To test the feasibility of TaDa for profiling Drosophila OSNs, we first focused on the Ir64a population, which comprises ~16 neurons that are housed in sensilla in the sacculus (a chamber within the antenna), where they detect acidic odors (Ai et al., 2010). We first verified that Ir64a promoter-Gal4 (hereafter, Ir64a-Gal4) induction of Dam or Dam:PolII did not impact the expression or localization of the IR64a receptor (Figure 1B) or the projection of these neurons to the antennal lobe (Figure 1C). We proceeded to collect triplicate samples of ~2000 adult antennae and processed these for TaDa (see Materials and methods). This experimental design is expected to integrate PolII occupancy patterns from the mid-pupal stage (when Ir-Gal4 expression initiates) to adults. We therefore anticipated identifying genes that are differentially expressed both during the second half of OSN development and in mature antennae. As a positive control, we first examined the mapping of reads of methylated DNA fragments to the Ir64a gene. Dam:PolII consistently accessed this gene region at higher levels than Dam alone, particularly near the 5’ end of the locus (Figure 1D), consistent with previously described genome-wide PolII occupancy patterns (Southall et al., 2013).

We extended the TaDa approach using drivers to target six additional individual OSN classes, including all acid-sensing Ir populations (Ir31a, Ir75a, Ir75b, Ir75c, Ir84a) (Grosjean et al., 2011; Silbering et al., 2011), as well as the hygrosensory Ir40a neurons (Enjin et al., 2016; Knecht et al., 2016); these populations range from ~8 to ~25 cells per antenna (Knecht et al., 2016; Prieto-Godino et al., 2017). Reproducibility among all triplicate datasets (average r² = 0.85) was consistent with previous TaDa studies (Southall et al., 2013). As expected, the enrichment of Dam:PolII access to the Ir64a gene was specific to the Ir64a neurons (Figure 1E). We quantified ‘PollI occupancy scores’ (OSs) by calculating the log₂ ratio of the normalized number of reads from the Dam:PolII samples compared to the normalized number of reads in control Dam samples across the entire gene body. In Ir64a neurons, the Ir64a gene OS was significantly positive (OS = 0.20; False Discovery Rate (FDR) = 4.26e-10), while it was not significantly different from zero in the other six neuron populations (Figure 1E, bottom).

Similarly, OSs for other Ir genes were significantly positive in the corresponding neuronal population, but not in those in which the receptors are not expressed (Figure 1F). We noted two exceptions: first, in Ir31a neurons, Ir40a had a positive OS (0.12; FDR = 1.93e-13), although it was much lower than that of Ir31a (OS = 0.41; FDR = 2.67e-6) (Figure 1F). Second, the Ir75a, Ir75b, and Ir75c genes had positive OSs in all three of the corresponding neuron populations (Figure 1F). As the proteins they encode (and the corresponding Ir-Gal4 drivers) are expressed in discrete neuron populations, this latter result is most likely due to these tandemly clustered genes having overlapping transcription units (Prieto-Godino et al., 2017).

To further examine the specificity of TaDa in reporting on neuron population-specific gene expression, we plotted OSs for all members of the Ir repertoire, as well as Or and Gustatory receptor (Gr) families. The vast majority of these genes are expressed only in other antennal sensory neuron populations, distinct chemosensory organs and/or life stages (Chen and Dahanukar, 2020; Joseph and Carlson, 2015; Sánchez-Alcañiz et al., 2018). Concordantly, only very few genes show significant OSs in any of the seven populations of Ir neurons analyzed by TaDa (Figure 1F, Figure 1—figure supplements 1–3). Two prominent examples are Ir8a and Ir25a, which encode co-receptors for subsets of tuning IRs (Abuin et al., 2011). Ir8a has a significant OS in all Ir populations except for Ir40a neurons, which is consistent with the selective expression and function of IR8a in acid-sensing OSNs but not hygrosensory neurons (Abuin et al., 2011; Figure 1F). IR25a is broadly expressed in most antennal neurons (Abuin et al., 2011; Benton et al., 2009) – although it may have a role in only a subset of these – which is reflected in a significant OS for Ir25a in all seven populations (Figure 1F). IR40a is co-expressed with another receptor, IR93a (Enjin et al., 2016; Knecht et al., 2016), although Ir40a neurons display significant occupancy of Ir93a, several other Ir populations also have, unexpectedly, a significant OS for this gene (Figure 1—figure supplement 1). TaDa signals of real transcription of Ir93a may be confounded by an intronic Doc transposable element.

Of the other chemosensory receptor genes displaying unexpected significant OSs in these Ir neurons (Figure 1—figure supplements 1–3), many are located in the introns of other genes (e.g. Ir47a, within the slowpoke 2 locus, which encodes a nervous-system expressed potassium channel [Budelli et al., 2016]), or have overlapping transcripts with other genes (e.g. Gr2a, which overlaps with futsch, a gene encoding a neuronal microtubule-binding protein [Hummel et al., 2000]). In these cases, calculation of a specific OS for the chemosensory genes is impossible, and we suspect that the OS reflects transcription of the non-receptor gene. A similar issue might affect receptor genes that are in very close proximity to (though not overlapping with) other genes, given that TaDa signals depend upon GATC genomic distributions and not transcript boundaries.

Olfactory receptors have lower transcript levels in the antenna compared to other classes of genes with broader neuronal or non-neuronal expression patterns (Menuz et al., 2014; Shiao et al., 2015). We extended the examination of the TaDa datasets to other genes that are known to be expressed (or not expressed) in the analyzed Ir neuron populations (Figure 1G). Two broadly-expressed neuronal genes, elav and nSyb, display significant OSs across populations, as does pebbled (peb), a molecular marker for OSNs (Sweeney et al., 2007). By contrast, oaz, a broad marker of PNs (Hong et al., 2009), is not significantly occupied. Many of the most highly expressed genes in the antenna are those encoding Odorant Binding Proteins (OBPs), which are thought to be transcribed exclusively in non-neuronal cells (Larter et al., 2016), often at >10 fold higher levels than olfactory receptor genes, as determined by bulk RNA-seq (Menuz et al., 2014; Shiao et al., 2015). (This idea was challenged recently by the OSN snRNA-seq analysis [McLaughlin et al., 2020], which will be considered further in the Discussion). Two of these, Obp59a and Obp84a are expressed in sensilla housing Ir neurons (Larter et al., 2016; Sun et al., 2018a), but neither of these genes displays significant OSs in any neuron population (Figure 1G). Such absence of occupancy is true for the Obp repertoire as a whole (Figure 1—figure supplement 4). Similarly, markers for other support cell types, such as nompA (thecogen cell) (Chung et al., 2001; Larter et al., 2016), and the glial marker repo are not significantly occupied (Figure 1G).

Electrophysiological experiments indicate that antennal OSNs are cholinergic (Kazama and Wilson, 2008). In line with this functional property, genes encoding the vesicular acetylcholine transporter (VAChT) and choline acetyltransferase (ChAT) have significant OSs in all populations, while glutamatergic and GABAergic neuron markers (vesicular glutamate transporter (VGlut) and glutamic acid decarboxylase 1 (Gad1)) do not (Figure 1G). We note, however, that a GABA receptor subunit (encoded by GABA-B-R2) – which is heterogeneously expressed and mediates population-specific presynaptic gain control in different Or OSNs (Root et al., 2008) – displays significant, but variable, OSs in different Ir populations (Figure 1G). Finally, of the multiple neuropeptides detected in the antennal lobe by proteomics, only one, short neuropeptide F (sNPF), originates from OSNs (Carlsson et al., 2010), where it has a role in glomerular-specific presynaptic facilitation (Root et al., 2011). Consistently, we observe significant (and heterogeneous) OSs for sNPF, but not for genes encoding neuropeptides whose presence in the antennal lobe derives from expression in central neurons (e.g. Tachykinin [Tk], Myoinhibiting peptide [Mip]) (Carlsson et al., 2010).

Together, these results indicate that TaDa is sufficiently sensitive and specific to detect PolII occupancy at transcriptionallyactive genes in very small populations of neurons embedded within the cellularly-complex antennal tissue.

We next surveyed OSs at a genome-wide level, to characterize patterns of transcriptional activity in different Ir neuron populations. The number of genes significantly occupied by PolII ranged from 3603 to 4775 across the seven datasets, representing 20–27% of all genes (Figure 2A and Supplementary files 1–3). To investigate patterns of similarity in global OSs among these populations, we carried out hierarchical clustering of the mean OSs for all genes in the genome (Figure 2B). This analysis revealed an interesting correspondence in the clustering of OSs with the phylogeny of the corresponding receptor genes (Prieto-Godino et al., 2017): neurons expressing the most recently duplicated receptor genes, Ir75b and Ir75c, were clustered together. The next most similar neuron population expresses Ir75a, which is the likely ancestral member of this tandem gene array (Prieto-Godino et al., 2017). Ir40a and Ir64a neurons clustered together away from the other acid-sensing populations, possibly reflecting the development of these neurons within the sacculus, rather than in sensilla located on the antennal surface.

Beyond the unique patterns of sensory receptor expression, little is known about the nature or neural specificity of other signaling pathways in OSNs. To gain insight into these properties, we examined the occupancy of genes encoding putative neuropeptides, G protein-coupled receptors (GPCRs) and ion channels (excluding chemosensory receptors) in the seven Ir populations (Figure 2C–E and Supplementary files 6–14). We observed that each Ir neuron population displays PolII occupancy of a distinct combination of 3–9 neuropeptide genes (out of 49 in the genome), 18–31 GPCR genes (out of 111), and 29–47 ion channel genes (out of 149). There were, however, only a few cases of genes that are unique to a single neuronal population (Supplementary files 8, 11 and 14).

OSNs are also distinguished by their morphological properties, notably their projections to discrete glomeruli within the antennal lobe. Genetic studies have identified several secreted and transmembrane proteins that play roles in different steps of the neuronal guidance process (Brochtrup and Hummel, 2011; Hong and Luo, 2014; Jefferis and Hummel, 2006), but our understanding remains incomplete and many key molecules likely remain to be discovered. Although our TaDa experiments profile the latter stages of OSN development – after glomerular targeting has occurred (Jefferis and Hummel, 2006) – the expression of known OSN axon guidance genes persists into adult stages (Barish et al., 2018; Menuz et al., 2014). We therefore reasoned our datasets could permit surveying of the potential complexity and neuron-type specificity of the molecular machinery underlying this process. We examined OSs of ~268 genes encoding molecules with known or implicated roles in neuron projection guidance (Figure 2F and Supplementary files 15–17). While each neuron population had a unique combination of many dozens of PolII-occupied genes, individual datasets had very few or no unique genes (Figure 2F and Supplementary file 17). These observations suggest that different OSN populations depend upon specific combinations of largelyoverlapping sets of guidance factors.

The expression of both chemosensory receptors and guidance molecules depends upon regulatory networks of transcription factors (Barish and Volkan, 2015; Jafari et al., 2012). We therefore assessed global patterns of PolII occupancy of the set of 754 predicted Drosophila transcription factors (Pfreundt et al., 2010) across the seven Ir populations. Half of these genes (379/754) displayed significant occupancy in at least one population of Ir neurons, with individual OSN classes having 190–286 occupied transcription factor genes (Figure 2G and Supplementary files 18–20). There was considerable variation in Dam:PolII occupancy of transcription factor genes across the neuron populations (Figure 2G), consistent with each neuronal subtype possessing a unique and complex gene regulatory network to support its specific differentiation properties. Here, Ir31a neurons stood out (45 uniquely occupied genes); the reasons for this are unclear, as the biology of this sensory pathway is poorly understood. By contrast, Ir75a neurons had a single uniquely occupied transcription factor gene (pou domain motif 3 (pdm3), investigated further below).

Many of the transcription factors with known functions in OSN development are broadly-expressed across OSNs, such as Acj6, Fer1, and Onecut (Clyne et al., 1999; Jafari et al., 2012); consistently, these genes display positive OSs in all seven neuron populations (Supplementary file 19). One exception is Engrailed (En), which is expressed and required in a subset of OSNs (Chou et al., 2010b; Song et al., 2012). In our TaDa datasets, we observed that en displays significant OSs in Ir31a, Ir40a, and Ir64a populations (Figure 2—figure supplement 1A), concordant with the robust detection of En protein in only these three Ir neuron classes (Figure 2—figure supplement 1B–C).

We next performed CATaDa analysis through investigation of the Dam-alone datasets, by using these to generate chromatin accessibility maps similar to ATAC-seq and FAIRE-seq (Aughey et al., 2018). Genome-wide visualization of the peaks of methylated sequences across the genome (see Materials and methods) revealed heterogeneous patterns of chromatin accessibility, with the expected overall decrease in open chromatin toward the heterochromatic centromeric regions of the chromosome arms (Figure 3A).

To examine chromatin accessibility patterns at a gene level, we focused on CATaDa signals at the Ir loci that distinguish different populations – that is, the seven tuning Ir genes and Ir8a – whose cell type-specific transcriptional activity is well-established (Abuin et al., 2011; Benton et al., 2009; Prieto-Godino et al., 2017). For each gene, peaks were present at the presumed promoter region, as well as in upstream regions, which may represent enhancers. Comparisons across neuron populations revealed that peak distribution was similar across all eight genes (Figure 3B). This observation indicates that neuron-specific differences in Ir gene transcription are not due to overt differences in promoter and enhancer accessibility. However, we noted the presence of peaks upstream of Ir75a and Ir75c that are unique to their respective neuron populations (q-value <0.05). These peaks lie within regions corresponding to those that, when cloned into reporter transgenes, are sufficient to recapitulate receptor expression patterns (Ai et al., 2010; Prieto-Godino et al., 2017; Silbering et al., 2011; Figure 3B). Such peaks may therefore reflect enhancers that are accessible/functional only in these Ir neuron populations. Peaks were also identified upstream of Ir64a in both Ir64a neurons and Ir40a neurons (Figure 3B). This observation suggests that these sacculus neuron populations share chromatin accessibility in this region but that other factors (e.g. trans-acting proteins) are required to determine the cell type-specific transcription of Ir64a.

The analyses of genes encoding chemosensory receptors, transcription factors and other categories of proteins demonstrate the use of TaDa OSs to identify potential developmental and functional determinants that are expressed within specific OSN populations. While differences in OS for a given gene across populations may suggest heterogeneous expression levels, we sought to test formally for PolII occupancy differences between populations. We implemented a different analysis method that uses variation in read-depth along the gene body in the triplicate paired Dam-alone/Dam:PolII TaDa datasets (Figure 4A and Materials and methods). We identified candidates that had at least two GATC motifs for which the excess of Dam:PolII read-depth compared to the Dam-alone read-depth significantly varied in the same direction between two or more neuron populations (see Materials and methods). With this criterion, 1694 genes exhibited significant variation across the seven populations, indicating that only a relatively small proportion of genes (~8%) differ among these neuron types. This number is likely to be an overestimate as it includes overlapping and/or intronic genes that, as described above, cannot be parsed further using TaDa data. Importantly, this subset encompasses all of the expected Ir genes (see Figure 4D). To examine how each individual Ir neuron dataset varied in comparison to the six others, we also quantified significant differences between each pair of neuron populations (Figure 4B). Despite the methodological differences with the TaDa analysis that produces OSs, the proportion of genes contributing to the total number of pairwise difference approximates the clustering of global OSs (Figure 2B). For example, comparisons among the closely clustered Ir75a, Ir75b, and Ir75c neuron datasets result in very few of the total differences (0.4–1.7%). Conversely, slightly more than half (51%) of the total differences in the Ir84a neuron dataset arise from comparisons with the distantly clustered Ir40a neuron dataset.

We investigated the general characteristics of the 1694 differentially occupied genes by performing Gene Ontology analyses, finding highly significant enrichment in neuron, neural development, and signaling categories (Figure 4C). This enrichment is consistent with these genes underlying the known unique anatomical and physiological properties of these neuron classes, and indicates that our comparative TaDa approach would be valuable in identifying candidates that contribute functionally to these properties.

To establish a priority list for follow-up experiments, we ordered the 1694 genes by the total number of differentially occupied GATC motifs, reasoning that this would highlight those displaying differential occupancy along a broad region of the gene body (Figure 4D). While this ‘ranking’ is biased toward longer genes (which on average will have more GATC motifs), we emphasize that this does not exclude interest in genes with fewer GATC motifs; for example, Ir genes are distributed widely in this list (Figure 4D). We subsequently focused on experimental validation of the top-ranked transcription factor gene, pdm3, and the top-ranked gene encoding a neuron guidance molecule, flamingo (fmi; also known as starry night (stan), flybase.org/reports/FBgn0024836), as described below.

The robust signal displayed by pdm3 for between-population differences is consistent with this gene having a significant OS only in Ir75a neurons (Figure 4E). Notably, from the global survey, pdm3 is the sole transcription factor gene that is uniquely occupied in this neuron population (Figure 2G and Supplementary file 20). Pdm3 is a nervous-system enriched transcription factor (Brown et al., 2014; Chen et al., 2012; Tichy et al., 2008), although its precise in vivo binding specificity is unknown. In another olfactory organ, the maxillary palp, Pdm3 is expressed in multiple classes of Or neurons and functions in controlling receptor expression and axon guidance (Tichy et al., 2008), but its role in Ir neurons is unknown. Using Pdm3 antibodies, we first analyzed protein expression in antennae in which we co-labeled each of the Ir populations with the driver lines used in the TaDa experiments. Only Ir75a neurons displayed robust nuclear Pdm3 immunofluorescence, although a subset of Ir40a neurons had above-background signals (Figure 5A–B).

To examine the role of pdm3 in Ir neurons we performed transgenic RNA interference (RNAi) using peb-Gal4. This driver is expressed broadly in OSNs from soon after their terminal divisions (~16 hr APF) throughout the initiation of olfactory receptor expression (~40–48 hr APF) (Li et al., 2020; Sweeney et al., 2007). peb-Gal4-driven pdm3^RNAi (peb>pdm3^RNAi) led to loss of most IR75a expression, with the remaining detectable neurons expressing very low levels of this receptor (Figure 5C–D). By contrast, robust expression of all other Irs was maintained (Figure 5C–D). We noted, however, that the number of cells expressing IR75b increased in pdm3^RNAi antennae (Figure 5D). These observations suggest that Pdm3 functions (directly or indirectly) to promote IR75a expression and repress IR75b expression.

The novel IR75b-expressing cells in pdm3^RNAi antennae are mainly located in the proximal region of the antenna near the sacculus, where Ir75a neurons are normally found (Figure 6A). This distribution suggested that loss of Pdm3 results in a switch of receptor expression from IR75a to IR75b. We investigated this possibility by examining the activity of transcriptional reporters for these receptors, encompassing minimal regulatory DNA sequences fused to CD4:tdGFP (hereafter, GFP) (Prieto-Godino et al., 2017). Both Ir75a-GFP and Ir75b-GFP transgenes are faithfully expressed in IR75a and IR75b protein-expressing neurons (Figure 6B–C). In pdm3^RNAi antennae Ir75a-GFP expression is greatly diminished, while Ir75b-GFP is expressed in many ectopic cells (Figure 6B–D).

We further used these Ir-GFP transgenes to examine the glomerular projection of Ir75a and Ir75b neurons in pdm3^RNAi animals. In controls, Ir75a-GFP and Ir75b-GFP label exclusively the DP1l and DL2d glomeruli, respectively (Figure 6E), concordant with previous analyses (Prieto-Godino et al., 2017; Silbering et al., 2011). In pdm3^RNAi animals the Ir75a-GFP DP1l signal is almost completely lost (Figure 6E), consistent with the reduction in reporter expression in antennae. By contrast, the Ir75b-GFP signal in the DL2d glomerulus is greatly intensified, and the glomerulus is enlarged (Figure 6E), presumably reflecting the increased number of Ir75b neurons (Figure 6C–D). The simplest interpretation of these observations is that in the absence of Pdm3, Ir75a neurons adopt Ir75b neuron fate including both receptor identity and glomerular innervation pattern.

While OSN fate is established in early pupae (Barish and Volkan, 2015; Jefferis and Hummel, 2006; Yan et al., 2020), the expression of Pdm3 in Ir75a neurons in adult antennae (Figure 5A–B) suggests that it has a persistent role in these cells. To test this idea, we used a temperature-sensitive repressor of Gal4 (Gal80^ts; active at 19°C, inactive at 29°C) to temporally control the onset of peb>pdm3^RNAi. In control conditions, when we continuously suppressed (19°C) or permitted (29°C) peb>pdm3^RNAi, we reproduced our initial observations (Figure 6F, compare to Figure 5D). When pdm3^RNAi was induced only in adults (through a 19°C to 29°C temperature shift after eclosion) the same increase in Ir75b expression and decrease in Ir75a expression was observed (Figure 6F). Notably, we detected several cells expressing both IR75b and IR75a, which are never observed in non-RNAi antennae (Figure 6G–H). These observations indicate that Pdm3 has a persistent role in mature Ir75a neurons to repress Ir75b expression and promote Ir75a expression. Qualitatively similar, although much weaker, changes in receptor expression were observed when we drove pdm3^RNAi using Ir8a-Gal4 (Abuin et al., 2011), which is active only from the second half of pupal development (~48 hr APF) (Figure 6—figure supplement 1). The ectopic expression of Ir75b in Ir75a OSNs was confirmed by examining the projection of Ir75b-GFP-expressing neurons: in adult-only pdm3^RNAi animals – but not in control pan-developmental pdm3^RNAi – Ir75b-GFP signal was detected both in DL2d and DP1l neurons (Figure 6I). This observation is consistent with late loss of pdm3 leading to misexpression of Ir75b-GFP in Ir75a neurons after they have targeted to DP1l.

fmi attracted our attention both for its heterogeneous occupancy across Ir neuron populations (Figure 4D–E), and because this gene encodes an atypical cadherin – comprising an extracellular cadherin-like domain and a GPCR-like domain – with diverse roles in axonal and dendritic targeting in other regions of the nervous system (Berger-Müller and Suzuki, 2011; Chen and Clandinin, 2008; Gao et al., 2000; Lee et al., 2003; Matsubara et al., 2011; Schwabe et al., 2013). To characterize the endogenous expression of Fmi in the olfactory system, we used an antibody to profile protein levels in OSNs. We first examined antennae from mid-pupal stages (48 hr APF) but observed only trace levels of immunoreactivity across OSN soma (Figure 7—figure supplement 1A). At 24 hr APF, this signal was slightly stronger, but much more prominent in the OSN axons as they left the antenna (Figure 7—figure supplement 1A), suggesting that this protein is predominantly transported to the axonal compartment of these neurons. Indeed, Fmi protein was strongly detected in the OSN axons as they enter the antennal lobe at 24 hr APF (Figure 7A). Fmi was still robustly expressed at 48 hr APF as OSNs coalesced onto individual glomeruli with broad, but slightly heterogeneous, expression in glomeruli across the antennal lobe (Figure 7A). Subsequently, Fmi levels began to decrease to undetectable levels in some glomeruli in the adult stage, while persisting in others (Figure 7B). Most of these glomeruli are innervated by Or neurons, but we could detect Fmi in DL2d (Ir75b) and DL2v (Ir75c), but not the neighboring DP1l (Ir75a) (Figure 7B), in partial concordance with the TaDa data (Figure 4E).

From later developmental stages (~48 hr APF), Fmi was also detected in a group of ~16 cells located on the antero-dorsal region of the antennal lobe (Figure 7A). These cells are likely to be LNs, based upon their expression of Elav (Figure 7—figure supplement 1B) and presence of stained processes that innervate the antennal lobe but do not project to higher olfactory centers. Numerous LN subtypes exist, and these often have broad innervation patterns across many glomeruli (Chou et al., 2010a; Liou et al., 2018). To examine the specific contribution of OSNs to the Fmi immunoreactivity observed in the antennal lobe, we depleted Fmi in LNs by RNAi using a combination of Gal4 drivers (1081 Gal4 and 449 Gal4 [Liou et al., 2018]) that cover the vast majority of these interneurons (Figure 7—figure supplement 1C). At mid-pupal development (48 hr APF), differential glomerular expression of Fmi was much more marked compared to control animals (Figure 7C). Although we could not confidently define all glomerular identities at this earlier developmental stage, prior to full maturation of the antennal lobe, Fmi protein levels did not appear to correspond precisely with the TaDa OSs (e.g. the VL2a glomerulus, innervated by Ir84a neurons, was not devoid of Fmi signal). This observation suggests that Fmi is dynamically expressed in different populations of OSNs during development. We note that the protein observed at 48 hr APF must be due to transcription of fmi at an earlier time point than captured by our TaDa analysis. It is also possible that different post-transcriptional regulation within OSN populations contributes to lack of strict correlation between transcript and protein levels with these cells, as noted in other tissues (Liu et al., 2016). Nevertheless, this analysis reveals developmentally dynamic and population-specific control of Fmi expression in OSNs.

To determine the function of Fmi in the developing olfactory system, we induced fmi RNAi in developing antennal tissue first using a constitutive driver (ey-Flp,act>stop>Gal4 [Chai et al., 2019]), which should deplete Fmi expression from the earliest stages of development, and examined adult brains stained with the general neuropil marker nc82 (Bruchpilot). The antennal lobes of these animals showed fully penetrant, severe morphological defects: the stereotyped glomerular boundaries observed in the lobes of control animals were mostly lacking, with only a few, large glomerulus-like subregions detected (Figure 8A). This phenotype was reproduced with an independent RNAi line that targets a distinct region of the coding sequence (Figure 8A).

We further examined this phenotype by visualizing the projection patterns of specific classes of OSNs labeled by receptor promoter-driven fluorescent reporters (Figure 8B–C). Neurons labeled by Ir75a-GFP, Ir75b-GFP, or Ir75c-GFP were no longer constrained to discrete glomeruli (Figure 8B), although the mistargeting was relatively mild within the antennal lobe as a whole. The limited defects of individual neuron populations suggests that the lack of Fmi does not disrupt long-range targeting properties of OSNs, but rather impacts local segregation of OSNs. Consistently, double labeling of neurons that normally project to adjacent glomeruli (Ir75a-GFP and Ir75b-Tomato (Tom); Ir84a-GFP and Ir31a-Tom) revealed frequent overlap between these normally-segregated populations (Figure 8C).

The constitutive driver is expressed broadly across the antennal disc (Chai et al., 2019). The observed phenotypes could therefore potentially be due to a requirement for Fmi in disc development rather than in OSNs. To limit fmi RNAi to OSNs, we used peb-Gal4, which produced an equally strong defect in antennal lobe glomerular segregation (Figure 8D). In these animals, the LN-expressed Fmi was still detectable but, in contrast to the expression in OSNs, Fmi was apparently homogeneously distributed across antennal lobe glomeruli (Figure 7—figure supplement 1D). This LN source of Fmi does not appear to contribute to glomerulus formation as LN>fmi^RNAi animals did not exhibit antennal lobe morphological defects (Figure 8D). Similarly, when we targeted fmi to the synaptic partners of OSNs, the PNs – which pre-pattern the glomerular organization of the antennal lobe (Jefferis and Hummel, 2006) – using the GH146 driver, antennal lobe morphology appeared normal (Figure 8D). The absence of defects is consistent with the lack of detectable expression of Fmi in these second-order neurons (Figure 7A). Together, these results indicate that Fmi functions principally, if not exclusively, within OSNs for correct glomerulus formation.

To understand the genesis of this defect during OSN development we used Gal80^ts to temporally-control the onset of peb>fmi^RNAi. When animals were shifted from 19°C to 29°C at 10 hr APF (prior to initial expression of peb-Gal4), all animals displayed antennal lobe morphological defects of varying degrees, and Ir75b-GFP-labeled neurons did not coalesce to a discrete glomerular unit (Figure 8E). When shifted at 20 hr APF, the severity of the phenotype was diminished (Figure 8E). When shifted at 40 hr APF, the antennal lobe and Ir75b neuron targeting were indistinguishable from controls (Figure 8E). These experiments indicate an early developmental function for Fmi.

We examined this early requirement in more detail by imaging OSN axons as they invade the antennal lobe. During 20–25 hr APF, pioneer OSNs reach the antennal lobe in a unique bundle that separates into ventromedial and dorsolateral bundles, which project around the lobe’s border (Hong and Luo, 2014; Joo et al., 2013; Figure 8F). In peb>fmi^RNAi animals, OSNs reach the antennal lobe, but bundle organization is highly disrupted, with subsets of axons apparently defasciculating prematurely in the central area of the lobe (Figure 8F).

Fmi (and its mammalian homologs) have roles in multiple biological processes in the nervous system as well as in the establishment of epithelial planar cell polarity (Berger-Müller and Suzuki, 2011; Keeler et al., 2015; Takeichi, 2007). Genetic and biochemical studies have identified a number of interaction partners of Fmi in different biological contexts. We tested mutations and/or multiple independent RNAi lines for several of these genes to determine whether they function with this cadherin in OSNs (Figure 8—figure supplement 1). Mutations in two core components of the planar cell polarity pathway, frizzled and dishevelled (Usui et al., 1999), had no apparent impact on antennal lobe formation. Similarly, loss of frazzled (encoding a Netrin homolog which function with Fmi in midline axon guidance [Organisti et al., 2015]), golden goal (with which Fmi collaborates in photoreceptor axon guidance [Hakeda-Suzuki et al., 2011]), or espinas (which encodes a LIM domain protein that binds Fmi intracellularly and functions in dendrite self-avoidance [Matsubara et al., 2011]) did not reproduce the fmi phenotype. Finally, although the intracellular signaling mechanism of Fmi is not well understood, the Fmi GPCR-domain has been suggested to act through the Gα_q subunit to regulate dendrite growth (Wang et al., 2016); however, we found no evidence that this signaling mechanism operates in OSNs (Figure 8—figure supplement 1). These observations suggest that Fmi functions through other cellular mechanisms to control OSN axon segregation.

Flies were maintained on a standard wheat flour/yeast/fruit juice diet at 25°C in 12 hr light:12 hr dark conditions. Published mutant and transgenic D. melanogaster strains are described in the Key Resources Table. For the temporal control of pdm3 RNAi with Gal80^ts, animals of the desired genotype were cultivated continuously at 19°C or 29°C, or shifted from 19°C to 29°C after eclosion to permit induction of RNAi in adults; for all conditions, antennae or brains were dissection from 1-week-old flies. For the temporal control of fmi RNAi with Gal80^ts, flies of the desired genotype were cultivated at 19°C; animals were staged by selecting white pupae (designated as 0 hr APF) and shifting these animals to 29°C after 10, 20, or 40 hr, to permit induction of RNAi at different time-points during pupal development. Unless otherwise indicated for specific experiments, antennae or brains were collected from both sexes.

New transgenes were constructed using standard molecular biological procedures and transgenesis was performed by BestGene Inc using the phiC31 site-specific integration system. Ir75a-CD4:tdGFP (Ir75a-GFP) was generated by subcloning the 5’ (7833 bp) and 3’ (1921 bp) genomic sequences from the Ir75a-Gal4 transgene (Silbering et al., 2011) to flank CD4:tdGFP in pDESTHemmarG (Addgene 31221) (Han et al., 2011), and integrated into the attP40 landing site. Ir75c-CD4:tdGFP (Ir75c-GFP) was generated by subcloning the 5’ (3072 bp) and 3’ (2545 bp) genomic sequences from the Ir75c-Gal4 transgene (Prieto-Godino et al., 2017) to flank CD4:tdGFP in pDESTHemmarG and integrated into attP40. Ir75b-CD4:tdTom (Ir75b-Tom) was generated by cloning the 299 bp genomic sequence immediately upstream of the Ir75b start codon (as in the previously-described Ir75b-GFP transgene [Prieto-Godino et al., 2017]) into pDESTHemmarR (Addgene 31222) and integrated independently into attP5 and attP2 (on chromosome II and III, respectively). Ir31a-CD4:tdTom (Ir31a-Tom) was generated by cloning the 2000 bp genomic sequence immediately upstream of the Ir31a start codon, via Gateway recombination, into pDESTHemmarR, and integrated into attP2. Ir84a-CD4:tdGFP (Ir84a-GFP) was generated by cloning the 1964 bp genomic sequence immediately upstream of the Ir84a start codon, via Gateway recombination, into pDESTHemmarG and integrated into VK00027.

Antennae from ~2000 flies (~1–8 days old, mixed sexes) of the desired genotype were harvested by snap-freezing animals in a mini-sieve (Scienceware, Bel-Art Products) with liquid nitrogen, shaking flies to detach and collect appendages in a Petri dish under the sieve containing Triton X-100 (0.1%). Third antennal segments were selected under a binocular microscope by pipetting and transferred to a 1.5 ml Eppendorf tube. After brief rinsing in PBS to remove detergent, antennae (occupying a volume of ~50–70 μl) were resuspended in 50 µl PBS and homogenized manually with a blue pestle (Sigma Aldrich Z359947-100EA) for at least 2 min. Antennal fragments were rinsed off the pestle with 100 µl PBS and 40 μl EDTA (100 mM) and 20 μl RNase (12.5 μg/μl) (Qiagen 19101) were added to the antennal lysate, mixed and allowed to sit for 3–5 min. 20 μl Proteinase K (Qiagen DNeasy Blood and Tissue Kit 69504) and 200 µl Buffer AL were added, and mixed by gently pipetting up and down with a blue tip ~50 times, before incubation in a 56°C heat block for 30 min. The digested lysate was centrifuged at 10,000 rpm for 30 s and the supernatant transferred to a new tube and allowed to cool before proceeding with DNA extraction.

Subsequent sample processing steps for TaDa – that is, DpnI treatment (which cuts at adenine-methylated GATC sites), DamID adaptor ligation, DpnII treatment (which cuts at non-methylated GATC sites, thereby digesting unmethylated DNA fragments), and PCR amplification and purification of methylated GATC fragments – were performed essentially as described (Marshall et al., 2016). PCR products were sonicated using a Covaris S220 to obtain ~300 bp average fragment size; DNA size and quality were verified by Qubit quantification. DamID adaptors were removed by Sau3AI digestion. Truseq Nano libraries were prepared at the Lausanne Genomic Technologies Facility and sequenced (SR 100 bp) on a HiSeq 2500.

To test for an enrichment of the genome occupancy of Dam:PolII relative to Dam alone, we used the ‘damidseq pipeline’ v1.4 (Marshall and Brand, 2015), with release 6 of the D. melanogaster genome as the reference. Within this pipeline, Bowtie v2.2.6 (Langmead and Salzberg, 2012) was implemented to align the Illumina sequence reads to the reference, and data processing called on SAMtools v1.2 (Li et al., 2009) and BEDTools v2 (Quinlan and Hall, 2010). This pipeline outputs log2 ratio files in bedgraph format. We calculated Pearson’s Correlation Coefficient among triplicate datasets based on the bedgraph files, and also used them to generate a single averaged OSs file for each of the seven neuron population data sets. The files containing the averaged OSs were then inputted into the ‘polii.gene.call’ script (https://github.com/owenjm/polii.gene.call; version 1.0.2, Marshall and Brand, 2015; Southall et al., 2013) for estimating an FDR for each gene’s OS. A file detailing the workflow and corresponding code is available on GitLab (https://gitlab.com/roman.arguello/ir_tada).

Ir, Or, Gr, and Obp genes analyzed are listed in Supplementary file 4. The set of genes encoding putative neuropeptides, GPCRs, ion channels (excluding Irs, Ors, and Grs), and neuron projection guidance proteins were based on annotated searches of Flybase (FB2020_06) using the following terms: ‘neuropeptide’, ‘ion channel’, ‘GPCR’, ‘neuron projection guidance’ (Supplementary files 6, 9, 12, 15). The set of genes encoding transcription factors was based on a custom filtered set from the FlyTF.org database v2 (Pfreundt et al., 2010; Supplementary file 18).

For testing the hypothesis that OSs are equal among datasets (neuron populations), we used DESeq2 (Love et al., 2014). Unlike the DamID pipeline, this approach focused on individual GATC fragments within each gene. To quantify read coverage of each fragment, we used the same bam files that were outputted by the DamID pipeline (see above). We converted these to bed files using BEDTools v2 ‘bamToBed’ utility and calculated the coverage of GATC fragments using the Bedtools’ ‘coverage’ utility (Quinlan and Hall, 2010). The threshold for significant PolII occupancy within a dataset was a log₂ fold-change of 1.5 and an adjusted p-value of 0.05. Differential occupancy was tested over the full dataset using a Likelihood-ratio test, where the full model specified neuron populations and DamID condition (Dam:PolII vs. Dam-alone) as factors, and an interaction term (between DamID condition and cell population): ~ cell.pop + cond + cond:cell.pop. The reduced model removed the interaction term (~ cell.pop + cond). This analysis yielded the 1694 genes analyzed in Figure 4C–D. We also carried out pairwise tests using a Wald test within DESeq2. Genes that were identified as candidates for differential expression were required to have ≥2 GATC fragments contributing to the difference (and to be significantly occupied by PolII).

Gene Ontology (GO) enrichment analyses were performed on the subset of genes that showed differential occupancy using clusterProfiler (v3.10.1) (Yu et al., 2012). We compared these genes against the three independent, controlled vocabularies provided by GO Consortium that model Biological Processes, Cellular Component, and Molecular Function (Ashburner et al., 2000).

CATaDa was performed on the seven populations of neurons using the same sets of sequencing reads (i.e. Dam-alone data) acquired for the TaDa analysis. The bioinformatics pipeline followed the general concepts of a previously described pipeline (Aughey et al., 2018), but with modifications. For each neuron population, the genomic sequencing reads from all three biological replicates were aligned to release 6.21 of the D. melanogaster genome with ‘very-sensitive’ settings (Bowtie2 v2.3.0). Following pooling of aligned reads from the replicates (samtools v1.7) (Li et al., 2009), the chromatin accessibilities were represented as the coverage of reads per bin of 10 bases, scaled to 1× average genomic coverage (~142.57 Mb) and averaged over a sliding window of 100 bases using bamCoverage within the deepTools library (v3.1.3) (Ramírez et al., 2014). The chromatin accessibility profile was plotted on genome browser tracks with pyGenomeTracks (v2.1) within the deepTools library. Significantly accessible regions were identified with peak-caller (MACS2, v2.1.1.20160309) (Gaspar, 2018) at an FDR cutoff (q-value) of 0.05. The correlation clustering of peaks among all assayed neuron populations was analyzed using the multiBamSummary program within the deepTools library (v3.1.3) with the Pearson method and bin size of 3000 bases.

Immunofluorescence and RNA FISH on whole-mount antennae or antennal cryosections were performed as described (Saina and Benton, 2013). Immunofluorescence on whole-mount brains was performed essentially as described (Ostrovsky et al., 2013; Silbering et al., 2011). Antibodies used are listed in the Key Resources Table. RNA probes for Ir31a and Ir84a were previously described (Benton et al., 2009).

Quantification of En and Pdm3 immunofluorescence signals in OSN soma nuclei (Figure 2—figure supplement 1B–C and Figure 5A–B) was performed in Fiji (Schindelin et al., 2012). In brief, regions-of-interest (ROIs) were manually defined using the GFP signal (which circumscribes the nucleus) in a single optical slice where the cross-section of a given nucleus has the largest area, followed by automated measurement of the average En or Pdm3 immunofluorescence signal within this ROI. Background fluorescence signals in the tissue were measured in the same way within equivalently sized, arbitrarily chosen ROIs that did not overlap with GFP signals or any other OSN nuclei (visualized with DAPI).

Preliminary experiments were used to assess variance and determine adequate sample sizes in advance of acquisition of the reported data; no statistical approaches were used to predetermine sample size. Details on specific statistical tests for individual experiments are provided in the corresponding figure legend. Unless stated otherwise, statistical significance thresholds were p<0.05 and the tests performed were two-tailed. Measurements were taken from distinct samples, unless otherwise stated, along with a corresponding correction for multiple tests.

TaDa sequencing data and experimental information have been deposited in the ArrayExpress database at EMBL-EBI (https://www.ebi.ac.uk/arrayexpress/) and are available under accession number E-MTAB-8935. Processed data and intermediate files, including the read coverage at GATC motifs, are available at GitLab (https://gitlab.com/roman.arguello/ir_tada). Raw image data for Figures 1 and 5–8, and the figure supplements are available upon request.

A detailed workflow for the TaDa analyses and associated code and data files are available at GitLab (https://gitlab.com/roman.arguello/ir_tada; Arguello, 2021; copy archived at swh:1:rev:2c181d8e15257d2344f4a0187ec3cf3be3e4b583).

All unique biological materials generated in this work are available from the corresponding author upon request.