Drosophila larvae are focused on feeding and have few neurons. Within these bounds, however, there still are behavioural degrees of freedom. This review is devoted to what these elements of flexibility are, and how they come about. Regarding odour–food associative learning, the emerging working hypothesis is that when a mushroom body neuron is activated as a part of an odour-specific set of mushroom body neurons, and coincidently receives a reinforcement signal carried by aminergic neurons, the AC-cAMP-PKA cascade is triggered. One substrate of this cascade is Synapsin, and therefore this review features a general and comparative discussion of Synapsin function. Phosphorylation of Synapsin ensures an alteration of synaptic strength between this mushroom body neuron and its target neuron(s). If the trained odour is encountered again, the pattern of mushroom body neurons coding this odour is activated, such that their modified output now allows conditioned behaviour. However, such an activated memory trace does not automatically cause conditioned behaviour. Rather, in a process that remains off-line from behaviour, the larvae compare the value of the testing situation (based on gustatory input) with the value of the odour-activated memory trace (based on mushroom body output). The circuit towards appetitive conditioned behaviour is closed only if the memory trace suggests that tracking down the learned odour will lead to a place better than the current one. It is this expectation of a positive outcome that is the immediate cause of appetitive conditioned behaviour. Such conditioned search for reward corresponds to a view of aversive conditioned behaviour as conditioned escape from punishment, which is enabled only if there is something to escape from – much in the same way as we only search for things that are not there, and run for the emergency exit only when there is an emergency. One may now ask whether beyond ‘value’ additional information about reinforcement is contained in the memory trace, such as information about the kind and intensity of the reinforcer used. The Drosophila larva may allow us to develop satisfyingly detailed accounts of such mnemonic richness – if it exists.
Behavioural paradigms for associative learning in larval Drosophila
The brain is not the maggot's most impressive organ (Fig. 1). It contains an estimated order of magnitude fewer neurons than that of adult Drosophila, and is correspondingly smaller: the entire antennal lobe of a stage 3 larva fits into a single glomerulus of an adult fly (and a fly brain fits into the antennal lobe of a bee, we hesitate to add). Indeed, a larva's brain is considerably smaller than its salivary glands, betraying that one way or another much of brain function is devoted to food uptake, while a number of other occupations, such as flying, courtship, copulation and egg-laying, obviously are of no concern to a larva. Thus, with the evolutionary ‘outsourcing’ of the feeding stage into a separate, larval form of life in the holometabolous insects, not only did various external structures such as wings, legs and sex organs become dispensable, but a number of brain structures did as well, such as the central complex. Given the relatively slow locomotion of larvae, sensory systems can operate at massively reduced cell numbers. This is particularly striking in the visual system, which in the adult is comprised of approximately 13,000 neurons (2×6000 photoreceptor neurons in the eyes, 3×80 in the ocelli and 2×6 in the Hofbauer–Buchner eyelet) as compared with 24 in larvae [2×12 (Steller et al., 1987)]; we note that in line with our above comment the gustatory, cephalic contact chemosensory system is relatively well equipped in the larva with 80 (Python and Stocker, 2002) to 90 (Colomb et al., 2007) neurons per body side. Therefore, the maggot brain is impressive – in both numerical simplicity and functional ‘focus’. Notably, within these bounds there still are behavioural degrees of freedom for the larvae. This review is devoted to what these elements of flexibility are, and how they come about. It aims at two goals: providing an overview of maggot learning as well as a detailed and comparative account of Synapsin function. This is because Synapsin provides an evolutionarily conserved experimental handle for unbroken chains of explanation between the behavioural and the molecular levels of mnemonic processing.
We restrict ourselves to Pavlovian conditioning. In this form of associative learning, animals learn the predictive relationship between two events, allowing them to behave according to the causal texture of the world (Dickinson, 2001). Regarding adult Drosophila, this research was pioneered by Benzer and colleagues (Quinn et al., 1974). In the presently used form (Tully and Quinn, 1985), the paradigm presents flies with an odour (A) together with electric shock (−), while another odour is presented without shock (B). After such A−/B training, the animals are offered a choice between A and B (A versus B). Because A signals upcoming shock, the flies escape from A and/or move towards the previously unpunished, safe odour B. In order to average out a possible innate preference for B, it is essential to run the experiment in a reciprocal manner (A/B−, test A versus B) in a second set of animals. Given that the two experimental groups differ in only the target parameter, namely the association between odours and shock, this allows one to calculate an associative performance index by comparing the A versus B choice after A−/B training to the A versus B choice after the reciprocal A/B− training (Tully, 1984) (paradigms and findings not using such reciprocal design and thus potentially confounded by non-associative effects are not considered here). Approximately a decade later this paradigm was supplemented with an appetitive version (A+/B, test A versus B as well as the reciprocal A/B+, test A versus B), where a sugar reward instead of shock punishment is used to induce a conditioned search for the sugar at the previously rewarded odour (Tempel et al., 1983). These two forms of learning attracted much experimental attention, and are, owing to the joint efforts of the Drosophila community, among the better understood study cases in the field of learning and memory to date (reviewed in Heisenberg, 2003; Gerber et al., 2004b; Margulies et al., 2005; McGuire et al., 2005; Schwärzel and Müller, 2006; Keene and Waddell, 2007; Heisenberg and Gerber, 2008; Zars, 2010; Davis, 2011).
Five years after the seminal paper on adult Drosophila, a corresponding paradigm for odour–shock associative learning in larvae was introduced (Aceves-Piña and Quinn, 1979), but compared with adult Drosophila the progress in understanding this form of learning has been slow (Tully et al., 1994; Khurana et al., 2009; Pauls et al., 2010a). Notably, it was as much as 25 years later that Scherer et al. (Scherer et al., 2003) introduced an appetitive version of this assay, using fructose as food reward (Gerber et al., 2010). Maybe thanks to the ecological validity of odour–food associations, this low-tech assay turned out to be robust and was adopted both in a number of research programmes and in teaching at graduate, undergraduate as well as high school levels. Subsequently, low concentrations of salt [~0.5 mol l−1 (Niewalda et al., 2008; Russell et al., 2011)] were likewise found to be effective as a reward, whereas high concentrations of salt as well as quinine (Gerber and Hendel, 2006; Niewalda et al., 2008; Selcho et al., 2009; Russell et al., 2011; Schleyer et al., 2011) and mechanosensory disturbances (Eschbach et al., 2011) can work as punishment. Last, but not least, an associative light–tastant learning paradigm is available (Gerber et al., 2004a), but it yields a less favourable signal-to-noise ratio as compared with odour–tastant learning (possibly because it is less ecologically valid), and therefore has received less experimental attention (Kaun et al., 2007; von Essen et al., 2011). In the following sections, we sketch the circuitry and molecular mechanisms for the acquisition and storage of odour–reward memory traces, and will discuss how these memory traces, once established, are used for behavioural organization.
Chemobehavioural circuitry of larval Drosophila
In general, the behavioural repertoire towards odours is largely restricted to orienting movements and subsequent search or escape behaviour, dependent on the presumed nature of the odour source. In other words, animals use odours ‘to guess what's over there’, to decide whether to search for or escape from the odour source, and to prepare for action in case the odour source is successfully tracked, or should escape fail. In order for this ‘guesswork’ to be as informed as possible, the olfactory sensory-motor loop features stages of processing with an enormous potential to discriminate odours, as well as to attach acquired meaning to them.
In contrast, contact chemosensation serves to behave in situations of immediate physical contact. Indeed, contact chemosensation is closely entangled with mechanosensation. Typically, contact chemosensory sensilla in addition to chemosensory receptor neurons also harbour one mechanosensory neuron (Falk et al., 1976; Awasaki and Kimura, 1997; Ishimoto and Tanimura, 2004), serving to integrate the chemosensory ‘what’ with the mechanosensory ‘where’. This may allow for fairly direct and often local, i.e. body-site specific, sensory-motor loops in diverse contexts such as predation and defence, kin and/or nestmate recognition, aggression, the pursuit of courtship and copulation, oviposition and pupariation. Furthermore, cephalic and oral contact chemosensation organizes eating and drinking behaviour, and it is this ‘taste’ system that is of relevance in the present review.
The cellular architecture of the chemobehavioural system in Drosophila was revealed by the systematic analyses of the Stocker group over the past 30 years (reviewed in Stocker, 1994; Stocker, 2001; Vosshall and Stocker, 2007; Stocker, 2008), providing a basis for the ongoing work at the molecular and behavioural levels. It turned out that this architecture in the larva is similar to the one in adults, but with reduced cell numbers. In simplified terms, odours activate specific combinations of the 21 olfactory sensory neurons housed in the dome of the dorsal organ of each body side, largely dependent on the ligand profile of the receptor expressed in the respective cell [for detailed reviews and reference to the relevant original literature, as well as for exceptions to the ‘rules of thumb’ presented here, please consult the following papers (Stocker 1994; Cobb, 1999; Tissot and Stocker, 2000; Gerber and Stocker, 2007; Melcher et al., 2007; Vosshall and Stocker, 2007; Stocker, 2008; Cobb et al., 2009; Gerber et al., 2009)]. Each of these cells expresses one member of the Or receptor gene family [plus the Orco (CG10609) gene product (Vosshall and Hansson, 2011) that is necessary for proper receptor trafficking and function], and projects to one glomerulus in the larval antennal lobe; these glomeruli can be identified by position. In turn, each receptor gene is expressed by one olfactory sensory neuron, and each glomerulus receives input from one of these sensory neurons. Whether the receptors of the Ir gene family (Benton et al., 2009) play a role for larval chemosensation remains to be investigated.
In the antennal lobe, lateral connections between the glomeruli [(Thum et al., 2011); this study also contains the description of hitherto unrecognized neurons connecting the antennal lobe and the subesophageal ganglion] shape the pattern of activation that then is carried forward by the output elements of the antennal lobe, the projection neurons. The projection neurons typically receive input in one glomerulus; correspondingly, a given antennal lobe glomerulus harbours the dendrites from typically one projection neuron. Thus up to this level in the circuit, the olfactory system lacks cellular redundancy. Notably, the axons of the projection neurons are branched: one axon collateral conveys odour information directly towards the lateral horn and presumably onto pre-motor circuits; the other collateral establishes a detour of olfactory information flow via the mushroom bodies. In the input region of the mushroom bodies, the calyx, the coding space is massively expanded, and then is massively reduced. That is, each of the approximately 21 projection neurons typically innervates one of the approximately 35 mushroom body glomeruli (thus, some mushroom body glomeruli apparently receive input from neurons other than the olfactory projection neurons). Different from the situation in adults, many of the mushroom body glomeruli can be identified by position, revealing a reliable relationship between the antennal lobe glomerulus, in which a projection neuron receives input, and the mushroom body glomerulus, to which it provides output. This allows the majority of the projection neurons to be individually identified. Each of the strikingly many mature mushroom body neurons [~600, of which 250–300 are embryonic-born and may be of particular relevance for associative function (Pauls et al., 2010b)] receives convergent input in an apparently random set of one to six mushroom body glomeruli and hence from one to six projection neurons (Fig. 2C). Depending on how many of these inputs need to be activated to drive the mushroom body neuron, it thus may sample up to a third of the larval odour space. In turn, each projection neuron diverges to 30–180 mushroom body neurons [for details of this approximation, see Gerber and Stocker (Gerber and Stocker, 2007)], thus reporting to a third of the mushroom body coding space (Fig. 2D). Assuming that mushroom body neurons need coincident excitatory input from more than one projection neuron to fire, this combined convergent–divergent architecture allows for combinatorial coding and can enhance small differences in input [for a review of temporal aspects of olfactory coding, see Laurent et al. (Laurent et al., 2001)]. Output from the mushroom bodies is drawn by relatively few neurons [a reasonable guess based on Pauls et al. (Pauls et al., 2010b) is that these may be an order of magnitude fewer neurons than mushroom body neurons], which each sample many, if not all, mushroom body neurons, and which project onto pre-motor circuitry.
It is important to note that the pre-motor systems, ill-characterized as they are, thus receive two kinds of olfactory signal, one from the direct pathway and one via the mushroom body detour. As will be argued below, the direct pathway mediates innate behaviour such as olfactory choice in experimentally naive animals, while the mushroom body detour mediates learned behavioural tendencies towards odours.
Compared with the olfactory system, the taste system is less well understood (for reviews, see Gerber and Stocker, 2007; Melcher et al., 2007; Cobb et al., 2009; Gerber et al., 2009; and references therein). It comprises approximately 2×80–90 gustatory sensory neurons, located in six paired sensory structures, three of them external (terminal organ, ventral organ and the bulge of the dorsal organ surrounding the central dome) and three internal (ventral, dorsal and posterior pharyngeal sense organs). The gustatory sensory neurons bypass the brain proper and project to the subesophageal ganglion in a way that depends on the receptor gene(s) expressed and their sense organ of origin (Colomb et al., 2007). Taste information then is relayed to modulatory neurons, which detour towards the brain and/or to the ventral nerve cord, as well as to (pre-)motor circuitry presumably in the ventral nerve cord to mediate taste-related behaviours. Such behaviours include preference for sugars and low salt concentrations, as well as avoidance of high salt concentrations and ‘bitter’ substances; both sugars and salt stimulate feeding at relatively low, but suppress feeding at higher concentrations. However, despite significant recent progress (e.g. Kwon et al., 2011), the principles of circuit organization underlying taste-mediated behaviours in the larva remains unresolved. In particular, we lack a comprehensive view of the entanglement of the taste system with mechanosensation and thus of the ‘what?’ and ‘where?’ of contact chemosensory processing, and of the role of chemosensory input from the gut and/or of metabolic feedback. Also, one can at present only suspect that, in analogy to the situation in adult flies, the ability of the larvae to discriminate between, for example, different kinds of sweetness or bitterness is limited, and that some of these limits are established already at the level of the sensory neurons. In other words, in the gustatory system the sensory-motor ‘watershed’ appears to be pushed relatively far out towards the sensory periphery, categorizing inputs early on into behavioural categories. This organization contrasts with the olfactory system, which features stages with an enormous capacity to discriminate odours, and the flexibility to tell apart or categorize between odours, depending on the task (Mishra et al., 2010).
In any event, from the discussion above it appears as if there is no connection between olfactory and taste pathways (but see Thum et al., 2011). How, then, can odours be associated with tastants? In a seminal paper, Martin Hammer (Hammer, 1993) identified the octopaminergic VUMmx1 neuron in the honeybee, which receives gustatory input likely in the subesophageal ganglion and provides output to the antennal lobe, the mushroom body and the lateral horn. This neuron thus ‘short-circuits’ taste with olfactory pathways, and is sufficient to mediate the rewarding function of sugar in honeybee olfactory learning. This neuron exists in adult (Busch et al., 2009) and larval Drosophila as well (Selcho et al., 2012). In larvae, the net effect of driving subsets of octopaminergic/tyraminergic neurons (TDC-Gal4) can substitute for reward in olfactory learning, while driving subsets of dopaminergic neurons (TH-Gal4) can substitute for punishment (Schroll et al., 2006). However, current research suggests that the mnemonic function of the neurons contained in these sets is not restricted to such reinforcement processing during training. For example, in adult Drosophila, the so-called dopaminergic MB-MP neurons seem to carry visceral and/or metabolic feedback information (‘satiety state’) towards the mushroom body to effectively inhibit mushroom body output (Krashes et al., 2009). Also, it turned out that some dopaminergic and conceivably also octopaminergic/tyraminergic neurons transgress a simple appetitive–aversive dichotomy (Schroll et al., 2006; Honjo and Furukubo-Tokunaga, 2009; Selcho et al., 2009; Kravitz et al., 2012; Liu et al., 2012). What is important for the present discussion is that the convergence of the olfactory pathway is not with the gustatory pathway itself, but with a modulatory signal branching off of the gustatory sensory-motor loop. At which point(s) in the circuit does such memory trace-inducing convergence take place? Or in other words, where in the circuit are odour–reward memory traces formed?
We used a three-step strategy towards memory trace localization (Gerber et al., 2004b; Heisenberg and Gerber, 2008): (1) find a gene that, when mutated, impairs neuronal plasticity and odour–reward learning; (2) test in which parts of the chemobehavioural circuits restoring this gene is sufficient to restore associative function; and (3) test whether these sites of sufficiency also are sites where the expression of that gene is necessary. As a candidate gene for such an analysis, we chose to focus on synapsin (Fig. 3 displays the synapsin gene of Drosophila).
Synapsins are highly abundant soluble neuronal phosphoproteins, located in nerve terminals and associated with the cytoplasmic surface of synaptic vesicle membranes [for reviews of the original literature, please see the following papers (Südhof et al., 1989; DeCamilli et al., 1990; Greengard et al., 1993; Südhof, 1995; Hilfiker et al., 1999; Kao et al., 1999; Ferreira and Rapoport, 2002; Fdez and Hilfiker, 2006; Cesca et al., 2010; Fornasiero et al., 2010; Benfenati, 2011)]. Vertebrates posses a three-member family of synapsin genes, while in invertebrates such as Drosophila only one synapsin gene is found (Klagges et al., 1996); from each gene, typically multiple protein isoforms are produced by alternative splicing. Synapsins are part of the presynaptic molecular network to fine-tune synaptic output, in particular to regulate the balance between the different presynaptic vesicle pools in a phosphorylation-dependent manner. That is, there are likely at least three functionally distinct (yet not necessarily spatially segregated) presynaptic vesicle pools (Rizzoli and Betz, 2004) [for a novel perspective regarding these pools, see the following papers (Denker et al., 2011a; Denker et al., 2011b)]: (1) a reserve pool bound to Actin filaments and not immediately available for release; (2) a readily releasable pool of docked vesicles at the presynaptic membrane primed for immediate exocytosis; and (3) the cycling pool, comprised of those vesicles engaged in the exo-endocytic cycle. Synapsins can bind to synaptic vesicles and to Actin, the latter in a phosphorylation-dependent way. Seminal work of Greengard and colleagues suggested a role of Synapsin in non-associative plasticity (see Benfenati, 2011). Their analyses indicate that phosphorylation of Synapsin reduces its affinity to Actin and thus recruits vesicles from the reserve pool for subsequent rounds of exocytosis. From a comparative perspective, it is noteworthy that domains A and C at the N-terminal side of the protein are shared by all known Synapsin isoforms, whereas additional domains (D to J) at the C-terminal side are less conserved. Phosphorylation sites in the A domain include an evolutionarily highly conserved PKA/CaM I/IV kinase consensus site; also, a candidate membrane-binding region was identified within this A domain. The C domain likewise contains such a candidate membrane-binding region, while other sites within the C domain appear responsible for binding to Actin filaments. These insights into Synapsin function were subsequently enriched in a number of ways (Benfenati, 2011):
Concomitant with the discovery of the requirement of Synapsin in non-associative short-term plasticity in Aplysia (Humeau et al., 2001), it was shown that serotonin induces Synapsin phosphorylation in Aplysia (Angers et al., 2002), likely via the cAMP-PKA pathway (Fiumara et al., 2004; see also Dolphin and Greengard, 1981) and/or the MAP kinase pathway (Giachello et al., 2010), and that these processes impact non-associative short-term plasticity (Fioravante et al., 2007; Doussau et al., 2010; Giachello et al., 2010; Valente et al., 2012). Strikingly, it was recently reported that the synapsin gene of Aplysia features a CRE-recognition site, that serotonin induces de novo synthesis of Synapsin via the CREB1 pathway, and that this is required for non-associative long-term plasticity (Hart et al., 2011).
The description of the crystal structure of a recombinant Synapsin C domain revealed a similarity to ATP-binding enzymes (Esser et al., 1998) and it was then shown that recombinant C domains of Synapsin I, II and III bind ATP in vitro (Hosaka and Südhof, 1998a; Hosaka and Südhof, 1998b). Interestingly, ATP binding by Synapsin I but not by Synapsin II is Ca2+-dependent, while Synapsin III is inhibited by Ca2+.
Additionally, the crystal structure of the C domain suggested a possible dimerization of Synapsins (Esser et al., 1998), which was subsequently confirmed (Hosaka and Südhof, 1999). This process is promoted by ATP and Ca2+ (Brautigam et al., 2004). Given that Actin depolymerization apparently does not disrupt vesicle clusters (Sankaranarayanan et al., 2003), whereas Synapsin ‘neutralization’ by antibodies does (Pieribone et al., 1995), it is speculated that Synapsins may bind vesicles not only to the Actin cytoskeleton but to each other as well (Bykhovskaia, 2011; Shupliakov et al., 2011).
Recently, an ALPS (amphipathic lipid packing sensor) motif in the B domain was discovered (Krabben et al., 2011) that, when mutated, affects the association of Synapsin with curved membranes, including synaptic vesicles, suggesting a novel way of how Synapsin acts in regulating vesicle function.
Additional implications of Synapsins, partially mediated by MAP kinases, include developmentally relevant aspects such as neurite elongation, synaptogenesis and synapse maturation (Fornasiero et al., 2010; Valtorta et al., 2011).
Last but not least, the mosaic expression pattern of different Synapsin isoforms in the vertebrate brain early on suggested a functional heterogeneity not only between domains and phosphorylation sites of Synapsin, but also between isoforms. Indeed, synapsin II, but not synapsin I, mutant mice were later reported to be impaired in associative learning (Silva et al., 1996). Mice lacking all three synapsin genes performed poorly in a number of tested reflexes and their ability to hang from a suspended wire; they also showed seizures upon disturbance, reduced piloerection and difficulties maintaining balance (Gitler et al., 2004). Regarding mnemonic function, these animals also performed poorly in a test for spatial memory, without reported task-relevant disturbances in motivation or motor ability. In humans, mutations in synapsin genes can apparently entail severe neurological, behavioural and/or psychiatric phenotypes (Garcia et al., 2004; Fassio et al., 2011; Porton et al., 2011).
Thus, the synapsin genes contribute to bewilderingly many functions, dependent on which particular species and gene is looked at, which isoform, domain or site is considered, how the respective protein is embedded into the local molecular network at a given presynapse, and the way the respective neuron is integrated into a functional circuit and ultimately into behavioural task(s). In other words, there does not seem to be anything like ‘the’ role of Synapsin. Assigning one particular behavioural role for a particular gene, isoform and molecular interaction, at a particular synapse of a given circuit obviously does not rule out other roles in other contexts. In turn, the discussion below will show that for a precisely defined behavioural task, a cellular site and a molecular mode of Synapsin action can be assigned with satisfying precision.
The presence of Synapsin-like immunoreactivity in invertebrates was known since the mid-1980s, but the first invertebrate gene coding for Synapsin (synapsin, syn, CG3985) was identified from Drosophila only in 1996 in a seminal project of the Buchner group to identify brain-specific proteins (Fig. 3) (Klagges et al., 1996; Hofbauer et al., 2009). The intron/exon structure of this gene exhibits similarity to synapsin genes of vertebrates (Klagges et al., 1996; Godenschwege et al., 2004) and, as in many of these genes, the gene for the tissue inhibitor of metalloproteinase (timp, CG6281) is nested within the syn gene on the antiparallel DNA strand (Pohar et al., 1999). Alternative splicing of the primary transcript produces two synapsin transcripts, leading to two protein isoforms with relatively low molecular mass (70–80 kDa); additionally, a full-length protein (~143 kDa) can be generated via read-through of an amber stop codon into the second open reading frame (Klagges et al., 1996; Nuwal et al., 2011); indeed, polyclonal mouse sera against this ORF2 specifically detect the 143 kDa band (Klagges et al., 1996). The mouse monoclonal antibody SYNORF1 binds in the conserved C domain of Synapsin (epitope LFGGMEVCGL, Fig. 3) (Godenschwege et al., 2004); it has turned out to be a particularly useful tool, because of its high-affinity binding, its suitability in histology and western blotting, and the fact that it can be used across species. Notably, the mentioned antisera against the full-length protein reveal staining patterns not obviously different from that of SYNORF1.
The Drosophila Synapsin protein contains domains homologous to the N-terminal A domain and the central, Actin- and vesicle-binding C domain (50% identity, 89% similarity of amino acids); additionally, the prolin-rich C terminus shows similarity to the E domain found in a-type vertebrate synapsins. Drosophila Synapsin contains a number of predicted phosphorylation sites, including the evolutionarily conserved PKA/CamK I/IV consensus motif RRFS at Ser6 (henceforth called the PKA-1 site), the evolutionarily non-conserved PKA/CamK I/IV consensus motif RRDS at Ser533 (henceforth called the PKA-2 site) and additionally suspect sites for other kinases, i.e. CamK II, prolin-dependent kinase and PKC (Nuwal et al., 2011). Drosophila Synapsin is expressed exclusively in most if not all neurons, starting at mid-embryogenesis. On a subcellular level, the SYNORF1 antibody reveals expression in type-Ib and type-Is boutons of larval motorneurons, but not in type-II and -III boutons (Klagges et al., 1996; Godenschwege et al., 2004). This situation, together with the role of Synapsins in synaptic function and plasticity in mammals and molluscs (see above), prompted the question of the function of Synapsin in Drosophila. To this end, the syn97 mutation was used, which shows a deletion including promoter regions, the first exon and part of the first intron of the synapsin gene (Fig. 3). Using SYNORF1 as well as a mouse polyclonal anti-synapsin antiserum, it was shown that the deletion leads to a complete lack of the Synapsin protein [adults (Godenschwege, 1999; Godenschwege et al., 2004); larvae (Michels et al., 2005; Michels et al., 2011)]. After removing the white1118 mutation (white, CG2759, which is used as genetic background to keep track of transgenes) by repeated outcrossing to wild type, syn97 mutant flies were found to be viable, fertile and normal in a number of simple behavioural tasks (initial olfactory jump response, basic optomotor response, basic walking parameters), as well as in presynaptic structure at the larval neuromuscular junction (i.e. the number of synaptic boutons, integrity of the T-shaped synaptic ribbon, number of vesicles immediately close to it, structure of the subsynaptic reticulum) and in basic transmission at this synapse (Godenschwege et al., 2004). Interestingly, the notion of Synapsin maintaining vesicles in place is consistent with the finding that the synaptic vesicles in the boutons of motoneurons are distributed over larger areas in syn97 mutant larvae than in wild-type larvae; this effect can be phenocopied by motoneuron stimulation in wild-type larvae (Angers et al., 2002; Akbergenova and Bykhovskaia, 2010). Fittingly, it was found that vesicles in syn97 mutants are more likely to be part of the non-reserve pools (Denker et al., 2011a), in turn implying that Synapsin maintains the reserve pool.
Importantly in the current context, it was found that in arguably more demanding and/or complex behaviour assays (Godenschwege et al., 2004) in adult Drosophila [ethanol tolerance, visual object fixation, conditioned courtship suppression, wing beat frequency during sustained flight, habituation of the olfactory jump response, performance in the heat-box conditioning paradigm, and in a paradigm where flies learn about the nutritional value of food (Fujita and Tanimura, 2011)], the syn97 mutant does show phenotypes (Godenschwege et al., 2004). Critically, the syn97 mutation entails an approximately 25% decrease in odour–shock associative function, but leaves intact task-relevant odour processing, shock processing and the respective behavioural faculties to behave towards these stimuli [the mutants also behave normally to odours after training-like exposure to either odours alone or to shock alone (Knapek et al., 2010)]. This prompted us to test the syn97 mutant for associative function at the larval stage, too.
In larval syn97 mutant animals, odour–sugar associative function is reduced by approximately half (Fig. 1B, Fig. 4) (Michels et al., 2005; Michels et al., 2011), an effect that can be phenocopied by reducing Synapsin levels throughout development by transgenically expressing RNAi with elav-Gal4 as the driver (Michels et al., 2011), while all behavioural control experiments (olfactory preference, sugar preference, susceptibility to training-like odour or sugar exposure) did not reveal a phenotype (Michels et al., 2005). As the syn97 mutation thus specifically affects associative function, we wondered when, in which cells and by which molecular mechanism Synapsin would function.
The reduced associative faculties of syn97 mutant larvae in the odour–sugar paradigm can be rescued by acutely restoring Synapsin throughout the nervous system (elav-Gal4; tub-gal80ts; UAS-syncDNA, syn97). Locally restoring Synapsin in the mushroom bodies is sufficient to fully restore associative ability (mb247-Gal4, UAS-syncDNA, syn97), whereas restoring Synapsin in the projection neurons is not (either GH146-Gal4 or NP225-Gal4; UAS-syncDNA, syn97) (in the two aforementioned cases and in all following cases, rescues were performed non-acutely, i.e. without tub-gal80ts). We note that in cases where no rescue is observed, it is necessary to test not only (by immunohistochemistry) whether the transgene of interest is indeed properly expressed, but also whether the Gal4-driver element per se may have an effect in the experimental animals, such that an actually successful rescue is obscured; indeed, no such effects of the Gal4-driver elements were observed in any of the cases of lack-of-rescue (Michels et al., 2011). In any event, if Synapsin is restored in wide areas of the brain excluding the mushroom bodies, learning ability is not restored either (elav-Gal4; mb247-Gal80; UAS-syncDNA, syn97), suggesting that Synapsin in the mushroom body may also be necessary for proper associative function. This latter conclusion of necessity remains tentative, however, because the mb247-Gal80 construct leads to Gal4 suppression also outside the mb247-Gal4 expression pattern and even outside of the mushroom bodies. Also, blocking synaptic output of mb247-Gal4 positive neurons with shibirets (Kitamoto, 2001) reportedly does not impair odour–sugar mnemonic processing (Pauls et al., 2010b), while broader mushroom body expression, arguably in embryonic-born mushroom body neurons in particular, does (201Y-Gal4, NP1131-Gal4). Thus, a Synapsin-dependent memory trace is located in the mushroom bodies, and this may turn out to be the only (i.e. necessary) site where such a trace is established in this particular paradigm. These findings prompted an enquiry into the molecular mode of action of Synapsin in memory trace formation, in particular into its possible role as a target of the AC-cAMP-PKA cascade. Indeed, with respect to odour–shock associative function in adult flies, it was found that the syn97 mutation is non-additive with rut2080, a hypomorphic mutation in the rutabaga gene (CG9533) coding for a type I adenylate cyclase (Knapek et al., 2010).
The role of the AC-cAMP-PKA cascade for olfactory short-term memory has been intensely studied in adult flies (Tomchik and Davis, 2009; Gervasi et al., 2010; and references therein). The rutabaga type I adenylyl cyclase acts as molecular coincidence detector between the reinforcer and the olfactory activation of the mushroom body neurons. On the one hand, the odour leads to presynaptic calcium influx, and hence to an activation of calmodulin, in the subset of mushroom body neurons that is activated by this odour. On the other hand, the reinforcer activates aminergic neurons and hence the respective G-protein coupled amine receptors in many if not all mushroom body neurons. In the larva, an octopaminergic/tyraminergic set of neurons, defined by TDC-Gal4, carries a net rewarding effect, whereas a dopaminergic set of neurons, defined by TH-Gal4, carries a net punishing effect (Schroll et al., 2006) (see final paragraph of Chemobehavioural circuitry of larval Drosophila for qualifying detail). Those subsets of aminergic neurons that target the mushroom bodies likely each connect to most if not all mushroom body neurons, with octopaminergic/tyraminergic and dopaminergic neurons innervating different domains of the mushroom bodies (Pauls et al., 2010b; Selcho et al., 2012). Critically, it is only by the simultaneous activation of the calmodulin (‘odour’) and G-protein pathways (‘reward’ or ‘punishment’, respectively) that the AC is substantially activated. Thus it is only in those mushroom body neurons that receive both odour and reinforcement activation that cAMP levels and PKA activity are boosted, and the respective protein substrates get phosphorylated. However, the behaviourally relevant substrates of PKA in flies remained obscure. Our working hypothesis was that one of these PKA substrates may be Synapsin, such that PKA-mediated phosphorylation of Synapsin recruits reserve-pool vesicles. Thus, a subsequent presentation of the learnt odour would result in enhanced output from an odour-specific set of mushroom body neurons, and ultimately in conditioned behaviour towards the learnt odour (Fig. 5). We therefore reasoned that transgenic mushroom-body expression of a Synapsin protein that carries dysfunctional PKA sites (mb247-Gal4, UAS-syncDNA-PKAS6A/S533A, syn97) will not be able to rescue the syn97 associative defect. This is indeed the case (Michels et al., 2011).
Interestingly, the evolutionarily conserved PKA-1 site, but not the non-conserved PKA-2 site, undergoes ADAR-dependent pre-mRNA editing (Diegelmann et al., 2006). This manifests itself in a discrepancy between the genomic sequence and the sequence of at least the majority of mRNAs, and ultimately the amino acid sequence of the protein: the amino acid motif RRFS, which should be optimal for phosphorylation by PKA, is altered via RNA editing into RGFS in both larval and adult flies. This edited RGFS form of Synapsin can hardly be phosphorylated, at least in an in vitro study using bovine PKA (Diegelmann et al., 2006). It should now be interesting to see in vivo whether and how RNA editing at Ser6 affects the efficacy of phosphorylation by PKA and/or other kinases, and how such fine regulation of phosphorylation impacts associative function. Along these lines we note that, in order to restore the wild-type situation, the rescue experiment displayed in Fig. 4A (left and middle panels) uses the edited form of Synapsin and that it does restore the defect in associative function.
As a working hypothesis, we thus propose that a Synapsin-dependent memory trace underlying attraction to the conditioned odour in the larva is localized within the mushroom body neurons. Whenever a mushroom body neuron, as a part of an odour-specific subset of mushroom body neurons, is activated coincidently with an aminergic reinforcement signal, the cAMP-PKA cascade is triggered. One substrate of this cascade is Synapsin. Phosporylation of Synapsin likely by PKA but possibly also by other kinases ensures an alteration of synaptic strength between this mushroom body neuron and its target(s). If the trained odour is encountered again at test, the same odour-specific subset of mushroom body neurons is activated again, such that now their modified collective output towards their target neuron(s) allows conditioned behaviour. In other words, the memory trace for the association of an odour with a reward consists of the pattern of changed synaptic weights across the array of the mushroom body neurons. But how, actually, is conditioned behaviour organized?
From memory trace to conditioned behaviour
At the moment of testing, the odour-reward memory trace manifests itself in a preference for the previously rewarded over the previously non-rewarded odour (Fig. 1A,B). To see how the memory-trace-to-behaviour process is organized, we manipulated the test conditions such that the larvae were tested either in the absence or in the presence of the reward (Fig. 6A,B) (Gerber and Hendel, 2006; Schleyer et al., 2011). Strikingly, despite identical training and thus in possession of the same odour–reward memory trace, the presence of the training reward prevents the larvae from showing appetitive conditioned behaviour (Fig. 6B, leftmost and middle groups). This was shown to be the case both for sugar and for low salt as reward. In turn, after aversive training, either with quinine or with high salt concentrations, larvae express aversive conditioned behaviour in the presence of the punishment, but not in its absence (Gerber and Hendel, 2006; Schleyer et al., 2011), and the same was found with respect to mechanosensory disturbance as punishment (Eschbach et al., 2011) (for a discussion of the generality of this effect see Schleyer et al., 2011). Importantly, tastants do not influence innate olfactory choice behaviour (Fig. 6C), and in turn odours do not affect innate gustatory behaviour (Fig. 6D). Thus, the gustatory situation specifically affects conditioned olfactory behaviour, while innate olfactory and innate gustatory behaviour seem to be mutually insulated.
These findings can be understood when conditioned behaviour, rather than as response to the odour, is regarded as an action in pursuit of its outcome (Dickinson, 2001; Elsner and Hommel, 2001; Hoffmann, 2003; Gerber and Hendel, 2006; Schleyer et al., 2011). Accordingly, appetitive conditioned behaviour is viewed as ‘informed search’ for food – which is abolished in the presence of the sought-for food. Conversely, aversive conditioned behaviour is viewed as ‘informed escape’ from punishment – which is disabled in the absence of punishment – much in the same way as we only search for things that are not present, and run for the emergency exit in a movie theatre only when there is an emergency. Thus, conditioned behaviour is not an automatic response: associative memory traces can be behaviourally expressed, or not.
Upon further analysis, it turned out that this organization of conditioned behaviour involves a comparative step: after training larvae with a particular concentration of sugar, they express conditioned behaviour only when the sugar concentration at the moment of testing is less than the sugar concentration during training. Given that the larvae of the four leftmost groups displayed in Fig. 6B had all been trained in the same way and thus will have established the same memory trace, it is obviously not the memory trace per se that determines the behaviour of the animals in the test. In turn, animals that are tested under the same sugar concentration but that have been trained with different concentrations also differ in behaviour: if the sugar concentration during training was higher than it is during testing, conditioned ‘search for more’ is expressed (Fig. 6B, rightmost group). Critically, at this very same test concentration of sugar, no conditioned behaviour is seen if the training concentration had been equally high as the testing concentration (Fig. 6B, middle group). Thus, the testing situation per se is not a sufficient determinant of conditioned behaviour, either [this is in contrast to effects of visceral and/or metabolic feedback (Krashes et al., 2009), which arguably shuts down mushroom body output altogether at a site ‘upstream’ of the comparative process under study here]. Rather, it is the comparison between the memory trace and the testing situation that needs to be considered (Fig. 6E): the larvae compare the value of the current situation (the ‘is state’, based on gustatory input) with the value of the odour-activated memory trace (the ‘could-get state’, based on mushroom body output). Clearly, for this comparison to happen, the memory trace has to be ‘read out’, i.e. the mushroom bodies need to actually output – but off-line from behaviour. The circuit towards conditioned search is only closed if the memory trace suggests a place better than the current one; in other words, if there is something to search for, if there is a gain to be expected from tracking down the learned odour. It is this outcome expectation, and not the value of the memory trace, that is the immediate cause of conditioned behaviour.
This flexible, open organization of conditioned behaviour contrasts with the more rigid, closed processing stream mediating innate olfactory behaviour, which is expressed regardless of the testing situation (Fig. 6C). This separation of olfactory processing streams corresponds to the bifurcation of the projection neurons: the direct pathway towards the lateral horn mediates innate, closed kinds of olfactory tendencies, whereas the mushroom body detour corresponds to an open, flexible olfactory processing stream that can be regulated to express conditioned behaviour – or not.
It seems that olfaction and olfactory learning in Drosophila are beginning to be understood satisfyingly well. Indeed, both in genuinely sensory aspects and in terms of olfactory memory trace formation, reasonable working hypotheses seem within reach. These working hypotheses are detailed enough to make the gaps in understanding obvious, with respect to, for example, the exact organization of the aminergic system in terms of mnemonic processing; the processing of odour intensity information and the way it is included in olfactory memories (Yarali et al., 2009); the change in valence of memory upon reversing stimulus timing during associative training (Tanimoto et al., 2004); the molecular and cellular architecture of gustation and the entanglement of gustation with mechanosensory processing; the role of visceral and/or metabolic feedback in associative processing; the molecular network organization of the presynapse; the role of the postsynapse in associative plasticity; the temporal dynamics of the memory trace(s) formed and their respective content; or the mechanisms of integration across sensory modalities.
It is precisely because the situation overall is already fairly satisfying that an extension of scope is now called for. For example, the internally motivating factors of behaviour remain distressingly unclear. Likewise, one may ask whether, beyond the mere ‘good’ and ‘bad’, information about the particular kind of reinforcer and its intensity is contained in the memory trace, whether the animals ‘know’ when and how often they had been rewarded, or whether they experience degrees of ‘certainty’ about what they have learned. The combination of carefully designed and fine-grained behavioural analyses, together the cellular simplicity and genetic tractability of the Drosophila larva, may allow us to develop circuit-level accounts of such mnemonic richness, if it indeed exists.
We thank Swidbert R. Ott, University of Cambridge, for comments on the manuscript; Erich Buchner, Universität Würzburg, for past and present cooperation and comments on the manuscript; Andreas Thum, Universität Konschtanz, and Dennis Pauls, Universität Würzburg, for discussion; the members of our group for their kind collegiality; and Martin Heisenberg and the members of the Lehrstuhl für Genetik und Neurobiologie, Universität Würzburg, for a decade of cooperation, support and inspiration. This contribution is meant to honour Reinhard F. Stocker for fostering the study, and the students, of the Drosophila chemobehavioural system.
We thank the Leibniz Institut für Neurobiologie (LIN) Magdeburg, the Wissenschaftsgemeinschaft Gottfried Wilhelm Leibniz (WGL), the Center for Behavioural Brain Science (CBBS), Magdeburg, the Universität Magdeburg and the Universität Leipzig for institutional support. We acknowledge current project funding from the Deutsche Forschungsgemeinschaft (DFG) via SPP 1392 ‘Integrative Analyses of Olfaction’ and CRC 779 ‘Motivated Behaviour’ (TP B 11), as well as from the Bundesministerium für Bildung und Forschung via the Bernstein Focus ‘Insect Inspired Robotics’. M.S. holds a fellowship of the Studienstiftung des Deutschen Volkes.