PHP.S that preferentially target CNS or PNS, respectively, through systemic administration [36]. Using similar approaches, an AAV2 variant named 7m8 and an AAV6 variant named ShH10 have been identified for enhanced selectivity for photoreceptors and Muller glia in the murine retina, respectively [37,38] (Figure 1B). Another strategy adopts a reverse principle to identify ancestral nodes or putative parents of currently available Cap proteins. Using this “back to future” approach, Anc80L65 was discovered and has showed strong potency for gene therapy of the inner ear [39,40]. We envision more variants of capsid proteins for targeting distinct neuronal cell types will be developed and optimized in the near future.

Identifying promoters/enhancers and other regulatory elements that can drive cell type-specific expression

Since neuronal cell types (or non-neuronal cell types in the nervous system) are determined by expression of marker gene(s), an approach to specifically label or manipulate distinct populations is to identify unique gene regulatory elements (promoters/enhancers) that fit into the packaging capacity of AAVs (4.7kb) to drive cell type-specific gene expression. Previous studies have identified a number of promoters useful for this purpose, such as human synapsin 1 (hSyn1) promoter for (pan-)neuronal specific expression [41], GFAP promoter for astrocytes [42], MBP for oligodendrocytes [43], tyrosine hydroxylase for catecholaminergic cells [36], L7/PCP2 for Purkinje cells [44], and FEV for serotonergic cells [45] (Figure 1C). In addition, since activity-dependent genetic methods are fruitful in tagging and manipulating functionally-activated ensembles of neurons [46–49], several studies have attempted to develop viral vectors using enhancers or promoters of immediate early genes, such as Fos or Arc [50] [51] [52] [53]. For example, an AAV vector in which the Fos promoter drives the expression of tTA was used to label different amygdala neurons activated by either aversive or appetitive stimuli [50], while AAVs containing the Arc promoter driving CreERT2 were used to label activated prefrontal cortex neurons [51]. The main caveat for AAVs using activity-dependent promoter/enhancers is a background basal expression which is independent of neuronal activity, and could therefore result in random labeling of non-functionally relevant neurons. To address this, Sorenson et al. developed AAVs with a Robust Activity Marking (RAM) system that consists of a synthetic neuronal activity-dependent promoter, which has very low expression in basal conditions, strong induction by neural activity and is under the temporal control of a modified Tet-Off system [54].

Suffice it to say, with the ever-increasing numbers of neuronal cell types being identified in the mammalian brain through the expansion of the single-cell RNAseq (scRNA-seq) technology, more enhancers/promoters that can drive specific expression using AAVs are urgently needed. Owing to recent advancement in scRNA-seq, Assay for Transposase-Accessible Chromatin using sequencing (ATAC-Seq) and other “-omic” methods, important progress has been made toward this end. Below we briefly review these methods. One method to identify enhancers is to search for evolutionarily-conserved DNA regulatory elements. By comparative genomic analysis, Visel and Pennacchino identified several hundred ultra-conserved non-coding elements in the human genome [55,56]. When tested in early embryos using transient transgenesis, many elements drove gene expression reproducibly in restricted regions of the embryonic brain [57]. Follow-up studies suggested many of these regulatory elements could act as enhancers for specific brain regions [58], and restrict expression to particular neuronal subtypes such as using the Dlx5/6 enhancer for GABAergic neurons [59]. However, many of the ultra-conserved enhancers are developmental enhancers, and fail to act in the adult brain (P. Williams and Z. He, unpublished data).

Second, mapping the open chromatin regions with advanced technologies has led to the identification of cell type-restricted gene regulatory elements (GREs) or enhancers. For example, Mich et al. used ATAC-seq to discover specific enhancer elements for diverse neural types within human cortex, which were then cloned into AAV vectors and tested in the mouse brain. Several enhancers were shown to enable labeling and perturbation of specific cortical neuronal subtypes [60] Furthermore, using the principles of massively parallel reporter assays (MPRA) with scRNA-seq, Hrvatin et al. designed a Paralleled Enhancer Single Cell Assay (PESCA) to identify and functionally assess the specificity of hundreds of GREs across the full complement of cell types present in the brain [61]. They identified GREs that confine AAV expression to somatostatin (SST)-expressing interneurons, which can be used to selectively modulate their neuronal activity [61]. We envision ATAC-seq and PESCA related approaches will generate a large database of neuronal subtype-specific enhancers.

Combining these different GREs may further optimize neuron type-specific AAV vectors. With this in mind, Jüttner et al. created a library of 230 AAVs, each with a different synthetic promoter designed based on identified “transcriptional code” for adult retinal cell types [62], and found that a number of these AAVs specifically target expression to neuronal and glial cell types in mouse and NHP retina in vivo and in the human retina in vitro [63]. Using logical combinations of different AAVs containing different synthetic promoters in an intersectional manner, they further achieved specific targeting of retinal cell types for manipulation and recording of neuronal function [63]. Thus, combinations of AAVs containing different GREs used in an intersectional manner will further improve the specificity of viral vectors for targeting desired neuronal types.

Third, promoters and enhancers may have sequences with the length beyond the packaging capacity of AAVs, but the recognition sites of microRNA (miRNA) are small (about 22 nucleotides) and can be easily packaged into viral vectors. Using the ability of miRNAs in restricting gene expression in certain neuronal types, AAVs with tandem copies of short miRNA-target sequences (miRNA-TSs) in their 3’ untranslated region of the genome have been characterized [64]. Recently, Keaveney et al. developed a miRNA-guided neuron tag (“mAGNET”) to restrict transgene expression to inhibitory (GABA⁺) neurons in the mouse neocortex (GABA mAGNET) [65]. In light of highly enriched expression of miRNAs in the mammalian brain, microRNA-based methods have huge potential. Mining the single-cell miRNA-seq database should facilitate the identification of new cell type-specific miRNAs.

Delivery-method controlled cell type-restricted targeting

In the nervous system, neuronal cells are defined by both their function and anatomical location. Thus, the method used to deliver viral vectors is another important consideration when attempting to restrict viral expression to precise neuronal cell types. The three most common virus delivery methods to target neurons are stereotaxic, intrathecal, and tail vein injection (Figure 1D), each with their own advantages and limitations, and which should be considered in parallel with the tropism of the particular viral vector of choice. For example, tail vein injection offers an accessible delivery method, but most viruses cannot cross the BBB. As discussed above, the newly developed AAV-PHP.B vectors are able to cross BBB in rodents, and are thus suitable for systemic injection to target CNS neurons [35]. Stereotaxic injection is most commonly used for targeting cells in different parts of the brain or spinal cord; however, it is a more invasive route of administration. In contrast, intrathecal injection is less invasive, but preferentially targets the dorsal root ganglia (DRGs) and spinal dorsal horn, with a decreasing expression gradient from lumbar to cervical spinal cord segments [66]. Such physical limitations to viral vector spread or entry can be addressed through, for example, adjusting body positioning post viral delivery [67]. More recently, bilateral intravenous injection of relatively large volumes of AAV9 into the transverse sinuses of early postnatal animals has been employed as a simpler way to achieve widespread and robust infection of the mouse brain [68].

In addition to this, different viruses display distinct infection profiles even when the same delivery method is chosen. For example, AAV2 has a more restricted, localized expression profile than other AAV serotypes when stereotaxically-injected into the CNS [30]. It is important to note that structural constraints can create boundaries for viral diffusion. This is nicely illustrated by distinct efficacies of different injection protocols in targeting different populations of retinal cells [69]. For example, intravitreal injection of AAVs is optimal for targeting inner retinal neurons/Müller glia, while subretinal AAV delivery is optimal for targeting the Retinal Pigment Epithelium (RPE) /photoreceptors [3]. Another example takes advantage of a transiently-compromised BBB after spinal cord injury. In this case, tail vein injection of AAV9 results in peri-lesion neuronal targeting [70]. Thus, the characteristic features and limitations of different delivery methods can be exploited to restrict infection of specific viral vectors to particular neuronal cell types and CNS regions of interest.

Other notes on using viral vectors to target specific cell types

While most efforts have aimed to achieve restricted and high-level gene expression with rAAV vectors, little attention has been paid to the toxicity associated with these AAV vectors, a crucial issue for their applications in human gene therapy [69,71]. Non-selective targeting of cell types represents a key issue. Based on the studies showing that AAV9 is able to transduce spinal motor neurons when administered intravenously at high doses [72,73], this approach was approved for expressing non-mutated survival of motor neuron (SMN) protein to treat infants with spinal muscular atrophy (SMA) [74]. However, recent studies have shown that high doses of AAV9 administration trigger sensory side-effects, with mononuclear cell inflammation in the dorsal root ganglion and cell death [75]. Based on such concerns, some of SMA clinical studies were halted (https://www.statnews.com/2019/10/30/novartis-zolgensma-gene-therapy-study-halted-by-fda-on-animal-safety-concerns/). Moreover, a recent clinical trial of an AAV9-mediated gene therapy for Duchenne muscular dystrophy was halted due to kidney injury and decreases in red blood cell counts in patients after intravenous viral infusion (https://www.the-scientist.com/news-opinion/trial-of-gene-therapy-for-duchenne-muscular-dystrophy-put-on-hold-66711). Similarly, by testing multiple AAV genome structures and capsid types using subretinal injection in mice, Xiong et al. found a strong correlation between cis-regulatory sequences and toxicity [76] . Interestingly, AAVs with any one of three broadly active promoters, or an RPE-specific promoter, were toxic, while AAVs with four different photoreceptor-specific promoters were not toxic at the highest doses tested. With increased efforts in using AAVs to express specific proteins to manipulate neuronal functions, distinguishing the true effects versus the noise/inflammation/toxicity associated with viral vectors should be considered with greater caution. Thus, refining these viral vectors will facilitate the development of new gene therapy protocols.

Finally, the optimizations of viruses for neuronal targeting discussed herein have limitations in their translational potential due to diversity in cross-species tropism of viruses. Indeed, the retinal cell types targeted by specific synthetic promoter AAVs varied widely between mouse and NHP/human retinas [63]. Moreover, the PHP.B virus discussed above, which crosses BBB and remarkably transduces CNS following intravenous infusion in mice [35], was not found to do the same in NHPs [77]. Therefore, in these cases, optimizing cell-type targeting in mice yields vectors that are unlikely to optimally target the same cell type in NHPs/humans. On the other hand, the ability of an AAV to target a neural cell group in NHPs is a good predictor for targeting the same cell group in humans [63], suggesting that phylogenetically closer species have more correlated responses to virus infection.

Concluding remarks

Developments in the past decade have shown that recombinant viral tools have started to transform our ability to map, observe and manipulate neuronal circuits in experimental models, and to design gene therapy strategies in humans. While this review has been mainly focused on targeting specific neuronal types, recombinant viral vectors have also been used for monosynaptic tracing of neuronal connections [1,78]. In brief, the vectors for retrograde trans-synaptic tracing include the SADB19-base vectors [79], and a less toxic CVS-N2c strain [80]. Recent studies also led to the development of anterograde trans-synaptic tracing vectors, such as the recombinant HSV-129 (Table 1) [81,82], AAV1 or AAV9-Cre [83], and VSV [84]. Furthermore, several studies have combined cell type-specific viral vectors with such “trans-synaptic” or “trans-neuronal vectors” to dissect input-output circuits [24].

In conclusion, it is an exciting time in neuroscience as we continue to make progress into uncovering the diversity of neuronal cell types. Using the methods (projection-based, serotype-restricted, GRE-dependent, and delivery-controlled) described here, their combinations, and other yet to be discovered methods to generate viral vectors, in the not too distant future it will be possible to target identified neuronal subtypes with even greater precision. This will allow us to determine the connectivity and functions of various types of neurons in the mammalian brain, and use such knowledge to develop therapies for brain disorders.