All animal protocols were carried out in accordance with the Carnegie Mellon University Institutional Animal Care and Use Committee guidelines. One of two lentiviral constructs, developed and characterized previously (Komai et al., 2006), were injected into 2-month-old C57B/L6 mice (Taconic Farms). The first viral vector contained a ubiquitin promoter followed by EGFP the second viral vector contained a tandem of U6 promoters, each expressing one shRNA, followed by the ubiquitin promoter expressing EGFP (see Figure 1A). Lentiviral injections were be performed as previously described (Dittgen et al., 2004; Cetin et al., 2006). Briefly, mice were anesthetized by IP injection of a mixture of ketamine (100 mg/kg) and xylazine (10 mg/kg) in water. Animals were kept deeply anesthetized as assessed by monitoring eyelid reflex, vibrissae movements, and pinch withdrawal. Body temperature was maintained at 37°C with a heating pad with digitally monitored feedback control. Two craniotomies 300–500 μm in diameter were drilled above the sub ventricular zone (final coordinates relative to Bregma: anterior −1 mm, lateral ±1 mm, depth 2.2 mm beneath the pia mater). Pulled glass quartz pipettes were broken to an outer diameter of ≈30 μm and inserted at a 27° angle caudal–rostral into the pia. 500 nL of lentivirus was injected slowly over ≈5 min.

Mice were anesthetized by IP injection of a mixture of ketamine (100 mg/kg) and xylazine (10 mg/kg) in water. After animals became non-responsive as monitored by tail pinch their external naris was closed by cauterization. A cautery tool (Fine Science Tools), consisting of a fine loop of platinum wire that is rapidly heated by passage of electrical current, was placed against the nostril until the skin flap that covers the naris was “melted” onto the inner surface. Immediately following the cautery, the area was swabbed with antibacterial/anesthetic gel and the animal was returned to a recovery cage where it was monitored until recovery over the following 48 h. Extent of naris closure was assessed when animals were sacrificed. Only tissue from animals in which the occluded naris remained completely closed was used in immunohistochemistry and analysis.

Mice were anesthetized as above and perfused transcardially with cold 1% NaCl in 0.1 M phosphate buffer (PB) followed by cold 4% paraformaldehyde in 0.1 M PB. The brains were dissected and fixed in 4% paraformaldehyde in PB overnight at 4°C and then sunk in 30% sucrose at 4°C for 24 h. Each hemisphere of the brain was sliced coronally at 300 μm using an SM2000R cryostat (Leica Microsystems). Incubations were done at room temperature with oscillation. Sections were first incubated in 2% Triton X-100 (Sigma) and 2% normal donkey serum (NDS; Jackson Immuno Research Laboratories) in PB for 1 h. Rabbit anti-GFP antibody (1:1000, Invitrogen) was added in the presence of 2% NDS and 0.1% Triton X-100 in PB and incubated for 1 h. The secondary antibody added was a conjugated Alexa Fluor 488 (1:600, Invitrogen), incubated for 1 h in 2% NDS and 0.1% Triton X-100 in PB. After the addition of the secondary antibody all incubations were performed in the dark. Sections were washed three times between each incubation step in PB for 5 min.

Sections were imaged in their entirety using a Carl Zeiss LSM 510 Meta NLO multi-photon microscope (Carl Zeiss). Excitation of EGFP was provided by a Chameleon Ultra-II femtosecond laser (Coherent) operating at 890 nm. Non-descanned detection was provided by two silicon photomultiplier tubes situated on the back of the fluorescence light train of the Axio Observer Z1 inverted microscope stand. A long-pass 680 nm blocking filter was placed in front of the detection assembly to prevent any stray two-photon excitation. An individual band pass barrier filter (505–550 nm) was placed in front of the detector for fluorescence emission selection. All images were produced using a 40× 1.3 NA Neofluar air objective (Carl Zeiss). Three hundred micrometer thick sections from the first coronal slice directly in front of the accessory OB were imaged in their entirety through the Z-plane (see gray bar in Figure 1A). Excitation and gain were set to maximize the resolution of dendritic spines in the center of the Z direction. Z-steps for each section were set at 0.8 μm.

Granule cells to be traced were chosen at random from the center 30% of the Z-plane of each slice. The volume used for this randomization was a series of image stacks from the lateral to medial edge of the OB. Image stacks were stitched together with the Neurolucida software package (MBF Bioscience). Cell bodies within the center 30% of the Z direction were then numbered and a random number list was generated based on the total number of cell bodies in the volume. The first five granule cells, corresponding to the order generated by the random number list, were then traced that met the following criteria were traced: (1) Cell body located in the granule cell layer, between the internal plexiform layer and the subependymal zone; (2) Dendritic tree completely contained within the slice; (3) Cells sufficiently bright under fluorescence to allow for tracing of all dendritic spines and fine dendritic processes. Cells not meeting all criteria were skipped and replaced by the following randomly numbered cell in the sequence. On average, the first 21.36 (±11.07) neurons from the random number list provided five cells fulfilling all criteria for tracing.

All slides and images were coded and the analysis was performed with the experimenter blinded to the conditions of the animal and individual brain hemispheres during analysis. Cells were analyzed using the Neurolucida Explorer software package (MBF Bioscience) quantifying aspects of the apical dendrite, such as the location, number, and branching order of spines and dendritic segments, as well as the location of the various lamina of the bulb. Further analysis was done using both Neurolucida Explorer and custom functions written in Igor Pro (Wavemetrics). A Wilcoxon Rank-Sum test followed by a Bonferroni correction for multiple measures was used for all statistical tests.

To visualize ABNs in the OB, we injected lentiviruses encoding EGFP into the SVZ of adult mice. This virus infected and labeled neural progenitors, which migrate along the rostral migratory stream (RMS) to the OB and integrate as inhibitory granule cells. To measure the effects of manipulating activity on ABN morphology, we considered two groups of animals (Figures 1C,D). The first group [control/small interfering RNA (siRNA), 4 animals, 20 cells; Figure 1C] received an EGFP-encoding lentivirus in one hemisphere (Figure 1A), and in the opposite hemisphere a lentivirus encoding for EGFP and a siRNA (Figure 1B) that knocked down sodium channel expression (Komai et al., 2006). These animals were otherwise unmanipulated (no naris occlusion). The second group (control/NO, 3 animals, 21 cells; Figure 1D) received injections of a lentivirus encoding EGFP only in both hemispheres followed by occlusion of one external naris on the same day (NO). Animals were allowed to survive for 5 weeks post-virus injection – during which expression of sodium channels increases substantially in control cells (Carleton et al., 2003) – before sacrifice and perfusion (Figure 1E). Following perfusion, thick sections (300 μm) were cut and completely labeled ABNs were imaged in their entirety throughout the depth of the section (Figure 1F). Manual 3-D cellular reconstructions were performed from the acquired image stacks (Figures 1G and 2).

Upon arrival in the OB, ABNs mature into olfactory granule cells (Saghatelyan et al., 2005). Olfactory granule cells are inhibitory interneurons which form dendrodendritic connections with olfactory mitral and tufted cells (Rall et al., 1966) through the formation of dendritic spines (Price and Powell, 1970). From our cellular reconstructions we quantified the number of spines on apical dendrites of mature ABNs to measure the degree to which these cells were able to integrate into mature OB circuitry. Genetic reduction of cell intrinsic activity (siRNA knock-down of sodium channels) led to significant reductions in the total number of apical dendritic spines formed by integrating ABNs (control: 139.8 ± 11.7; siRNA: 106.5 ± 8.4; p = 0.03; n = 20; Figure 3A). Network-level reduction in activity (NO) resulted in a similar decrement in total spine number (control: 122.9 ± 8.0; NO: 92.9 ± 5.9; p = 0.003; n = 21; Figure 3A). The two control groups were not significantly different (p > 0.31).

We also determined the location along the dendrite where these reductions occurred by plotting the number of spines (spine count) vs. the distance from the soma along the dendritic tree (Figures 3B,C). In siRNA knock-down animals (Figure 3B) we saw significant differences in spine count between control and siRNA knock-down cells beginning as little as 40 μm from the soma (at 40 μm: control: 9.1 ± 0.8 spines/bin, siRNA: 6.9 ± 0.06 spines/bin; p < 0.05; at 80 μm: control: 12.7 ± 1.9 spines/bin, siRNA: 6.9 ± 0.8 spines/bin; p < 0.05; n = 20 cells). Proceeding distally, in siRNA knock-down cells this significant reduction was followed by continued sparseness, although spine counts were similar to control cells in distal segments. Conversely in NO cells, reductions in spine count did not localize to any specific region of the dendrite.

We next evaluated whether changes in dendrite morphology could explain the observed decreases in spine count by comparing differences in average spine densities per unit of dendritic length between control and activity-reduced cells (Figure 4A). We also plotted the distribution of spine density in spines/μm of dendrite along the dendritic tree for both control and activity-reduced cells (Figures 4B,C). This normalization controls for changes in spine density due to changes in dendritic arborization. Comparison of spine density again revealed a small but significant reduction by siRNA knock-down (control: 0.25 ± 0.01 spines/μm; siRNA: 0.21 ± 0.01 spines/μm; p < 0.001; n = 20) and NO (control: 0.23 ± 0.01 spines/μm; NO: 0.21 ± 0.01 spines/μm respectively; p = 0.039; n = 21; Figure 4A). In cells from control/siRNA animals, the spine density distribution shows that this reduction is observed throughout the first 200 μm of the somato-dendritic axis (Figure 4B), with large differences in spine density in proximal (at 40 μm from soma: control: 0.23 ± 0.02 spines/μm; siRNA: 0.16 ± 0.01 spines/μm, p < 0.001; at 80 μm from soma: control: 0.28 ± 0.04 spines/μm, siRNA: 0.16 ± 0.02 spines/μm, p < 0.01; n = 20) regions of the dendrite. There was no significant difference in spine density in distal dendritic regions of siRNA cells (Figure 4B). In control/NO animals, differences in spine density distribution (Figure 4C) were small and not significant.

To understand how ABNs integrating into the mature circuit might sample input from potential presynaptic partners, we compared how each of the two types of activity reduction affect the morphology and complexity of the apical dendritic trees of integrating ABNs independent of spine distribution (Figure 5). Extensive and elaborate ramification of dendritic trees into zones of potential synaptic input would suggest successful integration of ABNs. Initial comparisons of total dendritic length, total number of branches, average terminal branch distance, and average terminal branch order showed no significant differences in either group (Tables 1 and 2), though in both experimental groups across these measures there was a trend toward reduced dendritic complexity in the activity-reduced groups. While these measures of overall dendritic complexity did not vary significantly, we wanted to carefully investigate the possibility of morphological changes specific to those regions where we saw changes in spine count and spine density.

To this end we performed a 3-D Sholl analysis (Figure 5A) to measure the total dendritic length or number of intersections in spheres of increasing radii (Figure 5A1). These measures are reported as a function of sphere radius and are indicative of the degree of branching in neuronal processes (Sholl, 1953), where more complex processes result in more intersections with the “Sholl” sphere as distance increases. Sholl analysis of total dendritic length indicates siRNA knock-down cells have reduced dendritic length beginning in medial regions of the dendrite, showing a significant difference at a radius of 120 μm (control: 42.3 ± 7.1 μm; siRNA: 20.4 ± 1.0 μm; p < 0.01; n = 20; Figure 5A2), indicating that control cells contain more or longer branches than siRNA knock-down cells. Unlike siRNA knock-down cells, NO animals failed to exhibit significant differences in any region of the dendrite (Figure 5A3).

Reductions in the length of dendrites as assayed by the 3-D Sholl analysis may be explained by two scenarios: (1) dendrites in activity-reduced groups are less tortuous than those in the control group, or (2) there is a reduction in the number of branches formed by activity-reduced cells. Qualitatively, there did not appear to be significant differences in the relative tortuosity between control and activity-reduced cells of either group (Figure 2). Therefore we suspected that the differences observed in the distribution of length could be accounted for by differences in branch pattern, i.e., the reduction in activity resulted in a reduction of high-order branching. To test this, we created modified Sholl plots that showed the number of intersections as a function of the distance from the soma along the dendritic tree (Figure 5B1) to form a more accurate picture of the magnitude of branching of the dendritic tree. We observed that siRNA reduced branching at medial distances (at 120 μm: control: 2.5 ± 0.17 branches; siRNA: 1.9 ± 0.03 branches; p < 0.01; n = 20; Figure 5B2). Thus, an exaggerated dendritic tree of an siRNA cell would resemble a control cell stripped of branches at all but the most distal locations. Global activity reduction in NO cells showed no significant difference in branching in our modified Sholl plot (Figure 5B3).

To examine global differences in dendritic complexity between control and activity-reduced cells, we plotted the average branch order against the distance from the soma along the dendritic tree (Figure 5C1). Though differences in total branch number and average terminal branch order were insignificant for both activity reduced groups (Table 1), observing these subtle differences together with the distributions of both the standard (Figure 5A) and modified (Figure 5B) Sholl plots suggest differences in cumulative branch number. Consistent with this prediction, there appeared to be an overall reduction of branching in siRNA knock-down cells over much of the dendrite (Figure 5C2). Branch order was significantly reduced at 100 μm (control: 0.71 ± 0.23; siRNA: 0.14 ± 0.05; p < 0.05; n = 20) although branching recovered at the furthest distal segments (Figure 5C2). However, global activity reduction appeared to have no effect on the distribution of branch order (Figure 5C3). Coupled with the modified Sholl plot (Figure 5B), these data suggest that while global activity reduction does not compromise branch elaboration, it does cause reductions in branch length.