Proliferative and transcriptional identity of distinct classes of neural precursors in the mammalian olfactory epithelium

The olfactory epithelium (OE) differentiates from the lateral surface ectoderm of the vertebrate head, and gives rise to three major neuron classes: olfactory receptor neurons (ORNs), vomeronasal receptor neurons (VRNs) and gonadotropin releasing hormone (GnRH) neurons. ORNs and VRNs detect and relay information about a vast range of chemicals essential for feeding, reproduction and social interactions (for reviews, see Ache and Young, 2005; Buck, 2000; Dulac and Torello, 2003). GnRH neurons migrate from the OE to the hypothalamus (Schwanzel-Fukuda and Pfaff, 1989; Wray et al., 1989), where they regulate reproduction (Foster et al., 2006). Despite the functional importance of ORNs, VRNs and GnRH neurons, the identity of OE precursors remains uncertain. Furthermore, it is not know how local inductive signals or transcription factors distinguish proliferative and neurogenic capacities. Solving these longstanding mysteries not only provides insight into developmental specification of these essential yet inadequately characterized precursors, it addresses a central issue in regeneration: embryonic OE precursors must give rise to adult counterparts that generate ORNs and VRNs throughout life (reviewed by Schwob, 2002). Thus, we determined cellular and molecular mechanisms that define identity, proliferation and neurogenic potential of the earliest OE neural precursors.

OE neuronal precursors, such as those in other nervous system regions (for reviews, see Gotz and Huttner, 2005; Jessell and Sanes, 2000; Livesey and Cepko, 2001), should be found in discrete locations, have characteristic gene expression, distinct proliferative kinetics and the unique capacity to generate all differentiated OE neuron classes. Cells expressing neurogenic bHLH transcription factors, particularly Ascl1, have been proposed as OE precursors (Beites et al., 2005; Cau et al., 1997); however, persistence of ORNs despite loss of Ascl1 function (Guillemot et al., 1993) indicates that other precursors can generate OE neurons. Accordingly, we determined relationships between additional transcription factors, and cells with distinct proliferative or differentiation capacity among the earliest OE precursors. We found substantial differences in OE precursor identity at midgestation: slowly dividing, self-renewing precursors expressing high levels of Meis genes populate primarily the lateral OE, whereas rapidly dividing neurogenic precursors expressing high levels of Sox2 and Ascl1 reside mostly in the medial OE. These identities are established in part by Fgf8, and a transcriptional network involving Sox2 dose, Meis1 activity and Ascl1 expression that regulates progression from multipotent precursor to transit amplifying neuronal progenitor to post-mitotic neuron. Our findings suggest that among primarily lateral, Meis-expressing OE precursors are stem cells whose presence guarantees initial genesis of ORNs, VRNs and GnRH neurons.

Mouse embryos were harvested from timed-pregnant mothers (plug day=0.5) maintained by the Department of Laboratory Animal Medicine at the University of North Carolina at Chapel Hill or Children's Hospital (Boston, MA, USA). The Sox2 indicator (Ellis et al., 2004) was bred into CF-1 females from Sox2^eGFP/+ males. Fgf8^neo/neo (Meyers et al., 1998) and Ascl1^–/– embryos (Guillemot et al., 1993) were generated from Fgf8^neo/+ or Ascl1^+/– males and females, respectively. Sox2^hyp/– embryos were generated from Sox2^eGFP/+ males and Sox2^hyp/+ females (Taranova et al., 2006). Dams were killed by rapid cervical dislocation, and embryos were collected and genotyped using appropriate PCR primers. Institutional Animal Care and Use Committees (IACUC) at UNC-CH and CHB approved all procedures.

Embryos were fixed with 4% paraformaldehyde, embedded and cryosectioned using standard methods. Primary antibodies were obtained commercially [NCAM (Chemicon/Millipore), PH3 (Chemicon/Millipore), TuJ1 (Babco), OMP (Wako), Pbx1/2/3, ACIIII (Santa Cruz), BrdU (Becton-Dickinson), IdU (Accurate Chemical and Scientific) and Ascl1 (Becton-Dickinson)] or as gifts [Meis1 and Meis2 (A. Buchberg, Thomas Jefferson University), and GnRH (S. Wray, NINDS)]. The Sox2 antiserum was produced by L. Pevny's laboratory, and TrpC2 antiserum by C. Dulac's laboratory. Images were obtained using a Leica DMR epifluorescence or Zeiss LSM510 laser-scanning confocal microscope.

We estimated cell cycle times using dual DNA synthesis labeling (Martynoga et al., 2005). Iodinated deoxyuridine (IdU) was injected initially (T0), intraperitoneally (i.p.; 70 mg/kg body weight) in pregnant dams followed by bromodeoxyuridine (BrdU; 50 mg/kg) 1.5 hours later (T1). After an additional 0.5 hours (T2), embryos are fixed for IdU/BrdU histochemistry. The mouse anti-BrdU antibody BrdU detects both IdU and BrdU; however, the rat anti-BrdU antibody is specific. Thus, cells remaining in S-phase during the 2-hour period are double-labeled; IdU-labeled cells exit the cell cycle. S-phase (TS) and total cell cycle length (TC) is calculated as: TS=1.5/(number IdU labeled/number double-labeled cells), TC=TS/(number double-labeled/number all cells – identified by nuclear staining). We divided each OE section into ten sectors representing equivalent parts of its total length, and calculated TS and TC for each sector in a full series (7-10 sections) from five E11.5 embryos. Statistical analysis was performed using analysis of variance (ANOVA) followed by Tukey's honestly significant difference test.

BrdU was injected i.p. at E9, E10 or E11, followed by 2-hour (DNA synthesis) or 5.5- to 6.5-day (birthdating) survival. For label retention, we adapted a long-term BrdU labeling protocol (Morshead et al., 1994); BrdU (50 mg/kg) was injected i.p. in pregnant dams at E9, with a second injection 4 hours later. Upon first injection, we provided 1 mg/ml BrdU as drinking water, and left this in place until E11.5. Thereafter, pregnancy continued, with no further BrdU exposure, until E16.5 when fetuses were collected for BrdU histochemistry.

The lateral and medial OE was microdissected from entire E11.5 litters (n=4 independent experiments/4 litters), and dissociated as described (Shen et al., 2002). Dissociated cells were plated at clonal density on poly-D lysine coated Terasaki plates. Some cell suspensions received Fgf8 (100 nM) prior to plating. Cultures were fixed after 20 hours, co-labeled for TuJ1 with Meis1 or Sox2, as well as bisbenzimide to identify nuclei. `Paired' bisbenzimide-labeled nuclei were identified in individual wells, and expression of neuronal and transcription factor markers was visualized and scored.

E11.5 lateral nasal processes (LNPs) were micro-dissected, and up to six LNPs at a time were transferred to an electroporation chamber (Protech International, San Antonio, TX) with 50 μl of 3 mg/ml endotoxin-free control or Sox2, Meis1 or Meis1/Pbx1 DNA plasmids (Sigma) in HBSS (Invitrogen). LNP luminal (OE) surfaces were electroporated with five 50 ms pulses at 30 V separated by a 950 ms inter-pulse delay using an ECM 830 electroporator (BTX). LNPs were transferred to membrane filters (luminal surface up), grown 72 hours in vitro, fixed and immunostained (Tucker et al., 2006). Images of five non-overlapping fields were collected at 63× from each explant with a Zeiss510 laser-scanning confocal microscope. Double-labeled cell frequencies were compared between conditions by a two-tailed unpaired Student's t-test.

At E11.5, shortly after the OE invaginates, graded expression of several transcription factors defines OE domains. Sox2, an essential SoxB1 factor in the OE (Donner et al., 2007), recognized with an antibody (Fig. 1A) or Sox2-eGFP indicator transgene (Fig. 1E) (Ellis et al., 2004) is expressed at low levels in the ventrolateral OE and progressively higher levels in the dorsomedial OE. The proneural basic helix-loop-helix transcription factor Ascl1 (formerly Mash1) is also expressed in a subset of dorsomedial OE cells where Sox2 levels are higher (Fig. 1B). Conversely, Meis1 (Toresson et al., 2000) (Fig. 1C), Meis2 (not shown) and Pbx1/2/3 (Fig. 1D) are concentrated primarily in the lateral OE (see also Fig. 2). Graded expression patterns of Sox2, Meis1 and Ascl1 appear related: Sox2 expression is typically lowest where Meis1 is highest (Fig. 1E). Ascl1 is seen in regions where Meis1 expression is diminished, and Sox2 expression enhanced (Fig. 1F). Finally, in the ventromedial OE, both Meis1 and Sox2 levels are elevated, and Ascl1 and Pbx are not seen (Fig. 1A-D). Thus, graded transcription factor expression defines a primarily lateral OE domain where Meis1 is elevated, Sox2 and Pbx1 expression varies, and Ascl1 is absent, a primarily medial domain where Ascl1 and Sox2 are elevated, Meis1 varies and Pbx1 is absent, and a ventro-medial domain where both Meis1 and Sox2 are elevated, and Ascl1 is not seen.

Neuronal markers identify cells with morphological and molecular characteristics of ORNs, VRNs and GnRH neurons in the nascent OE (Fig. 1G-K). Presumptive ORNs and VRNs, labeled with neuronal markers such as class III β-tubulin (TuJ1) (Fig. 1G,H) and NCAM (data not shown) have characteristic bipolar morphologies, including an immature apical dendritic knob and a single axon that extends from the OE through the mesenchyme (Fig. 1H and inset). VRNs are distinguished by co-expression of the TrpC2 channel, which is functionally associated with this subclass of chemosensory neurons (Stowers et al., 2002) primarily in the presumptive vomeronasal organ (pVNO) (Fig. 1I and inset). Finally, GnRH neurons are found in the pVNO epithelium (Fig. 1J) and in the adjacent mesenchyme (Fig. 1K), migrating toward the basal forebrain. Neuronal labeling begins in the dorsolateral OE (Fig. 1G), where Sox2/Ascl1 levels increase and Meis/Pbx levels decline (Fig. 1E,F, sector c), and ends in the ventromedial OE, where Sox2 is elevated (Fig. 1A, left of asterisk), Ascl1 is absent and Meis levels rise. Therefore, at E11.5, three major classes of neurons are concentrated around the medial OE, with presumed ORNs in dorsolateral through dorsomedial regions, VRNs most frequent in medial to ventromedial domains and GnRH neurons primarily in the ventromedial OE.

We next asked whether molecularly defined OE domains include precursors with distinct proliferative capacities. At E11.5, M-phase cells, recognized by phospho-histone 3 (PH3), are most frequent in the medial OE (Fig. 2A,B). PH3-labeled nuclei appear concentrated at the apical surface, with no apparent mediolateral differences in apical/basal distribution. Cells in the lateral OE are more likely to be in S-phase, based upon frequency of acutely BrdU-labeled cells (Fig. 2C,D). Most S-phase cells in the lateral OE also express Meis1 at high levels; however, few of the scattered medial OE cells that express low levels of Meis1 co-label with BrdU (Fig. 2C,D; see also Fig. 1C). The inverse relationship between S- and M-phase precursors suggests that cell cycle lengths may differ in register with molecular distinctions (Fig. 2E,F). Precursors in the ventromedial OE (particularly sector 1, Fig. 2E) that express Meis1, as well as high levels of Sox2 (Fig. 1A,C), have longer cell cycles and lower mitotic frequency, as do other Meis1-expressing lateral precursors (especially including sectors 2-4) in which Sox2 levels decline and Pbx is seen (Fig. 2F). In precursors in the remaining OE sectors, where Sox2 expression is more robust, Ascl1 is seen, the neurons are concentrated and have shorter cell cycle times. Apparently, Meis1 levels may have a more substantial influence on cell cycle length than Sox2, Pbx or Ascl1 levels.

S-phase appears constant across the OE (Fig. 2E,F), suggesting longer cell cycles in ventral and lateral OE precursors reflect longer G1-S transitions. Therefore, we asked whether lateral OE cells are more likely to retain BrdU for long periods, a signal that is characteristic of slowly dividing, multipotent neural stem cells (Morshead et al., 1994). Following chronic BrdU exposure (E9-E11.5) and additional 5-day survival (E16.5), a significant proportion of Meis1-labeled (Fig. 2G,I-K) as well as Sox2-labeled (Fig. 2H,L-N) basal cells in the lateral OE are also heavily labeled with BrdU (29% Meis1 cells/87 total cells counted; 22% Sox2 cells/18 total cells counted; n=4 animals). We found no medial Sox2 or Ascl1 cells heavily BrdU-labeled (Fig. 2O,P); however, a small number were faintly labeled (Fig. 2O, inset; 4% Sox2 cells/104 total cells counted; 2% of Ascl1 cells/83 total cells counted; n=4 animals). By contrast, a significant proportion of OMP-labeled ORNs in the medial OE are heavily BrdU labeled (54% OMP cells/31 total cells counted; n=4 animals). Thus, based upon frequency, kinetics and distribution of M- and S-phase markers, lateral and medial OE domains contain two distinct proliferative populations – one slowly dividing and another rapidly dividing – in register with differing levels of Meis1.

Meis1-positive OE precursors, concentrated in the lateral OE, may divide slowly and symmetrically to maintain multipotent precursors while the remaining precursors, primarily those in the medial OE, divide rapidly and terminally to expand the neuronal population. Accordingly, we determined the mode and consequences of single cell divisions from E11.5 micro-dissected lateral and medial OE using a pair cell assay (Shen et al., 2002). Pairs (two progeny from single OE cells after 20 hours in vitro) were classified as neuron-neuron (N-N; Fig. 3A, top), neuron-precursor (N-P; Fig. 3A, bottom) or precursor-precursor (P-P; Fig. 3B,C) based upon co-labeling for TuJ1, Sox2 and bisbenzamide or for TuJ1, Meis1 and bisbenzamide. TuJ1-positive cells were counted as neurons; Sox2- or Meis1-positive/TuJ1-negative cells were counted as precursors. Pairs stained with bisbenzimide only – unrecognized OE precursors, particularly those for non-neuronal or vascular cells, or mesenchymal cells that remain after incomplete removal of OE-adherent mesenchyme – were designated as precursor^*-precursor^* (P^*-P^*; Fig. 3D) and counted separately. The absence of Sox2 labeling in these P^*-P^* pairs suggests they are unlikely to be OE neural precursors. Meis1 (Fig. 3B) and Sox2 (Fig. 3C, top) pairs were found at similar frequencies in lateral OE cultures, and combined in P-P totals; medial cells, however, preferentially generated Sox2 pairs (Fig. 3C, bottom) and rarely labeled for Meis1. The primary mode of cell division for isolated lateral and medial OE precursors was symmetric (Fig. 3E). Lateral cells, however, generated significantly more P-P pairs (51.1±3.6% of 327 total lateral versus 7.6±0.4% of 251 total medial pairs from three independent experiments; P≤0.006, Mann-Whitney) and medial cells significantly more N-N pairs (49±2.1% medial versus 16.8±2.9% lateral, P≤0.006). Each region generated small percentages of N-P pairs (4.4±0.4% medial versus 2.8±0.7% lateral; no difference). Thus, lateral OE precursors are primarily self-renewing, producing additional precursors, and medial OE precursors are primarily neurogenic.

Self-renewal versus terminal neurogenesis may be regulated by exposure to inductive signals that influence initial OE differentiation in vitro or in vivo (Kawauchi et al., 2005; LaMantia et al., 2000). Based on previous observations that Fgf8 enhances neurogenesis in the OE (DeHamer et al., 1994; Kawauchi et al., 2005; LaMantia et al., 2000), we asked whether Fgf8 increases symmetric neurogenic divisions in lateral OE cells at the expense of divisions favoring self-renewal. When lateral OE precursors are treated with Fgf8, the frequency of N-N pairs increases twofold (Fig. 3E, bottom; 34±3.2% of 386 total pairs, three independent experiments versus 16.8±2.9% lateral control, 51% increase, P≤0.01) and P-P pairs decline similarly (28.1±4.9 Fgf8 versus 51.1±3.5% lateral control, 46% decline, P≤0.01). P^*-P^* pairs also increased in response to Fgf8 (42.9% of Fgf8 versus 29.4%, lateral control, 32% increase P≤0.045); however, this may reflect general mitogenic activity of Fgf8. Nevertheless, Fgf8 apparently enhances neurogenesis in lateral OE precursors from by promoting symmetrical, terminal, neurogenic divisions and diminishing self-renewal of precursors that express Meis1 or Sox2.

Based on these in vitro findings, we hypothesized that Fgf8, which is available from the medial OE in vivo (Bhasin et al., 2003; Kawauchi et al., 2005; LaMantia et al., 2000), establishes or maintains OE precursor distinctions. To test this genetically, we evaluated OE precursors in E11.5 hypomorphic embryos, in which Fgf8 levels are significantly reduced but not eliminated (Fgf8^neo/neo) (Meyers et al., 1998). Fgf8^neo/neo embryos display variably penetrant phenotypes as previously reported (Garel et al., 2003; Meyers et al., 1998). Three out of 6 Fgf8^neo/neo embryos had forebrain morphogenetic defects, including rostromedial extension of ventral telencephalic neuroepithelium (see Fig. S1 in the supplementary material) and no morphologically identifiable VNO; the remainder displayed relatively normal morphologies. Altered OE transcription factor expression was apparent in five embryos, most noticeably in those with forebrain dysmorphology (Fig. 3 and see Fig. S1 in the supplementary material). Sox2 is diminished in the medial OE of Fgf8^neo/neo embryos (Fig. 3F), and the wild-type gradient is difficult to discern (compare with Fig. 1A,E). Meis1 shifts medially (Fig. 3G,J; see Fig. S1D,F,I in the supplementary material) but declines dorsolaterally (Fig. 3G,J) where neurons are found (Fig. 3H). Neurogenesis, though attenuated, extends into regions where Meis1 is elevated (Fig. 3G,H). Ascl1 expression varies in Fgf8^neo/neo embryos (highest frequency, Fig. 3I,J; lowest frequency, see Fig. S1L in the supplementary material); however, remaining Ascl1 precursors appear diminished medially and shifted laterally (Fig. 3I; see Fig. S1F,L in the supplementary material). Apparently, Fgf8 promotes OE neurogenesis by enhancing Sox2 and Ascl1 expression, and restricting Meis1 expression.

OE precursors are distinguished by position-dependent expression of Sox2, Meis, Ascl1 and Pbx, suggesting that these transcription factors may be key regulators of their identity. Thus, we developed an in vitro E11.5 lateral OE preparation to ectopically express these factors and evaluate directly their influence on identity and neurogenic capacity (Fig. 4A). We first asked whether Sox2 dose regulates lateral versus medial precursor distinctions by electroporating a chicken β-actin/cytomegalovirus promoter that drives mouse Sox2 followed by an internal ribosome entry site (IRES) and eGFP (pCIG-Sox2; Fig. 4B,C). For quantification, five non-overlapping fields were sampled ventral to the transition zone where Sox2 and Ascl1/TuJ1 levels rise and Meis1 levels decline (Fig. 4A). Elevated Sox2 dose elicits three cell-autonomous changes within the ventrolateral OE. First, ectopic overexpression of Sox2 diminishes the frequency of Meis1-expressing cells (93±0.48%/100 control-electroporated cells versus 10±0.87%/100 Sox2-electroporated cells, P<0.00001; Fig. 4D-H). Second, overexpression of Sox2 increases the frequency of Ascl1 cells (3.1±1.18%/2217 control versus 33.1±3.06%/1585 Sox2-electroporated cells, P<0.00002; Fig. 4I-M). Third, raising Sox2 levels enhances the number of TuJ1-labeled OE neurons (1.1±0.86%/2021 control cells versus 13.2±1.27%/1161 Sox2-electroporated cells, P=0.0002; Fig. 4N-T). Thus, ectopic Sox2 overexpression imposes a more `medial' character on lateral OE precursors: Meis1 levels are reduced, Ascl1 expression is elevated and neurogenesis is enhanced.

We next asked whether Meis or Pbx regulates acquisition or retention of `lateral' OE characteristics. We over-expressed Meis1 alone (pCIG-Meis1) or in combination with Pbx1 (pCIG-Meis1-IRES-Pbx1) using a modified pCIG plasmid with an additional human EIF4G IRES (Wong et al., 2002) (Fig. 5A-C). We quantified five non-overlapping fields in the dorsolateral OE, dorsal to the transition zone where Sox2 levels rise, Ascl1 cells are seen, and Meis1 levels decline. This region contains neurons, as well as cells with more `medial' OE precursor characteristics (inset, Fig. 5A; see also Figs 1, 2). Ectopic elevation of Meis1, as well as Meis1/Pbx1, suppresses Ascl1 in dorsolateral OE cells (29±1.7% of 1632 control versus 15.4±1.7% of 718 Meis1 cells, P=0.001; 11.8±1.2% of 860 Meis1/Pbx1 cells, P<0.00005; Fig. 5D-I,N). There was no difference between Meis1 and Meis1/Pbx1, suggesting that Pbx co-factors are not required for this regulatory change. By contrast, elevated Meis1/Pbx1 levels do not suppress Sox2 (49.8±0.7%/903 control versus 49.2±1.5%/1125 Meis1/Pbx1 cells, P=0.7; Fig. 5J-M,O), suggesting Meis1-mediated Ascl1 suppression is independent of Sox2. We did not observe a change in TuJ1-labeled neurons (data not shown). This may reflect difficulties detecting changes among greater concentrations of already differentiated neurons in more dorsolateral OE regions, or the capacity of Meis1-expressing precursors that also express high levels of Sox to generate neurons – perhaps at a lower frequency independent of Ascl1 – as is seen in the ventromedial OE. Thus, a transcriptional network defined by Sox2 dose, Meis activity and antagonistic regulation of Ascl1 distinguishes primarily `lateral' from primarily `medial' OE precursors.

The transcriptional network defined by Sox2, Meis1 and Ascl1 provides a framework to confirm genetic relationships between OE precursor classes and OE neurogenesis. If Ascl1 is crucial for expanding but not specifying ORN, VRN and GnRH neuron populations, loss of Ascl1 function in Ascl1^–/– mutant embryos should diminish but not prevent genesis of all three neuronal classes. A high medial, low lateral Sox2 gradient and a substantial population of Meis1 cells are present in the Ascl1^–/– OE (Fig. 6A,B). Meis1 cell frequency is somewhat higher in the mutant medial OE, and fewer laterally positioned Meis1 cells are acutely labeled with BrdU (Fig. 6C, compare with Fig. 2B), suggesting proliferative characteristics may be altered in precursors that have differing combinations of Sox2 as well as Meis1 expression levels in the absence of Ascl1 function. Only a small population of TuJ1-positive OE neurons remains in the E11.5 mutant (Fig. 6D; 29% of wild type values; P≤0.02, n=4, Mann-Whitney); nevertheless, ORNs, VRNs and GnRH neurons differentiate (Fig. 6E-L). OMP-labeled ORN frequency, however, is diminished by E16.5 (Fig. 6M). The small compliment of ORNs probably reflects limited early neurogenesis that declines to undetectable levels within 1 day of terminal division of an initial precursor cohort (Fig. 6I-N).

In our electroporation experiments, we found that elevated Sox2 suppresses Meis1 and promotes Ascl1 as well as neurogenesis (Fig. 4); therefore, it seemed that reduced Sox2 dose in vivo (Sox2^hyp/–) (Taranova et al., 2006) should expand Meis1, reduce or eliminate Ascl1, and diminish neurogenesis. Indeed, cells expressing Meis1/2 as well as Pbx genes are seen throughout the Sox2^hyp/– OE (Fig. 7A-D; data not shown). Ascl1 cells are absent in the OE (Fig. 7E) but seen in the ventral forebrain (Fig. 7E, inset). S- and M-phase frequency declines in parallel with Meis1 precursor expansion (Fig. 7F-H). The number of TuJ1 neurons generated by E11.5 is significantly less than wild type (20%; P≤0.02, n=4, Mann-Whitney; Fig. 7I), similar to Ascl1^–/– OE values. Reduced neuronal differentiation and failed VNO morphogenesis in Sox2^hyp/– embryos parallels that in Fgf8^neo/neo embryos (Fig. 3 and see Fig. S1D in the supplementary material), suggesting a relationship between Fgf8 signaling, Sox2 dose, VNO morphogenesis and OE neurogenesis. Nevertheless, all neuronal classes generated by the OE – ORNs, VRNs and GnRH neurons – are seen in the Sox2^hyp/– at E11.5 as well as at E6.5 (Fig. 7J-L). Thus, Sox2 dose is crucial for restricting Meis1 in the lateral OE, enhancing Ascl1 in the medial OE, and sustaining genesis – but not specification – of ORN, VRN and GnRH neurons.

We identified distinct precursor populations in the nascent OE that comprise self-renewing neural stem cells as well as transit amplifying cells that give rise to ORNs, VRNs and GnRH neurons. Meis1-expressing OE precursors, primarily in the lateral OE where neurons are absent or sparse, proliferate slowly and symmetrically to generate additional precursors. Proliferative cells that express high levels of Sox2, as well as Ascl1, primarily in the medial OE among differentiating neurons, divide rapidly and symmetrically to expand the OE neuronal population. Meis1, Sox2 and Ascl1 define a novel transcriptional network that regulates progression between these precursor states. Sox2 dose, controlled in part by local Fgf8 signaling, promotes OE neurogenesis by suppressing Meis1 and enhancing Ascl1 expression. Elevated Meis gene dose, independent of Pbx co-factors, diminishes Ascl1, but not Sox2 expression, and reduces neurogenesis by maintaining a more `lateral' precursor identity. This Fgf8-regulated, dose-dependent network balances self-renewal and terminal neurogenesis in the early embryonic OE.

Transcription factor profiles distinguish slowly dividing self-renewing from rapidly dividing transit amplifying OE precursors (Fig. 8A). As in other CNS regions, including the cerebral cortex (Bani-Yaghoub et al., 2006; Schuurmans and Guillemot, 2002), spinal cord (Graham et al., 2003) and retina (Heine et al., 2008; Taranova et al., 2006), OE precursors express SoxB1, Meis and bHLH genes (Fig. 8A). Our analysis, although not exhaustive, illustrates that OE precursor classes are not identified by singular expression of any factor. Instead, relationships between graded expression – suggesting dose effects – define OE precursor identity. This expression-based identity of OE precursors corresponds approximately to `lateral' and `medial' OE regions; however, transcription factor gradients do not strictly respect these anatomical boundaries. Indeed, transition zones in the ventromedial as well as dorsolateral OE have cells with combinations of factors that may further distinguish proliferative capacity or fate; accordingly, there may be additional precursor classes than those defined here. Such observations may complicate planning or interpreting OE `transcriptional fate mapping' studies using recombination driven by single transcription factor loci: few OE precursors uniquely express any particular factor. Thus, alternative approaches may be needed to map OE lineages and fates, including local tracer injections combined with conditional recombination-based approaches.

The inductive signal Fgf8 and the transcription factor Sox2 regulate OE precursor identity and neurogenic capacity (Fig. 8B,C). Elevated Fgf8 levels alone are sufficient to elicit terminal symmetric neurogenic division from isolated `lateral' OE precursors – making them resemble `medial' OE precursors. When Fgf8 is reduced in the early OE by hypomorphic mutation, patterning is disrupted: Sox2 and Ascl1 levels decline, and Meis1 levels expand. Apparently, when Fgf8 is diminished, these cells acquire `lateral' precursor characteristics. The identity of `medial' OE precursors, and parallel morphogenesis of medial structures such as the VNO probably depend on maintaining high Sox2 levels, which in turn depend upon normal levels of Fgf8. Accordingly, both Fgf8 and Sox2 hypomorphic OE phenotypes include Meis expansion, diminished Ascl1, failed VNO morphogenesis (without complete loss of the VRN marker TrpC2) and diminished neurogenesis (see also Kawauchi et al., 2005), without apparent loss of distinct OE neuron classes. Thus, precursors in Fgf8 and Sox2 hypomorphs – mostly expressing Meis1 – must include fate-specified multipotent OE neural stem cells.

We defined a transcriptional network that regulates progression of slowly dividing self-renewing to rapidly dividing terminal neurogenic OE precursors (Fig. 8B,C). Our results place Sox2 at a crucial, but distinct, position in specification of peripheral chemosensory versus other sensory receptor lineages, including hair cells (Dabdoub et al., 2008), taste cells (Okubo et al., 2006) and retinal precursors (Taranova et al., 2006). In contrast to Sox2 function in the ear, tongue and eye, where dose influences genesis of differentiated neuron subclasses, Sox2 in the nose is a concentration-dependent regulator of precursor state for the entire OE neuronal lineage. Direct electroporation in vitro as well as genetic manipulation in vivo indicates that low Sox2 dose maintains Meis1-expressing OE precursors. High Sox2 dose, modulated by Fgf8, facilitates the transition to terminally neurogenic precursors by inhibiting Meis1 and promoting Ascl1. Thus, Sox2 and Meis1 antagonistically (but not fully reciprocally; Fig. 8C) regulate Ascl1, which governs quantitative neurogenic output, but not identity, in the OE. As in the retina (Taranova et al., 2006), higher Sox2 dose might also modulate Notch and Hes expression or activity in the medial OE (Carson et al., 2006; Cau et al., 2000; Schwarting et al., 2007); however, it may be difficult to distinguish Sox2 and bHLH function in Notch-dependent mechanisms (Cau et al., 2000).

Sox2 dose is a crucial regulator of OE precursors via its influence on Meis genes. Elevated Sox2 in the lateral OE in vitro results in an effective local loss of Meis1 and Ascl1. We confirmed the relationship between Sox2 and Ascl1 in vivo using Fgf8^neo/neo and Asc1^–/– mutants; however, evaluating this relationship for Meis1, independent of Sox2 manipulation, remains challenging. We found no molecular, proliferative or neuronal defects in the OE of E11.5 Meis1^–/– embryos (data not shown), perhaps because Meis1, Meis2 and Meis3, which are apparently redundant in other systems (Hisa et al., 2004), are co-expressed in the developing OE. Nevertheless, the effective gain of Meis1 function in the Sox2^hyp/– OE reinforces the relationship between Sox2, Meis1 and Ascl1, and is consistent with dose-dependent influences of Sox2 on distinct OE precursor classes.

In the OE, Meis1 supports slowly dividing self-renewing precursors, whereas in retinal (Heine et al., 2008) and myogenic precursors (Berkes et al., 2004), it influences rapidly proliferating terminal-differentiating transit-amplifying precursors, perhaps via interactions with Pbx/Hox, bHLH genes and other factors. This suggests a novel function for Meis genes in the OE: suppression of bHLH genes to maintain slowly dividing self-renewing multipotent stem cells. Meis1-expressing OE precursors (Fig. 8B) conform to many criteria that define tissue-specific stem cells. They undergo slow symmetric self-renewing cell divisions (Gage, 2000; Slack, 2000), are found in a `niche' or distinct location (Fuchs et al., 2004; Moore and Lemischka, 2006), and generate all differentiated cell classes in a particular tissue. No other OE precursors have these characteristics, including capacity for `label retention' that distinguishes stem cells in other epithelia (Borthwick et al., 2001; Cotsarelis et al., 1990; Wong and Wright, 1999). Finally, ORNs, VRNs and GnRH neurons are seen in Ascl1^–/– and Sox 2^hyp/– mutants, in parallel with expanded or potentially exclusive presence of Meis1-expressing precursors. This argues strongly that Meis1 OE precursor populations include fate-specified stem cells that yield all major types of OE neurons. We cannot, however, rule out additional heterogeneity among Meis1 precursors that leads to diversity of potential and fate as is the case for early retinal progenitors (for a review, see Marquardt and Gruss, 2002).

The lateral OE, perhaps owing to antagonism between lateral RA and medial Fgf8 signaling in the context of mesenchymal/epithelial (M/E) interaction (Bhasin et al., 2003; LaMantia et al., 2000), may provide a specific niche that maintains a substantial population of Meis1-expressing precursors. Only frontonasal M/E interactions establish this niche – probably owing to distinctive RA-producing neural crest that constitutes the lateral frontonasal mesenchyme (Anchan et al., 1997; LaMantia et al., 2000). Indeed, in vitro, excess RA diminishes Sox2 in the medial OE, whereas excess Fgf8 increases Sox2 in the lateral OE (Rawson et al., 2010). When Meis1 induction fails in the presumptive OE in vitro, due to altered M/E interactions (Rawson et al., 2010), neurons are generated, but ORNs, VRNs and GnRH neurons do not differentiate. The role of Meis1 OE precursors in generating non-neural supporting cells, which appear later in development (Asson-Batres and Smith, 2006), or ensheathing cells that migrate from the OE (Doucette, 1989) remains to be evaluated. Nevertheless, slowly dividing multipotent neural stem cells capable of generating ORNs, VRNs and GnRH neurons must be included among Meis1-expressing precursors found primarily in the lateral OE.

Our data provide new guidance for identifying adult OE neural stem cells, which have remained elusive despite their central importance in regeneration after OE injury (Nordin and Bramerson, 2008), and in psychiatric and neurodegenerative diseases (Feron et al., 1999; Hahn et al., 2005; Johnson et al., 1994). The distinct characteristics of embryonic OE precursors focus attention not only on molecular markers for adult counterparts, but on locations for a supportive niche, similar to those in other mature regenerating epithelia, including gut, lung and – in lower vertebrates – the retina (Borthwick et al., 2001; Hitchcock et al., 2004; Wong and Wright, 1999). If OE axes are systematically transformed between the embryo and the adult, neural stem cells should be sought along the lateral OE boundary with the respiratory epithelium. In this location, secreted factors (including RA) may maintain multipotent OE neural stem cells and antagonize signals involved in ORN maturation and zonal specification (Norlin et al., 2001). Indeed, cultures from human OE biopsies at this boundary generate cells with molecular and functional characteristics of ORNs (Borgmann-Winter et al., 2009). Thus, our results may facilitate study of nervous system regeneration, and provide insight into pathobiology of diseases that compromise olfaction as part of a broader spectrum of nervous system dysfunction.

In situ hybridization and confocal microscopy were performed in the UNC Neuroscience Center Expression Localization and Confocal Microscopy cores (National Institute of Neurological Diseases and Stroke, NS031768), with expert guidance from Y. Wu, M. Aita and R. Peterson. We are indebted to Thomas Sugimoto, Cliff Heindel and Amanda Peters, who provided excellent technical assistance in support of this work. National Institute of Child Health and Human Development grant HD029178 (A.-S.L./N.R.) from the NIH and a Sigrid Jusélius Foundation Fellowship (M.K.L.) supported this work. Deposited in PMC for release after 12 months.