Multiple lineages enable robust development of the neuropil-glia architecture in adult Drosophila

Across a wide range of species, neural remodeling is essential for the development of a functional nervous system (Luo and O'Leary, 2005). Thus, it is of great interest to understand the cellular and molecular mechanisms underlying the remodeling of brains. To date, neurons have been the main focus of studies to address this issue. However, apart from neurons (the key components of neural circuits), nervous systems also contain glial cells, which are indispensable for processing information and controlling animal behavior. However, little is known of how glial cells reorganize themselves in coordination with neuronal remodeling.

Given the abundance of powerful genetic tools available to address mechanistic underpinnings, the fruit fly Drosophila melanogaster is an excellent model organism for studying the remodeling of neural circuits (Awasaki et al., 2011; Yaniv and Schuldiner, 2016). The lifestyle of the fruit fly changes drastically (e.g. from crawling to flying) as it transitions from a larva to an adult. Accordingly, the larval brain undergoes alterations to prepare it for adult life. During metamorphosis, certain mature larval neurons that undergo remodeling and larval-born neurons that mature develop together to form adult neural circuits. Flies have two glial subtypes associated with neuropil, namely astrocyte-like glia (ALG) and ensheathing glia (EG). Similar to mammalian glia, these cells are crucial for neural functions. ALG are required to control various behaviors (e.g. circadian and olfactory) via neurotransmitter metabolism and modulation of neuronal activity (Jackson et al., 2015; Ma et al., 2016; Ryglewski et al., 2017). EG regulate locomotion and sleep through glutamate metabolism and transport of taurine, respectively (Otto et al., 2018; Stacey et al., 2010; Stahl et al., 2018). Similar to mammalian glia, neuropil-glia (both ALG and EG) respond to environmental changes (e.g. disease and injury), and phagocytose and/or proliferate (Doherty et al., 2009; Kato et al., 2009, 2011; Kato et al., 2018; Losada-Perez et al., 2016; MacDonald et al., 2006). Drosophila neuropil-glia are also vital to neural remodeling. Larval ALG lose their fine processes and transform into highly phagocytic cells, engulfing axonal branches to remodel neurons (Awasaki and Ito, 2004; Awasaki et al., 2006; Hakim et al., 2014; Omoto et al., 2015; Tasdemir-Yilmaz and Freeman, 2014).

The neuropil-glial architectures in larval ventral nerve cords (VNC) and brains are composed of a small number of neuropil-glia, generated during the embryonic stage (Omoto et al., 2015; Peco et al., 2016), whereas a large number of neuropil-glia form the glial architecture in the adult VNC (Enriquez et al., 2018) and brain (Awasaki et al., 2008; Kremer et al., 2017). The neuropil-glial architecture appears to change in concert with the remodeling of the brain. Omoto et al. (2015) proposed a model for the generation of the adult neuropil-glia architecture, in which both larval ALG and EG undergo programmed cell death. Others reported that the cell bodies of larval ALG persist during pupal life, and they re-infiltrate their fine process into the neuropil at the late pupal stage (Tasdemir-Yilmaz and Freeman, 2014). Neuropil-glia for an adult brain are generated from a small number of specific larval neuroblasts, termed as type II neuroblasts (Omoto et al., 2015; Ren et al., 2018; Viktorin et al., 2011). However, they are not accountable for the entire architecture of neuropil-glia in an adult brain; the superiormost and the inferior regions of a brain remain uncovered by neuropil-glia generated by type II neuroblasts (Ren et al., 2018). Thus, a broad conceptual view of how the architecture of neuropil-glia undergoes remodeling remains to be elucidated.

We investigated the fate of larval ALG, EG and glial cell missing-positive (gcm⁺) cells, and found that the larval EG dedifferentiate, proliferate and redifferentiate into adult ALG and EG. Together with the gcm⁺ lineage, the larval EG lineage generates the adult neuropil-glia architecture. Finally, to investigate the interaction between the lineages in the development of the neuropil-glial architecture, we evaluated whether one lineage compensates for the failure of gliogenesis in the other lineage.

In both larval and adult brains, cell bodies of neurons form the outer layer of a brain, whereas neuropils are located inside the brain (Fig. 1A). The cell bodies of neuropil-glia are located at the interface between the cell body layer and the neuropil in both larval and adult brains, and also between neuropils in adult brains. The neuropil-glia, consisting of ALG and EG, have been identified based on their unique morphology and distribution patterns revealed in GAL4 lines, such as alrm-GAL4 for ALG, and NP6520 and 56F03-GAL4 for EG (Fig. 1B,D-F,H,I, Fig. S1A-C) (Awasaki et al., 2008; Awasaki and Lee, 2011; Kremer et al., 2017; Omoto et al., 2015; Peco et al., 2016). We also identified another GAL4 line, 83E10-GAL4 (Jenett et al., 2012), for adult ALG (Fig. 1C,G, Fig. S1A). In larval and adult brains, membrane-tethered GFP (mGFP) driven by each GAL4 line showed that ALG project their fine processes into the neuropil, whereas EG wrap the neuropil (Fig. 1F-I). A single-cell labeling technique using the GAL4 lines and a flip-out technique clarified the characteristic morphology of each neuropil-glia subtype in both larval and adult brains (Fig S1D-G). We examined the distribution of established marker proteins in ALG and EG identified in the GAL4 lines to compare the larval and adult neuropil-glia. Glutamine synthetase 2 (Gs2), which is related to neurotransmitter recycling, is a marker for larval ALG and larval EG (Freeman et al., 2003; Thomas and van Meyel, 2007). Ebony, which is also related to neurotransmitter recycling, is a marker for both larva and adult ALG (Suh and Jackson, 2007). Nazgul (Naz), which is involved in tyramine metabolism (Ryglewski et al., 2017), is a marker for larval ALG (Stacey et al., 2010; von Hilchen et al., 2010). We found that Gs2, Ebony and Naz were also present in adult ALG (Fig. 1J,L). Similarly, adult EG were positive for Gs2; however, they were negative for Ebony and Naz (Fig. 1K,M). In summary, neuropil-glia are Gs2 positive (Gs2⁺) and ALG are Gs2⁺ Ebony⁺ Naz⁺⁺, whereas EG are Gs2⁺ Ebony^– Naz^– (Fig. 1N) in both larval and adult brains.

We subsequently investigated the changes to the larval neuropil-glia during the pupal life. Ebony, the ALG marker, disappeared during the early pupal stage and reappeared late in the pupal stage, immediately before eclosion (Fig. 2A,B). Similarly, the expression of the ALG- and EG-GAL4 lines (alrm-GAL4 and NP6520, respectively) ceased by the 25% pupa stage (Fig. S1A,B). Both GAL4 lines gradually resumed their expression from the 50% pupa stage, whereas the expression of 83E10-GAL4, an adult ALG-specific GAL4, was observed at the 100% pupa stage, before eclosion (Fig. S1A-C). The neuropil-glia marker Gs2 and the ALG marker Naz showed loss of ALG processes in the neuropil; however, positive cells continued to be detected on the surface of the neuropil at the 25% pupa stage (Fig. 2C,D). At this stage, Gs2 was distributed in the medial area of brains including the gnathal ganglion (Fig. 2C,E). On the other hand, the distribution of Naz was more obvious at superior neuropils and the ventrolateral neuropil area (Fig. 2D,F; white arrowheads), whereas only faint Naz⁺ cells were found at the gnathal ganglion area (Fig. 2D,F; yellow arrowheads). After the 50% pupa stage, Gs2⁺ and Naz⁺⁺ cells were found throughout the entire surface of the neuropil (Fig. 2C-F). These observations indicate that larval ALG and EG undergo marked changes in their features during metamorphosis.

To examine the fate of the larval neuropil-glia, we first traced larval ALG in the central brain using Histone::YFP, which localizes in the nuclei and allows us to identify the cells for a longer time than mGFP. This analysis suggested that larval ALG persist during the mid-pupal stage without increasing in numbers (Fig. 3A-C). The flip-out-based lineage tracing using mGFP (Fig. S2A), in which the expression of larval GAL4 was converted into the constant expression of actin-lexA, showed that larval ALG disappeared before eclosion (Fig. 3D). These results are consistent with those previously reported by Omoto et al. (Omoto et al., 2015).

In contrast to larval ALG, cells derived from larval EG (L-EG) persist into adulthood. The number of Histone::YFP-positive cells driven by larval NP6520 increased in the early pupal stage (Fig. 3C,E,F). This population contained cells labeled with phospho-Histone H3 (PH3), a marker for the mitotic phase, in which cytoplasm and duplicated DNA divides. This indicated that at least some larval EG undergo cell division (on average 2.2 cells of PH3⁺ NP6520⁺, s.d. 2.0/hemisphere, n=14 hemispheres from seven brains, 0-8% pupal stage) (Fig. 3G,G′). It has been suggested that larval EG are polyploid (Omoto et al., 2015). We measured the intensities of DAPI-stained DNA from larval EG, subperineurial glia and neuroblasts, and normalized each value to that from neurons nearby. Our analysis demonstrated that the DNA content of a few EG at the brain midline is comparable with that of polyploid subperineurial glia (SPG) (Fig. S3), whereas the DNA content of the other larval EG is similar to that of neuroblasts, which proliferate in the larval stage (Fig. S3). These results are consistent with our finding that at least some larval EG undergo cell division.

At the 25% pupa stage, mGFP, driven by NP6520, had almost disappeared (Fig. 3H,H′), whereas lineage-tracing experiments with NP6520 showed mGFP-positive cells (Fig. 3I,I′). The L-EG lineage cells persisted in the adult, occupying the anterior-inferior area (including the inferior neuropils, antennal lobe and gnathal ganglia) (Fig. 3J). The L-EG lineage cells were Gs2⁺ and faintly Naz⁺ at the 25% pupa stage (Fig. 3K), showing that these cells have a different profile versus original L-EG (Fig. 1M). At the 50% pupa stage and the adult-stage, L-EG lineage cells were Gs2⁺ and either Naz⁺⁺ or Naz^– (Fig. 3L,M). This indicates that they become ALG and EG. Single-cell analysis with subtype-specific GAL4 lines in pupal life demonstrated that adult neuropil-glia begin changing their morphology from the 50% pupa stage and gradually develop their characteristic morphology for adult brains (Fig. S1F,G). It is important to note that L-EG is differentiated and functional for larval behavior (Otto et al., 2018; Stacey et al., 2010; Stahl et al., 2018). Our results demonstrate that the L-EG change their character and proliferate during the early pupal stage, increase in number and differentiate into adult EG and ALG from ∼50% pupa stage. These results suggest that L-EG dedifferentiate, proliferate and redifferentiate into adult EG and ALG.

Although the L-EG give rise to the adult neuropil-glia, L-EG-derived cells did not make up the entire neuropil-glia architecture (Fig. 3J). Where do the remaining adult neuropil-glia originate from? Given that type II neuroblasts have been shown to generate gcm⁺ cells, which become adult neuropil-glia (Omoto et al., 2015; Ren et al., 2018; Viktorin et al., 2011), we examined their contribution to the adult neuropil-glia architecture. In wandering larvae, cells labeled with gcm-GAL4 contained a small group of neurons but were mainly Repo⁺ glial cells (Fig. 4A). Repo⁺ gcm⁺ glial cells were located along the surface of the neuropil and in the central complex area (Fig. 4A), and were Gs2^– and Naz^– (Fig. 4B). This indicates that gcm⁺ glial cells are a distinct population from larval ALG and EG. The expression of GFP driven by gcm-GAL4 was reduced at the 25% pupa stage compared with that observed at the wandering larvae stage (Fig. 4D,E). However, lineage tracing using mGFP showed that cells of this lineage appeared to increase in number by this stage (compare Fig. 4E′ with F). Consistent with this, gcm⁺ cells contained PH3-positive cells (on average 3.7 cells of PH3⁺ gcm-G4⁺, s.d. 2.1/hemisphere, n=14 hemispheres from seven brains, 0-8% pupa-stage) (Fig. 4G,G′), indicating that at least some of them undergo cell division. The gcm⁺ lineage cells persisted into adulthood and occupied the posterior-superior area (including the superior neuropils, ventrolateral neuropils and central complex) (Fig. 4H). We confirmed this using another GAL4 line (81B02-GAL4). This GAL4 line showed expression in larval gcm⁺ cells but less in neurons than did gcm-GAL4 (Fig. 4C), making it easier to observe the glial pattern (Fig. 4I). At the 25% pupa stage, the gcm⁺ lineage cells became weak Naz⁺ cells (Fig. 4J). At the 50% pupa stage and the adult stage, gcm⁺ lineage cells were Gs2⁺ and either Naz⁺⁺ or Naz^–, indicating that they become ALG and EG (Fig. 4K,L; Table 1). Together with the lineage tracing of both GAL4 lines, our data showed that the adult neuropil-glia (both ALG and EG) derived from gcm⁺ cells localized at the area that appeared roughly complementary to the area occupied by L-EG lineage cells (compare Figs 3J and 4H,I,M).

To examine whether the entire neuropil-glia architecture is generated by the two lineages, we quantified the number of cells derived from each lineage in four different brain areas: superior neuropil, central complex, antennal lobe and gnathal ganglia (Fig. 4N). The quantification was performed using another lineage-tracing technique (Fig. S2B) with cytoplasmic GFP at the 50% pupa stage, as the cells are larger and easier to count compared with the adult counterpart.

Our analysis revealed that each examined area comprised neuropil-glia derived from both lineages (Fig. 4N). In the antennal lobe area, the sum of the average number of traced cells in each lineage was slightly higher than the average of the total number of neuropil-glia (Repo⁺ cells at the surface of the neuropil), implying that some cells might be labeled by both GAL4 lines. In the other examined areas, the sums of the average number of traced cells in each lineage were slightly lower than the average of the total number of neuropil-glia (Fig. 4N). This may be caused by technical difficulties in labeling the whole population of lineages, which results in unlabeled cells (escaper cells). It is also possible that another population is involved in generating adult neuropil-glia. Nevertheless, our analysis suggests that the L-EG and gcm⁺ lineages are the primary source of adult neuropil-glia.

Our analysis also demonstrated that the distribution pattern of each lineage was not bilaterally symmetrical, suggesting that the boundaries between the lineages vary (Figs 3J and 4H,I,M). These observations imply that each lineage is not predetermined to generate neuropil-glia for specific neuropils.

To investigate the role of cell interaction in the development of adult neuropil-glia architecture, we manipulated gliogenesis in both of the lineages; we first examined whether homeoprotein Pros is required for the generation of adult ALG. It has been established that in embryonic and larval ventral nerve cords, Pros plays roles in the differentiation of interface glia (also termed Pros⁺ longitudinal glia in embryos and ALG in larvae) and maintain their proliferative ability (Griffiths and Hidalgo, 2004; Kato et al., 2011; Peco et al., 2016). Pros was detected in Naz⁺⁺ ALG in wandering larvae, in pupae at the 50% pupa stage and in adult flies; however, it was not detected in weak/faint Naz⁺ cells at the 25% pupa stage (Fig. 5A). Pan-glial expression of two different pros miRNAs during the pupal stage, using repo-GAL4 and temperature-sensitive tub-GAL80 (tubGAL80^ts), resulted in severe defects in the generation of adult ALG (Fig. 5B). Instead of Naz⁺⁺ cells, the adult brains had weak Naz⁺ cells with few fine processes projecting into the neuropil (Fig. 5C). These results suggest that Pros is involved in the differentiation of adult ALG.

To investigate the role of Pros in the lineages, we developed a gene knockdown (KD) system by modifying the flip-out lineage-tracing system, in which manipulated cells were labeled with cytoplasmic GFP (Fig. S2B). We examined the distribution of GFP, Repo and Naz, and quantified Naz⁺⁺/Naz⁺/Naz^– cells among GFP⁺/GFP^– Repo⁺ cells. These cells were counted on the surface of the neuropil (neuropil-glia), in the superior neuropil, antennal lobes, and gnathal ganglia (the areas indicated in Fig. 4N except for the central complex, which is in the center of brains and in which it is difficult to identify weak Naz⁺ cells). The Naz⁺⁺ ALG and Naz⁺ cells were distinguished based on their staining intensity. Because of the difficulties in labeling all of the lineage cells using this technique, there were always cells that escaped being labeled by the lineage tracer. The Naz⁺⁺ ALG derived from such escapers and/or unmanipulated lineages were used as an internal control to classify Naz⁺⁺ and Naz⁺ cells.

At the 50% pupa stage, the distribution of GFP⁺ cells in pros KD flies in both lineages appeared essentially normal (Fig. 6A,B), with a slight decrease in GFP⁺ cell numbers in the superior neuropil and the antennal lobe area (Fig. 6G). In contrast, the distribution of Naz⁺⁺ cells changed markedly in flies with pros KD of both lineages (compare Fig. 6A′ with B′). In all examined areas, the number of GFP⁺ Naz⁺⁺ cells decreased, whereas that of GFP⁺ weak Naz⁺ cells increased (Fig. 6H). Continuous expression of pros miRNA into adulthood yielded similar results: prominent distribution of weak Naz⁺ cells and fewer Naz⁺⁺ cells (compare Figs 5B left panel and Fig. 6C; Fig. 6M; Fig. S4A). Moreover, adult brains had fewer Ebony⁺ cells following the KD of pros in both lineages (Fig. 6N, Fig. S4B). Altogether, our findings suggest that Pros is required for the differentiation of weak Naz⁺ cells into adult ALG at approximately the mid-pupal stage.

To investigate the interaction between the lineages, we examined the distribution of ALG following failure of one lineage to develop ALG. At the 50% pupa stage, KD of pros in either lineage resulted in brains with an area devoid of Naz⁺⁺ cells, which was the area where GFP⁺ cells were distributed (Fig. 6D-E′). Continuous expression of pros miRNA into adulthood led to similar results, i.e. an area devoid of Naz⁺⁺ ALG (Fig. 6F). This was further supported by the quantification analysis. In the affected area, KD of pros resulted in a reduction in Naz⁺⁺ cells, and an increase in the number of weak Naz⁺ cells at the 50% pupa stage (Fig. 6J).

In some cases, pros KD resulted in fewer number of GFP⁺ cells than observed in the control (Fig. 6G, graph of superior neuropil and antennal lobe; and Fig. 6I, graph of superior neuropil). However, the number of neuropil-glia (sum of GFP⁺ Repo⁺ and GFP^– Repo⁺ cells, including Naz⁺⁺, Naz⁺ and Naz^– cells) was rescued to normal by the increase in GFP^– cells. This result indicates that the number of neuropil-glia can be compensated for by the intact lineage. The fact that the total number of Naz⁺⁺ ALG was not rescued to normal (Fig. 6J) implies that the number of neuropil-glia, including Naz⁺ cells, in an area is regulated. Collectively, our data suggest that, in pros KD flies, weak Naz⁺ cells fail to differentiate into Naz⁺⁺ cells and persist in the adult brains, and the insufficient numbers of Naz⁺⁺ cells are not compensated for.

The pros KD experiments in the lineages showed there was some degree of compensation for the total number of neuropil-glia. To further examine the compensatory capacity of the lineages, we first investigated whether the FGF receptor Heartless (Htl) is required for the proliferation of the lineages. It has been established that Htl is required in wide-ranging gliogenesis, which includes the proliferation of larval cortex glia and wrapping glia in eye disks (Avet-Rochex et al., 2012; Franzd; tir et al., 2009; Muha and Müller, 2013). We found that pan-glial expression of two different htl miRNAs during the pupal stage, using repo-GAL4 and tubPGAL80^ts, resulted in fewer Naz⁺⁺ ALG throughout the brain with both miRNAs (Fig. 7A-D).

Consistent with this, the expression of htl-GFP [driven from a fosmid insertion that contains the genomic region of the htl gene (Sarov et al., 2016)] was found in cells on the surface of the neuropil (Fig. 7E). At the wandering larval stage and 8% pupa stage, htl-GFP was expressed in various cells, including Repo⁺ gcm-lacZ⁺ cells (i.e. gcm⁺ cells) and Gs2⁺ glial cells around the gnathal ganglia (i.e. L-EG lineage), respectively (Fig. 7F,G). Subsequently, the expression became prominent in neuropil-glia; the expression of Htl-GFP at the 25% and 50% pupa stage was detected in Naz⁺ cells and Gs2⁺ Naz⁺⁺ cells, respectively (Fig. 7E,G). These results suggest that Htl is involved in the generation of adult neuropil-glia.

We subsequently examined the role of Htl in the lineages. There appeared to be fewer manipulated lineage cells in various brain regions following the KD of htl in both lineages at the 50% pupa stage (Figs 6A and 8A,C). In contrast to control flies, there was no increase in the number of lineage cells (GFP⁺ Repo⁺) in the early pupal stage (0-8% pupa stage) in flies with htl KD in the gcm⁺ lineage (Fig. 7H, Fig. S5A). Furthermore, expression of a constitutively active form of Htl in the gcm⁺ lineage from the mid-larval stage resulted in excess Naz⁺⁺ cells at the 50% pupa stage (Fig. 7I, Fig. S5B). Consistently, the number of GFP⁺ PH3⁺ cells exhibited a tendency to decrease in the gcm⁺ lineage cells with htl KD, whereas it was significantly increased in the gcm⁺ lineage cells with a constitutively active (CA) htl  in the brains from 0% to 8% pupa stage (Fig. 7J-L, Fig. S5C). These results suggest that Htl promotes glial proliferation in the early pupal stage of both lineages.

Finally, we investigated the interaction between the lineages in terms of glial proliferation. In flies with htl KD at the 50% pupa stage of either lineage, the net distribution of Naz⁺⁺ cells appeared normal, despite the severe deficiency in the distribution of GFP⁺ cells (Figs 6D,D′ and 8B,B′,D,E). Quantification showed that, although the number of GFP⁺ cells was severely decreased, the number of Repo⁺ neuropil-glia and Repo⁺ Naz⁺⁺ ALG among them did not differ from the control in most cases (Fig. 8F-H). This was attributed to the increase in GFP^– Repo⁺ neuropil-glia and GFP^– Naz⁺⁺ ALG (Fig. 8D,F-H), suggesting that the intact lineage (i.e. GFP^– cells) underwent compensatory proliferation. In contrast, htl KD in both lineages produced fewer neuropil-glia and Naz⁺⁺ cells in total (Fig. 8I), and resulted in the abnormal distribution of Naz⁺⁺ cells (Fig. 8A′,C). Although there are GFP^– cells, which might be escapers, or another lineage that is neither L-EG nor the gcm⁺ lineage (Fig. 8I), this result indicates that the L-EG and gcm⁺ lineages are the primary sources of adult neuropil-glia. Collectively, these results suggest that insufficient cell number is rescued by an increase in the number of intact cells by the 50% pupa stage, and subsequently the intact cells differentiate into adult ALG and EG. This leads to the robust development of adult neuropil-glia architecture, thereby ensuring a functional brain.

We demonstrate that the adult architecture of neuropil-glia is formed from two lineages: the differentiated larval EG lineage and the gcm⁺ lineage. Both lineages require Htl for proliferation and Pros for differentiation of ALG. Each lineage compensates for the failure of the other to proliferate. Thus, the architecture of adult neuropil-glia develops robustly to ensure a functional adult brain (Fig. 8J).

Previous studies have suggested that the adult neuropil-glia are derived from larval neuroblasts, and larval neuropil-glia (both L-EG and L-ALG) undergo programmed cell death during metamorphosis (Omoto et al., 2015). Given that neuroblasts give rise to gcm⁺ cells, which then generate mature glial cells, we traced the fate of gcm⁺ cells and showed that they generate adult neuropil-glia. Although this result is mostly consistent with previous reports, the area occupied by adult neuropil-glia derived from gcm⁺ cells was larger than that occupied by type II neuroblast-derived glia, as reported by Ren et al. (2018). This difference may be attributed to the involvement of type I lineage cells as previously suggested by Yu et al. (2013), although Omoto et al. (2015) indicated otherwise. Nevertheless, our results suggest that type II neuroblasts are not the sole origin of gcm⁺ cells that generate adult neuropil-glia. Furthermore, we demonstrate that L-EG also participates in the genesis of adult neuropil-glia. Collectively, our study demonstrates that adult neuropil-glia are generated from gcm⁺ cells and L-EG.

The L-EG and gcm⁺ lineages undergo proliferation at the early pupal stage to generate the architecture of neuropil-glia in the adult, which is more complex and has 100-fold more glial cells than the larva (Kremer et al., 2017; Omoto et al., 2015). Adult flies process a vast amount of sensory information and exhibit complex behaviors, such as courtship, aggression, flight and walking. Accordingly, the structure of the adult brain is more elaborate, with more subdivided neuropils and 20-fold more neurons than the larval brain (Simpson, 2009; Spindler and Hartenstein, 2010). Thus, the cell proliferation of both lineages leads to an increase in the number of glial cells, which likely occurs in coordination with the elaboration of adult neural circuits.

Neuron-glia interactions underlie the adjustment of glial cell numbers to neuronal structure through cell survival or cell proliferation in flies and vertebrates (Blaschuk and ffrench-Constant, 1998; Colombo et al., 2018; Griffiths and Hidalgo, 2004; Hidalgo and ffrench-Constant, 2003; Hidalgo et al., 2001; Wang et al., 1998). We showed that the FGF receptor Htl was required for cell proliferation in both L-EG and gcm⁺ lineages in early pupal life. In flies, the Htl ligand Pyramus, which is secreted from neurons, regulates the proliferation of Htl-positive cortex glia during the larval stage (Avet-Rochex et al., 2012). Similarly, it is possible that Htl ligands from neurons non-cell autonomously regulate the proliferation of lineage cells. Consistent with this notion, our data show that the total number of neuropil-glia in an area is limited, unless Htl is constitutively activated. Thus, such non-cell autonomous regulation may adjust the numbers of neuropil-glia in adult neural circuits, thereby enabling the complex behavior of adult animals.

ALG and EG were present in both larval and adult brains, and each cell type shares morphological features and the expression of certain markers between stages. Our data show the similarities in the developmental program of neuropil-glia for embryos/larvae and adults. We ascertain that Pros is required for the differentiation of adult ALG, as it is for the development of embryonic/larval ALG (Griffiths and Hidalgo, 2004; Kato et al., 2011; Peco et al., 2016). In embryos and larvae, Pros is also required to maintain the proliferative ability of ALG (Griffiths and Hidalgo, 2004; Kato et al., 2011). We show that the KD of pros in the lineages results in fewer GFP⁺ neuropil-glia at the 50% pupa stage in some areas of the brain. This implies that Pros is involved in the regulation of cell number in the development of adult neuropil-glia. The KD of pros in the cell lineages results in the appearance of Naz⁺ cells. The exact identity of these cells remains unknown. Rather than differentiating into adult ALG, the persisting weak Naz⁺ cells in adult brains may have acquired EG-like characteristics. However, it is difficult to assess this possibility because of the lack of markers that can clearly identify adult EG. Alternatively, they may be undifferentiated cells that have failed to differentiate into adult ALG. This notion is consistent with the fact that EG are Naz^– and the undifferentiated cells at the 25% pupa stage are weak/faint Naz⁺ cells.

We have established that Htl is required for the proliferation of the L-EG and gcm⁺ lineages. In the development of embryonic/larval neuropil-glia, the number of ALG in htl^AB42 null mutants is similar to that in the control, suggesting that Htl is not involved in cell proliferation in embryos (Stork et al., 2014). Instead, Htl is required for the proper organization of IG/ALG during embryogenesis (Shishido et al., 1997), and for extending the fine projections of larval ALG into the neuropil (Stork et al., 2014). In our analysis, the cell-proliferation phenotype of htl KD emerged in the early pupal stage; thus, we did not specifically investigate the role of Htl in later stages. We show that htl is also expressed in neuropil-glia at the 50% pupa stage, when their maturation is initiated. Thus, our results do not rule out the involvement of Htl in the maturation of adult ALG in later pupal stages. Nevertheless, they show that the developmental programs for larva and adult neuropil-glia partially differ.

The plasticity of glial cells in terms of their differentiation is well established in vertebrates. Radial glial progenitors generate neurons and, subsequently, some of them generate oligodendrocytes and astrocytes during development (Anthony et al., 2004; Gao et al., 2014; Kriegstein and Alvarez-Buylla, 2009; Malatesta et al., 2003). Radial glia progenitors persist in adults and transform into neural stem cells (Beattie and Hippenmeyer, 2017; Kriegstein and Alvarez-Buylla, 2009). Both astrocytes and NG2-glia [oligodendrocyte precursor cells (OPCs)] in adult brains proliferate after injury, and generate astrocytes (Buffo et al., 2008) and oligodendrocytes (Kato et al., 2015; McTigue et al., 2001; Nishiyama et al., 2009; Zhu et al., 2011), respectively. In some cases, astrocytes may even transdifferentiate into neurons after injury (Brulet et al., 2017; Duan et al., 2015; Magnusson et al., 2014; Noristani and Perrin, 2016). NG2-glia/OPCs also generate astrocytes in cell culture (Kondo and Raff, 2000), in development (Rivers et al., 2008; Zhu et al., 2008, 2011) and after injury (Hackett et al., 2018; Sellers et al., 2009; Tripathi et al., 2010). Some studies also reported that NG2 glia generate neurons in adult mice (Rivers et al., 2008; Robins et al., 2013) and after injury (Honsa et al., 2012; Tripathi et al., 2010). Drosophila ALG also proliferate in response to injury in larval ventral nerve cords (Kato et al., 2011; Losada-Perez et al., 2016). However, whether they differentiate into different glial subtypes or neurons is currently unexplored. Foo et al. (2017) reported the presence of adult neural progenitor cells in Drosophila that can generate glial cells and neurons in response to a defect in glial cells. In contrast, the changes of L-EG into the progenitor state, in which cells proliferate and then differentiate, are developmentally regulated. Thus, it may serve as an excellent model for the investigation of glial cell plasticity.

We showed that adult neuropil-glia are derived from two lineages: L-EG and gcm⁺. What is the significance of having two lineages to establish the architecture of adult neuropil-glia? The peculiar distribution patterns of the lineages may relate to the evolution of insect brains. In numerous hemimetabolous insects (e.g. locusts and cockroaches), the mandibular, maxillary and labial ganglia, which are mostly occupied by L-EG lineage cells in flies, are detached from the protocerebrum, deutocerebrum and tritocerebrum, and located more inferiorly to (i.e. below) the esophagus (Ito et al., 2014). In these insects, L-EG may generate neuropil-glia for the mandibular, maxillary and labial ganglia, whereas gcm⁺ cells may generate neuropil-glia for the protocerebrum, deutocerebrum and tritocerebrum. In contrast, in flies, the two different populations appear to generate neuropil-glia for one structure (i.e. a brain) as all of the areas are fused together. This notion is consistent with the idea that the segmental distribution pattern of specific embryonic neuroblasts is evolutionarily conserved between Drosophila and hemimetabolous insects (Scholtz and Edgecombe, 2006; Urbach and Technau, 2003).

We demonstrate that inhibition of glial proliferation in one lineage is rescued by the other lineage. This indicates that, regardless of evolutionary relevance, the multiple lineages (i.e. L-EG and gcm⁺ cells) ensure robust development of the adult neuropil-glia architecture. Such robust development of glial architecture has been reported in several contexts. In the thorax and brain, neuroblasts generate adult neuropil-glia and compensate for the failure of gliogenesis from other neuroblasts (Enriquez et al., 2018; Ren et al., 2018). We reveal that the ability to compensate for deficiencies in a lineage is not restricted to neuroblast-derived glia and is greater in scope. Each lineage (L-EG or gcm⁺) rescues the entire loss of the other lineage. A similar mechanism is involved in the development of mouse oligodendrocytes. Two lineages of oligodendrocyte precursor cells (a ventral and a dorsal population) generate oligodendrocytes in embryos and in postnatal animals. One lineage completely takes over the brain when the other fails to develop, preventing any locomotor defect (Kessaris et al., 2006). Therefore, multiple lineages with glial ability to adjust to the surroundings guarantee the robust development of glial architecture. Thus, we suggest that glial plasticity may be a widespread strategy for ensuring the robust development of functional brains.

The flies (Drosophila melanogaster) were maintained at 25°C with a standard medium, unless otherwise indicated. The alrm-GAL4 flies (Doherty et al., 2009) were a gift from M. Freeman (Oregon Health and Science University, OR, USA); UAS-histone::YFP flies were a gift from A. Hidalgo (University of Birmingham, UK); repo-GAL4 (Sepp et al., 2001), AY-GAL4 (Actin5C-FRT-yellow-FRT-GAL4), UAS-GFP(T2) (Ito et al., 1997), UAS-DsRedS197(C6) (Verkhusha et al., 2001), hs-FLP¹²²;; UAS-FRT-CD2-FRT:mCD8::GFP (GF51) (Wong et al., 2002) and NP6520 (Awasaki et al., 2008) flies were gifts from K. Ito (University of Cologne, Germany); 10xUAS-FLP (Awasaki et al., 2014), tubGAL80^tsx2 (McGuire et al., 2003), Act5C-FRT-stop-FRT-lexA (Harris et al., 2015), lexAop-myr-GFP (Pfeiffer et al., 2010) and gcm-GAL4 (Chotard et al., 2005) flies were gifts from T. Lee (Janelia Research Campus, Howard Hughes Medical Institute, VA, USA); tubGAL80^ts(2) (#7017), 56F03-GAL4 (#39157) (Peco et al., 2016 ), 83E10-GAL4 (#40362) (Jenett et al., 2012), 81B02-GAL4 (#40101), w¹¹¹⁸, UAS-mCD8::GFP(LL5) (#108068), UAS-RedStinger (#8546), UAS-htl-lambdaM (CA) (#5367) (Reich et al., 1999) and gcm^rA87 (gcm-LacZ, #5445) (Giangrande et al., 1993) flies were obtained from the Bloomington Stock Center. UAS-prospero-TRiP (JF02308 and HMJ02107) and UAS-htl-TRiP (HMS01437 and HMJ22375) flies were obtained from NIG-Fly (Japan). htl-GFP flies (#318120) (Sarov et al., 2016; Wu et al., 2017) were obtained from the Vienna Drosophila Resource Center. alrm-GAL4(II) wx5 and UAS-histone::YFP wx5 flies were generated by backcrossing with w¹¹¹⁸ five times. In the results and figures, we refer to alrm-GAL4(II) wx5 and UAS-histone::YFP wx5 flies as alrm-GAL4 and UAS-histone::YFP, respectively. 83E10-GAL4 and 81B02-GAL4 were identified as an adult ALG- and a larval gcm⁺- GAL4 driver, respectively, using the expression pattern database of GAL4 drivers created by the Janelia research campus (flweb.janelia.org/cgi-bin/flew.cgi).

The rate of development at 18°C was estimated as half of that at 25°C. Developmental rates at 30°C and 25°C were very similar and were regarded as the same, although the rate was slightly faster at 30°C. To classify the specific developmental stages regardless of temperature, pupal development was indicated by percentage, which is based on the rough estimation of pupal duration (Fig. 2A). The flies were allowed to lay eggs for 6 h at 25°C or at 18°C depending on the experiment being carried out. To stage pupal life, white pupae were collected as pupae at the 0% pupa stage (0 h after puparium formation, APF). To stage adult flies, newly eclosed flies were collected as flies at day 0.

The dissected brains were fixed with 4% formaldehyde (Polysciences) in PEM [100 mM PIPES, 2 mM EGTA, 1 mM MgSO₄ (pH 7)] or 4% paraformaldehyde phosphate buffer solution (FUJIFILM Wako, 161-20141) for 50 min at room temperature. After washing with 0.5% Triton X-100 in PBS, brains were blocked with 10% normal goat serum in PBS with 1% Triton X-100 for 30 min then incubated at 4°C overnight with primary antibodies or fluorescent-conjugated secondary antibodies with 1 µg/ml DAPI (Sigma). Brains were mounted with SlowFade antifade reagent (Thermo Fisher Scientific). The antibodies used in this study were: rabbit anti-Nazgul 1:1000 (a gift from B. Altenhein, University of Cologne, Germany) (von Hilchen et al., 2010), mouse anti-Pros 1:250 (Developmental Studies Hybridoma Bank, MR1A), chick anti-GFP 1:2000 (Aves, GFP-1020), mouse anti-Repo 1:100 (Developmental Studies Hybridoma Bank, 8D12); rabbit anti-Ebony 1:250 (a gift from S. Carroll, University of Maryland, MD, USA); mouse anti-GS 1:1000 (Millipore, Mab302), rabbit anti-beta-galactosidase 1:1000 (Cappel, 35976), rabbit anti-phospho-histone H3 1:250 (Upstate, 06-570) and anti-mouse Alexa Fluor 405- and anti-rabbit Alexa Fluor 647-conjugated secondary antibodies 1:500 (Invitrogen, A-31553 and A-21245, respectively); and anti-chicken Alexa Fluor 488- and anti-mouse Cy3-conjugated secondary antibodies 1:500 (Jackson ImmunoResearch, 103-545-155 and 115-165-166, respectively). Samples were scanned with an Olympus FV1000 confocal microscope. The images were processed with Fiji (Schindelin et al., 2012) and Photoshop (Adobe). The total pixel intensity density of DAPI staining was quantified as described in the supplementary Materials and Methods.

To genetically label neuropil-glia subtypes, UAS-mCD8::GFP, UAS-DsRedS197, UAS-RedStinger and UAS-histone::YFP wx5 strains were crossed with alrm-GAL4(II) wx5, 83E10-GAL4, 56F03-GAL4, NP6520 or gcm-GAL4 lines. We refer to DsRedS197(C6) as RFP.

For labeling single cells by the flip-out labeling technique, the following lines were crossed: hs-FLP¹²²;; UAS-FRT-CD2-FRT:mCD8::GFP (GF51) and alrm-GAL4(II) wx5, 83E10-GAL4 or NP6520. The flies raised at 25°C were heat-shocked at 37°C for 15 min for ALG and 10 min for EG sometime during L2 to the middle of L3 larval stage for the pupa samples, and 1 day before fixing for adult samples. The adult brains were fixed 5-7 days after adult eclosion. The single ALG morphology during pupal life was revealed by stochastic expression of alrm-GAL4(II) wx5.

For lineage-tracing experiments, 10×UAS-FLP; Act5C-FRT-stop-FRT-lexA, lexAop-myr-GFP; tubGAL80ts flies were crossed with the following: alrm-GAL4(II)wx5 flies for larval ALG tracing, NP6520 flies for L-EG tracing and gcm-GAL4 or 81B02-GAL4 flies for gcm+ cell tracing. They were maintained at 18°C, and the eggs were collected every 6 h and kept at 18°C. The F1 larvae were incubated at 30°C from second instar [48 h after larval hatching (ALH) at 18°C, roughly equivalent to 24 h ALH at 25°C] to 96 h ALH at 18°C (equivalent to 76 h ALH at 25°C), roughly 1 day before the wandering stage. Subsequently the larvae were returned to 18°C until 10 days after adult eclosion (which roughly corresponds to 5 days after eclosion at 25°C), after which the brains were fixed (Fig. S2A).

To test the role of Pros and Htl in pupal life, tubPGAL80tsx2; repo-GAL4 flies were crossed with the following: UAS-prospero-TRiP(III) (JF02308) flies, UAS-prospero-TRiP(II) (HMJ02107) flies, UAS-htl-TRiP(II) (HMJ22375) flies and UAS-htl-TRiP(III) (HMS01437) flies. The eggs were collected for 6 h at 18°C, and the flies were placed at 30°C at the third instar stage until the time of dissection. For gene manipulation in the lineages, 10xUAS-FLP; tubPGAL80^tsx2; NP6520 flies, 10xUAS-FLP; tubPGAL80^tsx2; 81B02-GAL4 flies and 10xUAS-FLP; tubPGAL80^tsx2; NP6520, 81B02-GAL4 flies were crossed with the following: AY-GAL4, UAS-GFP(T2) flies, AY-GAL4, UAS-GFP(T2); UAS-prospero-TRiP (JF02308) flies, AY-GAL4, UAS-GFP(T2); UAS-htl-TRiP (HMS0143) flies and AY-GAL4, UAS-htl-lambda.M (CA) flies. After 6 h of egg collection at 18°C, larvae were heat-shocked with 37°C for 30 min at the middle of the first instar, then placed at 30°C until the time of dissection (Fig. S2B).

For automatic whole-brain counting of Histone::YFP-positive cells, driven by alrm-GAL4 or NP6520, the ImageJ/FIJI plug-in DeadEasy Larval Glia (Forero et al., 2012) was used. The region of interest (ROI) was defined as a hemisphere of the region from the area around the brain neuropil to anterior of the group of exit glial cells in the first thoracic segment. Quantification of cells double positive for PH3 and nGFP/RedStinger (driven by gcm-GAL4 and NP6520, respectively)/number of GFP cells (in gcm⁺ lineage experiments) was performed manually using the cell count function in FIJI. The same ROI as used for the Histone::YFP⁺ cell counting was used. Quantification of the gene manipulation experiments in the lineages was carried out manually using the cell count function in FIJI. Counting cells in three dimensions with the combination of three markers, in which one of them was exclusively in cytoplasm, is extremely difficult. Therefore, we manually counted the cells in three optical sections, which were 5 µm apart from each other to ensure the cells do not overlap between sections yet stay in the same area in brains. Central complex areas were the exception because they were not thick enough, and thus; the counts were conducted in single optical sections. To choose the three optical sections, we first chose a middle section, then chose 5 µm-above and -below sections from the middle section. For the quantification in brains from the 50% pupa-stage and adults, the middle sections were chosen based on the following criteria. As a middle section for the superior neuropil and the central complexes area the section where the shape of the central complex appeared to be a pentagon was chosen. For the antennal lobes area, the optical section where the antennal lobes were distinct from other neuropils was chosen. For the gnathal ganglia area, the section where the glial cells associated with the peripheral nerve projecting into the gnathal ganglia was obvious was chosen. The Naz⁺⁺ ALG and Naz⁺ cells were distinguished based on their staining intensity. Because of the difficulties of labeling all of the lineage cells with this technique, there were always escapers. The Naz⁺⁺ ALG derived from such escapers and/or unmanipulated lineage were used as an internal control to classify Naz⁺⁺ and Naz⁺ cells. Using FIJI software, two binarized images with different thresholds for a Naz-staining image (one to classify Naz^– and one to classify Naz⁺⁺) were prepared based on this internal control, and used as a guide to classify Naz⁺⁺, Naz⁺ and Naz^– cells. For the quantification of Naz⁺⁺ cells in flies with Htl activation in the gcm⁺ lineage, only two rows of cells at the surface of neuropil were counted. The quantification in brains of 0% and 8% pupa stages were focused in the area corresponding to the superior neuropil area. The section where the neuropil appears largest was chosen as a middle section. The ROI was defined as a region corresponding to the superior neuropil to the anterior of the group of glial cells located at the boundary to the optic lobe. The number of samples, n, is based on the number of analyzed hemispheres, except for the central complex, where the number of samples is the number of analyzed brains.

Statistical analyses were performed using GraphPad Prism 6 software. The distributions of the numbers were either normal (D'Agostino-Pearson omnibus normality test, P>0.05) or could be assumed to be normal. Samples in experiments shown in Fig. S5C showed significantly different standard deviations (Brown-Forsythe test, P<0.05). Therefore, the multiple comparisons in the experiments were performed using the Kruskal–Wallis test, followed by Dunn's test to compare the most relevant genotypes. The comparisons between two genotypes, the control and the experiment, were performed as follows. When variances were not significantly different according to the F-test, the unpaired two-tailed Student's t-test was performed. When variances were significantly different between the samples, the unpaired t-test with Welch's correction was performed.

We thank B. Altenhein, S. Caroll, M. Freeman, A. Hidalgo, K. Ito, T. Lee, the NIG fly stocks of National Institute of Genetics, the Bloomington Drosophila Stock Center and the Vienna Drosophila Resource Center for reagents; M. Tomura, Y. Umeki and S. Sato for technical assistance; the lab members for support; and K. Oyama and Enago for editing the manuscript.