A neurotrophin functioning with a Toll regulates structural plasticity in a dopaminergic circuit

Experience shapes the brain, as neural circuits can be modified by neural stimulation or the lack of it. The molecular mechanisms underlying structural circuit plasticity and how plasticity modifies behaviour, are poorly understood. Subjective experience requires dopamine, a neuromodulator that assigns a value to stimuli, and it also controls behaviour, including locomotion, learning and memory. In Drosophila, Toll receptors are ideally placed to translate experience into structural brain change. Toll-6 is expressed in dopaminergic neurons (DANs), raising the intriguing possibility that Toll-6 could regulate structural plasticity in dopaminergic circuits. Drosophila neurotrophin-2 (DNT-2) is the ligand for Toll-6, but whether it is required for circuit structural plasticity was unknown. Here, we show that DNT-2 expressing neurons connect with DANs, and they modulate each other. Loss of function for DNT-2 or its receptors Toll-6 and kinase-less Trk-like kek-6 caused DAN and synapse loss, impaired dendrite growth and connectivity, decreased synaptic sites and caused locomotion deficits. By contrast, over-expressed DNT-2 increased dendrite complexity and promoted synaptogenesis. Neuronal activity increased synaptogenesis in DNT-2 and DANs, and over-expression of DNT-2 could mimic this effect. Altering the levels of DNT-2 or Toll-6 could also modify dopamine-dependent behaviours, including locomotion and long-term memory. We conclude that an activity-dependent feedback loop involving dopamine and DNT-2 labelled the circuits engaged, and DNT-2 with Toll-6 and Kek-6 induced structural plasticity in this circuit, modifying brain function.


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 [1][2][3][4].Such changes can be driven by experience, that is, neuronal activity or the lack of it [1,[3][4][5][6][7][8][9][10].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 [11][12][13].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 [14][15][16].They promote neuronal survival, connectivity, neurite growth, synaptogenesis, synaptic plasticity and Long-Term Potentiation (LTP), via their Trk and p75 NTR receptors [14][15][16].In fact, all anti-depressants function by stimulating production of BDNF and signalling via its receptor TrkB, leading to increased brain plasticity [17,18].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 [15].Remarkably, these latter functions are shared with neuroinflammation, which in mammals involves Toll-Like Receptors (TLRs) [19].TLRs and Tolls have universal functions in innate immunity across the animals [20], 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 [21][22][23][24][25][26].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.
Toll receptors are expressed across the Drosophila brain, in distinct but overlapping patterns that mark the anatomical brain domains [44].Tolls share a common signalling pathway downstream, that can drive at least four distinct cellular outcomes -cell death, survival, quiescence, proliferation -depending on context [44][45][46][47].They are also required for connectivity and structural synaptic plasticity and they can also induce cellular events independently of signalling [44,[46][47][48][49][50].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 [45,[51][52][53].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 [44].
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 [58,60,61].This associative circuit is required for learning, long-term memory and goal-oriented behaviour [56,59,62].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 [58].Thus, this raises the possibility that a link of Toll-6 to dopamine could enable translating experience into circuit modification to modulate behaviour.
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 [46,47,49].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.

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).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 1B,D).We additionally found post-synaptic signal at the SMP and pre-synaptic signal at the FLA/PRW (Figure 1D), suggesting bidirectional communication at both sites.Using Multi-Colour Flip-Out (MCFO) to label individual cells stochastically [70,71], single neuron clones revealed variability in the DNT-2A projections across individual flies (Figure 1C), consistently with developmental and activity-dependent structural plasticity in Drosophila [28,32,33,35,44,72,73].We found that DNT-2A neurons are glutamatergic as they express the vesicular glutamate transporter vGlut (Figure 1E, left) and lack markers for other neurotransmitter types (Figure S1).DNT-2A terminals overlapped with those of dopaminergic neurons (Figure 1F), 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 [74].Using reporters, we found that Dop2R is present in DNT-2A neurons, but not Dop1R2 (Figure 1E right, Figure S1C).Altogether, these data showed that DNT-2A neurons are excitatory 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 [47].To identify the cells expressing Toll-6 and kek-6 and explore further their link to the dopaminergic system, we used Toll-6Gal4 [44] and kek-6Gal4 [49] 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 1H, Figure S2D,F Table S1), whilst Kek-6+ neurons included PAM, PAL, PPL1, PPM2 and PPM3 dopaminergic clusters (Figure 1I, Figure S2B,E,F Table S1).DNT-2 can also bind various Tolls and Keks promiscuously [46,47] 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 [44] (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 1G).These cells belong to a circuit required for associative learning, long-term memory and behavioural output, and DANs are also required for locomotion [54,56,59,60,75].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 [76] (Figure 2A and Figure S3).DNT-2A RFP+ outputs included a subset of MB a'b' lobes, ab Kenyon cells, tip of MB b'2, DAL neurons, dorsal fan-shaped body layer and possibly PAM or other dopaminergic neurons (Figure 2A, see 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 [77].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.
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 [49].Importantly, DNT-2FL is spontaneously cleaved into the mature form [46,47] (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 in neighbouring DANs.
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 [78,79].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 FRETbased cAMP sensor, Epac1-camps-50A [80].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), whereas dopamine 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 [81].
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 [15,27].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 [44,82].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 purposely developed DeadEasy DAN software.DNT2 37 /DNT2 18 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, as there were fewer TH+ PAMs in DNT2 37 /DNT2 18 mutants (Figure 3D).Importantly, PAM cell loss was rescued by over-expressing activated Toll-6 CY in DANs in DNT-2 mutants (Figure 3D).Altogether, these data showed that DNT-2 functions via Toll-6 to maintain PAM neuron survival.
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 (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 (Figure 3H).This included Dcp-1+ PAMs (Figure 3H).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 [46,47]. Thus, DNT-2 and Toll-6 knock-down specifically in the adult brain induced apoptosis and PAM-neuron loss.
Altogether, these data showed that 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 PAM neurons, we found that DNT-2 18 /DNT-2 37 mutants completely lacked PAM synapses in the MB b,b' and g lobes (Figure 4B).Interestingly, PAM connections at a,a' lobes were not affected (Figure 4B).This means that DNT-2 is required for synaptogenesis and connectivity of PAMs to MB b,b' and g lobes.PAM-b2b'2 neurons express Toll-6 (Table S1) and their dendrites overlap axonal DNT2 projections.
To test whether this signalling system was required specifically in the adult brain, we used tubGAL80 ts to 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.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 [14,15,27,85].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 postsynaptic 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 [86].Here, the coincidence of increased pre-synaptic Syt-GFP from PAMs and post-synaptic Homer-GFP from DNT-2 neurons at SMP revealed that newly formed synapses were stable.PAM neurons also send an arborisation at the MB b, b', g 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.
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 b, b' 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 [87,88].We tested the effect of DNT-2 or Toll-6 and kek-6 loss of function in climbing, and both DNT-2 37 /DNT-2 18     mutants 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-6or a PAM-GAL4 neuron driver, climbing was also reduced (Figure 6B).Importantly, over-expressing activated Toll-6 CY in DANs rescued the locomotion deficits of DNT-2 mutants, showing that DNT-2 functions via Toll-6 in this context (Figure 6C).
We also tracked freely moving flies in an open arena [89].Interestingly, in that setting, locomotion of homozygous DNT-2 37 /DNT-2 18 mutants was similar to that of controls, but over-expression of Toll-6 CY in 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 [56], 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 (as in [90,91]).Remarkably, overexpression 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 dendrite complexity, and synaptogenesis, and it enhanced locomotion and long-term memory.
It was remarkable to find that dopaminergic neurons in the Drosophila adult brain require sustained trophic support, and this is functionally relevant as it can influence behaviour.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.These findings are reminiscent of the increased apoptosis and cell loss in adult brains with Toll-2 loss of function [44], and the support of DAN survival by Toll-1 and Toll-7 driven autophagy [92].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 [93].This did not result in neuronal loss -which was interpreted as due to Toll-6 functions exclusive to glia [93] -but instead of testing the optic lobes, neurons of the larval abdominal ventral nerve cord (VNC) were monitored [93].In the VNC, Toll-6 and -7 function redundantly and knock-down of both is required to cause neuronal loss in embryos [47], whereas in L3 larvae and pupae the phenotype is compounded by their pro-apoptotic functions [46].It is crucial to consider that the DNT-Toll signalling system can have distinct cellular outcomes depending on context, cell type and time, ie stage [44,46,94].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 [54,82,87,88].Dopamine is required for locomotion, associative reward learning and long-term memory [54-56, 58, 87, 88].In Drosophila, this requires PAM, PPL1 and DAL neurons and their connections to Kenyon Cells and MBONs [56,[58][59][60][61]. DNT-2 neurons are connected to all these neuron types, which express Toll-6 and kek-6, and modifying their levels affected 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 [32].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 [35].Prolonged odour exposure causes structural reduction at the antennal lobe and at the output pre-synaptic sites in the calyx, and habituation [37,95].Similarly, prolonged exposure to CO2 caused a reduction in output responses at the lateral horn and habituation [34].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 [31].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 [28].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 [7,10,18,96].Conversely, neurite, synapse and cell loss, correlate with ageing, neuroinflammation, psychiatric and neurodegenerative conditions [3,27,85,[97][98][99].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 (eg 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 [46,68,100].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 [46,47,68].As a result, overexpressed DNT-2FL does not induce apoptosis and instead it promotes cell survival [46].The same functions are played by over-expressed mature DNT-2CK as by DNT-2FL [46,49,68].In S2 cells, transfected DNT-2CK is not secreted, but when over-expressed in vivo it is functional [46,49,68] (and this work).In fact, overexpressed DNT-2CK also maintains neuronal survival, connectivity and synaptogenesis [46,49,68] (and this work).Similarly, over-expressed mature spz-1-C106 can rescue the spz-1-null mutant phenotype [101] and over-expressed DNT-1CK can promote neuronal survival, connectivity and rescue the DNT-1 mutant phenotype [68].Consistently, both DNT-2FL and DNT-2CK have neuro-protective functions promoting cell survival, neurite growth and synaptogenesis [46,49,68,69] (and this work).Importantly, over-expressing DNT-2FL enables to investigate non-autonomous functions of DNT-2 [49].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 motorneurons, where it induces synaptogenesis [49,69].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) [14,15,27,102,103].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 [99,104].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 [14,15,27].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 [99].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 [99,104].Long-term memory requires synaptogenesis, and in mammals this depends on BDNF, and its role in the protein-synthesis dependent phase of LTP [14,27,105].BDNF localised translation, expression and release are induced to enable long-term memory [14,27,105,106].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 [107,108].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 [109,110].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 by dopamine via 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, alterations in DNT-2 levels could cause seizures.Importantly, alterations in both NTs and dopamine underlie many psychiatric disorders and neurodegenerative diseases in humans [27,99,108,110,111].
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 [45].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 [49].Furthermore, Toll-2 is also neuro-protective in the adult fly brain, and loss of Toll-2 function caused neurodegeneration and impaired behaviour [44].There are six spz/DNT, nine Toll and six kek paralogous genes in Drosophila [49,[112][113][114], and at least seven Tolls and three adaptors are expressed in distinct but overlapping patterns in the brain [44].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, NFkB, JNK and CaMKII [15,27].Similarly, DNTs with Tolls and Keks also activate ERK, NFkB, JNK and CaMKII in the Drosophila CNS [45][46][47]49].However, alternatively, NTs may also use further receptors in mammals.Trks have many kinase-less isoforms, and understanding of their function is limited [115,116].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 [117][118][119][120][121][122][123][124].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 [46] and NGF also functions in immunity and neuroinflammation [125,126].Importantly, TLRs can regulate cell survival, death and proliferation, neurogenesis, neurite growth and collapse, learning and memory [23].They are linked to neuroinflammation, psychiatric disorders, neurodegenerative diseases and stroke [23,127,128].
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 [129][130][131][132].

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 dopaminedependent behavioural performance.These findings mean that DNT-2 is a plasticity factor in the Drosophila brain enabling 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.

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.

DECLARATION OF INTERESTS
The authors declare that they have no conflict of interest.

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.

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 1024x1024, 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 1024x512 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 3072x3072 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 1024x512, with 40x oil lens 2x zoom and 0.35mm step.TH counting in PAM were scanned with Zeiss 710 with a resolution 1024x1024, 40x oil objective, step 1 mm, speed 8. Zeiss LSM900 Airyscan with a resolution of 1024x1024, 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.

Automatic cell counting with DeadEasy plug-ins:
To automatically count cell number labelled with nuclear reporters (e.g.Histone-YFP, nls-tdTomato) and Dcp-1+ cells, we adapted the DeadEasy Central Brain ImageJ plug-in [44].DeadEasy automatically identifies and counts cells labelled with nuclear reporters in 3D stacks of confocal image in the adult central brain.Brains expressing these reporters were dissected, fixed and scanned without stainings.DeadEasy Central Brain was used with threshold set to 75.
To count the TH labelled PAMs, we purposely developed a newly plug-in called DeadEasy DAN.
Since TH-stained images had Poisson noise, a median filter was used to reduce noise, without having large losses at the edges.Then, given that cell walls of irregular intensity are frequently encountered, a 3D morphological closing was performed.Since each cell is identified as a dark area surrounded by pixels of higher intensity and that, in the background, due to noise, it is also possible to find local minima that can be incorrectly identified as cells, 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.Finally, a colour palette was used to visualize each cell more easily.DeadEasy DAN is accurate in controls and mutants, but less accurate with RNAi knock-down genotypes, for which it was not used and TH+ cells were counted manually instead, assisted by the ImageJ cell counter.

Dendrite analysis with Imaris:
To analyze dendritic complexity, image data were processed with Imaris using the "Filament Autodetect" function with the default algorithm and autopatch function.Setting of the threshold was consistent within the same experiment: the largest or smallest diameter of the dendrite, same threshold for background substation, local contrast.Number of dendritic branches, dendritic segments, dendritic branch points and dendritic terminal points were collected to compare the difference 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.Setting of the threshold was consistent across genotypes.Behaviour 1. Startle induced negative geotaxis Assay was carried out as described in [88].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.

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 framerate 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 [89].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 thermogenetic 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 [89] 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).

Appetitive long term memory test:
Appetitive long-term memory was tested as described in Krashes and Waddell [91].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 [91], 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 overexpression experiments with tubGAL80 ts ; DNT-2>DNT-2FL, flies were raised at 18°C until 7-9 days posteclosion.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.S2.S2.S2.S2.

Supplementary Figure S7
Altering DNT-2 levels induced seizures.Knock-down or over-expression of DNT-2 in the adult, using GAL80 ts .Fruit-flies were reared at 18°C from egg laying to adult eclosion, when they were transferred to 30°C and kept there for 5 days prior to testing.To test for seizures, we used the bang-sensitivity test.DNT2 37 /DNT2  S2.

Figure 3 DNT- 2
Figure 3DNT-2 and Toll-6 maintain PAM neuron survival in the developing and adult brain.

Figure 6 DNT- 2 -
Figure 6 DNT-2-induced circuit plasticity modified dopamine-dependent behaviour.(A) DNT-2 Tolls, keks and Toll downstream adaptors in cells related to DNT-2A neurons.Genes expressed in DNT-2 neurons, their potential and/or experimentally verified inputs and outputs, were identified with a combination of reporters (this work) and data from public single-cell RNAseq databases.

Table S2 Statistical
analysis.Table providing full genotypes, sample sizes, statistical tests, multiple comparison corrections and p values.