Introduction

The brain can change throughout life, as new cells are formed or eliminated, axonal and dendritic arbours can grow or shrink, synapses can form or be eliminated (Wiesel 1982, Feldman and Brecht 2005, Holtmaat and Svoboda 2009, Gage 2019). Such changes can be driven by experience, that is, neuronal activity or the lack of it (Wiesel 1982, Maguire et al. 2000, Cotman and Berchtold 2002, Feldman and Brecht 2005, Sur and Rubenstein 2005, Holtmaat and Svoboda 2009, Woollett and Maguire 2011, Chen and Brumberg 2021, Bharmauria et al. 2022). Structural changes result in remodelling of connectivity patterns, and these bring about modifications of behaviour. These can be adaptive, dysfunctional or simply the consequence of opportunistic connections between neurons (Kuner and Flor 2016, Leemhuis et al. 2019, Yang et al. 2020). It is critical to understand how structural modifications to cells influence brain function. This requires linking with cellular resolution molecular mechanisms, neural circuits and resulting behaviours.

In the mammalian brain, the neurotrophins (NTs: BDNF, NGF, NT3, NT4) are growth factors underlying structural brain plasticity (Poo 2001, Lu et al. 2005, Park and Poo 2013). They promote neuronal survival, connectivity, neurite growth, synaptogenesis, synaptic plasticity and Long-Term Potentiation (LTP), via their Trk and p75^NTR receptors (Poo 2001, Lu et al. 2005, Park and Poo 2013). In fact, all anti-depressants function by stimulating production of BDNF and signalling via its receptor TrkB, leading to increased brain plasticity (Casarotto et al. 2021, Castren and Monteggia 2021). Importantly, NTs have dual functions and can also induce neuronal apoptosis, neurite loss, synapse retraction and Long-Term Depression (LTD), via p75^NTR and Sortilin (Lu et al. 2005). Remarkably, these latter functions are shared with neuroinflammation, which in mammals involves Toll-Like Receptors (TLRs)(Squillace and Salvemini 2022). TLRs and Tolls have universal functions in innate immunity across the animals (Gay and Gangloff 2007), and consistently with this, TLRs in the CNS are mostly studied in microglia. However, mammalian TLRs are expressed in all CNS cell types, where they can promote not only neuroinflammation, but also neurogenesis, neurite growth and synaptogenesis and regulate memory – independently of pathogens, cellular damage or disease (Ma et al. 2006, Rolls et al. 2007, Okun et al. 2010, Okun et al. 2011, Patel et al. 2016, Chen, CY et al. 2019). Whether TLRs have functions in structural brain plasticity and behaviour remains little explored, and whether they can function together with neurotrophins in the mammalian brain is unknown.

Progress linking cellular and molecular events to circuit and behavioural modification has been rather daunting and limited using mammals (Wang et al. 2022). The Drosophila adult brain is plastic and can be modified by experience and neuronal activity (Technau 1984, Barth and Heisenberg 1997, Barth et al. 1997, Sachse et al. 2007, Kremer et al. 2010, Sugie et al. 2015, Linneweber et al. 2020, Baltruschat et al. 2021, Coban et al. 2024). Different living conditions, stimulation with odorants or light, circadian rhythms, nutrition, long-term memory and experimentally activating or silencing neurons, modify brain volume, alter circuit and neuronal shape, and remodel synapses, revealing experience-dependent structural plasticity (Heisenberg et al. 1995, Barth and Heisenberg 1997, Barth et al. 1997, Devaud et al. 2001, Gorska-Andrzejak et al. 2005, Sachse et al. 2007, Fernandez et al. 2008, Kremer et al. 2010, Bushey et al. 2011, Sugie et al. 2015, Duhart et al. 2020, Baltruschat et al. 2021, Vaughen et al. 2022, Coban et al. 2024). Furthermore, the Drosophila brain is also susceptible to neurodegeneration (Bolus et al. 2020). However, the molecular and circuit mechanisms underlying structural brain plasticity are mostly unknown in Drosophila.

Toll receptors are expressed across the Drosophila brain, in distinct but overlapping patterns that mark the anatomical brain domains (Li et al. 2020). Tolls share a common signalling pathway downstream, that can drive at least four distinct cellular outcomes – cell death, survival, quiescence, proliferation – depending on context (McIlroy et al. 2013, Foldi et al. 2017, Anthoney et al. 2018, Li et al. 2020). They are also required for connectivity and structural synaptic plasticity and they can also induce cellular events independently of signalling (McIlroy et al. 2013, Ward et al. 2015, McLaughlin et al. 2016, Foldi et al. 2017, Ulian-Benitez et al. 2017, Li et al. 2020). These nervous system functions occur in the absence of tissue damage or infection. This is consistent with the fact that - as well as universal functions in innate immunity - Tolls also have multiple non-immune functions also outside the CNS, including the original discovery of Toll in dorso-ventral patterning, cell intercalation, cell competition and others (Meyer et al. 2014, Pare et al. 2014, Anthoney et al. 2018, Tamada et al. 2021). The Toll distribution patterns in the adult brain and their ability to switch between distinct cellular outcomes means they are ideally placed to translate experience into structural brain change (Li et al. 2020).

We had previously observed that in the adult brain, Toll-6 is expressed in dopaminergic neurons (McIlroy et al. 2013). Dopamine is a key neuromodulator that regulates wakefulness and motivation, experience valence, such as reward, and it is essential for locomotion, learning and memory (Riemensperger et al. 2011, Waddell 2013, Adel and Griffith 2021). In Drosophila, Dopaminergic neurons (DANs) form an associative neural circuit together with mushroom body Kenyon cells (KCs), Dorsal Anterior Lateral neurons (DAL) and Mushroom Body Output neurons (MBONs) (Chen et al. 2012, Aso et al. 2014a, Boto et al. 2020, Adel and Griffith 2021). Kenyon cells receive input from projection neurons of the sensory systems, they then project through the mushroom body lobes where they are intersected by DANs to regulate MBONs to drive behaviour (Heisenberg 2003, Aso et al. 2014b, Boto et al. 2020). This associative circuit is required for learning, long-term memory and goal-oriented behaviour (Chen et al. 2012, Guven-Ozkan and Davis 2014, Adel and Griffith 2021). During experience, involving sensory stimulation from the external world and from own actions, dopamine assigns a value to otherwise neutral stimuli, labelling the neural circuits engaged (Boto et al. 2020). Thus, this raises the possibility that a link of Toll-6 to dopamine could enable translating experience into circuit modification to modulate behaviour.

In Drosophila, Toll receptors can function both independently of ligand-binding and by binding Spätzle (Spz) protein family ligands, also known as Drosophila neurotrophins (DNTs), that are sequence, structural and functional homologues of the mammalian NTs (DeLotto and DeLotto 1998, Weber et al. 2003, Hoffmann et al. 2008a, Hoffmann et al. 2008b, Zhu et al. 2008, Lewis et al. 2013, McIlroy et al. 2013, Foldi et al. 2017). Like mammalian NTs, DNTs also promote cell survival, connectivity, synaptogenesis and structural synaptic plasticity, and they can also promote cell death, depending on context (Zhu et al. 2008, McIlroy et al. 2013, Sutcliffe et al. 2013, Foldi et al. 2017, Ulian-Benitez et al. 2017). As well as Tolls, DNTs are also ligands for Kekkon (Kek) receptors, kinase-less homologues of the mammalian NT Trk receptors and are required for structural synaptic plasticity (Ulian-Benitez et al. 2017). Importantly, the targets regulated by Tolls and Keks - ERK, NFκB, PI3K, JNK, CaMKII – are shared with those of mammalian NT receptors Trk and p75^NTR, and have key roles in structural and functional plasticity across the animals (Park and Poo 2013, Foldi et al. 2017, Ulian-Benitez et al. 2017, Yang et al. 2020, Tamada et al. 2021).

Here, we focus on Drosophila neurotrophin −2 (DNT-2), proved to be the ligand of Toll-6 and Kek-6, with in vitro, cell culture and in vivo evidence (McIlroy et al. 2013, Foldi et al. 2017, Ulian-Benitez et al. 2017). Here we asked how DNT-2, Toll-6 and Kek-6 are functionally related to dopamine, whether they and neuronal activity – as a proxy for experience - can modify neural circuits, and how structural circuit plasticity modifies dopamine-dependent behaviours.

Results

DNT-2A, Toll-6 and Kek-6 neurons are integrated in a dopaminergic circuit

To allow morphological and functional analyses of DNT-2 expressing neurons, we generated a DNT-2Gal4 line using CRISPR/Cas9 and drove expression of the membrane-tethered-GFP FlyBow1.1 reporter. We identified at least 12 DNT-2+ neurons and we focused on four anterior DNT-2A neurons per hemi-brain (Figure 1A, B). Using the post-synaptic marker Denmark, DNT-2A dendrites were found at the prow (PRW) and flange (FLA) region, whereas axonal terminals visualised with the pre-synaptic marker synapse defective 1 (Dsyd1-GFP) resided at the superior medial protocerebrum (SMP) (Figure 1C, C’’). We additionally found post-synaptic signal at the SMP and pre-synaptic signal at the FLA/PRW (Figure 1C, C’), suggesting bidirectional communication at both sites. Using Multi-Colour Flip-Out (MCFO) to label individual cells stochastically (Nern et al. 2015, Costa et al. 2016), single neuron clones revealed variability in the DNT-2A projections across individual flies (Figure 1D), consistently with developmental and activity-dependent structural plasticity in Drosophila (Heisenberg et al. 1995, Kremer et al. 2010, Sugie et al. 2015, Mayseless et al. 2018, Li et al. 2020, Linneweber et al. 2020, Baltruschat et al. 2021). We found that DNT-2A neurons are glutamatergic as they express the vesicular glutamate transporter vGlut (Figure 1E, Figure S1A) and lack markers for other neurotransmitter types (Figure S1). DNT-2A terminals overlapped with those of dopaminergic neurons (Figure 1G), suggesting they could receive inputs from neuromodulatory neurons. In fact, single-cell RNA-seq revealed transcripts encoding the dopamine receptors Dop1R1, Dop1R2, Dop2R and/or DopEcR in DNT-2+ neurons (Croset et al. 2018). Using reporters, we found that Dop2R is present in DNT-2A neurons (Figure 1F, Figure S1B), but not Dop1R2 (Figure S1E). Altogether, these data showed that DNT-2A neurons are glutamatergic neurons that could receive dopaminergic input both at PRW and SMP. DNT-2 functions via Toll-6 and Kek-6 receptors, and Toll-6 is expressed in DANs (McIlroy et al. 2013).

Figure 1

To identify the cells expressing Toll-6 and kek-6 and explore further their link to the dopaminergic system, we used Toll-6Gal4 (Li et al. 2020) and kek-6Gal4 (Ulian-Benitez et al. 2017) to drive expression of membrane-tethered FlyBbow1.1 and assessed their expression throughout the brain. Using anti-Tyrosine Hydroxilase (TH) - the enzyme that catalyses the penultimate step in dopamine synthesis - to visualise DANs, we found that Toll-6+ neurons included dopaminergic neurons from the PAMs, PPL1 and PPL2 clusters (Figure 1I, Figure S2D,F Table S1), whilst Kek-6+ neurons included PAM, PAL, PPL1, PPM2 and PPM3 dopaminergic clusters (Figure 1J, Figure S2B,E,F Table S1). DNT-2 can also bind various Tolls and Keks promiscuously (McIlroy et al. 2013, Foldi et al. 2017) and other Tolls are also expressed in the dopaminergic system: PAMs express multiple Toll receptors (Figure S2F) and all PPL1s express at least one Toll (Figure S2F). Using MCFO clones revealed that both Toll-6 and kek-6 are also expressed in Kenyon cells (Li et al. 2020) (Figure S2I,J, Table S1), DAL neurons (Figure S2G,H, Table S1) and MBONs (Figure S2A-C). In summary, Toll-6 and kek-6 are expressed in DANs, DAL, Kenyon cells and MBONs (Figure 1H). These cells belong to a circuit required for associative learning, long-term memory and behavioural output, and DANs are also required for locomotion (Riemensperger et al. 2011, Chen et al. 2012, Aso et al. 2014b, Boto et al. 2014, Adel and Griffith 2021, Huang et al. 2024). Altogether, our data showed that DNT-2A neurons are glutamatergic neurons that could receive dopamine as they contacted DANs and expressed the Dop2R receptor, and that in turn DANs expressed the DNT-2 receptors Toll-6 and kek-6, and therefore could respond to DNT-2. These data suggested that there could be bidirectional connectivity between DNT-2A neurons and DANs, which we explored below.

Bidirectional connectivity between DNT-2A neurons and DANs

To verify the connectivity of DNT-2A neurons with DANs, we used various genetic tools. To identify DNT-2A output neurons, we used TransTango (Talay et al. 2017) (Figure 2A and Figure S3). DNT-2A RFP+ outputs included a subset of MB α’β’ lobes, αβ Kenyon cells, tip of MB β’2, DAL neurons, dorsal fan-shaped body layer and possibly PAM or other dopaminergic neurons (Figure 2A, Figure S3). Consistently, these DNT-2A output neurons express Toll-6 and kek-6 (Supplementary Table S1). To identify DNT-2A input neurons, we used BAcTrace (Cachero et al. 2020). This identified PAM-DAN inputs at SMP (Figure 2B). Altogether, these data showed that DNT-2A neurons receive dopaminergic neuromodulatory inputs, their outputs include MB Kenyon cells, DAL neurons and possibly DANs, and DNT-2 arborisations at SMP are bidirectional.

Figure 2

To further test the relationship between DNT-2A neurons and DANs, we reasoned that stimulating DANs would provoke either release or production of dopamine. So, we asked whether increasing DNT-2 levels in DNT-2 neurons could influence dopamine levels. For this, we over-express DNT-2 in full-length form (i.e. DNT-2FL), as it enables to investigate non-autonomous functions of DNT-2 (Ulian-Benitez et al. 2017). Importantly, DNT-2FL is spontaneously cleaved into the mature form (McIlroy et al. 2013, Foldi et al. 2017) (see discussion). Thus, we over-expressed DNT-2FL in DNT-2 neurons and asked whether this affected dopamine production, using mRNA levels for TH as readout. Using quantitative real-time PCR (qRT-PCR) we found that over-expressing DNT2-FL in DNT-2 neurons in adult flies increased TH mRNA levels in fly heads (Figure 2C). This showed that DNT-2 could stimulate dopamine production.

Next, we wondered whether in turn DNT-2A neurons, that express Dop2R, could be modulated by dopamine. Binding of dopamine to D2-like Dop2R (also known as DD2R) inhibits adenylyl-cyclase, decreasing cAMP levels (Hearn et al. 2002, Neve et al. 2004). Thus, we asked whether DNT-2A neurons received dopamine and signal via Dop2R. Genetic restrictions did not allow us to activate PAMs and test DNT-2 neurons, so we activated DNT-2 neurons and tested whether Dop2R knock-down would increase cAMP levels. We used the FRET-based cAMP sensor, Epac1-camps-50A (Shafer et al. 2008). When Epac1-camps-50A binds cAMP, FRET is lost, resulting in decreased YFP/CFP ratio over time. Indeed, Dop2R RNAi knock-down in DNT-2A neurons significantly increased cAMP levels (Figure 2D), demonstrating that normally Dop2R inhibits cAMP signalling in DNT-2A cells. Importantly, this result meant that in controls, activating DNT-2A neurons caused dopamine release from DANs that then bound Dop2R to inhibit adenylyl-cyclase in DNT-2A neurons; this inhibition was prevented with Dop2R RNAi knock-down. Altogether, this shows that DNT-2 up-regulated TH levels (Figure 2E), and presumably via dopamine release, this inhibited cAMP in DNT-2A neurons (Figure 2F).

In summary, DNT-2A neurons are connected to DANs, DAL and MB Kenyon cells, all of which express DNT-2 receptors Toll-6 and kek-6 and belong to a dopaminergic as well as associative learning and memory circuit. Furthermore, DNT-2A and PAM neurons form bidirectional connectivity. Finally, DNT-2 and dopamine regulate each other: DNT-2 increased dopamine levels (Figure 2E), and in turn dopamine via Dop2R inhibited cAMP signalling in DNT-2A neurons (Figure 2F). That is, an amplification was followed by negative feedback. This suggested that a dysregulation in this feedback loop could have consequences for dopamine-dependent behaviours and for circuit remodelling by the DNT-2 growth factor.

DNT-2 and Toll-6 maintain survival of PAM dopaminergic neurons in the adult brain

Above we showed that DNT-2 and PAM dopaminergic neurons are connected, so we next asked whether loss of function for DNT-2 or Toll-6 would affect PAMs. In wild-type flies, PAM-DAN number can vary between 220-250 cells per Drosophila brain, making them ideal to investigate changes in cell number (Liu et al. 2012). Maintenance of neuronal survival is a manifestation of structural brain plasticity in mammals, where it depends on the activity-dependent release of the neurotrophin BDNF (Lu et al. 2005, Wang et al. 2022). Importantly, cell number can also change in the adult fly, as neuronal activity can induce neurogenesis via Toll-2, whereas DANs are lost in neurodegeneration models (Feany and Bender 2000, Li et al. 2020). Thus, we asked whether DNT-2 influences PAM-DAN number in the adult brain. We used THGal4; R58E02Gal4 to visualise nuclear Histone-YFP in DANs (Figure 3A) and counted automatically YFP+ PAMs using a purposely modified DeadEasy plug-in developed for the adult fly brain (Li et al 2020). DeadEasy plug-ins were developed and used before to count cells labelled with sparsely distributed nuclear markers in embryos (Zhu et al. 2008, Forero et al. 2009, Forero et al. 2010a, Forero et al. 2010b, McIlroy et al. 2013), larvae (Kato et al. 2011, Forero et al. 2012, Losada-Perez et al. 2016) and adult (Li et al. 2020) Drosophila brains. Here we show that DNT2³⁷/DNT2¹⁸ mutant adult brains had fewer PAMs than controls (Figure 3B). Similarly, Toll-6 RNAi knock-down in DANs also decreased PAM neuron number (Figure 3C). DAN loss was confirmed with anti-TH antibodies and counted manually, as there were fewer TH+ PAMs in DNT2³⁷/DNT2¹⁸mutants (Figure 3D). Importantly, PAM cell loss was rescued by over-expressing activated Toll-6^CYin DANs in DNT-2 mutants (Figure 3D). Altogether, these data showed that DNT-2 functions via Toll-6 to maintain PAM neuron survival.

Figure 3

To ask whether DNT-2 could regulate dopaminergic neuron number specifically in the adult brain, we used tubGal80^ts to conditionally knock-down gene expression in the adult. PAMs were visualised with either R58E02LexA>LexAop-nls-tdTomato or THGal4; R58E02Gal4 >histone-YFP and counted automatically. Adult specific DNT-2 RNAi knock-down decreased Tomato+ PAM cell number (Figure 3E). Similarly, RNAi Toll-6 knock-down in DANs also decreased PAM neuron number (Figure 3F). Furthermore, knock-down of either Toll-6 or DNT-2 in the adult brain caused loss of PAM neurons visualised with anti-TH antibodies and counted manually (Figure 3G). Cell loss was due to cell death, as adult specific DNT-2 RNAi knock-down increased the number of apoptotic cells labelled with anti-Drosophila Cleave caspase-1 (DCP-1) antibodies compared to controls, including Dcp-1+ PAMs and other TH+ cells (Figure 3H). Dcp-1+ cells also included TH-negative cells, consistent with the expression of Toll-6 and kek-6 also in other cell types. By contrast, DNT-2FL over-expression in DNT2 neurons did not alter the incidence of apoptosis (Figure 3G), consistently with the fact that DNT-2FL spontaneously cleaves into the mature form (McIlroy et al. 2013, Foldi et al. 2017). Instead, and importantly, over-expression of DNT-2FL increased PAM cell number (Figure 3G). Thus, DNT-2 and Toll-6 knock-down specifically in the adult brain induced apoptosis and PAM-neuron loss, whereas DNT-2 gain of function increased PAM cell number.

Altogether, these data showed that PAM cell number is plastic, sustained PAM neuron survival in development and in the adult brain depends on DNT-2 and Toll-6, and a reduction in their levels causes DAN cell loss, characteristic of neurodegeneration.

DNT-2 and its receptors are required for arborisations and synapse formation

We next asked whether DNT-2, Toll-6 and Kek-6 could influence dendritic and axonal arbours and synapses of dopaminergic neurons (Figure 4A). Visualising the pre-synaptic reporter Synaptotagmin-GFP (Syt-GFP) in all DANs, we found that DNT-2¹⁸/DNT-2³⁷mutants completely lacked DAN synapses in the MB β,β’ and γ lobes (Figure 4B). Interestingly, DAN connections at α,α’ lobes were not affected (Figure 4B). This means that DNT-2 is required for synaptogenesis and connectivity of PAMs to MB β,β’ and γ lobes.

Figure 4

PAM-β2β’2 neuron dendrites overlap axonal DNT2 projections. Toll-6 RNAi knock-down in PAM - β2β’2 (with split-GAL4 MB301BGal4 (Aso et al. 2014a)), reduced dendrite complexity (Figure 4C). To test whether DNT-2 could alter these dendrites, we over-expressed mature DNT-2CK. DNT-2CK is not secreted (from transfected S2 cells), but it is functional in vivo (Zhu et al. 2008, Foldi et al. 2017, Ulian-Benitez et al. 2017). Importantly, over-expressed DNT-2CK functions cell-autonomously whereas DNT-2FL functions also non-autonomously, but they have similar effects (Zhu et al. 2008, Foldi et al. 2017, Ulian-Benitez et al. 2017). Over-expression of DNT-2CK in PAM-β2β’2 increased dendrite arbour complexity (Figure 4D). Thus, DNT-2 and its receptor Toll-6 are required for dendrite growth and complexity in PAM neurons.

To ask whether DNT-2 could affect axonal terminals, we tested PPL1 axons. PPL1-γ1-pedc neurons have a key function in long-term memory gating (Aso et al. 2012, Placais et al. 2012, Aso et al. 2014a, Boto et al. 2020, Huang et al. 2024) and express both Toll-6 and kek-6 (Table S1). Using split-GAL4 line MB320C-Gal4 to visualise PPL1 axonal arbours, RNAi knock-down of either Toll-6, kek-6 or both together caused axonal misrouting away from the mushroom body peduncle (Figure 4E,E’, Chi Square p<0.05, see Table S2). Similarly, DNT-2FL over-expression also caused PPL1 misrouting (Figure 4F, Chi-Square p<0.05, see Table S2). Thus, DNT-2, Toll-6 and Kek-6 are required for appropriate targeting and connectivity of PPL1 DAN axons.

To test whether this signalling system was required specifically in the adult brain, we used tubGAL80^tsto knock-down Toll-6 and kek-6 with RNAi conditionally in the adult and visualised the effect on synaptogenesis using the post-synaptic reporter Homer-GCaMP and anti-GFP antibodies. Adult-specific Toll-6 kek-6 RNAi knock-down in PAM neurons did not affect synapse number (not shown), but it decreased post-synaptic density size, both at the MB lobe and at the SMP dendrite (Figure 4G). These data meant that the DNT-2 receptors Toll-6 and kek-6 continue to be required in the adult brain for appropriate synaptogenesis.

Altogether, these data showed that DNT-2, Toll-6 and Kek-6 are required for dendrite branching, axonal targeting and synapse formation. The shared phenotypes from altering the levels of DNT-2 and Toll-6 kek-6 in arborisations and synapse formation, support their joint function in these contexts. Importantly, these findings showed that connectivity of PAM and PPL1 dopaminergic neurons depend on DNT-2, Toll-6 and Kek-6.

DNT-2 neuron activation and DNT-2 over-expression induced synapse formation in target PAM dopaminergic neurons

The above data showed that DNT-2, Toll-6 and Kek-6 are required for DAN cell survival, arborisations and synaptogenesis in development and adults. This meant that the dopaminergic circuit remains plastic in adult flies, consistently with their functional plasticity (Boto et al. 2014). Thus, we wondered whether neuronal activity could also induce remodelling in PAM neurons. In mammals, neuronal activity induces translation, release and cleavage of BDNF, and BDNF drives synaptogenesis (Poo 2001, Lu et al. 2005, Lu et al. 2013, Wang et al. 2022). Thus, we first asked whether neuronal activity could influence DNT-2 levels or function. We visualised tagged DNT-2FL-GFP in adult brains, activated DNT-2 neurons with TrpA1 at 30°C, and found that DNT-2 neuron activation increased the number of DNT-2-GFP vesicles produced (Supplementary Figure S4A). Furthermore, neuronal activity also facilitated cleavage of DNT-2 into its mature form. In western blots from brains over-expressing DNT-2FL-GFP, the levels of full-length DNT-2FL-GFP were reduced following neuronal activation and the cleaved DNT-2CK-GFP form was most abundant (Supplementary Figure S4B). These findings meant that, like mammalian BDNF, also DNT-2 can be influenced by activity.

Thus, we asked whether neuronal activity and DNT-2 could influence synapse formation. We first tested DNT-2 neurons. Activating DNT-2 neurons altered DNT-2 axonal arbours (Figure 5A) and it increased Homer-GFP+ synapse number in the DNT-2 SMP arbour (Figure 5B and Figure S5). Next, as DNT-2 and PAMs form bidirectional connexions at SMP (Figure 1, 2), we asked whether activating DNT-2 neurons could affect target PAM neurons. To manipulate DNT-2 neurons and visualise PAM neurons concomitantly, we combined DNT-2GAL4 with the PAM-LexAOP driver. However, there were no available LexA/OP post-synaptic reporters, so we used the pre-synaptic LexAOP-Syt-GCaMP reporter instead, which labels Synaptotagmin (Syt), and GFP antibodies. Activating DNT-2 neurons with TrpA1 increased the number of Syt+ synapses at the PAM SMP arbour (Figure 5C) and reduced their size (Figure 5C). This was consistent with the increase in Homer-GFP+ PSD number in stimulated DNT-2 neurons (Figure 5B). Neuronal activity can induce ghost boutons, immature synapses that are later eliminated (Fuentes-Medel et al. 2009). Here, the coincidence of increased pre-synaptic Syt-GFP from PAMs and post-synaptic Homer-GFP from DNT-2 neurons at SMP suggests that newly formed synapses could be stable. PAM neurons also send an arborisation at the MB β, β’, γ lobes, but DNT-2 neuron activation did not affect synapse number nor size there (Figure 5C). These data showed that activating DNT-2 neurons induced synapse formation at the SMP connection with PAMs.

Figure 5

Finally, we asked whether, like activity, DNT-2FL could also drive synaptogenesis. We over-expressed DNT-2FL in DNT-2 neurons and visualised the effect in PAM neurons. Over-expression of DNT-2FL in DNT-2 neurons did not alter Syt+ synapse number at the PAM SMP dendrite, but it increased bouton size (Figure 5D). By contrast, at the MB β, β’ lobe arborisation, over-expressed DNT-2 did not affect Syt+ bouton size, but it increased the number of output synapses (Figure 5D). This data showed that DNT-2 released from DNT-2 neurons could induce synapse formation in PAM target neurons.

Altogether, these data showed that neuronal activity induced synapse formation, it stimulated production and cleavage of DNT-2, and DNT-2 could induce synapse formation in target neurons.

Structural plasticity by DNT2 modified dopamine-dependent behaviour

Circuit structural plasticity raises the important question of what effect it could have on brain function, i.e. behaviour. Data above showed that DANs and DNT-2 neurons are functionally connected, that loss of function for DNT-2 or its receptors caused dopaminergic neuron loss, altered DAN arborisations and caused synapse loss or reduction in size, and that DNT-2 could induce dendrite branching and synaptogenesis, altogether modifying circuit connectivity. To measure the effect of such circuit modifications on brain function, we used dopamine-dependent behaviours as readout.

Startle-induced negative geotaxis (also known as the climbing assay) is commonly used as a measure of locomotor ability and requires dopamine and specifically PAM neuron function (Riemensperger et al. 2013, Sun et al. 2018). We tested the effect of DNT-2 or Toll-6 and kek-6 loss of function in climbing, and both DNT-2³⁷/DNT-2¹⁸mutants and flies in which DNT-2 was knocked-down in DNT-2 neurons in the adult stage had lower climbing ability than controls (Figure 6A). Similarly, when Toll-6 and kek-6 were knocked-down with RNAi in the adult using a Toll-6- or a PAM-GAL4 neuron driver, climbing was also reduced (Figure 6B). Importantly, over-expressing activated Toll-6^CYin DANs rescued the locomotion deficits of DNT-2 mutants, showing that DNT-2 functions via Toll-6 in this context (Figure 6C).

Figure 6

We also tracked freely moving flies in an open arena (Eyjolfsdottir et al. 2014). Interestingly, in that setting, locomotion of homozygous DNT-2³⁷/DNT-2¹⁸mutants was similar to that of controls, but over-expression of Toll-6^CYin their DANs increased locomotion as flies walked longer distances and spent less time immobile (Figure 6D). Adult flies over-expressing DNT2-FL walked faster (Figure 6E and Figure S5) and so did those where DNT-2 neurons were activated with TrpA1 (Figure 6F and Figure S6), consistently with the fact that neuronal activity increased DNT-2 production (Figure S4A) and that DNT-2FL increased TH levels (Figure 2C). Therefore, increased Toll-6^CY levels in DANs increase locomotion and increased DNT-2 levels are sufficient to boost walking speed. Interestingly, both loss and gain of function for DNT-2 also caused seizures (Figure S7). Thus, dopamine-dependent locomotion is regulated by the function of DNT-2, Toll-6 and Kek-6.

Next, as dopamine is an essential neurotransmitter for learning and memory (Adel and Griffith 2021), we asked whether DNT-2 might influence appetitive olfactory conditioning. Starved flies were trained to associate a sugar reward with an odour (CS+) while another odour was presented without sugar (CS-), and their preference for CS+ versus CS-was measured, 24h after training (Tempel et al. 1983, Krashes and Waddell 2008, 2011). Remarkably, over-expression of DNT-2FL in DNT-2 neurons in adults enhanced appetitive long-term memory (Figure 6G), consistent with the positive role of DNT-2 in synaptogenesis that we demonstrated above.

In summary, we have shown that alterations in DNT-2, Toll-6 and Kek-6 levels that caused structural phenotypes in DANs also modified dopamine-dependent behaviours, locomotion and long-term memory.

Discussion

Our findings indicate that structural plasticity and degeneration in the brain are two manifestations of a shared molecular mechanism that could be modulated by experience. Loss of function for DNT-2, Toll-6 and kek-6 caused cell loss, affected arborisations and synaptogenesis in DANs and impaired locomotion; neuronal activity increased DNT-2 production and cleavage and remodelled connecting DNT-2 and PAM synapses; and over-expression of DNT-2 increased TH levels, PAM cell number, dendrite complexity, and synaptogenesis, and it enhanced locomotion and long-term memory.

It was remarkable to find that the number of dopaminergic neurons in the Drosophila adult brain is plastic, and this is functionally relevant as it can influence behaviour. We showed that PAM cell number is variables across individuals, that adult-specific gain of DNT-2 function increases, whereas loss of DNT-2 or Toll-6 function decreases, PAM cell number. Loss of DNT-2 function in mutants, constant loss of Toll-6 function in DANS and adult-restricted knock-down of either DNT-2 (in DNT-2 neurons) or Toll-6 (in Toll-6 neurons and in DANs) all resulted in DAN cell loss, verified with three distinct reporters, and consistently with the increase in DAN apoptosis. Furthermore, DAN cell loss in DNT-2 mutants could be rescued by the over-expression of Toll-6 in DANs. Cell loss was also verified using two reporter types (ie GAL4 based nuclear reporters and cytoplasmic anti-TH antibodies), multiple GAL4 drivers and mutants, and multiple cell counting methods, including automatic cell counting with DeadEasy plug-ins for His-YFP and nls-Tomato (where the signal was of high contrast and sphericity) and software assisted manual cell counting for anti-TH (where the signal is more diffuse and less regular in shape). DeadEasy plug-ins have been used before for reliably counting His-YFP labelled cells in both larval CNS and adult brains, including Kenyon cells (Kato et al. 2011, Forero et al. 2012, Losada-Perez et al. 2016, Li et al. 2020, Harrison et al. 2021). Thus, the finding that loss of DNT-2 and Toll-6 function in the adult brain cause dopaminergic neuron loss is robust. Our findings are reminiscent of the increased apoptosis and cell loss in adult brains with Toll-2 loss of function (Li et al. 2020), and the support of DAN survival by Toll-1 and Toll-7 driven autophagy (Zhang et al. 2024). They are also consistent with a report that loss of function for DNT-2 or Toll-6 induced apoptosis in the third instar larval optic lobes (McLaughlin et al. 2019). This did not result in neuronal loss - which was interpreted as due to Toll-6 functions exclusive to glia (McLaughlin et al. 2019) - but instead of testing the optic lobes, neurons of the larval abdominal ventral nerve cord (VNC) were monitored (McLaughlin et al. 2019). In the VNC, Toll-6 and −7 function redundantly and knock-down of both is required to cause neuronal loss in embryos (McIlroy et al. 2013), whereas in L3 larvae and pupae the phenotype is compounded by their pro-apoptotic functions (Foldi et al. 2017). It is crucial to consider that the DNT-Toll signalling system can have distinct cellular outcomes depending on context, cell type and time, i.e. stage (Foldi et al. 2017, Li et al. 2020, Li and Hidalgo 2021). Our work shows that in the adult Drosophila brain DAN neurons receive secreted growth factors that maintain cellular integrity, and this impacts behaviour. Consistently with our findings, Drosophila models of Parkinson’s Disease reproduce the loss of DANs and locomotion impairment of human patients (Feany and Bender 2000, Riemensperger et al. 2011, Riemensperger et al. 2013, Sun et al. 2018). Dopamine is required for locomotion, associative reward learning and long-term memory (Riemensperger et al. 2011, Riemensperger et al. 2013, Waddell 2013, Sun et al. 2018, Boto et al. 2020, Adel and Griffith 2021). In Drosophila, this requires PAM, PPL1 and DAL neurons and their connections to Kenyon Cells and MBONs (Heisenberg 2003, Chen et al. 2012, Placais et al. 2012, Aso et al. 2014b, Placais et al. 2017, Boto et al. 2020, Adel and Griffith 2021, Huang et al. 2024). DNT-2 neurons are connected to all these neuron types, which express Toll-6 and kek-6, and modifying their levels affects locomotion and long-term memory. Altogether, our data demonstrate that structural changes caused by altering DNT-2, Toll-6 and Kek-6 modified dopamine-dependent behaviours, providing a direct link between molecules, structural circuit plasticity and behaviour.

We used neuronal activation as a proxy for experience, but the implication is that experience would similarly drive the structural modification of circuits labelled by neuromodulators. Similar manipulations of activity have previously revealed structural circuit modifications. For example, hyperpolarising olfactory projection neurons increased microglomeruli number, active zone density and post-synaptic site size in the calyx, whereas inhibition of synaptic vesicle release decreased the number of microglomeruli and active zones (Kremer et al. 2010). There is also evidence that experience can modify circuits and behaviour in Drosophila. For example, natural exposure to light and dark cycles maintains the structural homeostasis of presynaptic sites in photoreceptor neurons, which breaks down in sustained exposure to light (Sugie et al. 2015). Prolonged odour exposure causes structural reduction at the antennal lobe and at the output pre-synaptic sites in the calyx, and habituation (Devaud et al. 2001, Pech et al. 2015). Similarly, prolonged exposure to CO2 caused a reduction in output responses at the lateral horn and habituation (Sachse et al. 2007). Our findings are also consistent with previous reports of structural plasticity during learning in Drosophila. Hypocaloric food promotes structural plasticity in DANs, causing a reduction specifically in connections between DANs and Kenyon cells involved in aversive learning, thus decreasing the memory of the aversive experience (Coban et al. 2024). By contrast, after olfactory conditioning, appetitive long-term memory increased axonal collaterals in projection neurons, and synapse number at Kenyon cell inputs in the calyx (Baltruschat et al. 2021). Our data provide a direct link between a molecular mechanism, synapse formation in a dopaminergic circuit and behavioural performance. Since behaviour is a source of experience, the discovery that a neurotrophin can function with a Toll and a neuromodulator to sculpt circuits provides a mechanistic basis for how experience can shape the brain throughout life.

Importantly, in humans structural brain plasticity (e.g. adult neurogenesis, neuronal survival, neurite growth and synaptogenesis) correlates with anti-depressant treatment, learning, physical exercise and well-being (Cotman and Berchtold 2002, Woollett and Maguire 2011, Castren and Monteggia 2021, Cheng et al. 2022). Conversely, neurite, synapse and cell loss, correlate with ageing, neuroinflammation, psychiatric and neurodegenerative conditions (Holtmaat and Svoboda 2009, Lu et al. 2013, Wohleb et al. 2016, Forrest et al. 2018, Vahid-Ansari and Albert 2021, Wang et al. 2022). Understanding how experience drives the switch between generative and destructive cellular processes shaping the brain is critical to understand brain function, in health and disease. In this context, the mechanism we have discovered could also operate in the human brain. In fact, there is deep evolutionary conservation in DNT-2 vs mammalian NT function (e.g. BDNF), but some details differ. Like mammalian NTs, full-length DNTs/Spz proteins contain a signal peptide, an unstructured pro-domain and an evolutionarily conserved cystine-knot of the NT family (Zhu et al. 2008, Arnot et al. 2010, Foldi et al. 2017). Cleavage of the pro-domain releases the mature cystine-knot (CK). In mammals, full-length NTs have opposite functions to their cleaved forms (e.g. apoptosis vs cell survival, respectively). However, DNT-2FL is cleaved by intracellular furins, and although DNT-2 can be found both in full-length or cleaved forms in vivo, it is most abundantly found cleaved (Zhu et al. 2008, McIlroy et al. 2013, Foldi et al. 2017). As a result, over-expressed DNT-2FL does not induce apoptosis and instead it promotes cell survival (Foldi et al. 2017). The same functions are played by over-expressed mature DNT-2CK as by DNT-2FL (Zhu et al. 2008, Foldi et al. 2017, Ulian-Benitez et al. 2017). In S2 cells, transfected DNT-2CK is not secreted, but when over-expressed in vivo it is functional (Zhu et al. 2008, Foldi et al. 2017, Ulian-Benitez et al. 2017) (and this work). In fact, over-expressed DNT-2CK also maintains neuronal survival, connectivity and synaptogenesis (Zhu et al. 2008, Foldi et al. 2017, Ulian-Benitez et al. 2017) (and this work). Similarly, over-expressed mature spz-1-C106 can rescue the spz-1-null mutant phenotype (Hu et al. 2004) and over-expressed DNT-1CK can promote neuronal survival, connectivity and rescue the DNT-1 mutant phenotype (Zhu et al. 2008). Consistently, both DNT-2FL and DNT-2CK have neuro-protective functions promoting cell survival, neurite growth and synaptogenesis (Zhu et al. 2008, Sutcliffe et al. 2013, Foldi et al. 2017, Ulian-Benitez et al. 2017) (and this work). Importantly, over-expressing DNT-2FL enables to investigate non-autonomous functions of DNT-2 (Ulian-Benitez et al. 2017). We have shown that DNT-2FL can induce synaptogenesis non-autonomously in target neurons. Similarly, DNT-2 is a retrograde factor at the larval NMJ, where transcripts are located post-synaptically in the muscle, DNT-2FL-GFP is taken up from muscle by motoneurons, where it induces synaptogenesis (Sutcliffe et al. 2013, Ulian-Benitez et al. 2017). Importantly, we have shown that neuronal activity increased production of tagged DNT-2-GFP, and its cleavage. In mammals, neuronal activity induces synthesis, release and cleavage of BDNF, leading to neuronal survival, dendrite growth and branching, synaptogenesis and synaptic plasticity (ie LTP)(Poo 2001, Horch and Katz 2002, Lu et al. 2005, Arikkath 2012, Wang et al. 2022). Like BDNF, DNT-2 also induced synaptogenesis and increased bouton size.

It may not always be possible to disentangle primary from compensatory phenotypes, as Hebbian, homeostatic and heterosynaptic plasticity can concur (Forrest et al. 2018, Jenks et al. 2021). Mammalian BDNF increases synapse number, spine size and long-term potentiation (LTP), but it can also regulate homeostatic plasticity and long-term depression (LTD), depending on timing, levels and site of action (Poo 2001, Lu et al. 2005, Wang et al. 2022). In this context, neuronal stimulation of DNT-2 neurons induced synapse formation in PAM neuron SMP dendrites, whereas DNT-2 over-expression from DNT-2 neurons increased synapse size at SMP and synapse number in PAM outputs at the mushroom body lobes. These distinct effects could be due to the combination of plasticity mechanisms and range of action. Neuronal activity can induce localised protein synthesis that facilitates local synaptogenesis and stabilises emerging synapses (Forrest et al. 2018). By contrast, DNT-2 induced signalling via the nucleus can facilitate synaptogenesis at longer distances, in output sites. In any case, synaptic remodelling is the result of concurring forms of activity-dependent plasticity altogether leading to modification in connectivity patterns (Forrest et al. 2018, Jenks et al. 2021). Long-term memory requires synaptogenesis, and in mammals this depends on BDNF, and its role in the protein-synthesis dependent phase of LTP (Poo 2001, Minichiello 2009, Wang et al. 2022). BDNF localised translation, expression and release are induced to enable long-term memory (Poo 2001, Lee et al. 2004, Minichiello 2009, Wang et al. 2022). We have shown that similarly, over-expressed DNT-2FL increased both synaptogenesis in sites involved in reward learning, and long-term memory after appetitive conditioning.

The relationship of NTs with dopamine is also conserved. DNT-2 and DAN neurons form bidirectional connectivity that modulates both DNT-2 and dopamine levels. Similarly, mammalian NTs also promote dopamine release and the expression of DA receptors (Blochl and Sirrenberg 1996, Guillin et al. 2001). Furthermore, DAN cell survival is maintained by DNT-2 in Drosophila, and similarly DAN cell survival is also maintained by NT signalling in mammals and fish (Hyman et al. 1991, Sahu et al. 2019). Importantly, we showed that activating DNT-2 neurons increased the levels and cleavage of DNT-2, up-regulated DNT-2 increased TH expression, and this initial amplification resulted in the inhibition of cAMP signalling via the dopamine receptor Dop2R in DNT-2 neurons. This negative feedback could drive a homeostatic reset of both DNT-2 and dopamine levels, important for normal brain function. In fact, we showed that alterations in DNT-2 levels could cause seizures. Importantly, alterations in both NTs and dopamine underlie many psychiatric disorders and neurodegenerative diseases in humans (Hyman et al. 1991, Guillin et al. 2001, Berton et al. 2006, Forrest et al. 2018, Wang et al. 2022).

We have uncovered a novel mechanism of structural brain plasticity, involving a NT ligand functioning via a Toll and a kinase-less Trk-family receptor in the adult Drosophila brain. Toll receptors in the CNS can function via ligand dependent and ligand-independent mechanisms (Anthoney et al. 2018). However, in the context analysed, Toll-6 and Kek-6 function in structural circuit plasticity depends on their ligand DNT-2. This is also consistent with their functions promoting axonal arbour growth, branching and synaptogenesis at the NMJ (Ulian-Benitez et al. 2017). Furthermore, Toll-2 is also neuro-protective in the adult fly brain, and loss of Toll-2 function caused neurodegeneration and impaired behaviour (Li et al. 2020). There are six spz/DNT, nine Toll and six kek paralogous genes in Drosophila (Tauszig et al. 2000, MacLaren et al. 2004, Mandai et al. 2009, Ulian-Benitez et al. 2017), and at least seven Tolls and three adaptors are expressed in distinct but overlapping patterns in the brain (Li et al. 2020). Such combinatorial complexity opens the possibility for a fine-tuned regulation of structural circuit plasticity and homeostasis in the brain.

Drosophila and mammalian NTs may have evolved to use different receptor types to elicit equivalent cellular outcomes. In fact, in mammals, NTs function via Trk, p75^NTR and Sortilin receptors, to activate ERK, PI3K, NFκB, JNK and CaMKII (Lu et al. 2005, Wang et al. 2022). Similarly, DNTs with Tolls and Keks also activate ERK, NFκB, JNK and CaMKII in the Drosophila CNS (McIlroy et al. 2013, Foldi et al. 2017, Ulian-Benitez et al. 2017, Anthoney et al. 2018). However, alternatively, NTs may also use further receptors in mammals. Trks have many kinase-less isoforms, and understanding of their function is limited (Fryer et al. 1996, Stoilov et al. 2002). They can function as ligand sinks and dominant negative forms, but they can also function independently of full-length Trks to influence calcium levels, growth cone extension and dendritic growth and are linked to psychiatric disorders, e.g. depression (Ferrer et al. 1999, Yacoubian and Lo 2000, Ohira et al. 2006, Carim-Todd et al. 2009, Ernst et al. 2009, Ohira and Hayashi 2009, Fenner 2012, Tessarollo and Yanpallewar 2022). Like Keks, perhaps kinase-less Trks could regulate brain plasticity vs. degeneration.

A functional relationship between NTs and TLRs could exist also in humans, as in cell culture, human BDNF and NGF can induce signalling from a TLR (Foldi et al. 2017) and NGF also functions in immunity and neuroinflammation (Levi-Montalcini et al. 1996, Hepburn et al. 2014). Importantly, TLRs can regulate cell survival, death and proliferation, neurogenesis, neurite growth and collapse, learning and memory (Okun et al. 2011). They are linked to neuroinflammation, psychiatric disorders, neurodegenerative diseases and stroke (Okun et al. 2011, Figueroa-Hall et al. 2020, Adhikarla et al. 2021). Intriguingly, genome-wide association studies (GWAS) have revealed the involvement of TLRs in various brain conditions and potential links between NTs and TLRs in, for example, major depression (Sharma 2012, Mehta et al. 2018, Chan et al. 2020, Garrett et al. 2021). Importantly, alterations in NT function underlie psychiatric, neurologic and neurodegenerative brain diseases (Lu et al. 2005, Martinowich et al. 2007, Krishnan and Nestler 2008, Lu et al. 2013, Park and Poo 2013, Wohleb et al. 2016, Yang et al. 2020, Casarotto et al. 2021, Wang et al. 2022) and BDNF underlies the plasticity inducing function of anti-depressants (Lu et al. 2013, Casarotto et al. 2021, Wang et al. 2022). It is compelling to find out whether and how these important protein families – NTs, TLRs and kinase-less Trks - interact in the human brain.

Conclusion

To conclude, we provide a direct link between structural circuit plasticity and behavioural performance, by a novel molecular mechanism. The neurotrophin DNT-2 and its receptors Toll-6 and the kinase-less Trk family Kek-6 are linked to a dopaminergic circuit. Neuronal activity boosts DNT-2, and DNT-2 and dopamine regulate each other homeostatically. Dopamine labels the circuits engaged and DNT-2, a growth factor, with its receptors Toll-6 and Kek-6, drives structural plasticity in these circuits, enhancing dopamine-dependent behavioural performance. These findings mean that DNT-2 is a plasticity factor in the Drosophila brain, that could enable experience-dependent behavioural enhancement. Whether NTs can similarly functions with TLRs and Kinase-less Trks remains to be explored. As behaviour is a source of experience, this has profound implications for understanding brain function and health.

Acknowledgements

We thank our lab, Carolina Rezaval and Thomas Riemensperger for comments on the manuscript; Carolina Rezaval, Reinhard Wolf, Martin Heisenberg and Scott Waddell for advice on - Karina Piotrowska for help with - behaviour experiments; Xiufeng Li for help with programming; Serge Birman, Ann-Shyn Chiang, Ron Davis, André Fialá, Barret Pfeiffer, Xi Rao, Carolina Rezaval, Iris Salecker for flies; DSHB (Iowa) for antibodies; AddGene for plasmids; Bloomington Stock Centre for Drosophila stocks. This work was funded by Marie-Curie Sklodowska Post-Doctoral fellowship to J.S.; Science Without Borders-CAPES PhD Studentship BEX 13380/13-3 to SUB; BBSRC Project Grants BB/R00871X/1 and BB/P004997/1 to A.H.; and Wellcome Trust Investigator Award 223197/Z/21/Z to A.H.

Data availability statement

All data are contained within the manuscript; metadata and statistical analysis details including full genotypes, sample sizes, statistical tests and p values have been provided in Table S2. This work generated fly stocks and molecular constructs, which we will distribute on request. Raw data will be distributed on request.

Additional information

Author contributions

J.S., V.C and A.H. designed and/or executed experiments and/or analysed data; S.U-B, G.L, D.S, X.W., F.R.C, M.M. executed experiments and analysed data; M.G.F, S.C., A.H. developed tools; D.K., G.J, V.C., A.H. supervised; J.S and A.H. wrote the manuscript. A.H. conceived and directed the project. All authors provided feedback on, and contributed to improvements to, the manuscript.

Declaration of interests

The authors declare that they have no conflict of interest.

Materials and methods

Genetics

Mutants: DNT2³⁷ and DNT2¹⁸are protein null(Foldi et al. 2017, Ulian-Benitez et al. 2017). Toll6³¹is a null mutant allele(McIlroy et al. 2013). Driver lines: DNT2-Gal4 is a CRISPR/Cas9-knock-in allele, with GAL4 at the start of the gene (this work, see below). Toll-6-Gal4 was generated by RMCE from MIMIC Toll-6^MIO2127; kek6-Gal4 from MIMIC Kek6^MI12953(Ulian-Benitez et al. 2017, Li et al. 2020). MB320C-Gal4 (BSC68253), MB301B-Gal4 (BSC68311), R58E02-Gal4 (BSC41347), MB247-Gal80 (BSC64306), R58E02-LexA (BSC52740), R14C08-LexA (BSC52473) were from the Drosophila Bloomington Stock Centre (BSC). Dop1R2-LexA, Dop2R-LexA, Gad-LexA were kindly provided by Yi Rao; TH-LexA (gift of Ron Davis), TH-Gal4,R58E02-GAL4 (gift of Serge Birman); G0431-Gal4 (DAL-GAL4), VT49239 nls LexA (DAL LexA) (gifts from Ann-Shyn Chiang); tubGal80ts, Tdc-LexA. Reporter lines: UAS-CD8::GFP, for membrane tethered GFP; UAS-histone-YFP, for YFP-tagged nuclear histone; UASFlybow1.1, constant membrane tethered expression (gift of Iris Salecker); 13xLexAop-nls-tdTomato, nuclear Tomato (gift of B. Pfeiffer); UASCD8::RFP, LexAopCD8::GFP (BSC32229), for dual binary expression; UAS-homer-GCaMP for post-synaptic densities (gift of André Fialá), UAS-syt.eGFP (BSC6925) and LexAop-Syt-GCaMP (BSC64413) for presynaptic sites. For connectivity: UAS-DenMarkRFP, UAS-Dsyd1GFP (gift of Carolina Rezaval); TransTango: yw UAS-myrGFP,QUAS-mtdTomato-3xHA attP8; Trans-Tango@attP40 (BSC77124); BAcTrace 806 (w;LexAop2-Syb::GFP-P10(VK37) LexAop-QF2::SNAP25::HIVNES::Syntaxin(VK18)/CyO; UAS-B3Recombinase (attP2) UAS<B3Stop<BoNT/A (VK5) UAS<B3Stop<BoNT/A(VK27) QUAS-mtdTomato::HA/TM2): MCFO clones: hs-FLPG5.PEST;; 10xUAS (FRT.stop) myr::smGdP-OLLAS 10xUAS (FRT.stop) myr::smGdP-HA 10xUAS (FRT.stop) myr::smGdP-V5-THS-10xUAS (FRT.stop) myr::smGdP-FLAG (BSC64086). Optogenetic activation: 20x UAS-V5-Syn-CsChrimson td tomato (gift of B. Pfeiffer). For thermogenetic activation: UASTrpA1@attP2 (BSC26264) and UAS-TrpA1@attP216 (BSC26263). UAS gene over-expression: UAS-DNT2CK, UAS-DNT2FL-GFP, UAS-DNT2-FL-47C, UAS-Toll-6^CY(McIlroy et al. 2013, Foldi et al. 2017). UAS-RNAi knockdown: UAS-DNT2RNAi (VDRC49195), UAS-Toll6RNAi (VDRC928), y¹v¹; UAS-Toll6RNAi[Trip.HMS04251] (BSC 56048), UAS-kek6RNAi (VDCR 109681), y¹v¹; UAS-Dop2R-RNAi[TRiP.HMC02988](BSC50621).

Molecular biology

DNT-2GAL4 was generated by CRISPR/Cas9 enhanced homologous recombination. 1 kb long 5’ and 3’ homology arms (HA) were amplified by PCR from genomic DNA of wild type flies, using primers for 5’HA: atcgcaccggtttttacaggcaccccatgtctga containing AgeI cutting site and cttgacgcggccgcTGTCAATTCATTCGCCGTCGAT containing NotI cutting site. 3’HA primers were: tattaggcgcgccATGACAAAAAGTATTAAACGTCCGCCC containing AscI cutting site and tactcgactagtgaagcacacccaaaatacccagg containing SpeI cutting site. HAs were sequentially cloned by conventional cloning into pGEM-T2AGal4 vector (Addgene #62894). For gRNA cloning, two 20 nucleotide gRNA oligos (gtcGACAAGTTCTTCTTACCTATG and aaacCATAGGTAAGAAGAACTTGTC) were designed using Optimal Target Finder. BbsI enzyme sites were added: gtc(g) at the 5’ end of sense oligo, and aaac at the 5’ end of the antisense oligo. gRNA is located at 41 bp downstream of the start codon of DNT-2 within the first coding exon. The gRNA was cloned into pU6.3 using conventional ligation. The two constructs were injected in Cas9 bearing flies and red fluorescent (3xP3-RFP) transformants were selected and balanced, after which 3xP3-RFP was removed with CRE-recombinase.

qRT-PCR

qRT-PCR was carried out from 20 whole adult fruit-fly heads, frozen in liquid nitrogen before homogenising in 50 ml Trizol (Ambion #AM9738), followed by a stand RNA extraction protocol. RNA was treated with DNase treatment (ThermoFisher # AM1906) to remove genomic DNA. 200 ng of RNA was used for cDNA synthesis following Goscript Reverse Transcriptase (Promega #237815) protocol. Sample was diluted 1:4 with Nuclease free H₂O. Standard qPCR was performed with SensiFast syb Green Mix (Bioline #B2092020) in ABI qPCR plate (GeneFlow #P3-0292) and machine. To amplify TH mRNA the following primers were used: TH-F: CGAGGACGAGATTTTGTTGGC and TH-R: TTGAGGCGGACCACCAAAG. GAPDH was used as a housekeeping control. Reactions were performed in triplicate. Specificity and size of amplification products were assessed by melting curve analyses. Target gene expression relative to reference gene is expressed as value of 2^-ΔΔCt (where Ct is the crossing threshold).

Conditional expression

Multiple Colour Flip Out clones: DNT2-Gal4, Toll6-Gal4 and kek6-Gal4 were crossed with hsFLP::PEST;;MCFO flies, female offspring were collected and heat-shocked at 37°C in a water bath for 15 mins, then kept for 48h at 25°C before dissecting their brains. TransTango: DNT2-Gal4 or Oregon female virgins were crossed with TransTango males, progeny flies were raised at 18°C constantly and 15 days after eclosion, female flies were selected for immunostaining. Thermogenetic activation with TrpA1: Fruit-flies were bred at 18°C from egg laying to 4 days post-adult eclosion, then shifted to 29°C in a water bath for 24h followed by 24h recovery at room temperature for over-expressed DNT-2FL-GFP; for the other experiments, after breeding as above, adult flies were transferred to an incubator at 30°C, kept there for 24h and then brains were dissected. Conditional gene over-expression and RNAi knockdown: Flies bearing the temperature sensitive GAL4 repressor tubGal80^ts were kept at 18°C from egg laying to adult eclosion, then transferred to 30 °C incubator for 48h for Dcp-1+ and cell counting experiments and for 120h for TH+ cell counting.

Immunostainings

Adult fruit-fly female brains were dissected (in PBS), fixed (in 4% para-formaldehyde, room temperature, 20-30min) and stained following standard protocols. Primary antibodies and their dilutions were as follows: Mouse anti-Brp (nc82) 1:10 (DSHB); Rabbit anti-GFP 1:250 (Thermofisher); Mouse anti-GFP 1:250 (Thermofisher); Chicken anti-GFP 1:500 (Aves); Rabbit anti-FLAG 1:50 (Sigma); Mouse anti-V5 1:50 (Invitrogen); Chicken anti-HA 1:50 (Aves); Rabbit anti-VGlut 1:500 (gift of Hermann); Mouse anti-TH 1:250 (Immunostar); Rabbit anti-TH 1:250 (Novus Biologicals); Rabbit anti-DsRed 1:250 (Clontek); Mouse anti-ChAT4B1 1:250 (DSHB); Rabbit anti-5-HT 1:500 (Immunostar); Rabbit Anti-DCP-1 1:250 (Cell Signalling). Seconday antibodies were all used at 1:500 and all were from Thermofisher: Alexa Flour 488 Goat anti-mouse, Alexa Flour 488 donkey anti-rabbit, Alexa Flour 488 Goat anti-rabbit (Fab’)2, Alexa Flour 488 Goat anti-chicken, Alexa Four 546 Goat anti-rabbit, Alexa Four 546 Goat anti-mouse, Alex Four 647 Goat anti-rabbit, Alex Four 647 Goat anti-mouse.

Microscopy and Imaging

Laser scanning confocal microscopy

Stacks of microscopy images were acquired using laser scanning confocal microscopy with either Zeiss LSM710, 900 or Leica SP8. Brains were scanned with a resolution of 1024×1024, with Leica SP8 20x oil objective and 1 mm step for whole brain and DCP-1 stainings, 40x oil objective and 1 mm step for central brain. Resolution of 1024×512 was used for analysing PAM clusters with 0.96 mm step for cell counting; 0.5 mm step for neuronal morphology; and 63x oil objective with 0.5mm step for neuronal connections. Acquisition speed in Leica SP8 was 400Hz, with no line averaging. Resolution of 3072×3072 was used for single image analysis of synapses, using either Leica SP8 or Zeiss LSM900 and Airyscan acquisition with 40x water objective speed 6, and average 4, or with 1024×512, with 40x oil lens 2x zoom and 0.35mm step. TH counting in PAM were scanned with Zeiss 710 with a resolution 1024×1024, 40x oil objective, step 1 mm, speed 8. Zeiss LSM900 Airyscan with a resolution of 1024×1024, 40x water objective, speed 7, 0.7 zoom and 0.31µm step size was used for acquisition of optical sections of synapses in PAM neurons.

Optogenetics and Epac1 FRET 2-photon imaging

To test whether DNT-2 neurons can respond to dopamine via the Dop2R inhibitory receptor, we used the cAMP sensor Epac1 and 2-photon confocal microscopy. Epac1 is FRET probe, whereby data are acquired from CFP and YFP emission and lower YFP/CFP ratio reveals higher cAMP levels. DNT-2Gal4 flies were crossed to UAS-CsChrimson UAS Epac1 flies, to stimulate DNT-2 neurons and detect cAMP levels in DNT2 neurons. 1-3 day-old DNT2Gal4>UASCsChrimson, UAS Epac1 flies were collected and separated in two groups. Flies bearing DNT2Gal4 UASCsChrimson UAS Epac1 UASDop2RRNAi were fed on 50μM all-trans retinal food for at least 3 days prior to imaging and kept in constant darkness prior to the experiment.

Optogenetic stimulation of fly brains expressing CsChrimson in DNT-2 neurons was carried out using a sapphire 588nm laser, in a 2-photon confocal microscope. For acquisition of YFP and CFP data from Epac1 samples, a FV30-FYC filter was applied, using a 925nm laser for both YFP and CFP imaging. The stimulation laser was targeted onto DNT-2 neuron projections in SMP region for 20s. Acquisition ROI was at DNT-2A cell bodies with a frame rate of around 10Hz. The first acquisition started 10s before the 20s stimulation and consequential acquisition was done every 30s for 10 cycles.

Image analysis of Epac data was carried out using ImageJ. The two channels (YFP and CFP) were separated, and the ratio of YFP/CFP for each pixel was calculated using the ImageJ>Image Calculator by diving YFP channel by CFP channel. The obtained result of YFP/CFP ratio was saved and the mean ratio of YFP/CFP in the ROI was calculated for each time point and 11 time points were used. The 11 values represent the ratio of YFP/CFP change in the cell body upon stimulation, with 30s interval and repeated 10 times.

Cell counting

To count cells labelled with nuclear reporters (e.g. Histone-YFP, nls-tdTomato) and Dcp-1+ cells, where signal is of high intensity, contrast and sphericity, we adapted the DeadEasy Central Brain ImageJ plug-in (Li et al. 2020) for automatic cell counting in adult brains. DeadEasy plug-ins automatically identify and count cells labelled with nuclear reporters in 3D stacks of confocal image in the nervous system of embryos (Forero et al. 2010a, Forero et al. 2010b), larvae (Kato et al. 2011, Forero et al. 2012, Losada-Perez et al. 2016, Harrison et al. 2021) and adult (Li et al. 2020) Drosophila. DeadEasy plug-ins are accurate at counting cells sparsely labelled with nuclear markers, and importantly, treat all genotypes objectively and equally yielding reliable data. Here, adult brains expressing Histone-YFP or nls-td-tomato reporters were dissected, fixed and scanned without staining them. DeadEasy Central Brain was used with threshold set to 75.

To count the TH labelled PAMs, where both the signal and the labelled-cell shape are more irregular, we used assisted manual cell counting using two methods. First, we developed a plug-in called DeadEasy DAN as follows. A median filter was used to reduce Poison noise, without having large losses at the edges. Then, a 3D morphological closing was performed. Next, all very dark pixels were assigned a value of zero. To mark each cell, each chasm in the image was found using a 3D extended h-minimal transform. As more than one local minimum can be found within each cell, which would result in counting a cell multiple times, a 3D inverse dome detection was performed, and then labelled. Thus, each inverse dome was used as a seed to identify each cell. Once the seeds were obtained, a 3D watershed transformation was performed to recover the shape of the cells. Then, we ran DeadEasy DAN on our raw data, to obtain a results stack of images and formed a merged stack between the raw and result stacks, to manually add any missing cells. This assisted cell counting method was effective at producing accurate cell counts with less labour and time than conventional manual counting and worked well for some genotypes (Figure 3D). However, it was less effective with RNAi knock-down genotypes, where the signal can be less intense, for which TH+ cells were counted manually, assisted by the ImageJ cell counter instead.

Dendrite analysis with Imaris

To analyze dendritic complexity, image data were processed with Imaris using the “Filaments” module with the default algorithm and “Autopatch function”. A simple region of interest (ROI) with parallelepiped shape was delimited. Thresholds were set for the largest and the smallest diameter of the dendrite, and this was consistent across samples within the same experiment. The starting point threshold was adjusted to only represent the soma of the neurons, and the “Seed Points threshold” to match the branches of the neurons. “Remove Seed Points Around Starting Points” and “Remove Disconnected Segments” were chosen, keeping the default values. The threshold for background substation and local contrast was consistent across all samples within an experiment. The “Edit function” within the Filaments module was used to correct any inaccuracy detected in the resulting tracing. Number of dendritic branches, dendritic segments, dendritic branch points and dendritic terminal points were collected to compare differences between the groups.

Vesicles, synapses and PSD analysis with Imaris

To analyse the number and volume of Homer-GCaMP GFP+ post-synaptic densities (PSD), Syt-GCaMP GFP+ presynaptic sites and DNT-2FLGFP+ vesicles, optical section images of confocal stacks through the brain were processed with the Imaris “Spot function”. To analyze the number and volume of Homer-GCaMP GFP+ post-synaptic densities (PSD), the “Surface module” from Imaris was used to restrict a region of interest (ROI). Then, “Absolute Intensity Thresholding” method was applied to each sample choosing the same cutoff each time. The resulting surface was applied to mask the original scan. The masked image was processed using “Image Processing module” from Imaris. Background subtraction followed by threshold cutoff filters were applied. Afterwards, the “Spots module” was used as explained below.

An ROI was determined for Syt-GCaMP GFP+ presynaptic sites and DNT-2FLGFP+ vesicles using the “Surface module” with the “Edit Manually” option “Algorithm”. The ROI for the SMP region started in the slide immediately after the last slide where the soma of PAM neurons was detectable, and finished in the last one where the dendrite was visible. The SMP region laterally was delimited by the black space given by the α lobe position. For the MB lobe region, the ROI started in the first slide where γ5,β’2 and β2 was visible (Aso et al., 2014) and finished in the last slide where this structure was appreciable. The surface was used to create a “Masked Channel”, which a posteriori was used to determine the spots using the Spots module.

The “Spots module” algorithm was set to “Different Spot Sizes”. An Estimated XY Diameter was set according to each experiment group, using the same within an experiment. “Background Subtraction” option was selected. “Intensity Center Filter” was used. “Spot Region” type was determined from “Local Contrast”, and the “Region Threshold” according to the “Region Border”. Setting of the threshold was consistent across genotypes.

Behaviour

1. Startle induced negative geotaxis Assay

was carried out as described in (Sun et al. 2018). Groups of approximately 10 male flies of the same genotype were placed in a fresh tube one night before the test, after which flies were transferred to a column formed with two empty tubes 15cm long and 2cm and then habituated for 30mins. Columns were tapped 3-4 times, flies fell to the bottom and then climbed upwards. Multiple rounds of testing were performed 3-7 times in a row per column. The process was filmed and films were analysed. Flies were scored during the first 15s after the tapping, and those that climbed above 13cm and those that climbed below 2cm were counted separately. Results given are mean ± SEM of the scores obtained with ten groups of flies per genotype. The performance index (PI) is defined as 1⁄2[(ntot + ntop − nbot)/ntot], where ntot, ntop, and nbot are the total number of flies, the number of flies at the top, and the number of flies at the bottom, respectively. The assay was carried out at 25°C, 55% humidity. Flies with tubGal80^ts to conditionally overexpress or knock-down were shifted from 18 to 30°at eclosion and kept for 5 days at 30°C to induce Gal4. Experiments were carried in an environmental chamber at 31°C, 60% humidity or in a humidity and temperature-controlled behaviour lab always kept at 25°C.

2. Spontaneous Locomotion in an open arena

Male flies of each genotype were collected and kept in groups of 10-20 flies, in vials containing fresh fly food for 5-9 days. Before filming, three male flies from one genotype were transferred into a 24 mm well of a multi-well plate using an aspirator and habituated for 15 – 20 mins. The multi-well plate with transparent lid and bottom was placed on a white LED light pad (XIAOSTAR Light Box) and either inside a light-shielding black box (PULUZ, 40*40*40cm) in a room with constant temperature (25°C) and humidity (55%) to maintain stable environmental conditions (Figure 7D), or inside a temperature controlled environmental chamber at 18°C or 30°C (Figure 7E,F,). The locomotion behaviour of freely-moving flies was filmed with a camera (Panasonic, HC-V260) in the morning from ZT1-ZT4 and for 10 min at a frame-rate of 25 fps. The 10-min videos were trimmed (from 00:02:00 to 00:07:00) to 5-min videos for analysis using Flytracker software(Eyjolfsdottir et al. 2014). For the DNT-2 mutants and over-expression of Toll-6^CY experiments, flies were bred and tested at 25 °C. To test over-expression of DNT-2 with tubGAL80^ts; DNT2-Gal4>UAS-DNT-2FL, flies were raised at 18°C until eclosion, and controls were kept and tested at 18°C; test groups were transferred directly after eclosion to 30°C for 5 days and were tested in an environmental chamber kept at 30°C, 60%. For thermo-genetic activation of DNT2 neurons using TrpA1 (DNT-2GAL4>UASTrpA1), flies were bred at 18°C and kept at 18°C for 7-9 days post-eclosion. Following habituation at 18°c for 20 mins in the multi-well plates, they were transferred to the 30°C chamber 10 mins before filming to activate TrpA1 and then filmed for the following 10 minutes. Fly locomotion activity was tracked using FlyTracker and calculated (distance and speed) in Matlab(Eyjolfsdottir et al. 2014) using the raw data generated from the tracking procedure. The “walking distance” was calculated as the sum of the distance flies moved and the “walking speed” was the speed of flies only when they were walking, and it was calculated using only frames where flies moved above 4 mm/s (which corresponds to two body lengths).

3. Appetitive long term memory test

Appetitive long-term memory was tested as described in Krashes and Waddell(Krashes and Waddell 2011). The two conditioning odours used were isoamyl acetate, Sigma-Aldrich #24900822, 6mL in 8mL mineral oil (Sigma-Aldrich #330760) and 4-methylcyclohexanol (Sigma-Aldrich # 153095, 10mL in 8mL mineral oil). Groups of 80-120 mixed sex flies were starved in a 1% agar tube filled with a damp 20 x 60 mm piece of filter paper for 18-20 hour before conditioning. During conditioning training, one odorant was presented with a dry filter paper (unconditioned odour, CS-) for 2 minutes, before a 30 second break, and presentation of a second odorant with filter paper coated with dry sucrose (conditioned odour, CS+). The test was repeated pairing the other odorant with sucrose, with a different group of flies to form one replicate. After training, flies were transferred back to agar tubes for testing 24 hours later. Performance index (PI) was calculated in the same way as in Krashes and Waddell(Krashes and Waddell 2011), as the number of flies approaching the conditioned odour minus the number of flies going in the opposite direction, divided by the total number of flies. A single PI values is the average score from the test with the reverse conditioning odour combination. Groups for which the total number of flies among both odorants was below 15 were discarded. For the DNT-2 over-expression experiments with tubGAL80^ts; DNT-2>DNT-2FL, flies were raised at 18°C until 7-9 days post-eclosion. They were then either transferred to and maintained at 23°C (controls) or 30°C for 18-20h starvation, training, and up to testing 24h later.

Statistical analysis

Statistical analyses were carried out using Graph-Pad Prism. Confidence interval was 95%, setting significance at p<0.05. Chi-square tests were carried out when comparing categorical data. Numerical data were tested first for their type of distributions. If data were distributed normally, unpaired Student t-tests were used to compare means between two groups and One Way ANOVA or Welch ANOVA for larger groups, followed by post-doc Dunnett test for multiple comparisons to a fixed control. Two Way ANOVA was used when comparisons to two variables were made. If data were not normally distributed, non-parametric Mann-Whitney U-test for 2 two group comparisons and Kruskal Wallis ANOVA for larger groups, followed by post-hoc Dunn’s multiple comparisons test to a fixed control. Statistical details including full genotypes, sample sizes, tests and p values are provided in Supplementary Table S2.