Projection Neurons and Local Interneurons Are Made in Pairs from Common Ganglion Mother Cells in the Lateral Antennal Lobe Lineage
The lateral antennal lobe (lAL) lineage yields about 200 neurons during larval neurogenesis 
. Labeling the lAL progeny by conventional mosaic analysis with a repressible cell marker (MARCM) 
using a pan-neural nSyb-GAL4
revealed neuronal cell bodies packed along the lateral border of the AL. They elaborate densely in the antennal lobe (AL) and the neighboring antennal mechanosensory and motor center (AMMC) () and further innervate the inferior ventrolateral protocerebrum (IVLP), lateral horn (LH), superior medial protocerebrum (SMP), and some other brain regions (). It is not possible to determine the detailed “projectome” among the targets without single-neuron labeling.
lAL neurons are born as LN/PN pairs.
To reveal single-cell morphology and simultaneously determine the neuron birth order, we “sequenced” the larval lAL lineage using ts-MARCM 
. We determined ganglion mother cell (GMC) progeny born in 2-h windows throughout larval development (see Materials and Methods). We identified lAL GMC clones based on cell body positions and neurite trajectory patterns that match the lAL NB clones generated at various time points (unpublished data). Except near the lineage end (see below), both daughter neurons derived from each lAL GMC survived into the adult stage. Notably, one projection neuron (PN) consistently paired with one local interneuron (LN). They exist as twin clones when differentially labeled by ts-MARCM (). This confirms the previous hypothesis that the lAL lineage is composed of one PN hemilineage and one LN hemilineage 
. Moreover, we obtained neuron pairs with distinct characteristic morphologies following clone induction at different developmental times, indicating that the birth order of GMCs has governed neuronal diversification in the protracted lAL lineage (; see below).
The Lateral Antennal Lobe Projection Neurons Innervate Brain Regions That Are Involved in Different Sensory Modalities
The lAL PNs identified by ts-MARCM can be categorized into five classes based on their morphology: monoglomerular PN (mPN), unilateral PN (unPN), bilateral PN (biPN), AMMC PN, and suboesophageal ganglion (SOG) PN (). mPNs connect a single AL glomerulus to the mushroom body (MB) calyx and LH through the inner antennocerebral tract (iACT; ) 
. The mPNs target the VA4, VC2, VC1, DM1, DM2, VA5, VA7m, DA1, DL3, VM1, DA2, or DM5 AL glomerulus () and have been determined previously based on the GAL4-GH146
. The lack of additional mPNs using the more broadly expressed nSyb marker suggests this set was already complete. Unlike mPNs, unPNs and biPNs connect the AL(s) to various brain regions not yet implicated in olfaction, which include the posteriorlateral protocerebrum (PLP), inferior ventrolateral protocerebrum (IVLP), and superior medial protocerebrum (SMP). unPNs restrict their proximal elaborations to the ipsilateral AL, whereas biPNs show bilateral AL elaborations. Eight types of unPNs and six types of biPNs can be further distinguished based on (1) AL innervation patterns, (2) neurite trajectories, and (3) distal targets (; see Table S1
The lAL PNs can be grouped into five classes based on neuron morphology.
In addition to the AL PNs, we obtained 16 types of AMMC neurons and three types of SOG neurons that may account for the AMMC and SOG elaborations seen in the lAL NB clones (, compared to ). Most of the AMMC neurons acquire some bilateral elaborations across the brain midline. AMMC-1 to -11 connect the ipsilateral AMMC to the ipsilateral as well as contralateral IVLPs or posterior ventrolateral protocerebrum (PVLP) (AMMC-1 to -8), or only to the contralateral IVLP/PVLP (AMMC-9 to -11). AMMC-12 and -13 elaborate exclusively within the AMMC and wire the paired AMMC structures together. AMMC-14 to -16 show dendrite-like processes in the IVLP and axon-like projections in the AMMC, but AMMC-14 only targets the ipsilateral AMMC whereas AMMC-15 and -16 innervate both ipsilateral and contralateral AMMCs. Finally, the three types of SOG PNs show unique characteristic patterns of proximal elaboration in the SOG and further target distinct brain regions, including the PLP (SOG-1), the contralateral and ipsilateral clamp surrounding the MB peduncles (SOG-2), and the ipsilateral clamp and inferior bridge (IB) (SOG-3). Some proximal neurites of SOG-3 further innervate the vest, which is posterior to the AL. Please refer to Table S1
for more detailed description of these stereotyped AMMC and SOG neurons.
In sum, the lAL NB yields not only AL PNs but also AMMC and SOG neurons, which may contribute to distinct neural circuits (see Discussion). Does the GMC birth order guide the derivation of these multiclass neurons one group by another along the Notch-off hemilineage of the complex lAL pedigree 
Orderly, But Not Class-By-Class, Production of Distinct Lateral Antennal Lobe Projection Neurons
The twin single-cell clones collected for this study were induced in discrete 2-h windows to sample neurons born at different times from larval hatching to puparium formation. Notably, distinct lAL PNs were preferentially hit at different developmental times. To deduce their possible birth order, we attempted to arrange the identified lAL PN types chronologically based on when their precursors are susceptible to mitotic recombination. We first determined the primary window(s) of susceptibility for each lAL PN type (shaded boxes in ; boxes that account for less than 10% or less of the hits at the respective timing of clone induction are not shaded, except for rarely hit neuron types, including AMMC-9, and AMMC-15). All but two of the 45 identifiable PN types show a single narrow window of susceptibility that staggers in partially overlapping manners along the ~120 h of larval development. A tentative PN birth order can then be deduced based on the starts and/or ends of the susceptible windows as well as their prime times of appearance ().
The lAL PNs are produced in a specific birth order.
We also determined the sequence of birth for the mPNs through analysis of GAL4-GH146-labeled NB clones (–M″). We witnessed a sequential loss of the 12 glomerular targets from the NB clones of reducing sizes. As to their paired GMC clones, we observed the serial appearance of the VA4, VC2, VC1, DM1, DM2, VA5, and VA7m mPNs as they sequentially disappeared from the NB clones of reducing sizes (–H″). Then some DA1 mPNs were hit before the birth of DL3 mPNs, and additional DA1 mPNs arose later with the NB clones lacking DL3 mPNs (–K″). The remaining VM1, DA2, and DM5 mPNs then followed in the same sequence as they disappeared from the NB clones (–M″). In addition, certain NB clones apparently paired with GH146-negative progeny and existed alone when labeled with GAL4-GH146 (unpublished data). Ignoring those gaps, we derived the following birth sequence for the 12 types of lAL mPNs: VA4-VC2-VC1-DM1-DM2-VA5-VA7m-DA1-DL3-DA1-VM1-DA2-DM4. The same birth order was obtained from the analysis of nSyb-GAL4-labeled single-cell clones (). Notably, mapping the lAL lineage using a ubiquitous driver, like nSyb-GAL4, and through analysis of numerous serially derived single-cell clones (), further allowed us to (1) fill the gaps occupied by GH146-negative PNs, (2) resolve the mixing of DA1/DL3 mPNs, and (3) uncover the paired LNs (see below).
The complete birth order of larval-derived lAL PNs unveils several interesting points. First, distinct PNs are born in an invariant sequence. Second, different PN classes are born in an intercalated sequence with analogous PN types arising in separate windows. For example, the 12 mPN types derive in nine blocks that span nearly two-thirds of the larval development. During the same period of time, 14 other PN types, including 10 types of AMMC neurons, are made. Six additional AMMC types plus three types of atypical AL PNs are derived afterwards. In contrast with the late AMMC siblings, the majority of atypical AL PNs and all the SOG neurons are born prior to the mPN-producing windows. Third, the apparently arbitrary birth order is further complicated by the recurrent production of DA1 and DL3 mPNs. They are first born from 46 to 58 h after larval hatching and are also generated roughly 12 h later (). DA1 mPNs precedes DL3 mPNs during their initial contiguous production. By contrast, DL3 mPNs arise before DA1 mPNs in their second round of birth that is further separated by the production of two types of AMMC neurons. The early-born DA1/DL3 and later-derived DA1/DL3 mPNs are morphologically indistinguishable and are both positive for GAL4-GH146. Nonetheless, the DA1/DL3 mPNs born at different times pair with distinct LNs, and the early-born DA1 mPNs can be further divided into two groups based on their paired LNs (; see below).
Taken together, the lAL NB makes distinct PNs of diverse classes in a fixed arbitrary sequence. Neurons acquire specific fates based on their birth order, but the actual sequence of production reveals no obvious logic behind their stereotyped temporal deployment. Analogous neurons can arise at different times across the protracted lineage. Moreover, identical neurons can be born consistently in two waves. To uncover the genes that determine specific neuron classes versus the fates within a class will be critical for elucidating the molecular mechanisms underlying such orderly, but not class-by-class, production of distinct neuronal siblings.
Diverse Local Interneurons Pair with Distinct Projection Neurons to Account for 48 Serially Derived Neuron Pairs
Besides PNs, the lAL lineage yields LNs. For most of the lineage, PNs and LNs were made in pairs, as the mitotic recombination during GMC divisions consistently led to the labeling of one PN paired with one LN by ts-MARCM (). However, the final nine PN types were not paired with another neuron or a NB clone (). This indicates that either the paired cell died or could not be labeled with nSyb-GAL4. Notably, the longer PN hemilineage exhibits higher morphological diversity than its LN sister hemilineage.
The larval lAL lineage yields PNs and LNs in pairs except near the lineage end.
First, unlike PNs that innervate brain regions involved in multiple sensory modalities, their paired LNs exclusively innervate the ALs and should be selectively involved in olfaction. Second, many distinct PNs were paired with indistinguishable LNs. Nonetheless, the LNs can be grouped into four classes based on the extent of their AL elaborations. The pan-AL LNs densely innervate all the glomeruli in the AL; the lavish LNs occupy most, but not all, AL glomeruli; the patchy LNs invade many glomeruli in spotty patterns; the sparse LNs, by contrast, arborize locally within a few glomeruli (). Notably, except for the DA1 and DL3 mPNs, PNs of a given type consistently pair with a particular class of LNs. The DA1 mPNs may be born with lavish, patchy, or sparse LNs, and the DL3 mPNs can pair with patchy or sparse LNs. By contrast, the remaining 43 PN types show strict sisterhood with one of the four LN classes. Taking both PN and LN diversities into consideration, we have in total recovered 48 distinct PN/LN pairs () that arise sequentially from the lAL lineage as implicated from the invariant birth order of the PNs ().
Annotation of lAL LNs based on birth order and neuron morphology.
For those five PN/LN pairs (referred to as DA1/lavish, DA1/patchy, DA1/sparse, DL3/patchy, and DL3/sparse, respectively) whose distinction depends on the LN diversity, we refined the PN grouping and determined the subgroups' windows of production. We found that DA1/lavish, DA1/patchy, and DL3/patchy are born earlier in a contiguous sequence and that DL3/sparse and DA1/sparse are born later in separate windows (). When these 48 recognizable PN/LN pairs were chronologically arranged based on the derived birth order (), we noticed that, unlike PNs, the AL LNs of different classes have arisen in a more logical sequence with most pan-AL LNs () born before the lavish LNs (, AC–AD), which largely precede the patchy LNs (–AB) and ultimately transit to the sparse LNs (Figure 4AE–AM).
The pan-AL LNs paired with distinct PNs are morphologically indistinguishable from one another. They show analogous electrophysiological profiles 
, further indicating the homogeneity of the pan-AL class of LNs. How about the other three classes of LNs? Notably, the lavish or sparse LNs that associate with a particular PN type (thus born in a specific developmental time window) tend to avoid or innervate a characteristic set of AL glomeruli. To examine the LN diversity in further detail, we computed the average AL elaboration pattern of the LNs for each of the 48 sequentially derived PN/LN pairs. We manually annotated individual LNs' glomerular innervation patterns and then calculated the percentage of LNs, for a given PN/LN-pair type, whose neurites could be found within a particular glomerulus (). A uniform full pattern of elaboration was ascertained in the pan-AL LNs paired with distinct PNs (). By contrast, the patchy LNs innervate various glomeruli stochastically and may jointly tile the entire AL, as they collectively show a low-penetrant targeting to nearly all the AL glomeruli within any of the three PN/LN groups that carry patchy LNs (). Unlike the pan-AL and patchy LNs, the lavish as well as sparse LNs exhibit discriminative patterns of elaboration depending on the identity of the associated PNs. The lavish LNs selectively avoid certain glomeruli, while the sparse LNs preferentially innervate specific glomeruli (). The stereotyped patterns of AL glomerular innervation observed in the LNs, paired with distinct PNs and born at specific developmental times, argue for the presence of distinct types of lavish and sparse LNs. This is distinct from the lack of discernible cellular diversity among the pan-AL or patchy LNs.
The Projection Neuron and Local Interneuron Hemilineages Alter Temporal Identity Independently
Using the glomerular innervation frequencies to represent the LNs associated with a particular PN type and arranging them chronologically based on the deduced PN birth order revealed that the PN and LN hemilineages alter temporal identity independently. The PN hemilineage is longer and yields many more morphologically distinct neurons than the LN hemilineage. In addition, contrasting with PNs that arise in a rather complex sequence, the four LN classes are produced roughly in the order of pan-AL→lavish→patchy→sparse ( and ). During the production of the relatively homogeneous pools of pan-AL or patchy LNs, we witnessed multiple unilateral temporal fate changes in the PN hemilineage (). As to the lavish and sparse LNs that exhibit morphological subtypes, we found that LNs showing indistinguishable AL elaboration patterns are born in contiguous blocks that yield distinct PNs (). These observations collectively indicate that LNs alter temporal identity (that controls morphogenesis) at a slower tempo than PNs do, although they are derived from the same GMCs. Despite the presence of fewer LN fate transitions, the lavish-to-patchy LN fate switch consistently occurs without a concomitant PN fate change. It subdivides the window of DA1 mPN neurogenesis into two blocks that differ only on the LN side (, , and ). Taken together, PNs and LNs undergo independent temporal identity changes.
The independent PN/LN temporal fate specification is further evidenced by two unilateral PN fate duplications. The DA1 and DL3 mPNs were initially made at 46 to 58 h after larval hatching (ALH) paired with lavish or patch LNs. After that, the lAL NB switched to produce AMMC PNs paired with various LNs. Notably, around 70 to 84 h ALH, the lAL lineage yielded additional DL3 and DA1 mPNs in reverse birth-order and associated with sparse LNs ( and ,AF,AI). These phenomena collectively suggest that neuronal terminal fates are determined in hemilineage-specific manners.
Differential Notch Activity Governs the Differential Temporal Patterning Observed between Hemilineages
It is hard to image how the temporal fates of twin neurons can be differentially patterned, given that neuronal temporal identities are presumably conferred in the precursors by a set of sequentially and transiently expressed transcription factors 
. However, we have learned that the PN versus LN binary cell fates are determined through differential Notch signaling due to asymmetric segregation of Numb 
. We wondered if Notch merely specifies PN/LN binary fates or it also governs the differential patterning of PN and LN temporal fates. LNs were grossly transformed into PNs in the lAL NB clones that lacked Sanpodo (Spdo) (see below), a positive regulator of Notch 
. Analyzing the temporal fates for those PNs transformed from LNs due to loss of Notch should help elucidate the role(s) of Notch in specifying PN versus LN temporal fates.
We examined the PN composition of the PN-only spdo mutant lAL NB clones. We selectively marked the 12 types of mPNs, born in multiple clusters from 18 to 96 h ALH, with GAL4-GH146. We further checked distinct populations of AMMC neurons using GAL4-GR20C03 and GAL4-GR72G12. We found that the spdo mutant lAL NB clones, labeled with any of the three GAL4 drivers, show wild-type morphologies but carry two times more cell bodies (). These observations indicate a perfect duplication of the PN hemilineage in the spdo mutant clones and suggest that the transformed PN hemilineage undergoes the same temporal identity changes as the native PN hemilineage does. Such results argue that the differential Notch signaling not only promotes the LN or PN fate but also governs the differential manifestation of temporal identity changes in the LN versus PN hemilineage.
Complete duplication of the PN hemilineage in the absence of Spdo.
Notch Specifies the AMMC Fates in the Notch-Low Projection Neuron Hemilineage through a spdo-Independent Pathway
In contrast with the faithful duplication of diverse PNs in the spdo
mutant clones, the lAL NB clones homozygous for mutations in notch
or its co-activator Su(H)
exhibited abnormal PN compositions. Labeling entire clones with nSyb-GAL4
revealed missing of the AMMC neurite tracks specifically in the notch
mutant NB clones (, compared to ). There was no evidence for cell loss, given that we consistently counted around 200 cell bodies regardless of the clone genotype. To exclude changes in the pattern of lAL neurogenesis, we further determined the rate of proliferation at 30 h ALH when the lAL NB mainly produces mPNs and at 70 h ALH when the AMMC neurons are made. The sizes of wild-type and Su(H)
clones were comparable. Moreover, they carried analogous numbers of mitotic cells (Table S2
) as revealed with the mitosis marker phospho-histone H3 (PH3) 
. So the PN-only notch
mutant clones have made GMCs that yield viable neurons in correct numbers and at right timings, making us wonder if the prospective AMMC neurons have adopted other PN fates and acquired non-AMMC neurite trajectories. Given the prominence of AL neuronal elaborations in those clones lacking AMMC trajectories, we examined if the notch
mutant NB clones carry many more AL PNs at the expense of AMMC neurons. We found that notch
mutant lAL NB clones contain about five times more GH146-positive AL neurons than wild-type controls (, compared to ). A three times increase in the numbers of the later-born DA1, DL3, VM1, DA2, and DM5 mPNs, visualized with GAL4-GR83D12
, was also observed in Su(H)
mutant clones (). Note the exclusive dense innervation of the DA1, DL3, VM1, DA2, and DM5 glomeruli by the much enlarged Su(H)
mutant clones (), indicating an excessive production of normal-looking AL PNs by the lAL NB deficit in notch
. These observations suggest that the prospective AMMC neurons of notch
mutant clones might have aberrantly adopted the AL PN fates characteristic of siblings born at different times, reminiscent of some temporal cell fate transformation.
The specification of AMMC PNs requires a Spdo-independent Notch activity.
The majority of AMMC neurons are born after 60 h ALH (). If the prospective AMMC neurons had been transformed into AL PNs, one would expect that the supernumerary AL PNs were largely added during the second half of the lAL lineage. To verify this viewpoint, we examined when the GH146-positive mPNs were made in excess by the notch mutant lAL NB. We fed the larvae harboring GAL4-GH146-labeled wild-type or notch clones with EdU, a thymidine analog that labels proliferating cells, for 1 d at 0–24, 24–48, or 48–72 h ALH (). The pulse labeling of EdU first confirmed that the GH146-positive mPNs were mostly generated between 24 and 72 h ALH (). It further revealed that the majority of the excessive GH146-positive neurons in the notch mutant lAL NB clones were born after 48 h ALH when the prospective AMMC neurons were supposed to arise. Compared to wild-type controls, notch mutant clones yielded two times more GH146-positive neurons at 24–48 h ALH and up to four times more at 48–72 h ALH (). This increase was not due to an acceleration of NB proliferation, because the total numbers of the EdU-positive cells on the lateral side of the AL remained comparable to those of the wild-type controls (). And the 4-fold increase at 48–72 h ALH cannot be fully accounted for by the LN-to-PN fate changes. It argues instead that, on top of the binary cell fate transformation, most, if not all, of the PNs yielded during that period, including those that normally adopt the AMMC neuronal fates, have uniformly developed into GH146-positive mPNs.
In sum, Notch signaling underlies the specification of AMMC versus AL neurons in the Notch-low PN hemilineage. Interestingly, the positive regulator of Notch, Spdo, is essential for the binary cell fate decision between LNs and PNs but dispensable for the temporal fate specification of the AMMC versus AL PN fates.
Analogous Dynamic Chinmo Expression Governs PN and LN Temporal Fates in Hemilineage-Dependent Manners
Notch might regulate neuronal temporal cell fates through refining temporal codes or modulating postmitotic neurons' responses to Notch-independent transcriptional cascades. Chinmo and Br-C are dynamically expressed during larval neurogenesis 
. We wondered if such dynamic gene expressions exist in the developing lAL lineage and whether these temporal signatures vary depending on Notch activities. Consistent with previous reports 
, we could reliably detect a sequential birth-order-dependent expression of Chinmo and Br-C in the neuronal offspring of most, if not all, larval brain NBs. Chinmo preceded Br-C in the partially overlapping temporal gene expression, such that Chinmo(+)/Br-C(−) neurons consistently reside deeper in the cell body layer than their Chinmo(−)/Br-C(+) siblings (). We quantified the lAL offspring positive for Chinmo and/or Br-C at 70 h ALH when many AMMC precursors should already exist. We obtained comparable numbers of Chinmo(+)/Br-C(−), Chinmo(+)/Br-C(+), and Chinmo(−)/Br-C(+) neurons in the lAL NB clones regardless of the genotype of spdo
(). We conclude that the Chinmo→Br-C temporal expression takes place analogously in both PN and LN hemilineages and independently of Notch activities.
The temporal transition from Chinmo to Br-C is not affected by loss of Notch signaling.
We further examined the involvement of Chinmo in specifying neuronal temporal fates of PNs versus LNs. Using GAL4-GH146 to monitor the orderly production of the 12 types of mPNs with ts-MARCM, we demonstrated the requirement of Chinmo for proper specification of the VC2, VC1, DM1, and DM2 temporal fates (). All of them have aberrantly adopted the VA5 temporal fate following loss of Chinmo from respective GMCs, as evidenced by their targeting of the VA5 glomerulus and the branching of axons reminiscent of the wild-type VA5 mPNs ( for VC2; unpublished data for VC1, DM1, and DM2). We then examined Chinmo's requirement for their twin LNs. We created mutant LNs paired with wild-type PNs as isolated two-cell clones. Based on AL elaboration patterns, the chinmo mutant LN of the VC2 mPNs (LN2) has adopted the fate of LN3 (the twin LN of the VC1 mPN) rather than the fate of LN6 (the twin LN of the VA5 mPN) (). Compared to the wild-type LN2 innervating near all AL glomeruli (), the prospective LN2 homozygous for chinmo acquired a much more restricted pattern of neurite elaboration and resembled the next-born LN3 (). The transformed LN2 appears distinct from LN6 that normally pairs with the VA5 mPN (), although the chinmo−/− VC2 mPN has consistently adopted the VA5 mPN fate (). We did not observe chinmo-related temporal identity phenotypes for other LNs examined so far.
The PN and the LN hemilineages respond to chinmo loss-of-function differently.
Taken together, we identified chinmo as a temporal fating factor in the lAL lineage. Notably, the Notch-independent dynamic expression of Chinmo governs LN and PN temporal fates in hemilineage-specific (i.e., Notch-dependent) manners, arguing that Notch acts in parallel with or downstream of temporal fating factors to determine terminal temporal fates ().