Axonal wiring of guanylate cyclase-D-expressing olfactory neurons is dependent on neuropilin 2 and semaphorin 3F

The main and accessory olfactory systems of the mouse detect a large variety of chemical stimuli from the external environment. Chemoreception is mediated by several types of receptor molecules that are expressed by specialized sensory neurons (Axel,2005; Buck, 2005; Mombaerts, 2004). The information gathered by the sensory neurons in the periphery is relayed to the brain through anatomical structures called glomeruli, which reside in the main and accessory olfactory bulbs.

In the past decade, many studies have concentrated on the elucidation of the mechanisms of sensory neuron differentiation and axonal wiring in the mouse. Most of the efforts have been on the two major cell types of the olfactory system, olfactory sensory neurons (OSNs), expressing odorant receptor (OR) genes, and vomeronasal sensory neurons (VSNs), expressing vomeronasal receptor (VR) genes. Axons of OSNs expressing the same OR project to a small subset of glomeruli at reproducible positions, and ORs are intimately involved in axonal wiring of OSNs(Mombaerts, 2006).

There are several other olfactory subsystems, which have received much less attention. The first of these is an island of olfactory nature, the septal organ, which is located within the respiratory epithelium on the ventral aspect of the posterior septum(Rodolfo-Masera, 1943). This structure contains neurons that express a subset of ORs shared with OSNs that reside in the ventral regions of the main olfactory epithelium (MOE) and that project their axons to the ventral parts of the main olfactory bulb (MOB). The OR expression profile of the septal organ has similarities and differences with the MOE (Tian and Ma,2004). The specific function of the septal organ has not been established; its neurons may also have mechanosensory abilities(Grosmaitre et al., 2007). Second, the Grueneberg ganglion, located at the tip of the nose(Grueneberg, 1973), is now regarded as an olfactory subsystem (Fuss et al., 2005; Koos and Fraser,2005; Fleischer et al.,2006; Roppolo et al.,2006). Its function is not known. Third, a family of G-protein-coupled receptors, the TAARs, is expressed in a subset of sensory neurons of the MOE (Liberles and Buck,2006). Fourth, several types of neurons in the MOE project their axons to the necklace area, which demarcates the main from the accessory olfactory bulb. Several markers distinguishes these neurons from OR-expressing OSNs or VR-expressing VSNs: placental antigen X-P2(Shinoda et al., 1989; Shinoda et al., 1993),antigens to antibodies 2C6 and mAb213(Ring et al., 1997), and guanylate cyclase-D (GC-D) (Fülle et al., 1995; Juilfs et al.,1997). Neurons from the Grueneberg ganglion also project to the necklace area (Fuss et al.,2005). Thus, the necklace area is heterogeneous in axonal input. Little is known about the function, morphology and development of the necklace area, or about the various types of cells that project their axons to it.

We have demonstrated a major role for neuropilin 2 (Nrp2) in VSN axonal wiring and glomerular formation in the accessory olfactory bulb (AOB), and a minor role in OSN axonal projections to the MOB(Walz et al., 2002). The necklace area is also disorganized in our Nrp2 mutant mice. Several of these phenotypes were observed in another strain of Nrp2 mutant mice (Cloutier et al., 2002). Neuropilins are co-receptors for class III semaphorins, and Nrp2 binds Sema3b,Sema3c and Sema3f (Chen et al.,1997; Kolodkin et al.,1997; Takahashi et al.,1998).

Here, we assess the contributions of individual semaphorin ligands to the phenotypes that are observed in the necklace area of Nrp2 mutant mice. We labeled a subset of necklace glomeruli by generating two mouse strains with targeted mutations in the GC-D (Gucy2d) locus that produce bicistronic messages resulting in co-translation of GC-D with an axonal marker, tau-β-galactosidase or tauGFP (tau is also known as Mapt). A striking phenotype of Nrp2 or Sema3f knockout mice is that axons of GC-D-positive (GC-D⁺) neurons, which are not known to express ORs, coalesce into homogeneous glomeruli that are scattered widely at ectopic locations within the glomerular layer of the MOB, in addition to glomeruli in normal positions. Our findings provide genetic evidence for a Nrp2-Sema3f interaction as a determinant of axonal wiring of GC-D⁺neurons, and also for axons of some VSNs.

A probe corresponding to the last exon of the ORF of the rat GC-D gene was used to screen a 129/SvJ mouse BAC library (Genome Systems, St Louis, MO,USA). From the resulting BAC clone 112J14, a HindIII fragment containing the 3′ end of the Gucy2d gene was subcloned into pBS-SKII (Stratagene, La Jolla, CA, USA). To shorten its length for the right arm, a re-ligation between the XhoI site within pBS-SKII at the 3′ end of the subcloned HindIII fragment and an internal XhoI was performed. Next, an EcoRI fragment was inserted at the 5′ end to lengthen the left arm, to create a targeting vector with a 2.8 kb left arm and 6 kb right arm. A PacI site was generated 3 bp after the stop codon in exon 19 of the ORF by PCR mutagenesis and placed into the targeting vector by NcoI-SacI subcloning. A PmeI site was generated at the 5′ end of the targeting vector for linearization. A cassette consisting of IRES-tauGFP or IRES-taulacZ, and a self-excising pgk-neomycin expression cassette flanked by loxP sites (ACNF)(Bozza et al., 2002) was placed into the newly generated PacI site. The green fluorescent protein(GFP) version was EGFP-1 (Clontech, Palo Alto, CA).

To construct targeting vectors, the left and right arms were amplified using the GeneAmp XL PCR kit (Applied Biosystems, Alameda, CA) with genomic DNA of E14 cells as template. The following oligos were used: Sema3bright arm forward: GCGCGTTAATTAAGGACCACGTAGTGCAGCTCACTGGTGA, reverse:ATATAGGCGCGCCTAGAGCCCAGCAGTACAACATGGCCACC; Sema3b left arm forward:GGCCGTTTAAACGCCTTACCACCCCCTGGCACTAAGCTC, reverse:TATAAGCGGCCGCTAAAGGAGAGCCGAAGGCGGGGAAGGTT; Sema3f right arm forward:GCGCGCTTAATTAACAGTGATCGGCGCCGTGAGGTGAGTT, reverse:ATATAGGCGCGCCAGGGGCTGGAGAGATGGTCAGTGG; Sema3f left arm forward:GGCCGTTTAAACGGCAGTGGTTCTCAACCTTCTTAGCTGCCAC, reverse:TATAAGCGGCCGCGGTACCGGTGGCCTTAAGTTCTGTGG. The arms, which were between 3.8 kb and 4.4 kb, were subcloned into a modified pBS-SKII vector and sequenced for PCR errors. A PmeI site was generated at the 5′ end of the targeting vector for linearization of the construct. ACNF was placed in an XbaI site between the left and right arms.

Genomic DNA from G418-resistant ES colonies of E14 was digested with BamHI for Gucy2d, HindIII for Sema3b and BamHI for Sema3f and analyzed by Southern blot hybridization with a 5′ external probe for Gucy2d, and 3′ external probes for Sema3b and Sema3f. DNA of positive clones was digested with HindIII or EcoRI for Sema3b and Sema3f, respectively, and probed with 5′ external probes to verify that no large-scale genomic reorganizations had occurred. ES cells were injected into C57BL/6J blastocysts. Mice were in a mixed 129 × C57BL/6J background. The strains will be available from The Jackson Laboratory, as follows: strain name B6;129P2-Gucy2d^tm1Mom/MomJ and #6703 for GCD-ITL; strain name B6;129P2-Gucy2d^tm2Mom/MomJ and#6704 for GCD-ITG; strain name B6;129P2-Sema3b^tm1Mom/MomJ and #6705 for ΔB; strain name B6;129P2-Sema3f^tm1Mom/MomJ and #6710 for ΔF. The GCD-ITG mice have been described, in part, previously(Hu et al., 2007).

The Sema3c gene has previously been mutated by us(Feiner et al., 2001). We have here studied the Sema3c^tm1.1Jra strain in a genetic background that allows for postnatal survival.

Experiments were conducted according to guidelines and with approval from the Institutional Animal Care and Use Committee of The Rockefeller University.

For immunohistochemistry, mice were deeply anaesthetized and intracardially perfused with 4% paraformaldehyde (pH 7.4), postfixed on ice for 3 hours and finally sunk in 30% sucrose overnight. They were then frozen in OCT and 16-18μm sections were cut on a cryostat. Sections were blocked in 10% normal donkey serum (NDS) for 1 hour at room temperature followed by incubation with primary antibodies in blocking solution at 4°C overnight, except for MAP2 stainings which were incubated for 2 days. The following day the sections were washed in 0.1% Triton X-100 in PBS and incubated with secondary antibody for 1 hour at room temperature. After final washes in 0.1% Triton X-100 in PBS,sections were stored in PBS until viewed. Primary antibodies used are as follows: rabbit polyclonal anti-PDE2A at 1:100 (FabGennix, Frisco, TX, USA),rabbit polyclonal anti-β-galactosidase at 1:500 (MP Cappel, Solon, OH,USA) and chicken polyclonal anti-β-galactosidase at 1:500 (Abcam,Cambridge, MA, USA), donkey polyclonal anti-OMP at 1:5000 (Wako Chemicals USA,Richmond, VA, USA), and rabbit polyclonal anti-MAP2 at 1:1000 (Abcam). Secondary antibodies were: Alexa Fluor 488-conjugated donkey anti-rabbit,Alexa Fluor 546-conjugated donkey anti-rabbit, donkey anti-goat and donkey anti-chicken (Invitrogen, Carlsbad, CA, USA), and Cy5 donkey anti-rabbit(Jackson ImmunoResearch Laboratories, West Grove, PA, USA) all at 1:500.

For β-galactosidase whole-mount analysis, tissues were processed as described before (Mombaerts et al.,1996), except that the fixation was carried out on ice for 5 minutes. X-Gal-stained whole-mount specimens were examined with a Zeiss SV11 stereomicroscope, and images were taken with a Zeiss AxioCam CCD camera. For tauGFP/taulacZ whole-mounts, mice were sacrificed by CO₂asphyxiation and the olfactory epithelium exposed. Staining procedures were performed as described before (Feinstein and Mombaerts, 2004). Whole-mounts of tauGFP, GFP sections and immunostained sections were examined with a Zeiss LSM 510 confocal microscope. Image files were processed with Adobe Photoshop 8.0.

Measurements of GC-D⁺ glomeruli were performed on whole-mount bulbs of GCD-ITL homozygous mice that were lightly stained (<1 hour exposure to X-Gal at room temperature). The short staining period was necessary to avoid staining in axons that could obscure smaller glomeruli. Counting of glomeruli was performed with a Zeiss SV11 stereomicroscope, and size measurements were done on images taken with a Zeiss AxioCam CCD camera. The longest axis of individual glomeruli were measured with Adobe Photoshop 8.0 and Canvas 6.0. Student's t-tests were used to determine the significance of changes between glomerular counts and sizes.

To label specifically the subset of necklace glomeruli that receive axonal input from GC-D⁺ neurons(Fülle et al., 1995; Juilfs et al., 1997), we engineered two strains of mice to co-express GC-D with an axonal marker,tau-β-galactosidase or tauGFP. The Gucy2d gene is located on chromosome 7E1 (Yang et al.,1996), spans ∼35 kb, and contains at least 22 exons in its open reading frame (ORF) and 5′ untranslated region(Fig. 1A). By homologous recombination in embryonic stem (ES) cells, we created targeted mutations in the Gucy2d locus that leave the coding region intact and produce bicistronic messages (Fig. 1B). The mouse strains are referred to as GCD-ITL and GCD-ITG.

In GCD-ITL mice, cell bodies that express β-galactosidase activity,and therefore can be stained blue with X-Gal, reside in a punctate pattern within the MOE, at a higher density towards the ventral aspects of the MOE(Fig. 2A)(Fülle et al., 1995). X-Gal⁺ cells are also present in the septum, including the septal organ (Fig. 2B), in which a small number of neurons are known to express GC-D(Ma et al., 2003). As described previously for GC-D⁺ neurons(Juilfs et al., 1997), most X-Gal⁺ cells reside within the recesses of the ectoturbinates as opposed to the endoturbinates.

GC-D⁺ cells display the typical morphology of neurons(Fig. 2C,D) with the exception of somewhat shorter cilia. All GFP⁺ axons in GCD-ITG mice are immunoreactive for PDE2A (Fig. 6A), consistent with the expression of PDE2A by GC-D⁺cells in the MOE (Juilfs et al.,1997). In mice that are doubly heterozygous for GCD-ITL and GCD-ITG, GC-D⁺ neurons co-express β-galactosidase and GFP,demonstrating bi-allelic expression at the Gucy2d locus(Fig. 2E).

GC-D⁺ cell bodies appear before birth, reach peak numbers within the first postnatal week, and decline in the following weeks to reach minimal counts during adulthood (data not shown). Conceivably this decline in labeled cell bodies - which is not seen in labeled axons - is due to transport of tauGFP or tau-β-galactosidase along the ever longer axons, thus depleting the cell bodies of marker protein. We have reported a similar age-related phenomenon for the axons of mitral and tufted cells(Walz et al., 2006).

Axons of GC-D⁺ neurons course through the lamina propria along with OSN axons, penetrate the skull through the cribriform plate, and turn towards the caudal aspect of the MOB along a ventral trajectory(Fig. 3A-D). GC-D⁺axons spread out to form several glomeruli, which are located superficially. Often, clustering occurs in arrangements of several tightly packed glomeruli(Fig. 3E,F). GC-D⁺glomeruli form a complete ring around the stalk of the MOB and around the anterior part of the AOB (Fig. 3A-D), thus placing them in the necklace area. GC-D⁺axons interconnect these glomeruli to form the typical necklace appearance of beads on a string. None of the GC-D⁺ glomeruli could be identified unequivocally as belonging to the modified glomerular complex, a structure that is poorly defined. Overall, no stereotyped pattern in the positions of GC-D⁺ glomeruli was apparent among bulbs(Fig. 3E) except that a greater number and the larger glomeruli are located in the ventrolateral aspects of the caudal MOB (Fig. 3B). One exception is the conserved location of a glomerular cluster at the ventral-most position on the lateral side of the MOB(Fig. 3E).

GC-D⁺ glomeruli are not detectable on postnatal day 1 (P1),although GC-D⁺ axons are present in the MOB (data not shown). GC-D⁺ glomeruli form within the first postnatal week. Glomeruli in adults can be divided roughly into three size classes, as measured along the longest surface dimension: small (40-90 μm), medium (100-160 μm) and large (180-300 μm) (Fig. 4A). The number of GC-D⁺ glomeruli of each size class increases during the first postnatal weeks to reach a maximum at 6 weeks (6 wk), and remains constant at 12 wk (Fig. 4B).

What are the mechanisms that underlie the wiring of axons of GC-D⁺ neurons into this unusual arrangement of necklace glomeruli?We have reported that a mutation in the neuropilin 2 gene (strain NP2-Δ)perturbs the projection of some VSN axons to the AOB, and also produces a general disorganization of the necklace area(Walz et al., 2002). Most GC-D⁺ neurons express neuropilin 2 (Nrp2; Fig. 6B). Nrp2 has been proposed as a co-receptor for several class III semaphorins, most notably Sema3b, Sema3c and Sema3f (Chen et al.,1997; Kolodkin et al.,1997; Takahashi et al.,1998). To investigate the individual contributions of semaphorins in the wiring defects of GC-D⁺ axons seen in Nrp2 mutant mice, we generated Sema3b and Sema3f knockout mice by gene targeting.

The Sema3f and Sema3b genes are located on chromosome 9F2 within 80 kb of each other (Fig. 1C). We created individual null mutations for both Sema3f(henceforth called ΔF) and Sema3b (henceforth called ΔB)by deleting 85% (ΔF) or 100% (ΔB) of the ORF(Fig. 1D). Compared with other targeted mutations of Sema3f(Sahay et al., 2003) and Sema3b (Julien et al.,2005), ΔF and ΔB are null mutations.

The ΔF and ΔB mouse strains are viable as homozygotes. ΔF homozygous mice display many of the gross defects we observed in NP2-Δhomozygous mice (Walz et al.,2002), including reduced size of the mice during the first few weeks after birth and a smaller number of homozygous mice than expected from a Mendelian distribution (n=34 for wild type, n=51 for heterozygous, n=14 for homozygous). ΔB homozygous mice appear normal and are born at Mendelian frequencies (n=28 for wild type, n=47 for heterozygous, n=21 for homozygous).ΔB/ΔF double mutants could not be generated because of the linkage of these genes in the genome.

To evaluate a role for neuropilin-semaphorin signaling in axonal wiring of GC-D⁺ neurons, GCD-ITL mice were crossed with NP2-Δ,ΔB, and ΔF mice. GC-D⁺ axons fan out over a far greater territory of the MOB in a NP2-Δ homozygous background than in wild-type mice, and glomeruli form in inappropriate locations(Fig. 5). In addition to GC-D⁺ glomeruli that are located in the necklace area, many GC-D⁺ glomeruli are scattered ectopically across the MOB up to the rostral tip. There are fewer GC-D⁺ glomeruli in the necklace area compared to controls, but the combined number of necklace and ectopic GC-D⁺ glomeruli in the NP2-Δ homozygous background is significantly greater compared to wild-type mice(Fig. 4C). Large glomeruli form almost exclusively in the necklace area and are equally frequent in mutant and wild-type mice. Similar numbers of medium-sized glomeruli are found in mutant and wild-type mice, but some glomeruli are ectopic in mutant mice. Significantly more small glomeruli form ectopically in mutant mice, accounting for most of the increase of total number of glomeruli in mutant mice.

The defects in the NP2-Δ homozygous background are phenocopied in theΔF homozygous background (Fig. 5). By contrast, ΔB homozygous mice(Fig. 5) and Sema3chomozygous mutant mice (data not shown) have a normal GC-D innervation pattern among the necklace glomeruli.

To further investigate the composition of ectopic GC-D⁺glomeruli in NP2-Δ homozygous mice, we tested whether the ectopic glomeruli receive axonal input solely from GC-D⁺ neurons, or whether they may have heterogeneous innervation by neurons from the MOE. By double labeling ectopic GC-D⁺ glomeruli in the MOB with antibodies against OMP, no GC-D^-/OMP⁺ areas were detected within these glomeruli, indicating that axons of other types of OMP⁺neurons do not contribute to these glomeruli, at least not in substantial numbers (Fig. 6C). Ectopic GC-D⁺ glomeruli were found to reside at appropriate, deeper,locations within the glomerular layer of the MOB, instead of at the superficial locations in the necklace area(Fig. 6A,C). Furthermore, when ectopic GC-D⁺ glomeruli were counterstained with antibodies against MAP-2 to visualize mitral and tufted cell dendrites projecting to glomeruli,these glomeruli were doubly labeled (Fig. 6D). Thus, the homogeneous innervation of ectopic glomeruli by axons of GC-D⁺ neurons and the projections of dendrites of mitral and tufted cells into these ectopic glomeruli, suggest that these glomeruli may represent new functional units.

The NP2-ΔΔ mutation also perturbs the projection of VSN axons to the AOB (Walz et al.,2002). To evaluate a role of Sema3b, Sema3c and Sema3f in axonal wiring of VSNs, the targeted mutations in the corresponding genes were introduced into NP2-Δ (Walz et al.,2002) heterozygous, or V1rb2-IRES-taulacZ(Rodriguez et al., 1999)homozygous backgrounds to visualize, respectively, the Nrpn2⁺axonal population emanating from apical VSNs (expressing the entire repertoire of V1r genes), and the axonal population of V1rb2⁺ VSNs(expressing the V1r gene V1rb2). The NP2-Δ mutation is used in the heterozygous state merely as a marker for apical VSNs, and in the homozygous state both as a marker for apical VSNs and to assess the consequences of loss of Nrp2.

In the ΔF homozygous background, the cell bodies of Nrp2⁺VSNs were still located within the apical layer of the VNO, and the numbers and distribution of V1rb2⁺ VSNs were normal (data not shown). However, the axonal projections of Nrp2⁺ and V1rb2⁺ VSNs exhibited defasciculation across the medial surface of the MOB(Fig. 7D), as we reported for NP2-Δ homozygous mice (Fig. 7C) (Walz et al.,2002). By contrast, axonal projection patterns appeared normal inΔB and Sema3c homozygous mutant backgrounds(Fig. 7A,B, and data not shown).

No miswiring of V1rb2⁺ axons into the MOB was observed in whole-mount preparations of ΔF homozygous mice(Fig. 7D) or in sections (data not shown). Most Nrp2⁺ axons and all V1rb2⁺ axons eventually reached the AOB (Fig. 7C,D). Within the anterior AOB (aAOB), Nrp2⁺ axons erroneously project into the posterior AOB (pAOB) in a ΔF homozygous background (Fig. 7H),comparable to the phenotype of NP2-Δ homozygous mice(Fig. 7G). Likewise,V1rb2⁺ axons form glomeruli within the pAOB in a ΔF homozygous background (Fig. 7I-L). Interestingly, misprojections can be observed in some mice with a ΔF heterozygous background (18.2%, n=11; Fig. 7L). By contrast, ΔB homozygous and Sema3c homozygous mutant mice do not show abnormal innervation patterns for Nrp2⁺ and V1rb2⁺ axons into the pAOB (Fig. 7A,F, and data not shown). Axonal overshooting of OSNs observed for NP2-Δ homozygous mice(Walz et al., 2002) occurs inΔF homozygous mice, but not in ΔB or Sema3c homozygous mutant mice (data not shown).

Thus, as for GC-D⁺ neurons, the defects in VSN axonal wiring caused by a deficiency in Nrp2 can be phenocopied by loss of one of its ligands, Sema3f, but not of Sema3b or Sema3c.

This study provides genetic evidence for a functional interaction between Nrp2 and Sema3f in axonal wiring of GC-D⁺ neurons that produces the unusual arrangement of necklace glomeruli: knockout mice for either Nrp2 or Sema3f have very similar phenotypes. We have hereby extended our approach of targeted integration of axonal marker genes to GC-D⁺ neurons, which project their axons to the necklace glomeruli,an area of the olfactory bulb that is poorly characterized. In Nrp2or Sema3f knockout mice, some axons of GC-D⁺ neurons,which are not known to express ORs, coalesce into glomeruli ectopically within the glomerular layer across the MOB. Some VSN axons are also dependent on Nrp2 and Sema3f for their wiring.

Much of the focus of research on the olfactory system of the mouse has been on OSNs and VSNs. There has been relatively little progress in our knowledge of other subsystems such as the necklace area, a ring of interconnected glomeruli that surrounds the caudal end of the MOB and the aAOB in mouse and rat (Juilfs et al., 1997; Ring et al., 1997; Shinoda et al., 1989; Shinoda et al., 1993; Walz et al., 2002; Zheng et al., 1987). Several markers can be used to distinguish the necklace glomeruli from MOB glomeruli,including the high intensity cholinesterase glomeruli also known as atypical glomeruli (Le Jeune and Jourdan,1993; Zheng et al.,1987), the placental antigen X-P2 immunoreactive glomeruli(Shinoda et al., 1989; Shinoda et al., 1993), the 2C6 and mAb213-positive glomeruli (Ring et al., 1997) and GC-D and PDE2-expressing glomeruli(Fülle et al., 1995; Juilfs et al., 1997). Glomeruli can be positive for one or more of these markers, as well as for the presence or absence of NCAM or OMP (Ring et al., 1997), thus defining multiple classes. They aggregate into 9-16 larger glomerular complexes of differently labeled glomeruli(Ring et al., 1997; Shinoda et al., 1993). The Grueneberg ganglion projects to the necklace area(Fuss et al., 2005), adding to the heterogeneity.

In our GC-D-tagged mouse strains, the number of GC-D⁺ glomeruli(∼40 per bulb) was higher than other types such as mAb213- and 2C6-positive glomeruli [∼20 per bulb and ∼3 per rat bulb, respectively(Ring et al., 1997)], atypical glomeruli [∼15-20 per mouse bulb(Weruaga et al., 2001)] and X-P2 positive glomeruli [∼9 per rat bulb(Shinoda et al., 1989)]. These discrepancies may result from differences in methods to visualize or define glomeruli; it may be easier to miss glomeruli in serial sections compared to whole mounts, which may explain our higher counts. As observed with other markers for necklace glomeruli (Le Jeune and Jourdan, 1991; Shinoda et al., 1993), GC-D⁺ glomeruli only appear well after birth and are not fully formed until 6 weeks, at the onset of sexual maturity.

GC-D belongs to a class of receptor GCs that contains at least seven members, GC-A to GC-G (Kuhn,2003). All neurons in the MOE express some membrane GCs(Moon et al., 1998), but GC-D is unique to a small subset of neurons. GC-D⁺ neurons lack components of the signal transducing mechanism found in conventional OSNs,such as G_olf, high affinity cAMP-specific phosphodiesterase,adenylyl cyclase III and the cyclic nucleotidegated channel subunit CNGA2(Juilfs et al., 1997; Meyer et al., 2000). GC-D⁺ neurons may rely on cGMP as a second messenger(Fülle et al., 1995; Juilfs et al., 1997). It is possible that GC-D⁺ neurons do not use an OR for detection of chemosensory stimuli; they are not known to express ORs. GC-D itself may function as the primary mechanism for stimulus reception in these neurons;alternatively, another class of chemosensory receptor(s) is expressed in GC-D⁺ neurons.

We have recently shown that GC-D⁺ neurons respond specifically to carbon dioxide (Hu et al.,2007), but GC-D itself has not yet been implicated in this response. It is possible that these neurons also respond to other chemicals.

GC-D⁺ necklace glomeruli appear more like glomeruli in the AOB(Belluscio et al., 1999; Rodriguez et al., 1999) than in the MOB, and the necklace area is in the proximity of the AOB. Many of the necklace glomeruli, including GC-D⁺ glomeruli, develop postnatally,which is consistent with a role for complex behaviors in adult mice. Atypical glomeruli show a sexually dimorphic distribution, further suggesting a function in intraspecies behavior (Weruaga et al., 2001). It is, therefore, unlikely that necklace glomeruli play a role in the suckling behavior as was previously suggested(Shinoda et al., 1993). The modified glomerular complex expressing high intensity cholinesterase, by contrast, appears early on and may still be important for suckling in newborns(Greer et al., 1982), but another study suggested otherwise (Risser and Slotnick, 1987). We could not determine a contribution of GC-D⁺ glomeruli to the modified glomerular complex, a problem that may be due to the loose definition of this structure.

The most probable explanation for the miswiring of many GC-D⁺axons into ectopic glomeruli across the MOB may lie in the expression of Sema3f over most of the surface of the MOB(Cloutier et al., 2002; Cloutier et al., 2004; Walz et al., 2002). In wild-type mice, Sema3f would deflect the ventrally arriving axons away from dorsal areas such that these axons coalesce into glomeruli in ventral areas;this effect would occur on Nrp2⁺, OR-expressing OSNs and the Nrp2⁺ GC-D⁺ neurons. In the absence of Nrp2 or Sema3f,GC-D⁺ axons would be free to navigate across ectopic areas of the MOB, and conventional OSNs show an overshooting phenotype(Walz et al., 2002). It cannot be ruled out that Sema3f induces Nrp2⁺ axons to undergo specific differential adhesion. It is more likely that Sema3f exerts its function as a repellent, causing Nrp2⁺ growth cones to steer towards areas with lowest possible Sema3f concentration via a localized collapsing activity similar to the one shown for collapsin(Fan and Raper, 1995).

Many glomeruli still form within the necklace area in Nrp2 and Sema3f mutant mice. It is not clear whether the same GC-D⁺glomeruli are affected in Nrp2 and Sema3f mutant mice. Moreover, it is not known whether a single GC-D⁺ neuron sends a single unbranched axon to one glomerulus in wild-type and mutant mice, or whether it has projections to several glomeruli. Some cells in the septal organ also express GC-D, and they may behave differently than GC-D⁺neurons of the MOE in the absence of Nrp2 or Sema3f. Neurons from the Grueneberg ganglion project axons to the necklace area but do not express GC-D, yet they are Nrp2⁺ (Fuss et al., 2005). These axons are unlikely to contribute to the GC-D⁺ necklace glomeruli, but could be affected in Nrp2- or Sema3f-deficient mice.

Further studies will address the in vitro effects of Sema3f on GC-D⁺ axons. It will also be interesting to cross the NP2-Δand ΔF mutations, to determine if the phenotype of the double mutants is the same as that of the single mutants.

Our Sema3f knockout phenocopies our Nrp2 knockout for both necklace and vomeronasal glomeruli, whereas mutants in Sema3band Sema3c do not show obvious changes. Comparing our findings with a previous study investigating the role of Sema3f in the innervation of apical VSNs to the aAOB (Cloutier et al.,2004), there are some discrepancies. Whereas Cloutier et al. also reported the defasciculation of VSN axons along the medial surface in Nrp2 or Sema3f mutant backgrounds, they could see misplaced axonal growth and glomerular formation within the pAOB only in Nrp2mutants but not in Sema3f mutants(Cloutier et al., 2004). Moreover, the misplaced glomerular innervation of VSN axons into the MOB(Cloutier et al., 2004) was not detected by us.

One explanation is a difference in the nature of the targeted mutation in Sema3f: the mutation used by Cloutier et al.(Cloutier et al., 2004) removes exon 1 and 4 kb of presumptive promoter sequence, whereas our Sema3fmutation removes exons 2-15, or 85% of the ORF. Another, more interesting,explanation may lie in the vomeronasal receptor type 1 gene (V1r)that is expressed by the VSNs studied: V1rb2 in our case, and V1ra1 in the case of Cloutier et al.(Cloutier et al., 2004). There may be heterogeneity among populations of VSNs that express a given V1r from the repertoire of >100 V1r genes in mouse(Zhang et al., 2007). V1ra1⁺ VSNs may respond to other Nrp2 ligands such as Sema3b or Sema3c instead of Sema3f. Since neuropilins need other partners to transduce ligand binding, such as plexins (Tamagnone et al., 2001), cell adhesion molecules(Castellani et al., 2000; Julien et al., 2005), MICALs(Terman et al., 2002) or NIP(Cai and Reed, 1999),V1ra1⁺ VSNs could express a different receptor complex than V1rb2⁺ VSNs and respond differently to various semaphorin ligands. The misprojections of V1ra1⁺ but not V1rb2⁺ axons into the MOB could be explained by such molecular differences. It is less clear why V1ra1⁺ axons do not misroute into the pAOB of Sema3fmutant mice. Other explanations involve untested members of the class III semaphorin family, such as Sema3e(Miyazaki et al., 1999), or the unknown negative regulation of one class III semaphorin on the action of another to render it an attractant, as is the case for Sema3b or Sema3c acting to reverse the repulsive action of Sema3a on Nrp1-expressing growth cones(Takahashi et al., 1998).

A role for Sema3b and Sema3c cannot be ruled out completely. In the case of the anterior commissure (AC), Sema3f first appeared to be the sole ligand responsible for the absence of an AC in Nrp2 mutant mice(Sahay et al., 2003). In a later study, however, it became clear that Sema3b has a somewhat overlapping yet mostly complementary role in establishing the positioning of the AC as well as causing the attraction or repulsion of various axonal tracts contributing to the AC; the loss of Sema3b induced the defasciculation of parts of the AC (Julien et al.,2005). By contrast, no real function has been established for Sema3c in the nervous system. Some studies suggest that Sema3c can influence hippocampal (Steup et al.,1999), cortical (Gonthier et al., 2006) and cerebellar(Moreno-Flores et al., 2003)neurites in vitro but no investigations in vivo have been reported. Also, the expression of the various plexins can profoundly affect the ability of any member of class III semaphorins to act as either an attractant or repellant(Yaron et al., 2005). Thus a change in the plexin expressed in either Nrp2 or semaphorin mutant background may have occurred and consequently altered axon behavior.

The authors thank Julie Miwa, Stefan Fuss, and Takeshi Imai for reading the manuscript and useful discussions. P.M. thanks Melissa Berry from The Jackson Laboratory for advice on nomenclature. Supported by NIDCD grants to A.W. and P.M.