Olfactory Sensory Neurons Control Dendritic Complexity of Mitral Cells via Notch Signaling

Mitral cells (MCs) of the mammalian olfactory bulb have a single primary dendrite extending into a single glomerulus, where they receive odor information from olfactory sensory neurons (OSNs). Molecular mechanisms for controlling dendritic arbors of MCs, which dynamically change during development, are largely unknown. Here we found that MCs displayed more complex dendritic morphologies in mouse mutants of Maml1, a crucial gene in Notch signaling. Similar phenotypes were observed by conditionally misexpressing a dominant negative form of MAML1 (dnMAML1) in MCs after their migration. Conversely, conditional misexpression of a constitutively active form of Notch reduced their dendritic complexity. Furthermore, the intracellular domain of Notch1 (NICD1) was localized to nuclei of MCs. These findings suggest that Notch signaling at embryonic stages is involved in the dendritic complexity of MCs. After the embryonic misexpression of dnMAML1, many MCs aberrantly extended dendrites to more than one glomerulus at postnatal stages, suggesting that Notch signaling is essential for proper formation of olfactory circuits. Moreover, dendrites in cultured MCs were shortened by Jag1-expressing cells. Finally, blocking the activity of Notch ligands in OSNs led to an increase in dendritic complexity as well as a decrease in NICD1 signals in MCs. These results demonstrate that the dendritic complexity of MCs is controlled by their presynaptic partners, OSNs.

Introduction more than one glomerulus at P9, suggesting that embryonic Notch signaling is essential for proper formation of olfactory circuits. Furthermore, NICD1 was observed in nuclei of MCs in embryos. Misexpression in OSNs of a dominant negative form of Mib1 (dnMib1), which inhibits the activity of Notch ligands, led to increased dendritic complexity as well as a decrease in NICD1 in nuclei in MCs.
We examined MCs by labeling them with DiI injection into the lateral olfactory tract (LOT), which contains axons of MCs. Strikingly, MCs exhibited more complex dendritic morphologies in Maml1 −/− and Maml1 +/− than Maml1 +/+ embryos ( Fig 1A). To determine dendritic complexity, 3D structures of dendrites were reconstructed using the FilamentTracer software (Fig 1B), and the number of branch points, total dendrite length and length of the longest dendrite were quantified (Fig 1C-1E). They were significantly increased in Maml1 +/− and Maml1 −/− embryos. Maml1 +/− embryos demonstrated an intermediate phenotype between Maml1 +/+ and Maml1 −/− embryos, indicating that the dendritic complexity of MCs is affected by the dosage of the Maml1 gene.
To learn which stage is critical for the mutant phenotypes, we examined embryos at E15.5, at which stage many MCs have already finished radial migration to the MC layer and extended multiple dendrites [13]. No significant differences in the number of branch points, total dendrite length or length of the longest dendrite were detected in Maml1-deficient embryos (Fig 2), suggesting that the E18.5 mutant phenotypes were caused after the migration and initial dendrite outgrowth of MCs.

Conditional misexpression in MCs
To modulate gene expression in MCs at specific stages, we used exo utero electroporation [23,24] at E11.5, which was effective for transfection into MCs (S2 Fig), and the Tet-controlled gene expression system [25]. After cotransfection with CAG-rtTA, which carries the reverse Tet-controlled transcriptional activator downstream of the ubiquitous promoter CAG, a gene downstream of the Tet-responsive element (TRE) was specifically induced in the presence of doxycycline (Dox) (S3 Fig). caNotch and dnMAML1 altered dendritic complexity in MCs To examine whether Maml1 and the Notch pathway in MCs were crucial for their dendritic structures, we conditionally misexpressed caNotch [23] or dnMAML1 [26] in MCs at E16.5 (Fig 3A), at which stage MCs had already settled in the MC layer [13]. MCs positive for Venus or enhanced green fluorescent protein (EGFP), which was coexpressed with caNotch or dnMAML1, were chosen for analyses. TRE-enhanced cyan fluorescent protein (ECFP) was cotransfected to label cells, because nuclear localization of dnMAML1-EGFP made it difficult to match dnMAML1-EGFP-positive cells with DiI-labeled dendrites. 3D structures of dendrites were reconstructed as above. The number of branch points and total dendrite length were smaller than those of DiI-labeled MCs in Fig 1, probably because DiI spreads in the plasma membrane and is more sensitive in detecting thin membranous protrusions than ECFP, which is a cytoplasmic protein. Compared to misexpression of ECFP alone as a control, caNotch drastically reduced the number of branch points and total dendrite length (Fig 3B-3E). Conversely, dnMAML1 significantly increased them. The length of the longest dendrite was changed by misexpression of dnMAML1 (Fig 3F). These findings indicate that the Notch pathway in MCs is responsible for their dendritic complexity after their migration. MCs that were in the dorsomedial OB and located at 300 to 700 μm from the rostral tip of the OB were used for the analyses. The phenotypes were indistinguishable among embryos of the same genotype. Box plots indicate the median, 25th/75th percentiles (box) and the data range. *P < 0.005, **P < 0.05. The number and length of dendrites were reduced by presentation of Jag1 in vitro To evaluate whether a Notch ligand was able to directly affect neurons, dissociated cells were cultured at low density from the OB at E15.5, at which stage all MCs and some tufted cells have been already generated [12][13][14]. The vast majority (95.28 ± 1.28%) of the cells (n = 3365) were neurons and mostly not in contact with other cells. Almost all (99.76 ± 0.40%) of the neurons were positive for Tbx21, a marker of MCs and tufted cells [27,28], and extended multiple MAP2-positive dendrites and a single Tau-1-positive axon at 6 days in vitro (DIV) (Fig 4Aa-4Ac). MCs and tufted cells are aligned in different layers of the OB but are indistinguishable in vitro. They were cocultured for two days with control Nalm-6 cells (Fig 4Ad-4Af) or Nalm-6 cells overexpressing Jag1 (Fig 4Ag-4Ai), which activate the Notch pathway in vitro [29], and Tbx21-positive neurons were analyzed using the FilamentTracer software. The number of dendrites, total dendrite length and length of the longest dendrite of the neurons treated with the Jag1-overexpressing cells were smaller than those treated with the control cells, suggesting that Jag1 directly affects neurons but not through other cell types, such as glia.
To examine whether the aberrant morphology of MCs involved embryonic Notch signaling in MCs, MCs were analyzed at P9 after misexpression of caNotch or dnMAML1 at E16.5 ( Fig  6). Whereas almost all (92.2 ± 2.8%) of MCs (n = 56 from 4 mice) normally extended a single primary dendrite into a single glomerulus after misexpression of ECFP alone as a control, many MCs (61.4 ± 7.3%, n = 72 from 5 mice) extended multiple dendrites into more than one glomerulus after dnMAML1 misexpression. On the other hand, many MCs (65.3 ± 5.3%, n = 69 from 4 mice) had a single primary dendrite that did not reach any glomeruli after caNotch misexpression, suggesting that excess Notch signaling is deleterious to MCs. These findings suggest that proper strength of embryonic Notch signaling is necessary for establishing the mature morphology of MCs.
Next we examined whether postnatal perturbation of Notch signaling affected MCs, by inducing dnMAML1 at P4, at which stage many MCs still extend multiple dendrites [13,15,30]. MCs were not changed by the postnatal misexpression (Fig 7). Nearly all (94.2 ± 6.3%) of MCs (n = 62 from 5 mice) had a single primary dendrite targeting into a single glomerulus after dnMAML1 misexpression, equivalently to ECFP misexpression as a control (96.0 ± 5.3%, n = 51 from 4 mice). These findings suggest that Notch signaling is involved in early but not later stages of dendritic development in MCs.
Moreover, strong NICD1 signals were observed in MCs at E18.5 but not E15.5 or P4 (S5 Fig), suggesting that the canonical Notch pathway is active in MCs at E18.5.

OSNs activated the canonical Notch pathway in MCs
To determine which cells provided a signal to activate the Notch pathway in MCs, we conditionally misexpressed dnMib1, which inhibits the function of all the canonical Notch ligands, Delta and Jagged [31,32]. Because the Notch pathway is often activated between neighboring cells [10], we first misexpressed dnMib1 in MCs, by cotransfecting TRE-dnMib1-EYFP (enhanced yellow fluorescent protein) with CAG-rtTA at E11.5 and inducing its expression at E16.5 (Fig 8A), as performed in Fig 3   that misexpressed dnMib1 normally extended dendrites at E18.5 (Fig 8B and 8C). The number of branch points, total dendrite length and length of the longest dendrite of MCs that misexpressed dnMib1 were not significantly different from those which misexpressed EYFP alone as a control (Fig 8D-8F), suggesting that Notch is not activated by ligands that are expressed in the same MCs.
MCs that did not misexpress dnMib1 and were adjacent to those which misexpressed dnMib1 were also not affected (Fig 9A and 9B). Their number of branch points, total dendrite length and length of the longest dendrite were not significantly changed (Fig 9C-9E). These findings suggest that Notch ligands that activate the Notch pathway in MCs are provided by cells other than MCs.
Many OSN axons reach the OB at E14.5 [13,33]. To examine whether OSNs activated Notch in MCs, the olfactory epithelium (OE) was electroporated at E11.5 (Fig 10A and 10B), at which stage many OSNs are generated [34], and dnMib1 was induced at E14.5 ( Fig 10C). Error bars indicate SD. The percentages of MCs extending dendrites into more than one glomerulus (blue bar) and MCs with a truncated primary dendrite (grey bar) were significantly increased by dnMAML1 and caNotch, respectively: control and caNotch, P < 0.05; control and dnMAML1, P < 0.05. doi:10.1371/journal.pgen.1006514.g006 dnMib1 misexpression in OSNs increased dendritic complexity in MCs, compared with EYFP alone as a control (Fig 10D and 10E). The number of branch points and total dendrite length were significantly increased (Fig 10F and 10G). These results suggest that OSNs provide a signal(s) that controls the dendritic complexity of MCs.
Furthermore, NICD1 signals of Tbx21-positive cells were significantly decreased by dnMib1 misexpression in OSNs (Fig 10I and 10J). There were some Tbx21-negative but NICD1-positive cells in the MC layer (Fig 10I), probably reflecting that some MCs are negative for Tbx21 at this stage [28]. NICD1 signals of Tbx21-negative cells in the MC layer appeared to be also decreased. These findings suggest that OSNs directly activate the canonical Notch pathway in MCs. Error bars indicate SD. No significant changes in the percentages of MCs extending dendrites into more than one glomerulus (blue bar) or MCs with a truncated primary dendrite (grey bar) were detected between control and dnMAML1 (P ranged from 0.240 to 0.893). doi:10.1371/journal.pgen.1006514.g007

Homing behavior of Maml1 mutants
To learn whether Maml1 mutants have defects in processing of olfactory information at P9, we examined homing behavior, in which mouse pups show preferences for nest-specific odors  Fig  1: control, n = 18 from 6 mice; dnMib1, n = 16 from 6 mice. The phenotypes were indistinguishable among embryos transfected with the same genes. Scale bar: 20 μm. There were no significant differences in the number of branch points, total dendrite length or length of the longest dendrite of MCs between control and dnMib1 (P ranged from 0.905 to 0.972).
doi:10.1371/journal.pgen.1006514.g008 [35,36]. Most Maml1 +/+ mice preferentially went to and stayed on the side containing wood chips from their home cage, after they were placed in the middle of a test cage in which unused chips were present on the opposite side (Fig 11). Many Maml1 +/− mice spent less time on the side containing their home cage chips, although their walking appeared to be normal. 5-cm walk was completed in 1.02 ± 0.12 s by Maml1 +/+ mice (n = 191) and in 1.03 ± 0.17 s by Maml1 +/− mice (n = 209). These findings suggest that some Maml1 +/− mice have defects in the processing of nest-specific odors.

Discussion
MCs change their dendritic morphologies during development (Fig 12; [13 ,15,30]). MCs extend multiple dendrites at embryonic stages. At early postnatal stages, such as P4, some dendrites start to form glomerular tufts at their termini, and three types of MCs, which have multiple dendrites (embryo type), a primary dendrite with tufts in more than one glomerulus (transit type) and a single primary dendrite with a tuft only in a single glomerulus (mature type), are observed. The control of dendrite forms in embryonic MCs has been completely overlooked, because their dendrites continue to grow at embryonic stages. The effects of the Notch pathway on their dendritic morphologies has not been addressed, although Notch1 has been shown to be expressed in MCs, periglomerular cells and granule cells in the adult OB, and has been implicated in olfactory information processing in adult mice by using CamKII-Cre-mediated Notch1 knockout mice [37]. Here, we report for the first time that the dendritic complexity of embryonic MCs is controlled by the canonical Notch pathway, which is activated by their presynaptic partners, OSNs, and that Notch signaling is crucial for the formation of mammalian olfactory circuits.
Drosophila olfactory circuits have been shown to involve Notch, which controls the diversification of OSNs [38] and volume of glomeruli [39]. The Drosophila Notch pathway, however, operates in OSNs [39] and their precursors [38]. Thus, the cells in which Notch works in Drosophila are on the presynaptic side, in clear contrast to the postsynaptic side in the mouse that we report here.
In our cell culture assay, the number of dendrites, total dendrite length and length of the longest dendrite of the Tbx21-positive neurons were significantly decreased by the Jag1-overexpressing cells in comparison with 6 DIV (Fig 4B-4D), suggesting that Notch signaling might be involved not only in dendritic growth inhibition but also pruning. All the neurons appeared to respond similarly to the Jag1-overexpressing cells, suggesting that Notch signaling may also control tufted cells, which have a single primary dendrite extending into a single glomerulus [40].
It has been shown that dendritic morphologies of MCs are normal until P3 but their dendritic pruning from P4 to P6 is slowed in Cnga2 knockout mice, although the phenotype may be linked with a secondary effect by reduced body size [15]. The phenotype and the stages observed in the knockout mice are different from those that we found here by the perturbations of the Notch pathway. No effect of the postnatal perturbation of Notch signaling on MCs suggests that there may be at least two stages, Notch-dependent early and Notch-independent later stages, in the dendritic development of MCs. At P9, many MCs did not show the mature morphology in Maml1 +/− mice, raising the possibility that maturation of MCs may be delayed in the mutants. Even at P35, however, significant number (7.9 ± 0.7%) of MCs (n = 38 cells from 3 mice) had a primary dendrite with tufts in two glomeruli in Maml1 +/− mice, whereas no MCs did in Maml1 +/+ mice (n = 41 cells from 4 mice) (S6 Fig). This finding may imply that the P9 mutant phenotype was not simply caused by delayed maturation of MCs. The Notch- dependent early stage may be necessary for reducing dendritic complexity so that the mature morphology of MCs could be properly shaped at the Notch-independent later stage.
The effects of Notch on dendrites in the cerebral cortex have been complicated. Notch has been reported to have a positive effect on the length and complexity of dendrites in vivo [41] and a negative effect on the length of neurites, probably containing both axons and dendrites, in vitro [7,8]. The complexity of neurites has been affected by Notch negatively [7] and positively [8]. These discrepancies may be explained by the pleiotropic functions of Notch and numerous neuronal types, which have not been examined, and by the non-canonical Notch pathway, which has been shown to control the volume of glomeruli in Drosophila [39]. The distinction of neuronal types and stages in which Notch works may be necessary. It remains to be determined if the Notch-dependent early stage is unique to MCs or used in other neuronal types.

Materials and Methods Animals
All experimental procedures were approved by the Animal Care and Use Committee (Chiba University) and conducted in accordance with the Guidelines for Use of Laboratory Animals (Japan Neuroscience Society). ICR and C57BL/6J mice were purchased from CLEA, Japan. Generation of Maml1-deficient mice has been previously described [20,42]. Mutant mice were backcrossed to C57BL/6J mice at least 10 times before use in the experiments. The noon of the day when a vaginal plug was detected was designated E0.5. The day of birth was designated P0.

Size analysis of the brain, haematoxylin-eosin (HE) staining and DiI labeling
Dissected brains were placed horizontally, and the lengths and widths of the OB and cerebral cortex were measured along and perpendicularly to the longitudinal fissure, respectively. For HE staining, the brains were fixed in 6% formalin neutral buffer solution (pH 7.4), embedded in paraffin and sectioned sagittally at 6 μm. The sections were de-waxed, re-hydrated with decreasing concentrations of ethanol and stained with HE. For DiI labeling, the brains were fixed in 4% paraformaldehyde in phosphate-buffered saline (PBS), and a small crystal of DiI (Life Technologies) was placed superficially into the LOT. They were incubated in PBS containing 1 μM leupeptin (Roche) at 37˚C, embedded in 3% low-melting-point agarose (Lonza) in PBS and coronally sectioned at 100 μm using a vibratome (Dosaka EM). Plasmids pCAG-EYFP [23] and pCAG-rtTA [25] carry EYFP and rtTA, respectively, downstream of the CAG promoter. pTRE-mCherry was purchased from Clontech. pTRE-mCherry, pTRE-EYFP [25], pTRE-ECFP and pTRE-dnMAML1-EGFP carry mCherry, EYFP, ECFP [23] and dnMAML1-EGFP [26], respectively, downstream of TRE. pTRE-caNotch-Venus carries caNotch, which lacks the coding region of the extracellular domain of Notch1 [23], an internal ribosomal binding site and Venus [43] downstream of TRE. pTRE-dnMib1-EYFP carries dnMib1, which encodes only the N-terminal 767 amino acids but not the C-terminal RING fingers of the mouse Mib1 and acts dominant-negatively [31,44], fused with EYFP in frame, downstream of TRE.

In vivo electroporation
To transfect DNA into a limited portion of the nervous system, we used exo utero electroporation (S2A Fig), in which electrodes are exactly placed over a targeted site of the clearly visible embryo in the yolk sac. E11.5 was chosen, because electroporation at this stage was the most effective for transfection into MCs and OSNs. E11.5 ICR mouse embryos were electroporated as described [23,24], with slight modifications. In brief, a pregnant mouse was anesthetized with intraperitoneal injection of the solution (20 μl per g body weight) that contained 15 μg/ml medetomidine (Kyoritsu Seiyaku), 160 μg/ml midazolam (Astellas Pharma) and 250 μg/ml butorphanol tartrate (Meiji Seika Pharma) in saline. For misexpression in MCs and OSNs, DNA was injected into the rostral portion of the telencephalon and the nasal cavity using a microinjection needle ("m" in S2A Fig and Fig 10A), respectively. Then, five square pulses of 50-ms duration with a 950-ms interval at 22 V were delivered by placing the electrodes as illustrated in the figures. The concentrations of pCAG-rtTA and pTRE-vectors were determined for efficient induction in the presence of Dox and for repression in the absence of Dox as described previously [25]. Dox (Clontech) was administered to pregnant mice via drinking water containing 2 mg/ ml Dox. For P4 pups, Dox was intraperitoneally injected at the dose of 100 μg per 30 g of body weight [25].

Cell culture
Dissociated neurons were prepared from the OB of ICR mouse embryos and cultured as described previously [30] with slight modifications. In brief, after trypsinization and gentle trituration of the OB, cells were plated at a density of 3.0 × 10 3 cells/cm 2 onto polyethylenimine-coated glass coverslips [45]. Half of the media was replaced at 1 DIV and 4 DIV. At 6 DIV, Nalm-6 human B-cell acute lymphoblastic leukemia cells overexpressing Jag1 or control cells that were X-irradiated with 600 rad [29] were added at a density of 2.5 × 10 5 cells/cm 2 . Cells on the coverslips were fixed at 8 DIV.

Image analysis and quantification
Large views of the brains were obtained with a fluorescence stereomicroscope (MZ16F, Leica). Images of HE-stained sections and cell cultures were obtained with a microscope (BX60, Olympus). Other fluorescent images were acquired on the LSM5 Exciter confocal microscope (Zeiss).
Because MCs are not generated synchronously [14], we focused on MCs in the middle portion along the anteroposterior axis of the dorsomedial OB, which are mainly born at earlier stages [14], although the same phenotypes were observed throughout the OB. 3D reconstruction and dendrite analyses were performed using the AutoPath and AutoDepth modes of IMARIS FilamentTracer 7.4 (Bitplane).
To quantify NICD1 signals in the MC layer, mean fluorescence intensity (FI) of Alexa Fluor 633 (reacting with anti-NICD1) in Tbx21-positive-nuclei was measured using the Histo tool in the ZEN software (Zeiss) with the confocal setting in which the pixel intensity of the brightest sample was below saturation. As internal controls, NICD1-positive Tbx21-negative cells below the MC layer that may be migrating neurons as described previously [47,48] were used. The mean background FI was obtained by measuring Alexa Fluor 633-negative areas on the same section. The relative FI of NICD1 of each Tbx21-positive cell was defined as 100 times the ratio of the background-subtracted mean Alexa Fluor 633 intensity of the cell to the average of the background-subtracted mean Alexa Fluor 633 intensities of the internal controls on the same sections. The data in Fig 10 were obtained from three brains each for misexpresion of EYFP (control) and dnMib1.

Behavioral test
We have developed a homing behavior test by modifying unidirectional homing tests [35,36]. After mating of a Maml1 +/+ C57BL/6J female with a Maml1 +/− C57BL/6J male, a pregnant female was housed individually. At P5, the dam and its litter were transferred to a clean cage containing unused wood chips. The littermates did not undergo any previous behavioral tests. At P9, the naive littermates and wood chips of their home cage were used. A pup was placed with its body axis parallel to the demarcation lines in the middle of a disposable 20.0-cm × 13.5-cm plastic cage (ICM) on a sheet of paper demarcated into three areas. The areas where plastic dishes containing 0.2 g of wood chips from the home cage and unused chips (as a control) were placed were designated H and C, respectively. The M area was between H and C. The cumulative time spent by the pup's snout in each area was measured in a 3 min test. Each pup was tested only once. The plastic cage was replaced with a new one after every test. The ANY-maze video tracking system (Stoelting) was also used to analyze the pup's movement as shown in Fig 11B. Walking was evaluated by measuring the time taken to walk 5 cm. After the test, the genotypes of pups were determined by PCR.

Statistics
Data in the text are mean ± SD. P-values were determined using Mann-Whitney U test. (B, C) Ventral view of the brain (B) and a coronal section of the OB immunostained for EYFP and Tbx21 (C), 7 days after electroporation of EYFP at E11.5. An arrow indicates the LOT positive for EYFP. Similar immunostaining patterns were reproducibly observed in all examined embryos (n = 3 mice). The OB and LOT were strongly positive for EYFP. Most EYFP-positive cells were aligned in the MC layer and were positive for Tbx21, as previously described by Imamura and Greer [47], who performed in utero electroporation. Scale bars: 0.5 mm in B; 10 μm in C.  A, B, C). MCs and their postmitotic precursors have been shown to be positive for Tbr1 [47]. NICD1-positive cells, which may be migrating neurons as described previously [47,48], were observed below the MC layer at E15.5 and E18.5 (arrows). The ventricle is to the bottom. Strong NICD1 signals were observed in the cerebral cortex of the same P4 brain on the same slide glass (the bottom panel in C). Similar immunostaining patterns were reproducibly observed in all examined embryos (n = 3 mice). Scale bar: 20 μm.