Introduction

The nervous system contains a diverse ensemble of neuron types, which are generated during development in appropriate numbers under tightly controlled processes. A central question in developmental neurobiology concerns how diverse neuron types acquire unique cell identities with characteristic morphologies, distinct gene expression profiles and specific functionalities. Kenyon cells (KCs) are the intrinsic neurons in the Drosophila mushroom body (MB) [1] and can serve as an excellent model system for investigations into the molecular mechanisms of neuron subtype diversification. Around 2,000 KCs generated by four neuroblasts are grouped into three sequentially born subtypes, including γ, α’/β’ and α/β neurons [1]. Functionally, these KC subtypes play different roles in short-term memory, acquisition, stabilization, and retrieval of memory [2, 3]. Morphologically, both α/β and α’/β’ neurons have axons with two branches that project dorsally and medially into respective α and β, or α’ and β’ lobes, whereas γ neurons send out single axon branches that project medially into the γ lobe [1]. These three KC subtypes can be also distinguished by differentially expressed marker genes. For instance, broad-complex, tramtrack and bric-ȧ-brac zinc finger transcription factor (BTBzf TF) Abrupt (Ab) is specifically expressed in γ neurons [4–6], whereas cell adhesion molecule Fasciclin II and Rho guanine nucleotide exchange factor Trio are expressed in different subsets of KC subtypes [4]. Therefore, specifying KC subtypes with proper cell numbers, unique cell identities and distinct functions is a key aspect of MB formation.

Previous studies have revealed that the other BTBzf TF, Chronologically inappropriate morphogenesis (Chinmo), exhibits a graded expression pattern in KCs and plays a crucial role in diversifying KC subtypes [7]. Chinmo is highly expressed in early-born γ neurons, but its expression level is low in later-born α’/β’ and absent in α/β neurons [7]. Notably, Chinmo controls the expression of the third BTBzf TF, Maternal gene required for meiosis (Mamo), functioning to specify the identities of KC subtypes at the larval stage via a fine-tuning process. In particular, a high expression level of Chinmo inhibits Mamo expression in γ neurons, whereas low Chinmo expression promotes Mamo production in α’/β’ neurons [4]. As a gradual reduction of Chinmo expression occurs during development, Mamo, adding to its function in α’/β’ neurons, takes over Chinmo’s role of regulating the differentiation of γ neurons at the pupal stage [4, 5, 7]. In the absence of chinmo and mamo, molecular and morphological characteristics of γ and α’/β’ neurons are shifted towards those of α/β neurons, constituting a cell identity transformation [4, 5, 7]. Therefore, the late-born α/β neural identity is hypothesized to be a default status upon the loss of the neural identity determinants for early-born KCs [4, 7]. However, it is also possible that key specification regulators of α/β neurons exist but have not yet been identified.

In this study, we leverage RNA-seq databases on KCs to identify subtype-specific markers as readouts for the cell identity of γ, α’/β’ and α/β neurons [8, 9]. By doing so, we identified a phylogenetically conserved Pipsqueak domain-containing TF, Ecdysone-induced protein 93F (Eip93F/E93), with preferential expression in α/β neurons. Loss-of-function of E93 not only downregulated α/β-specific gene expression, including a subunit of T-type like voltage gated calcium channel Ca-α1T, but it also upregulated γ-specific Ab in late-born KCs, implying that an identity shift towards early-born KCs had occurred. Intriguingly, RNAi knockdown of E93 or Ca-α1T in α/β neurons further compromised animal behaviors, including foraging-related and night-time activities. In contrast, E93 overexpression precociously turned on Ca-α1T expression in early-born KCs at the expense of abolishing expression of early-born KC markers, such as Ab and Mamo. Notably, E93 was upregulated in early-born KCs in the absence of chinmo and mamo but diminished in late-born KCs upon Ab overexpression. Taken together, our results suggest that a hierarchical genetic network among chinmo, mamo, E93 and ab with potential feedback loops controls the identity and function of KC subtypes during the construction of functional MBs.

Results

Identification of KC subtype-specific markers

By leveraging information from published RNA-seq studies showing preferentially expressed genes in adult KC subtypes [8, 9], we sought to identify a collection of KC subtype marker-expressing lines (available at stock centers) with GFP transgenes at genes of interest [10–12]. The results of this screen for KC subtype marker GFP lines are depicted in S1 Fig; highlights of certain KC subtype marker lines are described below. First, Abrupt (Ab)-GFP, a BAC clone-engineered GFP line, can be used to replace an excellent Ab antibody (generated by Dr. Crews laboratory but no longer available) for labeling γ neurons [4–6] (Fig 1A-B and S2 Fig). In addition, Lachesin (Lac; an Ig superfamily protein[13])-FSVS, a GFP-trapping line, expresses GFP-fused Lac specifically in α’/β’ neurons starting from the early pupal stage (Fig 1C-D and S3A Fig). Notably, we also identified two other GFP-trapping lines, Ecdysone-induced protein 93F (E93)-GFSTF and Calcium channel protein α1 subunit T (Ca-α1T)-GFSTF, that express GFP-fused proteins enriched in α/β neurons. Enrichment in these neurons is evidenced by patterns of complementary expression to Trio, a marker of γ and α’/β’ neurons [4] (Fig 1E-H). Consistent with the notion that generation of most of α/β neurons occurs at the pupal stage [1], E93-GFSTF and Ca-α1T-GFSTF were not detectable in KCs at the wandering larval (WL) stage (Fig 1E, 1G and S3B Fig). With this set of useful reagents, we set out to further explore the molecular mechanisms underlying cell identity specification of KC subtypes.

Fig 1.

Loss-of-function of E93 suppresses the α/β-neural identity and has behavioral consequences

Since E93, a Pipsqueak-domain containing TF, functions crucially in various biological processes [14–17] and E93-GFSTF was preferentially expressed in α/β neurons (Fig 1F), we sought to investigate whether E93 acts as a critical regulator for specifying the α/β-neural identity. First, we found that the expression of a α/β-specific marker, Ca-α1T-GFSTF, was diminished in the context of E93 mutation (E93^Δ11) and in a line with E93 knockdown due to overexpression of E93 RNAi by a pan-KC driver, GAL4-OK107[1] (Fig 2A-B and S4-S5 Fig). In contrast, the same RNAi knockdown elicited no discernible effects on the expression levels of Trio and Lac-FSVS in γ and α’/β’ neurons (Fig 2A-B and S6 Fig). Since Ca-α1T encodes a subunit of a T-type like voltage gated calcium channel, which regulates sleep behavior [18], we wondered whether the loss of E93 and Ca-α1T in α/β neurons could cause behavioral defects. To examine the behavior, we utilized a monitoring system to film and analyze the activities of individual flies (S7A-B Fig). The animal’s speed of locomotion usually peaks around the day-to-night shift and gradually becomes reduced at night. In control samples (wild-type, RNAi-only, and GAL4-only lines), it took around eight hours for the moving speed to reach a minimum during the night period (Fig 2C and S7C-D Fig). Interestingly, night-time activity was significantly impaired (with a sharp reduction of moving speed to a minimal value at around two hours into the night period) when Ca-α1T and E93 were knocked down using an α/β neural driver, c739-GAL4 [19] (Fig 2C and S7C-D Fig). Although the overall moving speed was lower and more variable in E93 knockdown samples than in controls, it is possible that E93 also regulates factors that add on to causal effects from Ca-α1T. In addition to the effects on night-time activity, RNAi knockdown of Ca-α1T and E93 appeared to compromise the foraging-related behavior. Compared to control flies, Ca-α1T- and E93-deficient animals tended to explore regions without fly food during the day-time period (Fig 2C and S7E Fig). Taken together, these results suggested that Ca-α1T (expression is regulated by E93) acts in α/β neurons to potentially control animal behavior.

Fig 2.

In addition to the observed reduction of Ca-α1T-GFSTF, E93 knockdown in KCs substantially compromised the expression of another α/β-specific marker, myr::GFP driven by R44E04-LexA [5] (Fig 2D-E). Although these results seemed to indicate the loss of α/β neural identity due to loss of E93 function, the findings also raised a possibility that the reduction of Ca-α1T-GFSTF and R44E04-LexA might simply be due to the absence of late-born KCs. However, this explanation was ruled out by the observation of relatively intact morphology of α and β lobes when we examined the expression of a third α/β-specific marker (myr::GFP driven by R70F05-LexA) upon E93 knockdown in KCs [5] (Fig 2F-G). Moreover, compared to wild-type samples, the cell number of R70F05-LexA-positive neurons was not significantly altered in KCs when E93 was knocked down (Fig 2H-I and S8 Fig). Taking advantage of R70F05-LexA as a putative α/β-neural marker, we further found that more than half of the R70F05-LexA-positive neurons ectopically expressed γ-specific Ab-GFP in E93 knockdown samples (Fig 2H-I and S8 Fig), implying that α/β neurons had transformed into γ-like neurons. These results taken together suggested that E93 is required for specifying KCs towards the cell identity of α/β neurons, which are responsible for specific functions governing animal behaviors.

E93 overexpression promotes α/β neural traits in early-born KCs

We next asked if E93 indeed plays a crucial role in specifying α/β neural identity, i.e., does gain of E93 function transform the cell identity of other KC subtypes to α/β neurons? In contrast to the undetectable Ca-α1T-GFSTF expression in wild-type with major KC subtypes of γ and α’/β’ neurons at the WL stage, Ca-α1T-GFSTF expression was precociously upregulated in KCs when E93 was overexpressed by GAL4-OK107 (Fig 3A-B). Similarly, a portion of putative γ neurons exhibited α/β-neural like features (myr::GFP driven by R70F05-LexA) with axonal defects when E93 was overexpressed by the γ-neural driver, GAL4-201Y [1] (Fig 3C-D). Consistent with this observation, we further found that E93 overexpression driven by GAL4-OK107 (but not Worniu-GAL4, a pan-neuroblast driver [16]) compromised the expression of specific markers in γ neurons at various developmental stages, including Ab-GFP, the receptor for insect molting hormone ecdysone (EcR-B1) and Mamo isoforms (Fig 3E-J and S9-S10 Fig). Along with the effect of impairing γ-specific marker expression, we further found that E93 overexpression driven by GAL4-OK107 impaired the expression of markers in α’/β’ neurons, including Mamo^D∼G isoforms and Lac-FSVS (Fig 3I-L). Its overexpression also caused morphological defects by perturbing the remodeling process that normally occurs in early-born KCs at 24-hour after puparium formation (APF; Fig 3K-L). These results taken together suggested that E93 when overexpressed is sufficient to cause specification of KCs toward the α/β neural identity.

Fig 3.

Hierarchical genetic network of chinmo, mamo, E93 and ab controls the cell identity of KC subtypes

Since Chinmo and E93 act as key TFs to regulate the cell identity of KC subtypes, we wondered whether chinmo and E93 might form a genetic network to diversify KC subtypes. Since chinmo controls the expression of γ-specific Ab [4, 5] (S2 Fig), we decided to test whether the loss of Ab-GFP by E93 overexpression (seen in Fig 3F) was caused by compromised Chinmo expression. However, we did not observe a reduction of Chinmo level upon E93 overexpression at the first instar larval stage (the most abundant Chinmo expression stage in the wild-type [7]), even though the same manipulation did block Ab-GFP expression (Fig 4A-B). In contrast, E93-GFSTF expression was precociously upregulated in early-born KCs in a chinmo mutation line (chinmo[1]) and in the chinmo knockdown line (overexpression of chinmo RNAi driven by GAL4-OK107) at the WL stage (Fig 4C-D and S11 Fig). In line with the facts that microRNA let-7 and RNA binding protein Syncrip (Syp) inhibit Chinmo expression [20, 21], we further found that E93-GFSTF expression was abolished by syp RNAi knockdown and partially promoted by let-7 overexpression (S12 Fig). These results suggested that the adoption of γ neural identity by early-born KCs is in part due to suppression of the α/β-neural regulator E93 via Chinmo.

Fig 4.

After the diminishment of Chinmo at the early pupal stage [7], we wonder whether E93 would be disinhibited in γ neurons and whether their neural identity would be transformed if this disinhibition indeed occurs? Since γ neurons exist in adult brains and their neural identity is crucially regulated by Mamo at the pupal stage [5], we then tested whether Mamo takes over Chinmo’s role to suppress E93 expression in γ neurons. As such, mamo RNAi knockdown indeed caused the upregulation of E93-GFSTF in KCs (within enriched Trio expression) of adult brains but not brains at the WL stage (Fig 4E-G), suggesting that Mamo could inhibit E93 expression in γ neurons to ensure their neural identity. Since the E93-mediated α/β neural identity is accompanied by absence of certain γ-specific traits, such as Ab-GFP (Fig 2I and 3F), we next sought to explore whether Ab also plays a crucial role in controlling the KC identity. Intriguingly, we found that the expression of α/β-specific Ca-α1T-GFSTF and E93-GFSTF was not detectable upon Ab overexpression (Fig 4H-J and S13 Fig). This was accompanied by the expansion of Trio in the cytosol in almost all KCs, indicating the transformation of KCs into γ neuron-like cells. We also noted that RNAi knockdown of ab neither abolished Trio expression nor upregulated the expression of Ca-α1T-GFSTF caused by E93 knockdown in early-born KCs (S14-S15 Fig). Nonetheless, these results together suggested that chinmo, mamo, E93 and ab form a hierarchical genetic network with potential feedback loops to control the identity of KCs (Fig 5).

Fig 5.

Discussion

Drosophila MBs are primarily constructed by three sequentially generated KC subtypes, γ, α’/β’ and α/β neurons [1]. Previous studies have revealed that BTBzf TFs, Chinmo and Mamo, function crucially in the specification of early-born KC subtypes into γ and α’/β’ neurons [4, 7]. In this study, we revealed the mechanisms of KC subtype specification by identifying and characterizing Pipsqueak domain-containing TF E93 as a driver of KC identity towards α/β neurons. Our findings that E93 regulates the expression of calcium channel Ca-α1T to subsequently control animal behaviors also provide an illustration that neural identity specification is associated with acquisition of specific functions among KC subtypes. We further showed that the specification of KC subtypes is controlled via a genetic network formed by chinmo, mamo, E93 and ab, which mainly controls the KC identity of γ and α/β neurons during construction of the adult MB (Fig 5).

Despite that reciprocal regulation between BTBzf TFs and E93 has been reported in multiple cell types during the Drosophila development [22, 23], here, we disclosed a novel strategy of using genetic networks of BTBzf TFs and E93 to specify distinct cell identities among neurons derived from the same neuroblasts (Fig 5). To diversify KC subtypes, Chinmo and Mamo take turns to inhibit the E93 expression, thereby establishing the cell identity of γ neurons, as suggested by Ab expression [4] (Fig 2I and 4G). Without the inhibitory effects of Chinmo and Mamo on E93 expression, late-born KCs specify into α/β neurons by turning on the expression of α/β-specific genes and suppressing the expression of γ-specific genes, Ab included (Fig 2). Although how the genetic network of BTBzf TFs and E93 specifies α’/β’ neural identity has not yet been fully resolved, the Mamo expression promoted by the low level of Chinmo, which is regulated by the transforming growth factor beta (TGF-β) signaling in MB neuroblasts, is crucial for KCs to adopt the α’/β’ neural identity [4, 24]. Intriguingly, our results revealed that E93 overexpression can abolish the expression of Mamo to block KCs from adopting the α’/β’ neural identity (Fig 3J and 3L). All these results taken together portray that the genetic network of BTBzf TFs and E93 is very likely to control the KC identity during construction of the adult MB. In light of the phylogenetic conservation of the E93 family and tramtrack (TTK)-type BTBzf TFs, including chinmo, mamo and ab, as Arthropoda-specific genes [25, 26], one cannot help but wonder whether TTK-type BTBzf TFs might have been evolutionally introduced to intervene with E93 to regulate the development of multiple arthropodan tissues; perhaps future investigations may delineate their relationship on neuron-type diversification in the nervous system.

Once the identities of neuron subtypes are determined by key regulators, how the cells acquire specific functions in the nervous system to eventually participate in animal behaviors? Previous findings together with our current studies might shed some light on this topic. First, the expression of Chinmo and Ab is inhibited in late-born KCs, in part due to the preferential expression of let-7 in α/β neurons [21, 27]. In addition, the neuronal (and KC) contribution of let-7 regulates day– and night-time sleeping behaviors in developmental– and adult-restricted manners [28]. Furthermore, Chinmo and let-7 potentially regulate the expression of E93, for which expression in α/β neurons governs the expression of a phylogenetically conserved sleep-modulatory calcium channel Ca-α1T to regulate animal behaviors [18] (Fig 2B-C, 4D and S11A Fig). These results together with the previous reported function of E93 in circadian rhythm provide an intriguing link between E93 and sleep-associated behaviors [29]. Since KC subtypes were reported to function crucially in sleep/arousal and alcohol-induced sleep deficit behaviors [30–32], our finding of KC identity specification and Ca-α1T expression under the regulation of E93 suggests that this process may in some way endow α/β neurons with the capacity to control animal behaviors related to sleep. Since the intricate relationship between sleep and memory is well established [33–36], future deeper investigations on cell-type specification of memory-associated neurons are expected to provide insights into how to acquire for unique functions among neurons to regulate these two crucial traits of animals.

Materials and Methods

Experimental model and subject details

Flies were cultured in a room maintained at 25°C (± 1.5°C) and 50-65% humidity for all experiments. For most experiments, flies were used with no selection for sex; therefore, roughly equal numbers of males and females were used. However, females were selected for the analysis of Ca-α1T-GFSTF in Fig. 1G-H, 2A-B, 3A-B and 4I-J and S5A-B, S13 and S14D Figs due to the cytolocation of the Ca-α1T gene on the X chromosome. In addition, males were selected for the behavioral assay to mitigate complications of egg laying behaviors on locomotion tracking.

Fly strains

The fly strains used in this study were as follows. Most strains are available from either Bloomington Drosophila stock center (BDSC), Kyoto Drosophila stock center (DGGR) or Vienna Drosophila stock center (VDRC). (1) Ab-GFP (BDSC38626); (2) Lac-FSVS (DGGR115308); (3) E93-GFSTF (BDSC59412); (4) Ca-α1T-GFSTF (BDSC61800); (5) hs-FLP¹²,UAS-mCD8::GFP (BDSC28832); (6) tubP-GAL80,FRT^40A(BDSC5192); (7) chinmo[1],FRT^40A(BDSC59969); (8) GAL4-OK107 (BDSC854); (9) UAS-mCD8::RFP [attp40] (BDSC32219); (10) UAS-mCD8::RFP [attp2] (BDSC32218); (11) UAS-E93 RNAi (BDSC57868); (12) UAS-E93 RNAi (VDRC104390); (13) UAS-Ca-α1T RNAi (BDSC39029); (14) GAL4-c739 (BDSC7362); (15) 44E04-LexA::P65 (BDSC52736); (16) 70F05-LexA::P65 (BDSC523629); (17) lexAop2-myr::GFP[37]; (18) lexAop2-mCD8:: RFP [37]; (19) FRT^82B,tubP-GAL80 (BDSC5135); (20) FRT^82B (BDSC86313); (21) E93^Δ11 (BDSC93128); (22) E93(EP) (BDSC30179); (23) UAS-E93-A [VK37] (this study); (24) UAS-E93-B [VK37] (this study); (25) GAL4-201Y (BDSC4440); (26) mamo^H/I-HA[38]; (27) mamo^D∼G-HA [38]; (28) UAS-chinmo RNAi [VK37] [5]; (29) UAS-LUC-let7 (BDSC41171); (30) UAS-Ssyp RNAi (VDRC33011); (31) UAS-mamo RNAi (BDSC44103); (32) UAS-ab (BDSC23639); (33) UAS-ab-HA (FlyORF000705); (34) UAS-Dcr2 (BDSC24651); (35) UAS-ab RNAi (VDRC104582). The UAS-E93-A and UAS-E93-B transgenes were generated using standard molecular biology methods to clone cDNA fragments derived from fully-sequenced EST clones, GH10557 and LP08695 (available from Drosophila Genomics Resource Center), carrying E93-A and E93-B isoforms into the attB-UAST vector. The generation of UAS-E93-A and UAS-E93-B transgenes and fly stains was performed by WellGenetics, Inc.

RNAi knockdown and overexpression experiments and MARCM clonal analyses

UAS-RNAi and UAS-transgene lines were crossed to GAL4-107 and GAL4-201Y for knockdown and overexpression of genes of interest in KCs. Mosaic clones for the MARCM studies were generated as previously described [39]. In short, mosaic clones of chinmo[1] and E93^Δ11 mutations were induced by 35 min of heat shock using hs-FLP[12] in newly hatched larva. Dissection, immunostaining and mounting of adult brains were performed as described in a standard protocol [39]. Primary antibodies used in this study included guinea pig antibody against Chinmo (1:1000, Sokol laboratory [40]), rat monoclonal antibody against mCD8 (1:100, Thermo Fisher Scientific), rabbit antibody against GFP (1:750, Thermo Fisher Scientific), and mouse monoclonal antibodies against EcR-B1 (1:50, DSHB), Fas2 (1:100, DSHB) and Trio (1:50, DSHB). Secondary antibodies conjugated to different fluorophores (Alexa 488, Alexa 546 and Alexa 647; Thermo Fisher Scientific and Jackson ImmunoResearch Lab, Inc.) were used at 1:750 dilutions. Immunofluorescence images were collected by confocal microscopy on a Zeiss LSM 700, projected using the LSM browser and processed in Adobe Photoshop CS6. All data are representative of more than 3 brains per genotype.

Distance measurement in the behavioral assay and statistical analysis

Individual fly activities were recorded and analyzed using the activity monitor system developed by DroBot, Inc. This system used an open-resourced software pySolo to track individual moving flies [41]. Average speed and standard deviation of individual flies were calculated according to total traveling distance within 30 min periods from the first day to the fifth day, as shown in Fig 2C. Box-plots and statistical analyses in S7C-D Fig were conducted using two online tools [(https://shiny.chemgrid.org/boxplotr/) and (https://astatsa.com/OneWay_Anova_with_TukeyHSD/)]. One-way ANOVA with post-hoc Tukey HSD test was used to compare datasets with six groups in S7D Fig. Student’s t-test was used for statistical analysis to compare datasets with two groups in S8 Fig.

Acknowledgements

We thank the TRiP at Harvard Medical School (NIH/NIGMS R01-GM084947) for providing the transgenic RNAi fly stocks used in this study. We also thank DroBot Inc. for designing the behavioral assay system used in this study. This work was supported by National Science and Technology Council (NSTC-112-2311-B-001-029) and Thematic Research Program of Academia Sinica (AS-TP-113-L01), Taiwan.

Additional information

Author contributions

Conceptualization, P.C.C. and H.H.Y.; Methodology, P.C.C., K.Y.K., S.Y.H. C.C., and H.H.Y.; Formal Analysis, P.C.C., C.C. and H.H.Y.; Investigation, P.C.C. and H.H.Y.; Writing: P.C.C. and H.H.Y.; Visualization, P.C.C., C.C. and H.H.Y.; Supervision: H.H.Y., Funding Acquisition, H.H.Y.

Materials and correspondence

Correspondence and requests for materials should be addressed to Hung-Hsiang Yu.