Activity-Induced Remodeling of Olfactory Bulb Microcircuits Revealed by Monosynaptic Tracing

The continued addition of new neurons to mature olfactory circuits represents a remarkable mode of cellular and structural brain plasticity. However, the anatomical configuration of newly established circuits, the types and numbers of neurons that form new synaptic connections, and the effect of sensory experience on synaptic connectivity in the olfactory bulb remain poorly understood. Using in vivo electroporation and monosynaptic tracing, we show that postnatal-born granule cells form synaptic connections with centrifugal inputs and mitral/tufted cells in the mouse olfactory bulb. In addition, newly born granule cells receive extensive input from local inhibitory short axon cells, a poorly understood cell population. The connectivity of short axon cells shows clustered organization, and their synaptic input onto newborn granule cells dramatically and selectively expands with odor stimulation. Our findings suggest that sensory experience promotes the synaptic integration of new neurons into cell type-specific olfactory circuits.


Introduction
The mammalian brain ensures adaptive behavior through its large capacity for cellular and circuit plasticity. The diverse scales of neural plasticity range from single synapse modification [1][2][3] to network remodeling that accompanies ongoing neurogenesis [4][5][6][7]. Plasticity mechanisms accommodate changing environmental stimuli that are continuously relayed to the brain via multiple sensory modalities. Among sensory systems, the olfactory system possesses a large capacity for circuit plasticity through continued generation of new neurons in adult life. Such continuous incorporation of new neurons implies persistent, large-scale remodeling of synaptic connections, the nature of which is not well known. Within the olfactory system, the axons of olfactory sensory neurons (OSNs) expressing the same odorant receptor [8] converge onto discrete glomeruli in the main olfactory bulb (MOB) [9,10]. Organized around glomeruli, groups of mitral/ tufted cells, as well as various interneurons, form connected networks that extend into all layers of the olfactory bulb [11]. These networks likely represent unitary modules for odor information processing [11][12][13][14] and may be functionally analogous to barrels in the somatosensory cortex or ocular dominance columns in the visual system.
The functional organization within and between MOB glomerular units has been the subject of intense investigation. Lateral interactions between glomeruli are mediated primarily by dendrodendritic synapses between mitral cells and granule cells [15][16][17][18][19][20], and the electrophysiological properties of these synapses have been well characterized [13,21,22]. Although mostly studied as singly recorded neurons or synaptically coupled pairs, these experiments support the notion that populations of neurons associated with multiple glomeruli are highly interconnected. Among the most studied forms of intrabulbar circuitry, granule cells provide inhibitory feedback onto spatially distant glomeruli by forming synapses with the lateral dendrites of mitral cells [13,15]. In addition, synaptic inputs from both local short axon cells (SACs) and distant cortical neurons provide direct regulation of granulemitral cell synapses [23][24][25][26]. Despite a central role in olfactory processing, the relative connectivity of individual granule cells to different cell types, the spatial organization of granule cell synaptic partners, and the regulation of granule cell connectivity by sensory stimulation remain unclear.
New GABAergic granule and periglomerular cells in the MOB are continually generated throughout adulthood [27][28][29]. Whereas many adult-born neurons fail to establish and maintain dendrodendritic synapses and ultimately undergo apoptosis [30][31][32], granule cells born during early stages of postnatal development tend to be long-lived and form stable synaptic connections [33]. We thus sought to define the patterns of cellular connectivity formed by postnatal-born granule cells in the MOB and determine how new granule cell microcircuits are influenced by sensory input. In the present study we employed monosynaptic circuit tracing using pseudotyped rabies virus together with a conditional red-fluorescence mouse reporter strain to label newborn olfactory bulb interneurons and their presynaptic partners in vivo [34]. We show that postnatal-born granule cells make synaptic connections with cortical inputs and multiple olfactory bulb cell types. The pattern of monosynaptic connectivity shows a clustered organization that is characterized by extensive presynaptic inputs from anatomically distinct short axon cells. Moreover, increased sensory experience by odor enrichment enhances SAC connectivity onto postnatal-born granule neurons. These results define the presynaptic repertoire of novel inputs onto newborn granule cells, and support a model whereby clustered patterns of organization in the olfactory bulb extend from local short axon cells to cohorts of deep granule cells that span the laminae of the olfactory bulb. The identification of numerous short axon cells presynaptic to new granule cells reveals unanticipated cellular interactions that occur during granule cell (GC) synapse development. Experience-driven changes in SAC-GC connectivity provides a circuit basis for refining or remodeling synaptic inputs upon exposure to a complex sensory environment through ongoing neurogenesis.
Although much is known about the cellular patterns of MOB interneuron development [35,43,44], the patterns of synaptic connectivity onto new granule cells are poorly understood. To determine the postnatal patterns of synaptic connectivity onto newborn granule cells, we performed in vivo monosynaptic circuit tracing [34]. For this we generated a conditional reporter mouse harboring a Cre/loxP-dependent allele (Figs. 1a, S1a, and S1b) capable of driving high levels of cytosolic tdTomato expression upon introduction of the plasmid G-IRES-TVA-IRES-Cre (Fig. 1b) that encodes components for targeted rabies virus (RV) infection, monosynaptic virus propagation, and Cre-mediated conditional reporter activation. In this design, neurons are genetically targeted for infection by expression of the TVA receptor, which binds selectively to RV particles pseudotyped with EnvA coat proteins [45]. Since the engineered RV mutant lacks the G coat protein, plasmid expression of the wildtype rabies-G coat protein enables precisely one round of ''live'' virus packaging and trans-synaptic infection of presynaptic cells [34,[46][47][48][49][50]. Retrograde spread of virus is strictly monosynaptic since only cells that receive G-IRES-TVA-IRES-Cre are capable of synthesizing live virus, whereas presynaptic targets do not contain the G protein and thus are incapable of producing infective RV particles. Replacement of the viral gene encoding the G capsid protein with an EGFP reporter renders presynaptic partners of RV-targeted cells brightly labeled [34,51].
To test the functionality of our tracing system, we injected the G-IRES-TVA-IRES-Cre tri-cistronic construct into the lateral ventricles of embryonic day 14.5 (E14.5) ROSA26-stop flox -tdTomato mice, electroporated neuronal progenitors in the ventricular zone, and made ex vivo cortical slice cultures [52] (Figs. S1c-i). After two days, EnvA-pseudotyped SADDG-EGFP RV particles [34] were applied to cultured brain slices to infect neurons expressing G-IRES-TVA-IRES-Cre. Three days later, we observed widespread tdTomato expression throughout layers 4 and 5 and in the walls of the ventricles (Figs. S1c, S1f, and S1i), indicating neurons expressing Cre recombinase via the G-IRES-TVA-IRES-Cre cassette. In addition, we observed EGFP expression in a small number of tdTomato-positive neurons directly infected by SADDG-EGFP RV, which appeared yellow (Figs. S1h and S1i). Finally, we observed a large cohort of EGFP-expressing neurons lacking tdTomato expression, showing trans-synaptic spread of SADDG-EGFP RV to presynaptic partners of RV-infected cells (Figs. S1g-i).
After confirming that our tricistronic rabies G-IRES-TVA-IRES-Cre construct (Fig. 1b) functioned in conjunction with the ROSA26-stop flox -tdTomato mouse line (Fig. 1a), we implemented this model system to examine the synaptic connectivity formed onto postnatal-born granule cells in the olfactory bulb. To target newborn granule cells for RV infection and subsequent monosynaptic tracing, we injected the G-IRES-TVA-IRES-Cre construct into the SVZ of postnatal day 2 (P2) mice and applied an external voltage across the brain to electroporate neuronal progenitors (Fig. 1c, left) [53]. After 30 d, at which point many of the electroporated cells have migrated to the MOB and formed functional synaptic connections [44,54], EnvA-pseudotyped SADDG-EGFP rabies virus was injected into the granule cell layer of the olfactory bulb to infect newly incorporated neurons expressing G-IRES-TVA-IRES-Cre (Fig. 1c, right). This experimental paradigm allowed us to probe the monosynaptic connectivity onto postnatal-born granule cells by directly visualizing vital fluorescence in neurons susceptible to pseudotyped RV infection (red), those that became infected (red and green, thus yellow), and presynaptic target neurons (green) (Fig. 1d). To target the electroporated granule cells for infection and monosynaptic tracing, we injected 250 ml the SADDG-EGFP rabies virus 750 mm below the surface of the olfactory bulb, midway from the anterior and posterior boundaries. Although the methods of in vivo electroporation and viral infection are variable, by counting the number of infected source cells and determining the average bulbar volume spanned by these cells, we estimated viral spread and infection to encompass a spherical domain 3006200 mm in diameter (Fig. 1e, n = 10 labeled bulbs from 10 mice). To count labeled source cells, we made 100 mm slices through the entire bulb and identified all GCs that expressed both EGFP and tdTomato (n = 20-25 slices each, with the middle 3-5 slices harboring most of the labeled cells). Occasionally we infected periglomerular cells, which are also continually generated from electroporated SVZ [43]. In these instances we observed labeling of the cell types that contribute to the spherical glomerular structures (data not shown). To avoid infecting newborn periglomerular cells, we included a small volume (,20 nl) of mineral oil in the tip of the injection pipette to prevent exposure of the virus while passing through the superficial bulb layers en route to the granule cell layer.
One week after RV infection, we processed the olfactory bulb and imaged fixed sections for reporter expression. Similar to what we observed in cortical slice explants (Fig. S1c-i), in vivo conditional reporter activation and trans-synaptic viral labeling was robust, and showed distinct patterns of presynaptic labeling (Figs. 1f-m). Although we began to observe viral-mediated EGFP in presynaptic target cells by 3 d following infection, high and stable levels of labeling were routinely achieved by 7 d. By 14 d post-infection, numerous presynaptic neurons showed abnormal morphologies, likely due to the extremely high levels of reporter expression driven by the RV genome. Thus, for all subsequent experiments, we chose 7 d post infection to investigate our presynaptic labeling. Since electroporation relies on generating a voltage across the ventricular space, injected DNA reproducibly showed unilateral patterns of reporter expression in the brain (Fig. 1f). This phenomenon was most obvious when imaging the conditional expression of tdTomato following transient introduction of Cre into the SVZ, which triggered tdTomato expression in neuronal progenitors that remained active in all daughter cells born from an electroporated lineage ( Fig. 1g and 1h). Although this approach resulted in clonal labeling of many newborn granule cells with tdTomato, we were able to small numbers of postnatalborn granule cells for SADDG-EGFP RV infection ( Fig. 1i and 1j). This approach allowed us to resolve distinct clusters of presynaptic targets to electroporated granule cells (Fig. 1k-m). Based on the sparse infection by SADDG-EGFP RV, the lack of infection points in close vicinity to one another, and the absence of clonal sectors (unlike those observed in cortical slices, Fig. S1), we reasoned that only a very small number of electroporated cells showed stable integration of the electroporated G-IRES-TVA-IRES-Cre plasmid. To further test this, we performed monosynaptic labeling of postnatal born granule cell networks while marking neuronal lineages born after electroporation with bromodeoxyuridine (BrdU) (Fig. S2a). At 14 d after electroporation, mice were supplemented with BrdU in their drinking water for 2 weeks to label all neurons born after that time, followed by RV infection and subsequent tissue processing. By comparing the number of green presynaptic inputs to the number of BrdU labeled green cells, we rarely ever observed newborn neurons that were doubly labeled ( Fig. S2b-e). These data show that only a very small fraction of the tdTomato marked cells harbor the monosynaptic tracing components, and suggest that stable integration of our expression plasmid is rare and likely occurs in postmitotic neuronal precursors. By separately labeling primary points of infection (yellow source cells) and the cohort of neurons that provide their presynaptic input (green) (Fig. 1d), this modular virus-based molecular genetic system thus allowed us to directly visualize the synaptic microcircuits associated with postnatal-born granule cells.

Cohorts of Mitral Cells and Short Axon Cells Synapse Onto New Granule Cells
The full complement of neurons that contact newly integrated granule cells is unknown. In addition to dendrodendritic synapses formed between mitral/tufted cells and granule cells, it has recently been shown that granule cells form functional connections with multiple local interneuron cell types [24,26]. After genetically targeting postnatal granule cells for SADDG-EGFP RV infection, we observed extensive labeling throughout the granule cell, mitral cell, and external plexiform layers (EPL) in neurons presynaptic to electroporated granule cells. Interestingly, we noted clustered patterns of connectivity (Fig. 2a). These clustered networks contained mitral cells [55] (Fig. 2b) and other inframitral layer cells that morphologically corresponded to both superficial and deep short axon cells (SACs) (Figs. 2c and 2d). By restricting rabies-G, TVA, and Cre expression to postnatal-born interneurons through timed in vivo electroporation, we never observed conditional tdTomato reporter expression in cell types other than periglomerular and granule cells. Further, we never observed trans-synaptic labeling between granule cells born after electroporation ( Fig. S2 and data not shown), arguing against nonspecific RV uptake through proximity spillover.
Varying the amount of pseudotyped RV that was injected into the MOB allowed us to control the number of virally infected source cells. Levels of viral infection ranged from very low (100 nl) (Figs. 1k-m) to high (500 nl) (Fig. 2a), and thus could be adjusted for optimal microcircuit analysis. For our experiments, we used conditions (250 nl) that achieved a moderate level of labeling (see Fig. 1e and Experimental Procedures) to facilitate source cell and presynaptic partner identification. To validate the retrograde trans-synaptic transport of SADDG-EGFP RV, we examined EGFP expression in cortical brain regions known to send longrange centrifugal inputs to the MOB and make synapses onto granule cells [13,56]. Consistent with the known presynaptic transport of RV, we observed strong SADDG-EGFP expression in neurons of the anterior olfactory nucleus (AON), horizontal limb of the diagonal band nucleus (HDB), and piriform cortex (Fig. S3). Moreover, a complete absence of EGFP label in olfactory sensory neurons (OSNs), which heavily innervate the MOB and synapse onto mitral cells, confirmed that viral labeling stops after one presynaptic connection (data not shown). Together these data show that postnatal-born granule cells can be precisely targeted for RV infection, demonstrate both short-and long-range synaptic connectivity onto postnatal-born granule cells, and reveal clustered patterns of organization intrinsic to granule cell microcircuits.
To further investigate the repertoire of cells presynaptic to postnatal born granule cells, we performed immunohistochemical analysis on thin sections of MOB and counted the number of EGFP-positive, RV-infected neurons that were negative for tdTomato (hence, presynaptic to Cre-expressing granule cells) using molecular markers expressed by olfactory bulb cell types [57]. To begin to elucidate the molecular identity of the cells presynaptic to newborn granule cells, we made 50 mm sections through the labeled domain of the MOB (average 6 sections per bulb), performed immunohistochemistry, and counted EGFP labeled cells to determine the percentage of cells that showed overlapping expression of various interneuron markers. Whereas only a subset of EGFP-positive cells expressed the interneuron markers calretinin (4665% EGFP-positive cells, n = 100 cells in 18 sections from 4 mice; Figs. 2e-g) and parvalbumin (5567%; Figs. 2h-j), we routinely observed co-expression of the GABA A receptor subunit a1 (8464%; Figs. 2k-m), which is highly and selectively expressed in SACs residing within inframitral cell layers [57]. Thus, through immunohistochemical analysis against interneuron markers in the olfactory bulb, ,50% of the SADDG-EGFP labeled cells expressed parvalbumin and/or calretinin, whereas .80% of the local presynaptic inputs expressed GABA A receptor subunit a1. We never detected SADDG-EGFP expression in cell types that express tyrosine hydroxylase in the upper EPL or glomerular cell layer (Figs. 2n-p). Interestingly, we occasionally observed SADDG-EGFP expression in proximate glial cells that were directly contacting source granule cell (GC) dendrites, as shown by colocalization with glial fibrillary acidic protein (GFAP) (Figs. 2q-s), suggesting a potential role for glial contact in synapse remodeling of newborn granule cells [58][59][60]. Together, these data show that, in addition to forming welldescribed contacts with centrifugal inputs and dendrodendritic synapses with mitral and tufted cells, postnatal-born granule cells receive extensive input from local SACs in the MOB.
To more closely examine the connectivity made between granule cells and their presynaptic partners, we took advantage of the dual-color labeling scheme of our experimental system to count the number and types of neurons that were labeled with SADDG-EGFP RV by monosynaptic transfer. Given that all granule cells targeted for G-IRES-TVA-IRES-Cre electroporation express tdTomato (red), and only a subset of those cells became infected by injected SADDG-EGFP RV (red and green, thus yellow), presynaptic targets can be clearly identified by the presence of only EGFP (green) (Fig. 1d). From our immunohistochemical data identifying the majority of non-mitral cell presynaptic neurons as SACs (Figs. 2c and 2d), we sought to determine the relative connectivity of SACs onto new granule cells. We counted doubly labeled (yellow) granule source cells and all of their green presynaptic neuronal partners in 100 mm thick serial sections of entire olfactory bulbs. On average we identified 3368.1 doubly labeled source cells per bulb. Interestingly, we found that the ratio of presynaptically labeled short axon cells to RV-infected granule cells was quite high (4.660.8, SEM, n = 7 olfactory bulbs), revealing a previously unappreciated level of local SAC input to newborn granule cells.
We next investigated the clustered architecture of SACs presynaptic to newly integrated GCs. For this, we selected olfactory bulb tissue that had relatively sparse SADDG-EGFP RV infection and presynaptic labeling in order to facilitate the imaging of labeled SAC to GC microcircuits with cellular resolution (Figs. 3a-c). To determine the relative numbers, locations, and types of presynaptic neurons that contribute to the clustered network, we prepared semi-thick brain slices (150-200 mm) that were optically cleared in glycerol for z-stack confocal imaging and 3-dimensional volume rendering [61]. To quantify the inputs to granule source cells, we took serial image planes through the slices at 2 mm intervals, counting all doubly labeled and green-only neurons to determine the ratio of presynaptic inputs to granule source cells. In addition to clearly identifiable mitral cells (Fig. 2a-d), reconstructed image stacks revealed two main populations of SACs which have been previously characterized as deep SACs and superficial SACs [24,57,62]. Quantitative analysis revealed 5.661.5 SACs per resolvable clustered network (Fig. 3d), and an average width of the network (including SAC cell bodies and proximal dendrites) of 148.5640.6 mm (n = 10 SAC/GC networks from 4 bulbs, of 4 mice 6 SEM). Neurons presynaptic to individual GCs were categorized as SACs by morphology, molecular marker analysis, and electrophysiological properties ( Fig. 2 and Fig. 3d-f), whereas the widths of GC-SAC networks were determined by the measuring the lengths of clearly resolvable SAC dendrites. Consistent with previous reports [24,63], presynaptic deep SACs showed low frequency trains of action potentials with depolarizing current injections, whereas superficial SACs responded with high frequency patterns of firing ( Fig. 3e-f). It is unlikely that our viral labeling identifies the full complement of functional synapses onto postnatal-born granule cells due to the unknown efficacy of particle transfer. In addition, it is likely that we underestimated the width of a functional MOB cluster by not including distal SAC axonal arbors in our characterization due to the lack of fine resolution imaging in thick brain slices. Altogether, these data reveal an unexpected and extensive local connectivity between SACs and newly integrated granule cells in olfactory bulb.

Odor Enrichment Increases SAC-GC Connectivity in the Olfactory Bulb
To examine how sensory experience influences granule cell microcircuits, we analyzed the patterns of connectivity between granule cells and their presynaptic targets following olfactory stimulation. We concentrated on synaptic input from SACs as these cells were robustly labeled using RV monosynaptic tracing (Figs. 2, 3). To provide a broad palette of odors for long-term sensory enrichment, we designed a robotic system for continuous cycled delivery of multiple odorants to freely exploring mice (see Fig. S4 and methods). For odor enrichment, ROSA26-stop flox -tdTomato mice were subjected to in vivo SVZ electroporation with G-IRES-TVA-IRES-Cre and reared with their mothers for 30 d in cages ported for odor delivery (Fig. S4). After the 30 d odor stimulation period, SADDG-EGFP RV was injected into the granule cell layer of the olfactory bulb, and 7 d later the olfactory bulb was dissected and sectioned to count dual-labeled (yellow) granule cells and their single-labeled (green) presynaptic partners (Fig. 4a). Odor exposure induced a dramatic increase in the number of SACs presynaptically coupled to new granule cells compared to the non-odor enriched control group (Figs. 4b and 4c). Specifically, odor stimulation tripled the connectivity ratio of SACs onto source GCs (control, 4.660.8; odor-enriched, 13.861.0; n = 3 bulbs from 3 mice, 6 SEM) (Fig. 4d). On average, we counted 48614 doubly labeled source cells per bulb in mice subjected to odor enrichment. The increase in labeled presynaptic partners was not simply a reflection of an increased absolute number of new granule cells since the quantification normalizes to the number of labeled granule cells. Thus, odor enrichment enhances SAC connectivity onto new granule cells.
One possible explanation for the increased labeling of presynaptic cells upon odor stimulation is that elevated neuronal activity enhances RV transfer at existing synapses. To address this issue, we prepared olfactory bulb explants from mice that had been electroporated with a plasmid encoding the monosynaptic tracing components, infected with the SADDG-EGFP RV, and cultured in the presence of pharmacological agents to block synaptic transmission. We found that blocking SNARE-dependent neurotransmitter release, action potentials, or fast glutamatergic neurotransmission had no significant effect on the number of monosynaptically labeled cells compared to untreated controls (Fig. S5). Thus, trans-synaptic transfer of RV is insensitive to activity manipulations over several days in vitro, and we conclude that the odor-induced expansion in granule cell circuit labeling in vivo (Figs. 4c and 4d) is attributable to changes in the number of synaptic inputs per granule cell, rather than to alterations in the effectiveness of RV transfer between neurons following olfactory stimulation. Consistent with this notion, we observed morphological differences in the granule cells themselves. Doubly labeled granule source cells in the bulbs of mice that were reared in odor enriched environments showed a significant increase in the number of dendritic protrusions (control, 8.460.7 per 25 mm dendrite, n = 17 doubly labeled neurons from 3 mice; odor, 12.760.9; n = 18 doubly labeled neurons from 4 mice; p,0.01; Figs. 5a and 5b). We also noted a significant increase in the number of inhibitory synapses on the dendrites of source granule cells by staining for the inhibitory synapse marker gephyrin (control, 7.860.9 gephyrin clusters per 35 mm dendrite; odor, 11.961.3; n = 12 neurons from 3 bulbs each; p,0.02; Figs. 5c and 5d). To restrict our analysis to synaptic changes in postnatal born neurons, gephyrin-positive puncta were only counted if they could be clearly colocalized within doubly labeled granule source cells. Corresponding to enhanced SAC input onto granule cells following odor enrichment, we also noted increased mitral cell input (data not shown). Given these findings, we cannot rule out that a portion of the gephyrin-positive puncta may indeed correspond to increased centrifugal input from other inhibitory cell types. Nonetheless, these data support increased presynaptic input onto newborn neurons following odor enrichment, and reveal activity-induced expansion of SAC circuitry.
SACs provide GABAergic inhibitory input onto MOB granule cells [24,26,64]. We next tested whether the odor-induced increase in both RV-labeled SACs presynaptic to GCs and gephyrinlabeled inhibitory synapses onto GCs corresponds with changes in functional synaptic connectivity. To this end, we performed wholecell patch clamp recordings in acute brain slices from labeled MOB granule cells following in vivo postnatal electoporation and measured miniature inhibitory postsynaptic currents (mIPSCs). Consistent with the observed increase in gephyrin puncta (Figs. 5c,d), odor stimulation significantly increased mIPSC frequency in postnatal born granule cells (Fig. 6). Whereas mIPSC amplitudes were similar between experimental groups (control, 15.261.4 pA; odor, 14.461.3 pA; Fig. 6b), mIPSC frequency was increased in odor-exposed animals (control, 0.5560.06 Hz, n = 9 cells from 3 mice; odor, 0.89 Hz60.07, n = 10 cells from 4 mice, p,0.01, unpaired t-test; Fig. 6a). Taken together, these data show that sensory experience expands connectivity between short axon cells and new granule cells, defining a novel cell type-specific reorganization of olfactory circuits in response to odor-induced activity.

Discussion
In the present study we combined genetic circuit tracing and in vivo labeling technologies to map monosynaptic connections made between postnatal-born granule cells and their presynaptic input neurons in mouse olfactory bulb. We found that, in addition to previous known connections with mitral/tufted cells, postnatalborn granule cells show extensive connectivity to short axon cells, and these short axon cell inputs contribute to clustered architecture in the olfactory bulb. In addition, we have found that increased sensory experience in the form of odor stimulation expands circuit connectivity made onto newborn granule cells. This increase in synaptic connectivity manifests as a threefold expansion in presynaptically coupled short axon cells, and is accompanied by a corresponding increase of granule cell inhibitory synapses and mIPSCs, as well as changes in dendritic morphology.

Short Axon Cell Input onto Granule Cells
Implementing a highly selective in vivo genetic targeting strategy, we investigated the synaptic patterns of connectivity that are made between postnatal-born granule cells and their presynaptic targets using engineered rabies virus for monosynaptic circuit tracing. Our experimental design enabled us to selectively target granule cells for rabies infection and propagation via in vivo electroporation [53], allowing direct determination of the cell types presynaptic to targeted granule cells. Using this approach, we observed discrete synaptic networks associated with granule cell microcircuits. Surprisingly, the majority of neurons making local presynaptic inputs onto postnatal-born granule cells were identified as SACs in the granule cell and external plexiform layers of the olfactory bulb.
Recent morphological and electrophysiological studies have shown that SACs elaborate axonal arborizations throughout all layers of the olfactory bulb, are presynaptic to resident granule cells, and provide GABAergic input that modulates granule cell firing [24,26,57]. Interestingly, the superficial SACs we have identified in this study display morphological and electrophysiological characteristics similar to Van Gehuchten neurons, which express many of the same molecular markers as deep SACs, but are thought to make contacts on both GCs and mitral cells, and are axon-less [62][63][64][65][66]. These properties suggest that SACs provide spatial refinement of local inhibitory circuits that sculpt the firing properties of mitral cells. In other brain areas, GABAergic input during neuronal development contributes to cell differentiation, survival, and circuit maturation [67][68][69][70][71]. Given ongoing adult neurogenesis, resident short axon cells in the mature olfactory bulb may similarly contribute to dynamic remodeling, differentiation, or survival of newly incorporated granule cells [43,72]. Thus, in addition to providing inhibitory control over dendrodendritic synapses in the mature circuit, SACs may regulate synaptic integration of newborn granule cells through developmental signaling mechanisms. In such a scenario, odor enrichment would not only modify network activity in the bulb, but also promote GABAergic input onto new GCs from select cohorts of SACs. These odor networks may serve as hubs of activity that are more favorable or attractive for newborn neuron synapse formation and survival. More refined modes of circuit activity manipulations using cell type-specific chemical genetic or optogenetic manipulations will be required to fully address this notion [46].
Our experimental approach was effective in demarcating presynaptic cell types known to make functional connections onto postnatal-born granule cells, including centrifugal input from olfactory cortex, mitral/tufted cells, and SACs. The extensive connectivity made by SACs onto granule cells raises the question of the identity of neurons that are presynaptic to SACs. What cell types drive the inhibitory influence that SACs in turn relay onto granule cells? Although it has been proposed that mitral cells could serve this role [24], direct evidence in support of this model is lacking. Future genetically targeted monosynaptic tracing would be well suited to address this question. Another intriguing observation was the viral labeling of certain sparse glial cells found to be in direct physical contact with granule ''source'' cells. Such labeling could be due to uptake of viral particles at neuronglia contacts. Indeed newborn neocortical neurons are known to form transient gap junction contacts onto radial glia [73], and functional neuron-glia synapses have been described [74,75]. Alternatively, glial labeling could arise by selective engulfment or uptake of cell remnants during apoptosis or synapse pruning [76,77]. It will be important for future studies to examine the electrophysiological and ultrastructural nature of this neuron-glia interaction.

Odor Experience Expands Olfactory Bulb Circuitry
SACs are born during embryonic development and are resident prior to either sensory input or granule cell integration [35,78].
Thus, the clustered organization onto granule cells may reflect developmental patterning events that occur during forebrain maturation, or alternatively may represent local connectivity that is elaborated or pruned in response to patterns of sensory experience or glomerular activity. In our experimental system, we observed clustered SAC inputs onto granule cells in both unstimulated controls and odor-enriched groups. However, in addition to the known enrichment of postnatal-born neurons that become integrated into olfactory bulb circuits following odor stimulation [7,31], we have found that the number of neurons forming functional presynaptic inputs onto individual granule cells is also dramatically increased following odor enrichment. This increased level of presynaptic input onto postsynaptic granule source cells was revealed by greater numbers of labeled presynaptic targets, morphological changes in postsynaptic spine structures and inhibitory scaffolds, as well as increased inhibitory drive. The presence of more presynaptic inputs onto newborn GCs does not necessarily imply a linear increase in cell-to-cell connectivity. That is, in response to odor stimulation additional presynaptic cell types could form synaptic connections without an overall change in the absolute number of synapses onto a GC, provided that the overall connectivity ratio for any given presynaptic target cell decreases. Although the focus of our analysis was on the increased number of inputs onto GCs by SACs, we also observed an increase in the number of labeled presynaptic mitral cells (data not shown). It is likely that the changes we observe in cellular morphology and electrophysiological output in response to odor enrichment reflects a composite of expanded connectivity from not only the SACs, but also mitral cells and centrifugal inputs. In the present study, we have shown that presynaptic connectivity made onto newborn granule cells by SACs is significantly increased in response to sensory stimulation, suggesting they play a pivotal role in the capacity for postnatalborn neurons to integrate within the intact brain. It will be important for future studies to determine, in detail, how the complete repertoire of presynaptic inputs is tuned by sensory experience. Thus, in addition to increasing the number of newly incorporated granule cells [4,7,31,79], sensory experience recruits additional presynaptic elements that contact each granule cell. Such augmented integration may contribute to the high degree of cellular plasticity in this region of the brain and the improved sensory discrimination observed upon odor enrichment [4].
Our slice explant experiments suggest that synaptic transfer of RV is independent of action potentials, VAMP-mediated synaptic release, and fast glutamatergic neurotransmission and thus reflect changes in physical network connectivity. Indeed, odor stimulation increased both spine density and inhibitory synapses on new granule cells, consistent with an expansion of presynaptic input. It remains to be determined if the circuit changes we observed in our experimental paradigm result in long-term structural changes that persist throughout the life of an animal, or represent transient connections that become pruned or lost in the absence of continued odor experience. Our current method of labeling presynaptic inputs is irreversible, and it will be important for future long-term tracing studies to examine the temporal course of granule cell circuit plasticity in response to odor enrichment or deprivation. Such an analysis may require new technologies for repeated genetic labeling of presynaptic partners onto defined cell populations over time.
Our present study has demonstrated the power of new genetic technologies to define neural circuits in vivo. The olfactory system is an unusually plastic region of the mammalian brain, capable of continued addition and removal of cell cohorts, with accompanied synapse and circuit remodeling into adulthood. Our results indicate that sensory experience promotes the synaptic integration of new neurons into extensive, spatially organized, cell typespecific olfactory circuits, providing a tractable in vivo model to better understand how activity influences complex neural circuit formation. Moreover, adult-born neurons normally destined for the olfactory bulb can reroute to lesion sites in the neocortex and striatum [80,81], suggesting that the mechanisms of plasticity innate to this renewable cell type may be harnessed for tissue repair. More broadly, monosynaptic tracing in vivo provides a powerful approach to map global connectivity of newly integrated neurons during development, plasticity, and regeneration.

Methods
All experimental procedures and reagents used for this study were approved by the Institutional Animal Care and Use Committees at Baylor College of Medicine and Duke University Medical Center.

Expression Plasmid Construction
To generate the pCAG-Rabies G-IRES-TVA construct, the cDNA encoding rabies G was excised from pHCMV-RabiesG [82] and cloned into a modified pCIG backbone [83]. For bicistronic expression of the TVA receptor, the TVA cDNA from pCMMP-TVA800 [84] was PCR amplified and cloned downstream of an IRES2 element (Clontech, Mountain View, CA) for insertion 39 to rabies G. For tri-cistronic expression of Cre recombinase, IRES-Cre [85] was inserted downstream of rabies G-IRES-TVA to generate pCAG-rabies G-IRES-TVA-IRES-Cre.
Generation of ROSA26-stop flox -tdTomato Mice A 1.6 kb cDNA fragment encoding the tdTomato protein (provided by Roger Tsien, UCSD) was PCR amplified and cloned upstream of the polyA signal sequence of pCRII. The tdTomato-polyA construct was verified by sequencing. We next constructed a shuttle vector by first cloning the tdTomato-polyA cDNA into the pBigT vector [86] and inserting the CAG promoter element using PacI upstream of the loxP-stop-loxP sequence. We then moved the CAG-loxP-stop-loxP-tdTomato-polyA cassette into the pROSAacceptor targeting plasmid as previously described [87] to generate the ROSA-CAG-loxP-stop-loxP-tdTomato targeting vector. This targeting construct was linearized and electroporated into E14 ES cells [88]. Following selection, clones were picked and screened for the recombined allele by southern blotting. For southern blots, ES cell genomic DNA was digested with EcoRV, transferred, and hybridized with an external probe to the ROSA26 locus. The wildtype allele gave an 11.5 kb fragment, whereas a 5.7 kb band detected the mutant allele. Using standard procedures [89], positive ES cell clones were used to generate gene-targeted mice. The resulting offspring were genotyped by PCR. To detect both the wildtype and targeted alleles, the following PCR primers were designed for multiplexing: Rosa/01, 59-CACTTGCTCTCC-CAAAGTCG -39; Rosa/02, 59-TAGTCTAACTCGCGA-CACTG -39; CAG/02, 59-GTTATGTAACGCGGAACTCC -39. The wild type allele produced a ,560 bp fragment with Rosa/ 01 and Rosa/02 primers, whereas the mutant allele was detected by a ,300 bp fragment with Rosa/01 and CAG/02 primers.

In Vivo Electroporation and RV Labeling of Postnatal-Born Granule Cells
Newborn mice were anesthetized by brief hypothermia. Then, 500 nl of 1 mg/ml endotoxin-free plasmid DNA was injected unilaterally into the right lateral ventricle using a Hamilton syringe with a custom beveled 33-gauge needle (Hamilton Company, Reno, NV). Electroporation was carried out by applying multiple voltage pulses across the width of the newborn heads, just posterior to the eyes, using circular 7 mm tweezertrodes and a BTX ECM 830 square wave electroporator (Harvard Apparatus, Holliston, MA). The electroporation parameters included 5 pulses of 150 V for 50 ms each with 950 ms intervals. Immediately after the procedure, electroporated mice were returned to a heated home cage and monitored until recovery. At 30 d following the electroporation procedure, 250 nl (6610 3 particles/ml) of pseudotyped SADDG-EGFP RV was injected into the granule cell layer of the olfactory bulb using glass injection pipettes and a Nanoject II (Drummund Scientific Company, Broomall, PA). We targeted injection at the midpoint of the olfactory bulb 750 mm below the surface of the brain. This yielded a volume of infection spanning ,300 mm in diameter6200 mm (see Fig. 1e). At 7 d post-infection, olfactory bulbs were dissected and prepared for image analysis.

Electroporation and Organotypic Slice Cultures
Organotypic brain slices were prepared and cultured as previously described [52] with minor modifications. For ex vivo slices, dorsal telencephalic progenitors were labeled by injecting pCAG-Rabies G-IRES-TVA-IRES-Cre plasmid DNA (0.1 mg/ ml) diluted in a 0.1% Fast Green solution into the lateral ventricles of decapitated E14.5 ROSA26-stop flox -tdTomato mouse heads. The solution was delivered with a small glass capillary pipette attached to a Picospritzer II (General Valve Corp., Fairfield, NJ) using five 15-psi pulses lasting 4 ms each. Electric potentials were generated across intact heads using gold-coated electrodes attached to an ECM 830 electroporator with the following parameters: four 100 ms 45 V pulses separated by 100 ms intervals. Immediately after electroporation, brains were dissected, vibratome sectioned at 250 mm, and maintained as interface organotypic cultures prior to fixation and immunohistochemical labeling. For acute olfactory bulb slices, wildtype newborn mice were electroporated as described above with a DNA plasmid encoding EF1a-tdTomato-P2A-Rabies G-IRES-TVA-WPRE-pA. 10 d later, brains were dissected, vibratome sectioned at 250 mm, infected with virus, and maintained as interface organotypic cultures. The next day pharmacological agents including one or more of TTX (1 mM, Sigma), botulinum toxin A (50 nM, Sigma), tetanus toxin (50 nM, Sigma), APV (50 mM, Tocris), or CNQX (50 mM, Tocris) were added to the culture media and re-administered every 24 h for 5 d. Slices were then fixed, imaged, and counted for fluorescently labeled cells.

Confocal Imaging and Immunohistochemistry
Experimental mice were sacrificed, perfused with 4% paraformaldehyde in phosphate buffered saline, dissected to remove intact brains, and post-fixed for 1 h at 4uC. Brain tissue was embedded in O.C.T and either sectioned to 12 mm on an upright Leica cryostat, or 50-100 mm slices were cut on a cooled stage microtome. Tissue sections were mounted on slides and imaged using an upright Zeiss 510 scanning confocal microscope (Carl Zeiss Inc.). For immunohistochemistry, sections were incubated with blocking solution (10% normal goat serum, 2% BSA, 0.1% Triton X-100 in PBS pH 7.4) and incubated at 4uC for 2 h. 3D fluorescent image reconstruction. To generate threedimensional images of RV labeled microcircuits in the olfactory bulb, LSM image files comprising 50-75 z-stack image planes from 150 mm cleared brain slices [61] spaced 1.5 mm apart were processed using Amira segmentation and volume rendering software (Visage Imaging, San Diego, CA). Image planes were captured at 206 full field magnification. Isolated presynaptic networks comprising identifiable single source granule cells and their presynaptic SACs were reconstructed. After tracing all EGFP expressing cell types and performing segmentation, EGFP labeled mitral cells and occasional gial cells were masked from the image reconstruction to offer better image resolution and measurement of SAC networks. Due to the uncertainty of origin for all of the fine neurites extending from the imaging field, measurements of SAC networks was constrained to clearly identifiable dendrites.
Analysis of dendritic protrusions. To quantify the number of protrusions we observed on the dendrites of postnatal born granule cells following odor enrichment, we sacrificed mice from odor stimulated and control groups, performed intracardial perfusion, then postfixed for 1 h in 4% paraformaldehyde in PBS. Brain tissue was cryoprotected in 30% sucrose/PBS and frozen in O.C.T. Thin sections (20 mm) were cut on a cryostat and mounted on Superfrost Plus slides. Using an upright Zeiss 510 confocal microscope, doubly labeled (granule source cell) dendrites were first identified extending from RV-infected granule cells under 406 magnification. Imaging and analysis for protrusion counts was performed in randomly sampled doubly labeled dendrite segments 25 mm in length within the internal EPL. Upper and lower boundaries of doubly labeled dendritic segments were identified, and multiple confocal image planes were collected 250 nm apart to generate individual 40-plane, 10 mm thick z-stack image files, so as to span the entire thickness of the dendrite. All counts of dendritic protrusions were performed blind to the experimental manipulation. The morphological criteria established for analysis was that protrusion length (axis perpendicular to dendritic shaft) was $ to protrusion width. This was determined in 3 dimensions by manually visualizing each plane of the serial z-stacks for each dendrite segment included in analysis. Data were reported as 6 SEM of spine number per 25 mm dendrite on granule cells in mice exposed to cycled odorants (odor) compared to non-odor exposed controls. p,0.001, Student's t-test. For display, representative images of dendritic segments used for protrusion counts were generated as maximal Zstack projections.
Analysis of synaptic puncta. All counts of gephyrin positive puncta were performed blind to the experimental condition. To quantify changes in the number of synaptic puncta following odor stimulation, we prepared virally-labeled olfactory bulb tissue as described above for spine analysis. Briefly, 20 mm sections were cut and mounted on slides. Using an upright Zeiss 510 confocal microscope, doubly labeled dendrites were identified extending from RV-infected granule cells under 406 magnification. Upper and lower boundaries of doubly labeled dendritic segments were identified, and multiple confocal image planes were collected 250 nm apart to generate individual 40-plane, 10 mm thick z-stack image files, so as to span the entire thickness of the dendrite.
Gephyrin-positive puncta were identified and included in our analysis only if they were clearly resolved within the doublylabeled granule source cells. This was determined by manually visualizing each plane of the serial z-stacks for each dendrite segment. Puncta were considered colocalized and counted if gephyrin staining was observed in $2 independent image planes through the doubly labeled dendrite. Gephyrin positive scaffolds were excluded from analysis if they were $3 mm in any dimension. Data were reported as 6 SEM of gephyrin positive puncta per 35 mm dendrite on granule cells in mice exposed to cycled odorants (odor) compared to non-odor exposed controls. p,0.02, Student's t-test. For display, representative images of dendritic segments used for gephyrin counts were generated as maximal Zstack projections.

Odor Enrichment
Odor enrichment was carried out using similar parameters as previously described [90]. For controlled odor delivery, a liquiddispensing robot (model 7200; I & J Fisnar, Fair Lawn, NJ), was programmed to cycle through 42 vials containing different odorant mixtures continuously for 4 weeks following electroporation. Odorants were presented for 5 s each, followed by a 1 min clean air exposure at a flow rate of 0.2 L/min. Odorants were diluted in mineral oil according to their individual vapor pressures to give a nominal headspace concentration of 100 ppm. Further flow diluted the odorants to a nominal final vapor phase concentration of ,10 ppm. For mixture stimuli, compounds were diluted to the same final delivery concentration.
Whole cell recordings. Short axon cells or source newborn granule cells were visually identified with IR-DIC optics and then targeted for recordings through either EGFP, or dual EGFP and tdTomato fluorescence, respectively. Patch pipettes were pulled from thick-walled borosilicate glass with open tip resistances of 2-7 MV and were filled with (in mM) 120 K-gluconate, 5 KCl, 2