Clonal analysis of Drosophila antennal lobe neurons: diverse neuronal architectures in the lateral neuroblast lineage

Proper function of the nervous system requires diverse types of neurons. Distinct neurons acquire different morphological features and exhibit discrete electrophysiological or neurochemical properties. To understand how the brain develops and operates, one needs to identify all the neuron types, elucidate how they are specified and wired into functional circuits, and ultimately determine the functions of individual neuron types that control organismal behaviors.

In the adult Drosophila olfactory circuitry, many neuron types have been identified; characterization of them has greatly advanced our understanding of olfaction (reviewed by Jefferis and Hummel, 2006; Stocker, 1994; Vosshal and Stocker, 2007) (Fig. 1A). First, odors are sensed by olfactory receptor neurons (ORNs) in the antennae and maxillary palps. There are ∼50 classes of ORNs. ORNs express specific odorant receptors and target axons to specific glomeruli in the antennal lobe(AL) (Couto et al., 2005; Fishilevich and Vosshall,2005; Gao et al.,2000; Stocker et al.,1990; Vosshall et al.,2000). Second, olfactory information is relayed from ORNs to higher brain centers through AL projection neurons (PNs). Most PNs elaborate dendrites in only one of the 50 AL glomeruli, extend axons through the iACT(inner antennocerebral tract), and acquire subtype-specific patterns of axon arborization in the mushroom bodies (MBs) and the lateral horns (LHs). Although there probably exist 50 distinct types of uniglomerular PNs for conveying ORN activities with point-to-point correspondence to the high brain centers, not all the expected PN classes have been identified(Jefferis et al., 2001; Jefferis et al., 2007; Lin et al., 2007; Marin et al., 2002; Stocker et al., 1990; Tanaka et al., 2004; Wong et al., 2002). In addition, some PNs do not limit dendrites to just one of the AL glomeruli and display different axon trajectories. These multiglomerular PNs are poorly characterized, in part due to their lack of readily identifiable topographic projection patterns (Marin et al.,2002; Stocker et al.,1990). Third, there are excitatory and inhibitory local interneurons that make connections among various subsets of AL glomeruli. Identifying distinct types of AL interneurons should help elucidate how olfactory information may be first processed in the AL through the intricate inter-glomerular network (Olsen et al.,2007; Sachse et al.,2007; Shang et al.,2007; Stocker et al.,1990; Stocker et al.,1997; Wilson and Laurent,2005).

Among those identified AL neurons, the PNs are the mostly well characterized because of the enhancer trap line GAL4-GH146, which has been shown to label ∼90 PNs that might cover two-thirds of total PNs(Stocker et al., 1997). The clonal analysis with GAL4-GH146 further demonstrated that these PNs were derived from three neuronal lineages(Jefferis et al., 2001). Distinct subsets of specific uniglomerular PNs are made by the anterodorsal PN(adPN) and lateral PN (lPN) lineages, respectively. By contrast, the ventral PN (vPN) lineage produces both uni- and multi-glomerular PNs(Jefferis et al., 2001; Marin et al., 2002). Birthdating of the uniglomerular PNs in the adPN lineage further demonstrates that these different PN types are sequentially made by the adPN Nb(Jefferis et al., 2001). However, many PN types remain undescribed; and little is known about the diversity of the AL local interneurons, not to mention the mechanism(s) by which multiple types of interneurons are derived(Jefferis et al., 2001; Stocker et al., 1990; Stocker et al., 1997). Given that specific neuron types are derived from specific lineages at specific times, identifying the neuronal lineages and examining their neuron type compositions through birthdating analysis of post-mitotic neurons should allow one to determine all the neuron types in any neural circuitry of interest(e.g. Jefferis et al., 2001; Lee et al., 1999; Schmid et al., 1999; Zhu et al., 2006).

Cell lineage and birthdating analysis in the adult Drosophilabrain is made possible by MARCM (mosaic analysis with a repressible cell marker), a positive-labeling genetic mosaic technique(Lee and Luo, 1999). MARCM allows the determination of lineage relationship by examining multi-cellular clone, and identify neuron types based on detailed morphological features of single-cell clone (Lee and Luo,2001) (Fig. 1B,C). Furthermore, inducing clones (mitotic recombination) at specific developmental times further allows one to determine the sequences in which distinct neurons are derived (e.g. Jefferis et al.,2001; Lee et al.,1999; Zhu et al.,2006). However, in the complex brain where many neuronal lineages remain uncharacterized, informative lineage analyses by MARCM have relied on the use of subtype-specific GAL4 drivers to focus on certain neuron types at one time (e.g. Jefferis et al.,2001; Lee et al.,1999; Zhu et al.,2006). As one GAL4 driver may not cover an entire lineage,thorough analysis of a complex neuronal lineage first requires identification of diverse GAL4s that respectively label different neuron types of the same lineage and together reconstitute the entire lineage(Jefferis et al., 2007; Lai and Lee, 2006). Here, we describe how we used the MARCM and the newly established dual-expression-control MARCM (Fig. 1B,C) (Lai and Lee;2006) to re-characterize the adPN, lPN and vPN lineages for more thorough lineage analysis.

A modified `transposon swap' strategy(Sepp and Auld, 1999) was used to generate the LexA::GAD-GH146 (LG-GH146). The swapping P-element P[LGawB] was constructed by replacing LexA::GAD(Lai and Lee, 2006) with GAL4 in P[GawB] (Brand and Perrimon, 1993) and then injected into y^-,w^- strain to generate transgenic flies (Spradling and Rubin,1982). Each single male fly with the genotype P[LGawB]/Y;GAL4-GH146/CyO;Δ2-3,Sb/+was then crossed with UAS-EcR^DN(Cherbas et al., 2003) to cause lethality of the flies carrying P[GawB]. The expression patterns were then examined by fluorescent microscopy in the progeny flies, which were produced from the cross of the male candidates and the reporter strain lexAop-rCD2::GFP (Lai and Lee,2006).

Larvae collected within 2 hours of hatching were cultured at the density of 80 larvae per vial at 25°C. Wild-type MARCM Nb and/or single-cell clones were induced at various developmental stages by heat shock at 38°C and then examined in adults. The fly strains used for various experiments were as follows: (1)Acj6-GAL4 (Bourbon et al., 2002; Komiyama et al.,2003); (2)GAL4-GH146(Stocker et al., 1997);(3)GAL4-GH298 (Stocker et al.,1997); (4)GAL4-KL107(Shang et al., 2007);(5)GAL4-MZ699 (Ito et al.,1997); (6) GAL4-NP6115 (N. K. Tanaka and K.I.,unpublished); (7) UAS-mCD8;Pin/CyO,Y; (8) tubP-LG;FRT^G13,UAS-mCD8,lexAop-rCD2::GFP/CyO,Y; (9) FRT^G13,hs-FLP,tubP-GAL80/CyO,Y; (10) hs-FLP,UAS-mCD8::GFP;FRT^G13,tubP-GAL80/CyO; and (11) FRT^G13,LG-GH146,lexAop-rCD2::GFP/CyO,Y.

Dissection, immunostaining and mounting of adult brains were carried out as described (Lee and Luo, 1999). Because GAL4-KL107 and acj6-GAL4 also labeled the olfactory neurons, and the stereotyped connections within the olfactory circuitry were not affected by the deprivation of olfactory input(Berdnik et al., 2006; Tanaka et al., 2004; Wong et al., 2002), the antennae and the maxillary palps of these newly eclosed adult flies were surgically removed to allow the ORN axons to degenerate to reveal the morphologies of ALNs. Primary antibodies used in the study include rat monoclonal antibody to mCD8 (1:100, Caltag), rabbit antibody to GFP (1:100,Molecular Probes), mouse monoclonal antibody to Acj6 (1:100, DSHB), Elav(1:200, DSHB), nc82 (1:100, DSHB), rCD2 (1:100, Serotec) and Repo (1:100,DSHB). Immunofluorescent signals were collected by confocal microscopy and then processed using Adobe Photoshop.

To extend analysis of AL neurons beyond the GAL4-GH146-positive PNs, we sought to determine how many GAL4-GH146-negative neurons exist in the three neuronal lineages of interest(Fig. 2). We used the dual-expression-control MARCM with GAL4-GH146 plus a ubiquitous LexA::GAD driver to label AL clones using two independent reporters. We identified the Nb clones that contain specific clusters of GH146-positive PNs based on the expression of UAS-mCD8, and visualized all the progenies of the clones using lexAop-rCD2::GFP (Lai and Lee, 2006) (Fig. 2B,D-F). When the labeled clones, obtained through induction of mitotic recombination in newly hatched larvae (NHL), were co-labeled with the antibody against neuronal transcription factor Elav(Robinow and White, 1991) or glial specific transcription factor Repo(Xiong and Montell, 1995), we found that all three lineages produced only neurons(Lai and Lee, 2006) (data not shown). Analysis of those clones revealed averages of 73.2 (s.d.=3.4, n=4), 193.3 (s.d.=5.7, n=8) and 49.2 (s.d.=6.2, n=5) cells in the anterodorsal, lateral and ventral lineages,respectively.

Comparing the neurite projection patterns of the entire clones with those of the GH146 subsets further revealed that the adPN(Fig. 2D) and vPN(Fig. 2F) lineages appear to consist of uniglomerular and mixed PNs, respectively, as exemplified by the GH146-positive subsets. Their axons constitute the iACT and mACT, respectively(Fig. 2D,F). By contrast,diverse novel neurite trajectories probably exist among the GH146-negative progenies of the lateral lineage (referred to as the lAL lineage hereafter)(Fig. 2E). One whole lAL Nb clone has fully populated the entire ipsilateral AL lobe(Fig. 2E), significantly innervated the contralateral AL, and connected the ALs with various neuropils in addition to the MB and LH of the same side (see below). Furthermore,although the POU domain transcription factor Acj6(Ayer and Carlson, 1991) is still expressed in every adPN and not detectable in the vPNs, despite the inclusion of GH146-negative progenies, cells that express Acj6 near PNs of the lateral lineage (Komiyama et al.,2003) clearly belong to the lAL lineage(Fig. 2C,G-I). Besides, some Acj6-positive neurons that are not included in the anterodorsal Nb clone and juxtapose to the adPN lineage could be embryonic born or derive from non-AL lineages. Thus, it is intriguing that while Acj6 distinguishes GH146-positive adPN from lPN, it is not a lineage-specific transcription factor when all neuronal types are taken into consideration. These findings have not only substantiated presence of many undetected neurons, but also demonstrated the possibility of identifying novel types of neurons within the previously characterized AL lineages.

A hierarchical strategy was then taken to determine neuron type compositions of the GH146-negative progeny. Briefly, we first searched for additional GAL4 drivers that, like GAL4-GH146, selectively label subsets of AL neurons. We further determined their relationships with GAL4-GH146, then reconstituted the lineages of interest using multiple subtype-specific GAL4 drivers, and ultimately labeled single neurons within each GAL4 sublineage by MARCM.

To target the GH146-negative AL neurons for single-cell analysis by MARCM,we first identified a number of AL GAL4 drivers that selectively label various subsets of the GH146-negative AL neurons. We collected AL GAL4 drivers that label neurons in similar locations but with different patterns of neurite projections. We determined their relationships with GH146-positive PNs by labeling them differentially in the same organism. A LexA::GAD counterpart of GAL4-GH146 was obtained by P-element swap (see Materials and methods)(Fig. 2J-J″). When various GAL4 drivers were combined with LG-GH146, we demonstrated that the same clusters of PNs can be dually labeled by GAL4-GH146 and LG-GH146 (Fig. 2J-J″), and that other AL GAL4 drivers we subsequently characterized (except acj6-GAL4) exclusively mark GH146-negative neurons (see below). We further examined the lineage origins for the AL neurons that are positive for specific AL GAL4 drivers. Dual-expression-control MARCM with an AL GAL4 driver plus tubP-LG or LG-GH146 allowed us to assign unambiguously the individual groups of GAL4-positive neurons to the identifiable adPN, lAL or vPN lineage. Briefly,distinct clusters of acj6-GAL4-positive neurons constitute the entire adPN lineage and a subset of ventrally positioned lAL neurons, respectively(Fig. 3). GAL4-MZ699labels many, if not all, GH146-negative vPNs(Fig. 4). By contrast,consistent with the notion that lAL neurons are highly diverse, to reconstitute the lAL lineage requires five distinct GAL4 drivers, including GAL4-GH298, GAL4-KL107 and GAL4-NP6115 in addition to GAL4-GH146 and acj6-GAL4(Fig. 5).

We further determined the neuron type compositions within the individual groups of GH146-negative AL neurons through analysis of single neuron morphologies. Single-cell clones of GAL80-minus neurons were generated in specific 12-hour windows throughout larval development. We used one GAL4 driver at one time to label the serially derived single neurons. This allowed us to readily separate the neuron types that express the same GAL4 but have been born at different times. Systematic analysis of such single-cell clones successfully led to identification of many novel distinct neurons in the adult Drosophila ALs. Below, we provide detailed description of these cell lineages.

Because all the adPNs are positive for Acj6, we used the acj6-GAL4(Bourbon et al., 2002), which expresses in all Acj6-positive cells (data not shown) to characterize the adPN lineage (Fig. 3). We first confirmed that acj6-GAL4 permits labeling of the entire adPN lineage,as acj6-GAL4 marked all the cells in the adPN Nb clones that have been independently labeled by tubP-LG (n=15)(Fig. 3A). The adPN Nb clones,labeled with acj6-GAL4 or GAL4-GH146, looked comparable:their dendrites targeted subsets of AL glomeruli and the axons projected through the iACT to form synapses in the MB calyx and the LH (compare Fig. 3D with Fig. 2D). However, analysis of the adPN Nb clones, which were simultaneously labeled with acj6-GAL4and LG-GH146, revealed the presence of ∼24 adPNs that were marked only by acj6-GAL4 (data not shown). Furthermore, six AL glomeruli were only labeled by acj6-GAL4 and not innervated by GH146-positive adPNs at all. We identify them as DL2d, DP1l, VC3, VC4, VM4 and VM5v(Fig. 3B,C)(Laissue et al., 1999). Consistent with the notion that adPNs and lPNs target different glomeruli,none of the newly identified adPN glomerular targets is innervated by GH146-positive lAL PNs either. These observations suggest that all the adPNs,irrespective of GH146 expression, acquire the neurite projection pattern characteristic of most known uniglomerular PNs(Stocker et al., 1990; Jefferis et al., 2001).

We then examined single adPNs by MARCM with acj6-GAL4 as the driver. Single-cell clones of adPNs uniformly targeted dendrites to one of the 50 AL glomeruli and extended axons to the MB and LH through the iACT(n=86 in 86 single-cell clones) (e.g. Fig. 3E). Intriguingly, adPNs with different glomerular targets acquired different stereotyped patterns of axon arborization in the LH (e.g. Fig. 3F-H), suggesting that specific types of uniglomerular PNs are made by the adPN lineage (Marin et al.,2002). Furthermore, specific adPNs were born at specific times. For example, whereas all the single-cell clones generated within the first 36 hours after larval hatching (ALH) innervated the DL1 glomerulus (n=18 in 18 single-cell clones), the majority of single-cell clones induced in the window of 108 to 120 hours ALH developed into DL2d-targeted PNs (n=12 in 14 single-cell clones) and very few DL2d PNs were obtained outside this window of induction (n=2 in 14 single-cell clones). We also observed a stereotyped pattern of axon arborization characteristic of DL2d-targeted PNs, one representative type of GH146-negative adPNs(Fig. 3F,G). These results collectively suggest that the adPN lineage consistently makes uniglomerular PNs and sequentially yields different types of uniglomerular PNs, including those negative for GAL4-GH146.

There are ∼50 cells in the post-embryonic-born vPN lineage (see above). Although only six of them express GAL4-GH146(Jefferis et al., 2001), an average of 45.2 vPNs (s.d.=5.2, n=4) are positive for GAL4-MZ699. Concurrent use of GAL4-MZ699 and LG-GH146 showed that no cell was simultaneously labeled by these two drivers in the vPN lineage (Fig. 4A). This indicates that the MZ699-positive cells are distinct from the GH146-positive cells and together they potentially cover the entire vPN lineage (Fig. 4B).

Analysis of vPN Nb clones, labeled with GAL4-MZ699 versus GAL4-GH146, revealed analogous patterns of neurite projections despite a huge difference in the numbers of labeled cells. Both innervated the ipsilateral AL and extended neurites to the LH through the mACT (compare Fig. 4C with Fig. 2F). The ventral GH146-PNs are composed of PNs that either innervate single glomeruli or the entire unilateral antennal lobe (Jefferis et al.,2007; Marin et al.,2002). The GAL4-MZ699-positive ventral AL neurons can also be largely categorized as uniglomerular PNs (n=12 in 45 single-cell clones) or multiglomerular PNs (n=33 in 45 single-cell clones), and both types of neurons target their axons towards the LH bypassing the MB calyx (n=45 in 45 single-cell clones) (e.g. Fig. 4D-F). The uniglomerular PN targets its dendrites to only one glomerulus in the AL, such as DA1(n=7 in 45 single-cell clones)(Fig. 4D) and VL1 (n=3 in 45 single-cell clones) (data not shown), which are also innervated by ventral GH146-PNs (Marin et al.,2002; Wong et al.,2002). The axon termini arborization patterns of the MZ699-PNs are similar to those of GH146-positive uniglomerular vPNs(Fig. 4D)(Marin et al., 2002). The neurites of the multiglomerular PNs innervate the partial ipsilateral AL and do not form apparent glomerular shapes, and several different stereotyped projection patterns could be observed. Therefore, we categorized these multiglomerular PNs based on their dendrite and axon projection patterns. For example, the dendrites of one type of multiglomerular PN spread in the medial dorsal AL as one big cluster, and its axons project along the ventral side of LH, turn dorsalwards at the lateral side of the LH and terminate at the dorsal side of the LH (n=8 in 45 single-cell clones)(Fig. 4E). Another example of a multiglomerular PN concentrates its dendrites in the middle of the AL and targets its axons along the ventral side of the LH (n=6 in 45 single-cell clones) (Fig. 4F).

The distinct types of vPNs are apparently specified based on when they were generated by the vPN progenitor. Analysis of single-cell clones of vPNs that were derived at different times of development revealed that specific types of vPNs were born at specific times and to obtain them as single-cell clones required induction of mitotic recombination at specific times (during the neuron-producing mitoses). For example, single-cell clones of DA1-targeted uniglomerular vPNs (Fig. 4D)were mostly obtained following mitotic recombination within the period of 24 to 48 hours ALH (n=6 out of 7 clones). By contrast, the multiglomerular vPNs that targeted dendrites to the medial dorsal region of the AL (Fig. 4E) were apparently born during the window of 12-24 hours ALH (n=8 out of 8 clones). Another type of vPN (Fig. 4F) was generated between 96 and 120 hours ALH (n=6 out of 6 clones). These phenomena suggest that distinct vPNs acquire different neurite projection patterns according to their temporal identities.

The lAL lineage is the most prominent among the three GH146-PN-containing AL lineages (Fig. 5). It is composed of about 200 cells, occupies the entire lateral side of the AL and probably derives from the lateral Nb, one of the five actively proliferating Nbs per brain lobe, including the four MB Nbs, since larval hatching(Ito and Hotta, 1992; Stocker et al., 1995; Truman and Bate, 1988). We used five enhancer trap GAL4 lines, GAL4-GH146, GAL4-GH298, GAL4-KL107,GAL4-NP6115 and acj6-GAL4, to reconstitute this lineage(Fig. 5B,E,H,K,N). GAL4-GH298 and GAL4-KL107 have been used to characterize inhibitory and excitatory AL local interneurons, respectively, though they might label some common interneurons(Shang et al., 2007; Wilson and Laurent, 2005). Other lAL GAL4s apparently label different subsets of lAL neurons judging from projection patterns of marked neurons (Fig. 5C,F,I,L,O), and none of them co-labels GH146-positive PNs(Fig. 5D,G,J,M). Adding the numbers of marked cells together, except those of the GH298 lineage because of its potential overlap with KL107, we have about 90% of the entire progeny covered for single-cell analysis of the largest AL lineage(Fig. 5A).

There are 21.0 (s.d.=3.0, n=7) GAL4-GH298-positive neurons in the lAL lineage (Fig. 5E) (Stocker et al.,1997). Consistent with earlier studies, they are GABAergic local interneurons with projections restricted to the ipsilateral AL (data not shown; Fig. 5F)(Wilson and Laurent, 2005). Single-neuron labeling revealed a general neurite elaboration pattern characteristic of all GH298-positive interneurons: each neuron ramifies throughout the entire ipsilateral AL but does not outline individual glomeruli(n=90 in 90 single-cell clones)(Fig. 6B)(Stocker et al., 1997).

The GAL4-KL107 group contains about 46.8 (s.d.=2.9, n=6) cells that consist purely of AL local interneurons, as evidenced by the restriction of neurites to the ipsilateral AL (Fig. 5I). However, individual glomeruli are discernible in the KL107-labeled AL (Fig. 5I). Detailed single-cell analysis showed that GAL4-KL107-positive AL neurons can be categorized as two types. The first type looks indistinguishable from the GAL4-GH298-positive lAL neurons (n=109 in 223 single-cell clones), and it is named as type A local interneuron (type A LN). Their neurites do not form obvious glomerular shape (Fig. 6B). It is unknown yet whether GAL4-GH298-positive AL neurons are subsets of GAL4-KL107-positive type A LN or vice versa. The other type of GAL4-KL107-positive ALN is termed as type B LN, and its neurites form obvious glomerular shape in the ipsilateral AL(Fig. 6C,D) (n=114 in 223 single-cell clones). We also noticed that some type B LNs innervate the entire AL and form synapses within each glomerulus (n=54 in 114 single-cell clones) (Fig. 6C)and other type B LNs make connections with subsets of glomeruli (n=60 in 114 single-cell clones) (Fig. 6D). Whether the targeted glomeruli are stereotyped remains to be determined.

GAL4-NP6115, which labels a distinct subset of iACT neurons in the lateral cell bodies cluster (N. K. Tanaka and K.I., unpublished), expresses in 13.0 (s.d.=0.9, n=13) neurons in the lAL lineage(Fig. 5K). The lAL Nb clones,when labeled with GAL4-NP6115, displayed novel neurite projection patterns (Fig. 5L). Single-cell labeling showed that GAL4-NP6115-positive lAL neurons are all PNs(n=24 in 24 single-cell clones) and can be classified as uniglomerular PNs (n=4 in 24 single-cell clones) and atypical multiglomerular PNs (n=20 in 24 single-cell clones). The projection patterns of the uniglomerular PNs are similar to those of GH146-PNs. They target dendrites specifically to the DL3 glomerulus, one of the glomerular targets of GH146-positive lPNs (Marin et al., 2002) (data not shown). The atypical multiglomerular PNs target dendrites to multiple glomeruli in the AL but do not elaborate the dendrites in glomerular shape. Their axons pass through iACT to various brain centers beyond the MB and the LH. For example, one type of atypical multiglomerular PN targets dendrites to the posterior ventral side of both the ipsi- and contralateral ALs through the inter-antennal connective(Strausfeld, 1976). Its axon projects towards the posterior inferior protocerebrum, turns ventrally around the ventral lateral protocerebrum, passes through the ventral lateral protocerebrum and terminates in the posteriorlateral protocerebrum(n=6 in 24 single-cell clones)(Fig. 6E). Another example of the atypical multiglomerular PN targets its dendrites to the medial half of the ipsilateral AL, and project axons to the superior protocerebrum(n=6 in 24 single-cell clones) (data not shown).

In the lAL lineage, acj6-GAL4 expresses in 82.7 (s.d.=4.8, n=7) cells whose cell bodies cluster ventral to other lAL neurons(Fig. 5N). The projection patterns of acj6-GAL4-labeled lAL Nb clones display several unique features (Fig. 5O). Analysis of serially derived single-cell clones showed that the acj6-GAL4-positive neurons include at least three different types of cells (n=89 single-cell clones). The first type of acj6-GAL4-positive lAL neuron (type A Acj6-lALN) targets its neurites to the partial ipsi-lateral AL, though these neurites do not form obvious glomerular shape. Its axon projects through the oACT to the posterior inferior lateral protocerebrum (n=8 in 89 single-cell clones)(Fig. 6F). The second type of acj6-GAL4-positive lAL neuron (type B Acj6-lALN) projects the neurites to the both sides of the ALs by passing through the inter-antennal connective. The neurites concentrate at the posterior ventral side of the ALs and they do not form glomerular shapes (n=13 in 89 single-cell clones)(Fig. 6G). The third type of Acj6-GAL4-positive neurons (type C Acj6-lALN) does not target their neurites to the AL, even though these neurons are also derived from the lAL lineage(n=68 in 89 single-cell clones). This type of neuron extends neurites between the ipsi- and contralateral deutocerebra by passing through the great commissure, and the neurites showed bilateral asymmetrical distribution, i.e. the neurites on the ipsilateral side elaborate more exuberantly than the contralateral ones (Fig. 6H).

In summary, unlike the anterodorsal and ventral AL lineages, which are homogeneously composed of PNs, the lateral AL neuroblast generates diverse types of neurons, which include interneurons, uniglomerular PNs, atypical multiglomerular PNs and the neurons that innervate other parts of the deutocerebrum instead of the ALs or MB and LH. In addition, besides the presence of distinct uniglomerular PNs innervating different specific glomeruli, there apparently exist multiple subtypes of neurite elaboration patterns within the interneurons, atypical PNs, as well as non-AL neurons(Fig. 6). To resolve the complex lAL lineage completely requires larger-scale and more systematic single-cell analysis.

With respect to the major types of lAL progenies, including the type A and type B LNs, uniglomerular PN, atypical multiglomerular PN and three types of Acj6-lALN, we wondered whether they are specified based on neuronal temporal identity. We determined whether specific lAL neurons were derived at specific times. We analyzed the birth timings of different types of neurons by calculating the frequency of single-cell clones of particular types per 100 brains across the larval development from 0 to 132 hours ALH and summarized the results in the Table 1. We found that specific types of neurons were indeed born at specific times.

Although some neuron types are preferentially made at a given time, we observed that different types of neurons could be produced within the same developmental periods (italics in Table 1). The temporal overlap in the generation of single-cell clones of different types suggests that the lAL Nb may not finish producing early types of neurons before switching to make later types. This is in great contrast to the sequential non-overlapping production of distinct neuron types in the well-characterized MB lineages (Lee et al., 1999; Zhu et al.,2006). However, the current protocol of single-cell analysis(synchronization in NHL followed by induction of mitotic recombination at various later time points) might fail to resolve the absolute birth order in the complex lAL lineage, as multiple ganglion mother cells probably co-exist around a rapidly dividing Nb, Flipase activity can last for a while despite a transient induction, and initially synchronized larvae may grow at variable rates and thus receive heat shock at slightly different ages.

To determine whether multiple lAL neuronal types were indeed generated in overlapping windows, we sought to characterize the co-labeled two-cell clones,as well as the neuron type compositions in mid-sized partial lAL Nb clones,both of which were derived from Nbs that should be in the middle of producing some intermediate neuron types. We selectively focused on the period between 48 and 72 hours ALH when both type A LNs and uniglomerular PNs could be readily targeted for labeling in single-cell clones. However, we were not able to use current genetic techniques to distinguish the two-cell clones that were generated from one GMC versus the single-cell clones that were derived from two co-existing GMCs. Therefore, we mainly characterized the lAL Nb clones that were generated from the intermediate multicellular Nb clones. If a lAL Nb never made the next types of neurons until completion of all earlier types(sequential production), we expect to detect no early-type neurons in such mid-sized Nb clones (Fig. 7A,left panel). Conversely, if it involved orderly but overlapping production of multiple neuron types (simultaneous production), one should detect concurrent decreases in the cell numbers of several neuron types among gradually reduced Nb clones (e.g. Jefferis et al.,2001) (Fig. 7A,right). We hope to determine whether such intermediate multicellular Nb clones already lost some GH146-positive PNs but still contained cells that are positive for GAL4-GH298 (type A LNs).

We obtained three mid-sized lAL Nb clones from 156 mosaic brains; and, by dual-expression-control MARCM, GH146-positive uniglomerular PNs (uPNs) and GH298-expressing type A LNs in the GAL80-minus clones were simultaneously labeled with lexAop-rCD2::GFP and UAS-mCD8, respectively. Partial reduction in the cell numbers of both types of IAL neurons were observed in all the three cases. When compared with the presence of 35 uPNs and 21 type A LNs in most full-sized lAL Nb clones(Fig. 7B,B′,B″),the three later-derived Nb clones carried 30 uPNs and 18 type A LNs (data not shown), 21 uPNs and 13 type A LNs (Fig. 7C,C′,C″), and 12 uPNs and 5 type A LNs (data not shown), respectively. The reduction in uPNs was also evidenced by the presence of fewer GH146-labeled glomeruli in the mid-sized clones(Fig. 7C″). Analysis of the remaining glomerular innervation patterns revealed missing of common glomerular targets DM1, VA4 and VA7m. This suggests that, as in the adPN lineage, distinct lAL uPNs were derived in an invariant non-overlapping sequence. Nevertheless, the simultaneous reduction of uPNs and type A LNs in the later-derived Nb clones indicates that distinct types of lAL neurons are indeed produced in overlapping windows. Additional mechanisms are likely involved in increasing neuronal diversity in the much more complex lAL lineage.

More thorough lineage analysis of three previously characterized AL PN lineages has allowed us to identify novel types of AL neurons in addition to recovering more uniglomerular PNs, detecting diverse multiglomerular PNs and better categorizing the already identified intra-AL interneurons (Figs 3, 4, 6). The novel types of AL neurons include type A and type B Acj6-lAL neurons as well as NP6115-positive atypical PNs. Type B Acj6-lAL neurons are the only inter-AL interneurons in these three AL lineages. Type A Acj6-lAL neurons connect the ipsilateral AL glomeruli with the posterior inferior lateral protocerebrum through the oACT,whereas NP6115-positive atypical PNs may connect the ALs with additional brain centers, including the posteriorlateral protocerebrum. The posteriorlateral protocerebrum is one of the brain regions that receive visual signals from the optic lobe (Otsuna and Ito,2006). The networking among multiple brain centers through the ALs may help integrate the diverse types of information that an organism has received at the same time. Alternatively, these atypical PNs may relay non-olfactory information from special kinds of ORNs to the brain centers involved in the processing of other environmental cues, such as humidity,amine or carbon dioxide (Yao et al.,2005; Jones et al.,2007; Kwon et al.,2007). In addition to uncovering novel types of PNs, we found it more feasible to identify stereotyped multiglomerular PNs through analysis of single cells born at specific times. We reproducibly labeled the same identifiable multiglomerular PNs following stage-specific induction of single-cell clones (Fig. 4, Table 1). These results suggest the presence of specific types of polyglomerular PNs, and imply that they,like distinct uniglomerular PNs, are specified based on temporal cell fates. Analysis of the intra-AL interneurons born at different times further allowed us to detect interneurons that exhibit distinct patterns of connectivity(Fig. 6, Table 1). Finally, we identified novel uniglomerular GH146-negative adPNs(Fig. 3). Although more uniglomerular PNs remain to be identified to cover all the AL glomeruli, it is interesting that distinct types of uniglomerular adPNs have been found to target the same glomeruli, as well the uniglomerular vPNs. For example, both GH146- and NP6115-lPNs innervate the DL3 glomerulus, and GH146- and MZ699-vPNs innervate DA1 or VL1 glomeruli (Figs 4, 5). These observations suggest the presence of a much more complex topographic map than we have imagined even in the first olfactory brain center: the AL. Identifying additional AL lineages to characterize every single progeny and describe its connectivity is essential for resolving the entire topographic map in the AL for better understanding the mechanisms of olfaction.

Specific neuron types derive from specific AL neuronal lineages, suggesting involvement of lineage identity in the diversification of AL neurons. For example, distinct Nbs in the Drosophila ventral ganglion are specified according to the Cartesian coordinates of their positions in each hemisegment within the two-dimensional developing neuroepitheilium. This generally involves spatial patterning along the anteroposterior and dorsoventral axes (reviewed by Bhat,1999; Skeath,1999). Similar mechanisms probably govern the acquisition of different lineage identities in the AL Nbs that are located in stereotyped positions, although the details are poorly understood (reviewed by Urbach and Technau, 2004). In addition, several molecules have been observed to confer lineage identity to targeting specificity, such as Acj6, Cut, Chip and Drifter(Komiyama et al., 2003; Komiyama and Luo, 2007). For example, Acj6 is required for adPNs to target dendrites to pre-specified glomeruli, and our more thorough lineage analysis revealed that many different types of Acj6-positive neurons, although not uniglomerular PNs, are made by the lAL Nb (Fig. 6). These observations suggest more complex mechanisms for lineage identity specification and neuronal morphogenesis.

Specific neuron types are born at specific developmental times in all the lineages analyzed here. This suggests that post-mitotic neurons of the same lineage acquire different temporal cell fates (Figs 3, 4, Table 1) (reviewed by Ito and Awasaki, 2008; Pearson and Doe, 2004; Yu and Lee; 2007). In the adPN lineage, which homogeneously consists of uniglomerular PNs, one can unambiguously demonstrate that distinct adPNs are made by a common progenitor in an invariant non-overlapping sequence via analysis of the glomerular innervation patterns of serially derived Nb clones(Jefferis et al., 2001). Although Chinmo, a determinant for MB neuronal temporal cell fates, has been shown sufficient to alternate adPN neuron identity(Zhu et al., 2006), it is unknown whether Chinmo governs adPN temporal identities using the same mechanism. Similar analysis suggested that uniglomerular PNs of the lAL lineage are also possibly made one subtype after another(Jefferis et al., 2001), the major types of lAL neurons are apparently yielded in overlapping windows(Table 1). It remains to be determined whether the co-production of multiple neuron types involves derivation of distinct neurons from the same ganglion mother cells and requires asymmetric Notch/Numb signaling following the neuron-producing mitoses (Artavanis-Tsakonas et al.,1999; Endo et al.,2007; Guo et al.,1996; Skeath and Doe,1998; Spana and Doe,1996). A thorough single-cell analysis of the entire vPN lineage is also needed for determining if diverse vPNs, including many distinct multiglomerular PNs, are actually born in a fixed non-overlapping sequence, a possibility that remains viable after birthdating of additional identifiable vPNs (Fig. 4). Besides, it remains to be determined whether other known temporal cell fate determinants,such as Hb, Pdm and Cas, are involved(Isshiki et al., 2001; Komiyama and Luo, 2007).

Taken together, we have established means for more thorough analysis of complex neuronal lineages. Re-characterization of three previously examined AL lineages has led to the identification of many novel types of neurons. Identifying additional AL lineages and characterizing them accordingly should allow one to determine the detailed neural map connectivity of this Drosophila olfactory relay center. Furthermore, thorough lineage analysis lays the essential foundation for elucidating how numerous types of neurons can be derived from a limited number of progenitors. Intriguingly,multiple types of lAL neurons are made by a common progenitor in orderly but overlapping windows (Table 1),suggesting possible involvement of post-mitotic patterning in addition to temporal cell fate specification in the development of Drosophilaolfactory circuitry.

We thank R. F. Stocker for GAL4-GH298, G. Miesenbock for GAL4-KL107, L. Luo for acj6-GAL4, N. Perrimon for P[GawB] plasmid and N. K. Tanaka for sharing GAL4-NP6115 prior to publication. We thank B. Leung, L. Luo and members of the Luo laboratory (Stanford University) for critical reading of the manuscript. This work was supported by the US National Institutes of Health and March of Dimes Birth Defects Foundation.