In order to analyse amos regulation, we characterised its expression pattern. Expression in the antennal imaginal discs has been described previously 1. Here, expression begins at puparium formation in three distinct semicircles in the future third segment. By 8 h after puparium formation (APF), this expression merges into a single large crescent and continues until 16 hours APF. This expression is responsible for the third wave of olfactory SOP specification that takes place in the third antennal segment, which forms the precursors of the sensilla basiconica and trichodea 120.

In the embryo, amos RNA is expressed transiently in a segmentally repeated pattern of presumed proneural clusters and SOPs at stages 10/11 221. In the trunk, Amos protein expression begins at stage 10 in single small cluster in the dorsal ectoderm of each abdominal segment and of thoracic segments T2 and T3 (Fig. 1A). This cluster is absent from T1. Shortly afterwards, Amos expression ceases here. At this time, in abdominal segments A1–A7 only, transient expression begins in a second small cluster of cells adjacent and ventral to the first cluster (Fig. 1B). Subsequent analysis (below) shows that the first clusters give rise to the dbd neurons and the second clusters to the dmd1 neurons. Interestingly, the presumed SOPs that derive from both these clusters (but not the clusters) also expresses the related proneural protein, Atonal (Fig. 1A,B). The expression of Ato in these SOPs requires amos but the converse is not true, confirming that Amos provides the proneural function for these SOPs (Fig. 1C,D and data not shown).

In the stage 10 embryonic head, Amos is expressed in large ectodermal clusters in the antennal, maxillary, and labial segments (Fig. 1A). These are rather more ventral than the clusters in the trunk. Slightly later, expression appears in small clusters in the maxillary, mandibular and labial primordia. Expression also appears in small clusters that appear homologous to those in the trunk. The head expression suggests that amos may function in the formation of a variety of head sense organs.

To identify the cis-regulatory sequences of amos, the intergenic regions upstream and downstream were tested for their ability to drive expression of a GFP reporter gene (Figs 2,3, and Table 1). A construct with a 3.5-kb upstream fragment (amos-3.5-GFP) supports GFP expression in the third antennal segment. The expression pattern was characterised by co-labelling with antibodies to Amos and Senseless (Sens) as a marker of SOPs 22. Over a period of 0–8 h APF, GFP expression was largely coincident with Amos protein (Fig. 2B,C), suggesting that the fragment contains a major enhancer for expression of Amos in the antennal disc. Whilst generally co-expressed, GFP is not observed strongly in the cells in which Amos is most recently activated. This appears to represent a slower induction of GFP synthesis and maturation relative to endogenous proneural proteins 23. A subset of GFP-expressing cells (with deeper nuclei) also express Sens, thus representing the amos-dependent olfactory SOPs themselves (Fig. 2C). GFP-negative SOPs are also present, which are likely to represent the earlier waves of ato-dependent olfactory SOPs. Thus, amos-3.5-GFP contains a major enhancer that drives Amos expression in most or all cells of the olfactory PNCs. Strong expression in the resulting SOPs suggests that the fragment drives expression in both the PNCs and SOPs, although it is also possible that GFP expression in SOPs represents perdurance of expression driven by the PNC enhancer. Previously, the enhancer activity of a 56-bp larger fragment had been tested by cloning into a Gal4 expression vector 1. The expression pattern of amos-3.5-GFP described here differs from that described for amos-3.6-Gal4 in that the latter appeared largely SOP-specific 1. It is possible that this might represent a strongly delayed onset of GFP expression from this construct.

amos-3.5-GFP also supports expression in the embryo. The pattern of GFP closely resembles that of Amos protein although the appearance of GFP is delayed (Fig. 3A–C). Owing to the transient nature of Amos expression, this means that relatively little overlap of Amos and GFP expression are observed. Between stages 10 and 11, GFP expression is detected in the same sequence of cell clusters as Amos in the head and trunk. Expression begins in the head antennal and maxillary segments. It is then observed in the other head clusters and the first thoracic and abdominal clusters. In the latter, GFP expression appears in the first clusters as Amos expression disappears from them and is replaced by expression in the second more ventral cluster in A1–7 (Fig. 3B,C).

By stage 12, amos expression has been turned off. However, perdurance of GFP expression was used to follow the fate of the different amos clusters. There is a complex network of sensory neurons in the trunk of the embryo, but amos is responsible for only two of the multidendritic neurons, the dbd and dmd1 neurons 121. In the abdominal segments of late embryos, amos-3.5-GFP expression is observed strongly in the dorsally located dmd1 neuron in abdominal segments A1–7 (as marked by the sensory neuron marker 22C10 (anti-Futsch)) (Fig. 3E–G). Weaker and variable expression is observed in the dbd neuron and its associated glial cell in segments T2,3 and A1–8/9. In addition, some ectodermal cells also express GFP, which is consistent with perdurance in some of the PNC cells. GFP is not expressed in neurons in T1, which is consistent with observation that amos is not expressed in this segment. Interestingly, the lack of expression in T1 suggests that this segment does not possess a dbd neuron. Similarly, only segments A1–7 appear to have a GFP-expressing dmd1 neuron. Consistent with this, a marker of the dbd and dmd1 cells (anti-Pdm) detects no neurons in T1, and only the dbd neuron in T2,3 (data not shown). Interestingly, the dorsoventral locations of the dbd and dmd1 neurons appear reversed relative to their proneural clusters, suggesting that one or both neurons undergo migration (cf Fig. 3B and 3G).

GFP expression also perdures in the head region. This is particularly associated with the complex clusters of sense organs that form the antennomaxillary complex and pharynx-associated sense organs of the larva. Perdurance of expression confirms that in the antennal segment amos contributes to the larval olfactory organ, the dorsal organ (do) (see 24 for nomenclature). The main expression in the maxillary segment contributes to the ventral organ (vo), and perhaps the papilla organ (pao), whose neurons are reported to resemble dbd neurons 24. Equivalent expression in the labial segment contributes to the labial sense organ (lbso).

amos-3.5-GFP is expressed in all locations of Amos expression except for three small areas dorsally in the maxillary and labial segments (arrowed in Fig. 3A,C). To locate the enhancer sequences responsible for these areas, a 1-kb region downstream of amos was also tested for enhancer activity (construct amos-D-GFP; Fig. 2A). This region supports GFP expression in these three groups of cells (Fig. 3D). Their location and GFP perdurance suggests that these appear to contribute neurons of the terminal organ (to), labial organ (lbo), and labial sense organ (lbso). Between them, the upstream and downstream flanking regions contain enhancers that can account for the entire amos expression pattern.

The complex nature of amos expression suggests that the 3.5-kb fragment may contain different enhancers for different aspects of pattern. The 3.5-kb fragment was subdivided into three smaller fragments (A, B, C) measuring 1.68 kb, 961 bp and 893 bp. amos-A-GFP is expressed solely in the trunk of the embryo: there is no GFP expression in either the antennal disc or the head region of the embryo (Fig. 4A and data not shown). This suggests that an enhancer element responsible for amos expression in the dbd and dmd clusters is present in fragment A. In late embryos, perduring GFP was seen in the trunk in a similar pattern to that of amos-3.5-GFP (Fig. 4B).

amos-B-GFP also supported expression in the embryo but not the antennal disc. In this case, GFP was observed in the head region in a pattern similar to that supported by amos-3.5-GFP (Fig. 4C). Thus, an enhancer(s) for amos expression in the head is present in fragment B, and this appears to be required for most of the components of the amos head pattern, as confirmed by 22C10 staining of late embryos (Fig. 4D). amos-B-GFP is also expressed in a pattern of ectodermal clusters in the trunk, but this expression appears to be ectopic: the clusters are more ventral than those for amos-3.5-GFP or for Amos itself, and the expression does not perdure into cells associated with the PNS (Fig. 4C,D). Interestingly, this ectopic expression appears to resemble in its segmental location the amos-B-GFP pattern observed in the head (Fig. 4E). This suggests that the ectopic trunk pattern represents the inappropriate activity of the head enhancer present in B. In contrast, a construct combining fragments A and B (amos-AB-GFP) shows correct head and trunk expression (data not shown). There may therefore be an inhibitory sequence in A that normally restrictsthe activity of the B enhancer to the head.

amos-C-GFP does not support expression in the embryo at the time that amos is normally expressed (Fig. 4F). However, as the neurons start to differentiate, GFP expression is switched on in an inconsistent subset of cells marked by 22C10 (data not shown). This presumably represents artefactual expression. In contrast, fragment C drives expression in the antennal imaginal disc (Fig. 2D). GFP expression is present in the amos dependent SOPs and in some cells of the proneural cluster. There is no discernible difference between the location of GFP expression as driven by the amos-3.5-GFP fragment and that driven by amos-C-GFP, although expression from the latter is generally weaker.

In summary, distinct enhancer sequences are required for amos expression in the embryo and antennal disc. More than one enhancer is responsible for amos expression in the embryo, and head and trunk enhancers appear to be separate. The presence of enhancer modules for different expression locations is consistent with the findings of other proneural genes. However, our analysis found only one enhancer for each location of amos expression.

For sc and ato, experiments investigating transgene rescue of sensillum loss in mutants showed that substantial phenotypic rescue is achieved if a transgene includes enhancers for both phases of expression 111725. In contrast, a transgene driven by a single enhancer (for either the first or second phase of expression) rescues poorly. In the case of amos, we have found no separation of enhancers for temporal phases. Although distinct phases of amos expression can be discerned, they appear to be driven by a single element in each location. It seems unlikely that further subdivision will reveal such separable enhancers, nor that other enhancers exist farther up- or downstream of amos. We therefore determined whether the 3.5-kb region contained all elements responsible for amos regulation in the antennal disc in a rescue experiment. The amos-3.6-Gal4 line was used to drive UAS-amos expression in amos mutant flies (amos¹/Df(2L)M36-S6), and antennae from such flies were examined for types and numbers of sensilla on the third segment. Mutation of amos results in the loss of all sensilla trichodea and sensilla basiconica, as well as the appearance of ectopic sensory bristles 1. Expression of amos driven by the amos-3.6-Gal4 line resulted in a substantial rescue of this defect. Ectopic bristles were almost completely suppressed (average of 1.66, cf c.20 for amos¹/Df(2L)M36-S6). Sensilla trichodea were present in numbers close to that expected in wildtype (66.7 ± 11.0). Substantial numbers of sensilla basiconica were also present, although less than half the number expected for wildtype (84.0 ± 10.0). Although quantitatively not complete, the degree of rescue suggests that all major patterning elements necessary for amos expression in the antenna are present in amos-3.6-Gal4 and also, by inference, in the 3.5-kb fragment. Lack of complete rescue may reflect the delay in onset of Gal4-driven expression in this system. Interestingly, the numbers of ato-dependent sensilla coeloconica are reduced compared to wildtype (34.7 ± 4.2). Such reduction is also seen when amos is misexpressed in the wildtype antenna (S. Maung and APJ, in prep.). It is possible that perdurance of Gal4-driven expression of amos interferes with endogenous ato function.

Where known for other proneural genes, the second phase of expression involves direct autoregulation, with proneural/Daughterless protein heterodimers binding to E box sequences within an autoregulatory enhancer. We asked whether autoregulation is also important in amos regulation, concentrating on the antennal disc expression. If amos is autoregulatory, one might expect functional E box binding sites to be present within the amos-C antennal regulatory region. Four potential E box sequences (CANNTG) are present in this fragment, two of which are conserved between Drosophila melanogaster and D. pseudoobscura (data not shown). None of these sequences (ttCAAGTGa, aaCAATTGt, gtCATATGg, gtCATTTGg) conform completely to the consensus sequences reported for Sc (gCAG(G/C)TG(g/t)) or Ato (a(a/t)CA(G/T)GTG(g/t) (Singson et al., 1994; Powell et al., 2004). However, although Amos protein is predicted to function via E box binding, no such site has yet been characterised. Therefore, we investigated whether any of these E box sequences are important for amos-C enhancer function by mutating all four E boxes within the amos-C-GFP construct (amos-Cmut-GFP). However, no clear reduction in expression was observed for amos-Cmut-GFP compared with the unmutated construct (data not shown). In case autoregulation lies outside fragment C, the whole of the 3.5-kb sequence was scanned for E box sequences. No further E boxes matching the known consensus sequences for Sc or Ato were found. The closest match is a site of atCAGGTGa (differing from the Ato consensus sequence in its 3' flanking base). This sequence is conserved in D. pseudoobscura. However, when mutated within amos-3.5-GFP, no difference in GFP expression pattern was observed in the antenna or embryo (data not shown).

Autoregulation may occur indirectly via the regulation of an intermediate factor. To find evidence for indirect autoregulation, we determined whether misexpression of amos results in ectopic induction of amos-3.5-GFP. In the embryo, no ectopic expression was observed from amos-3.5-GFP when UAS-amos was driven in the ectoderm by a sca-Gal4 driver (Fig. 3H). Using the Gal4^109-68 line in imaginal discs in third instar larvae 26, no ectopic amos-3.5-GFP expression was observed upon amos misexpression, except for a small number of GFP-expressing cells in antennal discs. However, variable numbers of these cells were also visible in control antennal discs that lacked the UAS-amos, and so appear to represent a genetic background effect (data not shown).

amos-3.5-GFP expression was also examined in amos mutant embryos to look for loss of GFP expression that might indicate the need for autoregulation. In such embryos, no clear difference from wild type was observed in the GFP expression pattern (data not shown). In mutant antennal discs, ectodermal GFP expression appeared unchanged, although SOP expression was lost as would be expected from the absence of such cells (data not shown). In summary, no part of the amos expression pattern could clearly be seen to depend on endogenous amos expression.

Functional dissection of the regions around the amos gene has shown that it is regulated via an array of separable enhancers, both upstream and downstream. Thus, a complex expression pattern results from the sum of a number of simpler sub-patterns driven by specific enhancers. In this respect, amos conforms well to the regulation characteristics of other Drosophila proneural genes. Vertebrate proneural gene homologues also have modular enhancer arrangements 2829. One notable feature is that the separation of head and trunk enhancers appears to be unique to amos.

At least in the case of ac/sc, individual modules respond to different combinations of upstream (prepattern) factors 1330. For amos, one likely prepattern factor is the runt domain transcription factor, Lozenge, whose function is required for amos expression in the antennal disc 2. The presence of three potential Lozenge binding sites in the amos fragment C suggests that amos may be under the direct control of Lozenge (unpublished observations).

Investigation of sc regulation has led to a paradigm in which it is regulated in two phases via different enhancers. In the first PNC phase, sc is regulated via enhancers that respond to upstream prepattern factors. In the SOP phase, it is regulated via a single SOP enhancer, with positive autoregulation as a major input 15. For ato, there is also evidence for two separable regulatory phases, although separation is less clear. For instance, an autoregulatory enhancer has been described, but this only functions in the specific context of SOP recruitment via EGFR signalling 19. Furthermore, embryonic expression has been described as being initiated by a 3' 'prepattern' enhancer, and then maintained by a 5' enhancer with an autoregulatory component 17. However, the 5' enhancer is not SOP-specific, and there is no direct evidence that it responds to autoregulation (S. Cachero and APJ, unpublished observations). Nevertheless, whilst differing in important details from the sc paradigm, it is clear that ato is regulated in different temporal phases via distinct enhancers 17.

Given these observations, it is surprising that we found no evidence that amos is regulated in distinct temporal phases via separate enhancers. Although it is possible that the enhancers identified here could be subdivided further, there is currently no obvious separation of PNC and SOP enhancers. This suggests either that enhancers for separate phases of expression are not fundamental to proneural gene function in neurogenesis, or that amos regulation and function differs substantially from that of other proneural genes.

For sc regulation, the rationale for separate temporal enhancers is that the mechanism of neurogenesis depends on a progression from an initial competence phase in the PNC to a commitment phase in the SOP. Separable enhancers allow the first phase to be achieved by an array of prepattern enhancers and the second phase by a positive autoregulatory enhancer. The latter contains two E-box binding sites responsible for the autoregulation 1011. A similar enhancer seems to exist for ac: a 0.9 kb enhancer upstream of ac has three E-boxes that are important for its function 103132.

For amos, we found no evidence for autoregulation. From the close similarity of their bHLH sequences, it is expected that Amos and Ato proteins would have similar DNA binding characteristics. However, no E box could be identified as being required for amos enhancer function. It is possible that autoregulation is indirect, or occurs directly via a protein-protein interaction with another DNA binding cofactor. However, no ectopic enhancer activity could be observed after amos misexpression. Whilst contrary to the regulatory trends established for sc and ato, the lack of evidence for amos autoregulation nonetheless correlates with the lack of separable temporal enhancers.

The lack of amos autoregulation is surprising. The autoregulatory elements of sc and ato are the target of other regulatory inputs, so that modulation of autoregulation (facilitation or inhibition) during the PNC to SOP transition is an important factor in limiting SOP determination. The sc SOP enhancer is thought to be a direct target of E(spl) proteins and other factors during lateral inhibition 11. The ato recruitment enhancer has a binding site for Pointed protein next to the functional E box, which is crucial for allowing autoregulation 19. This ensures that autoregulation (and SOP commitment) is triggered when a competent cell (in the PNC) receives EGFR signalling input. For sc and ato, modulation of autoregulation allows precise control neurogenesis. Lack of amos autoregulation may reflect the fact that amos function results in large numbers of SOPs, at least in the antenna, and so such precise limitation on its activity may not be necessary. Interestingly amos also appears to be a more powerful proneural gene in other assays (S. Maung and APJ, in prep)3334. Another intriguing possibility is that the need for proneural autoregulation correlates not with SOP commitment, but instead with the need to maintain expression in the SOP in situations where SOP expression is relatively prolonged. At least in the embryo, Amos expression in SOPs indeed appears to be very transient (and hence the limited overlap with GFP in reporter gene lines); such transient expression may not require an autoregulatory input.

Our findings suggest that autoregulation is not a universal feature linking proneural gene function to neurogenesis. For ato, only in the case of one enhancer has direct positive autoregulation been demonstrated so far 19. For vertebrate proneural homologues, direct autoregulation via a conserved E box has been demonstrated for Math1 35. On the other hand, there is no evidence of positive autoregulation for Mash1 36 or neurogenin 37. It is far from clear how general the occurrence of autoregulation is in proneural gene function.

Fly stocks used were amos¹ 1, Df(2L)S6 1, UAS-amos 2, UAS-ato 5, 109-68Gal4 26 and scabrous-Gal4.

Primers were designed to amplify a 3.5-kb fragment upstream of the amos gene (5'-GGAGTGCAACCGGATTTAACC and 5'-CCCGATGCCAACCTCTTGA). Three further amplifications subdivided this fragment into three sections: amos-A (1.68 kb) (5'-GGAGTGCAACCGGATTTAACC and 5'-CCTAGCGAAAGCGGAGAATT), amos-B (961 bp) (5'-AATTCTCCGCTTTCGCTAGG and 5'-CGAGGAGTTCGCTGAATTTC), and amos-C (893 bp) (5'-GAAATTCAGCCAACTCCTCG and 5'-CCCGATGCCAACCTCTTGA). Primers were also designed to amplify a 1-kb fragment downstream of the amos gene (amos-D) (5'-GTATGAAATGGTGGAGTTGG and 5'-CGAACTCAGGTGTCTTTAGA). All five fragments were cloned into pHStinger 38 to give amos-3.5 GFP, amos-A-GFP, amos-B-GFP, amos-C-GFP and amos-D-GFP for germ line transformation.

E boxes were mutated using the Stratagene Quikchange mutagenesis kit. In each case the sequence was changed to GGATCC.

Transformation plasmids were injected into w¹¹¹⁸; Δ2–3 flies. Transformants were selected and outcrossed to w¹¹¹⁸ to remove the Δ2–3 element. At least two independent lines were analysed for each construct.

Antibody staining of pupal antennae was carried out as previously described 2. Pupae were staged by collecting at the time of puparium formation and then aging on moist filter paper at 25° before dissection. For misexpression analysis, imaginal discs were dissected from wandering third instar larvae and fixed in 3.7% formaldehyde (10 min at room temperature). Embryos were collected, fixed and stained according to standard procedures. Incubations with primary and secondary antibodies were carried out according to standard procedures. Primary antibodies used were mouse anti-22C10 (1:200; obtained from the Developmental Biology Hybridoma Bank, Iowa City, Iowa), mouse and rabbit anti-GFP (1:500; Molecular Probes), guinea-pig anti-Sens (1:6250; 22, and rabbit anti-Amos 1:1000 1). Secondary antibodies (1:1000) were obtained from Molecular Probes. Confocal microscopy analysis was carried out using a Leica TCS SP2 or a Zeiss PASCAL microscope.