Olfactory memories are intensity specific in larval Drosophila

Stimuli can differ in kind and/or intensity. At the sensory level, stimulus kind is defined by the type of receptor activated and its connectivity, while the level or latency of activation of the receptor codes for the intensity of the stimulus. Thus, processing of stimulus-kind and stimulus-intensity is entangled: one cannot conceive of a receptor that is activated, but at no particular level. In turn, a given level of activation must always be the level of activation of a particular receptor. At the perceptual level, however, the entanglement of quality and intensity can be resolved: it is possible to refer to a smell as ‘fruity’ without specifying the intensity of the olfactory impression, or to regard the olfactory impression of a perfume as ‘too much’, without specifying its kind. Clearly, such disentanglement of intensity from quality is a feature of perception, brought about by post-receptor computations. It is one of the more challenging tasks to understand these computations neurobiologically.

In this context, we decided to study intensity processing in olfactory associative function; that is, olfactory discrimination learning can rely either on intensity differences, quality differences, or both. While the coding of odour quality is often proposed to be combinatorial along the olfactory pathway (see Discussion), and although a fairly explicit working hypothesis about short-term odour quality memory trace formation is available (see Discussion), it is less obvious how odour intensity information is organized. In the present paper, we focus on the question whether odour intensity information is included in olfactory memory traces at all.

We tackle this issue using odour–sugar associative conditioning in larval Drosophila (Fig. 1A) (Scherer et al., 2003; Michels et al., 2005; Neuser et al., 2005; Mishra et al., 2010; Chen et al., 2011; Michels et al., 2011; Saumweber et al., 2011a; Saumweber et al., 2011b) (for reviews, see Gerber and Stocker, 2007; Gerber et al., 2009; Diegelmann et al., 2013). This is a suitable system for such a study due to its simplicity in terms of cell number, its genetic tractability and the robustness of the paradigm. Last, but not least, the circuit architecture of the olfactory pathway of the larva (as of insects in general) is functionally analogous to that in vertebrates (for reviews, see Hildebrand and Shepherd, 1997; Strausfeld and Hildebrand, 1999; Korsching, 2002; Davis, 2004; Ache and Young, 2005; Bargmann, 2006; Ramdya and Benton, 2010; Wilson, 2008; Galizia and Rössler, 2010), rendering experimental as well as computational studies of insect olfaction potentially inspiring on a broader scale.

Our approach follows the one advocated for adult flies (Yarali et al., 2009) [that paper also includes a discussion of alternative approaches (Xia and Tully, 2007; DasGupta and Waddell, 2008; Masek and Heisenberg, 2008)]. A distinguishing feature of this approach is that first the dose–effect curves of learnability are described. This allows choosing odour intensities appropriate for an intensity generalization type of experiment (see Fig. 1B), where we train larvae to a medium intensity, but test them with either a lower or higher intensity of the trained odour. The rationale of this experimental design is that if associative testing scores turn out to increase when the testing intensity is higher than the training intensity, this must be because a higher intensity is judged by the larvae as ‘more of the trained’ odour (Fig. 1Bi). If, in contrast, the larvae regard a higher intensity as ‘something different’, we should observe a generalization decrement for the higher testing condition (Fig. 1Bii). This latter result would imply that the memory trace established by the larvae during training is parametrically specific for the trained intensity of the odour.

Third-instar, feeding-stage Drosophila larvae (5 days after egg laying) of the Canton Special wild-type strain were used. The flies were kept in mass culture under a 14 h:10 h light:dark cycle at 25°C and 60–70% relative humidity. For the learning assay, a spoonful of medium containing larvae was placed in an empty Petri dish, and 30 larvae were collected and washed in distilled water. Experiments complied with applicable law.

One day prior to the experiment, Petri dishes of 85 mm inner diameter (Sarstedt, Nümbrecht, Germany) were filled either with a solution of 1% agarose (electrophoresis grade) or 1% agarose with 2 mol l⁻¹ fructose (both from Roth, Karlsruhe, Germany). Once the agarose had solidified, dishes were covered with their lids and left until the following day.

As odours, we used 3-octanol (3-Oct), n-amyl acetate (AM), 1-octen-3-ol (1-Oct-3-ol), linalool (Lin) and 1-octanol (1-Oct) (all from Merck, Darmstadt, Germany; CAS: 589-98-0, 628-63-7, 3391-86-4, 78-70-6, 111-87-5), and hexyl acetate (HA), benzaldehyde (BA) and 4-methylcyclohexanol (MCH) (from Sigma-Aldrich, Steinheim, Germany; CAS: 142-92-7, 100-52-7, 589-91-3). Odours were diluted in paraffin oil (Merck) as described in the Results section. In each case, 10 μl of odour solution was applied to custom-made Teflon containers with an inner diameter of 5 mm, and a perforated cap with seven holes of 0.5 mm diameter each.

For AM, 3-Oct, 1-Oct-3-ol and BA we describe the dose–effect functions of learnability and probe for the intensity specificity of the memory trace; for the remaining odours, we restrict ourselves to documenting the dose–effect functions of learnability and the underlying preferences (supplementary material Fig. S2), either because learnability is undesirably low (1-Oct, MCH, Lin), or in the case of HA because a high similarity, both physico-chemically and perceptually (Chen et al., 2011), seems to make HA redundant to AM. A summary of the dose–effect functions of learnability for all odours can be found in supplementary material Fig. S3.

Data were collected in parallel for all the groups to be statistically compared, using non-parametric analyses throughout. Kruskal–Wallis (KW) tests were used to compare across multiple groups; in case of significance, we separately tested the scores of single groups against zero using one-sample sign (OSS) tests. The significance level for these tests was set to 0.05, maintaining an experiment-wide error rate of 5% by a Bonferroni correction. That is, in a case where for example five groups are to be compared individually with zero, the critical P-level is set to 0.05/5=0.01. The Mann–Whitney U-test (MWU) along with the Bonferroni correction was used to compare two groups with each other. All statistical analyses were performed with Statistica (version 8.0, StatSoft, Tulsa, OK, USA) on a PC.

Performance indices are presented as box plots with the median as mid-line, box boundaries as the 25/75% quartiles and whiskers as the 10/90% quantiles. Sample sizes are given in the figures.

Using AM as odour, we find an optimum function for associative performance indices across odour intensities (Fig. 2Ai; KW test: H=47.4, d.f.=7, P<0.05). Specifically, at intermediate intensities significant associative scores are obtained, whereas the lowest intensity used is apparently not learnable; notably, also at the highest intensity performance indices do not formally differ from chance (Fig. 2Ai; OSS tests with P<0.05/8 as criterion for significance). This is probably because at such high intensity the relatively strong innate preference for AM hinders revealing an associative memory (see supplementary material Fig. S1A). We therefore restrict our choice of odour intensities up to the 1:10 dilution.

To probe for a possible intensity specificity of the AM memory trace, we used an intensity that supports about half-maximal associative performance indices (Fig. 2Aii), allowing us to detect both increases and decreases in scores. Specifically, we chose 1:10⁴ as the medium intensity for training, and then tested larvae either at lower (1:10⁵, 1:10⁶) or higher (1:10³, 1:10², 1:10) intensities. It turns out that as the testing intensities deviate from the training intensity towards either higher or lower intensities, performance indices approach zero (Fig. 3A: OSS tests with P<0.05/6 as criterion for significance; the KW test across all groups yields P<0.05, H=29.4, d.f.=5). Thus, in order to support full retention, the testing intensity needs to match the training intensity; this follows scenario ii in Fig. 1B. Given that for 3-Oct and 1-Oct-3-ol we obtain the same results (Fig. 2B,C, Fig. 3B,C; for statistics, see legends), we conclude that as a rule olfactory associative learning establishes intensity-specific memory traces in larval Drosophila.

In the adult, BA memories do not seem to be intensity specific as assayed in an odour–electric shock associative paradigm: higher-than-trained BA intensities support higher associative performance indices than the actually trained intensity [see fig. 4D in our previous paper (Yarali et al., 2009); for a replication within the present study, see supplementary material Fig. S5), thus following scenario i in Fig. 1B. We therefore include BA in our analysis concerning the larva as well.

In the dose–effect description of the learnability of BA, associative performance indices increase as odour intensity is increased (Fig. 2Di; KW test: H=43.3, d.f.=6, P<0.05). We chose high, medium and low intensities from this dose–response curve (Fig. 2Dii), and trained the larvae with the medium intensity. Different groups of larvae were then tested with either the same medium, with lower, or with higher intensities, respectively. As expected, when lower intensities are used for testing, associative performance indices are lower than when the trained medium intensity is presented at test (Fig. 3Di; MWU test: U=30.0, P<0.05/3). However, associative performance indices remain unaltered if medium-trained larvae are tested with higher or even much higher intensities (Fig. 3Di; MWU tests: U=90, 64.0, P=0.51, 0.12) (the corresponding KW test yields P<0.05, H=13.11, d.f.=3). This result is not conclusive regarding the question whether BA memory traces are intensity specific (compare the data of Fig. 3Di to the two scenarios presented in Fig. 1B).

To overcome this deadlock, we trained larvae with a low intensity and tested them with either that very same trained low intensity, or the medium, or the higher odour intensity (note that in this experiment the latter two testing intensities are both higher-than-trained). Clearly, if the training intensity is relatively low, overall performance indices are lower (compare the medium-medium condition in Fig. 3Di to the low-low condition in Fig. 3Dii); more critically, we found that associative performance indices are decreased as testing intensities are strongly elevated above the trained low intensity [Fig. 3Dii; train low, test low versus the groups tested with medium (MWU test: U=271.0, P=0.42), higher (MWU test: U=203.0, P=0.2), or tested with much higher intensities (MWU test: U=124.0, P<0.05/3)] (the corresponding KW test yields: P<0.05, H=9.16, d.f.=3). Thus just as for all other assayed odours, these results fit the scenario in Fig. 1Bii: BA memories are intensity specific in larval Drosophila (Fig. 3E).

We provide an analysis of whether intensity can be a distinctly learnable parameter of an odour. Indeed, for adult flies (Xia and Tully, 2007; Masek and Heisenberg, 2008; Yarali et al., 2009) and bees (Bhagavan and Smith, 1997; Wright et al., 2005; but see Pelz et al., 1997), such intensity specificity of memory has been reported. Here, we show that in a system as simple as larval Drosophila, too, there is intensity learning (Fig. 3E). Interestingly, in a corresponding study in adult Drosophila, three of the four odours used (namely AM, 3-Oct and 4-methylcyclohexanol) support intensity learning, but apparently BA does not (Yarali et al., 2009) (for a replication of this latter result, see supplementary material Fig. S5). This discrepancy between the intensity specificity of larval and adult BA memories may be caused either by the difference in learning paradigms used (e.g. kind of reinforcer, number and duration of training trials, etc.) or by the fact that in adult Drosophila the genetic and neuronal basis for BA responsiveness differs from those of other odours (Helfand and Carlson, 1989; Ayer and Carlson, 1992; Keene et al., 2004; Yarali et al., 2009), while this is not apparently the case in the larvae. Also, while many investigators have found that 4-methylcyclohexanol can be learned well in adults (Yarali et al., 2009), this is not the case in larvae (supplementary material Fig. S2C). Actually, larvae seem behaviourally little responsive to 4-methylcyclohexanol (supplementary material Fig. S2C). Given that the general circuit architecture between larvae and adults is rather similar (Gerber et al., 2009), it is tempting to speculate that these discrepancies between larvae and adults may be based on different receptor repertoires of the two life stages (Kreher et al., 2005; Hallem and Carlson, 2006).

With respect to larval Drosophila, nothing is known as yet about the mechanisms of intensity learning. Trivially, the recognition of a particular odour intensity at test as being different from the trained one is possible only if the neuronal activity induced by a given odour intensity differs in at least some respect from the activity induced by other intensities of that same odour. At which stage along the olfactory pathway may such dissociation be found? We first briefly review the architecture of the olfactory pathway (for reviews, see Vosshall, 2000; Gerber and Stocker, 2007; Vosshall and Stocker, 2007; Stocker, 2008; Gerber et al., 2009; Masse et al., 2009; Touhara and Vosshall, 2009; Diegelmann et al., 2013) and then suggest two alternative scenarios for intensity learning.

Different odours initially activate partially overlapping subsets of olfactory sensory neurons in the olfactory organs, dependent on the ligand profile of the olfactory receptor protein expressed. In the larva, each of the 21 olfactory sensory neurons expresses only one ligand-specifying Or receptor gene, and in turn each receptor gene is expressed in only one sensory neuron. Each sensory neuron then innervates one of the 21 glomeruli in the antennal lobe. In analogy to the situation in adults (Wilson, 2008), the pattern of activity in the antennal lobe is probably shaped by local interneurons (Thum et al., 2011). The resulting glomerular activity pattern is picked up by typically uni-glomerular projection neurons and is relayed both to pre-motor centers as well as to the Kenyon cells of the mushroom bodies, which in turn have access to pre-motor areas as well. Thus, dependent on the ligand profiles of the receptors and the connectivity in the system, odour quality could be combinatorially encoded along the olfactory pathway.

As for odour intensity, activity patterns in sensory and projection neurons seem to broaden with increasing intensity [larva (Asahina et al., 2009); adult (Ng et al., 2002; Wang et al., 2003; Root et al., 2007)]; notably, however, at successive processing stages activity patterns become more and more intensity invariant (Voeller, 2009). Such nested representations clearly could not accommodate intensity learning. Suppose that during training a memory trace was laid down in those neurons that are activated by the particular odour intensity used. In the subsequent test, a higher intensity of the same odour would activate, among other neurons, always all these same neurons, probably even more strongly than the trained intensity does, hence inducing at least as strong conditioned behaviour as the trained intensity. It therefore seems unlikely that the traces of intensity memories are laid down at a level of processing where olfactory representations are nested, such as is probably the case for sensory or projection neurons. At the next level of olfactory processing, mushroom body Kenyon cells show different levels of intensity invariance in their responses [locust (Stopfer et al., 2003); adult Drosophila (Wang et al., 2004; Voeller, 2009)]; critically, the activity pattern evoked by a low intensity of an odour is not always fully nested within that evoked by a higher intensity of the same odour [e.g. for ethyl acetate, see fig. 3 in Wang et al. (Wang et al., 2004)]. It remains unclear what kind of a connectivity scheme could transform nested representations at the projection neuron level to intensity-specific representations at the Kenyon cell level. In any event, taking this scenario to its logical extreme, training with a particular intensity could lay down a memory trace in a set of Kenyon cells that, as a set, is specifically activated only by that same odour at that same intensity. Obviously, this implies an entangled storage of quality and intensity information in the Kenyon cells (Fig. 4A).

Alternatively, quality and intensity might be encoded separately, enabling independent learning and retrieval of each (Fig. 4B). While the quality of an odour may be coded for memory formation by the unique set of Kenyon cells it activates, its intensity may be coded for example by the level of activity summed across all antennal lobe glomeruli, as previously argued with respect to adult Drosophila as well as the honey bee (Borst, 1983; Sachse and Galizia, 2003; Yamagata et al., 2009; Schmuker et al., 2011). Both larval and adult antennal lobes harbour inhibitory interneurons innervating various numbers of glomeruli; also, excitatory interneurons with similarly wide connectivity are found in the adult antennal lobe [larva (Python and Stocker, 2002a; Python and Stocker, 2002b; Asahina et al., 2009; Thum et al., 2011); adult (Wilson and Laurent, 2005; Olsen et al., 2007; Shang et al., 2007; Chou et al., 2010; Huang et al., 2010; Seki et al., 2010; Yaksi and Wilson, 2010); for a particularly detailed anatomical analysis, see Tanaka et al. (Tanaka et al., 2012)]. Finally, particular adult projection neurons with yet unknown response characteristics connect multiple glomeruli to Kenyon cells and pre-motor centers in the lateral horn (Marin et al., 2002; Lai et al., 2008; Tanaka et al., 2012). Any or all of these multi/omni-glomerular neurons could sum up the activity across broad aspects of the antennal lobe, and might thus contribute to encoding odour intensity. Note, however, that even at the level of a set of omni-glomerular neurons differing in sensitivity, the representation of a low intensity would be nested within that of a higher intensity. In order to lay down an unambiguous intensity-specific odour memory trace, one would need an additional layer of neurons. These would need to receive, for example, excitatory input from a highly sensitive omni-glomerular neuron and inhibitory input from an intermediately sensitive omni-glomerular neuron to become activated specifically at low but not at medium intensity ranges (Fig. 4B). It would be in these neurons where a memory trace for specifically a low odour intensity could be established. Note that, at its logical extreme, this scenario implies that odour intensity is encoded entirely independent of odour quality. It is as yet unclear whether such a circuit exists, and if so, whether and how such an intensity memory trace is then neuronally and behaviourally integrated with the odour quality memory trace.

To summarize, we show that in a system as simple as the one of larval Drosophila, olfactory memory traces are intensity specific. This reveals an unexpected richness of olfactory processing in the larva, and defines the demands on cellular accounts and computational models of associative olfactory function – indeed, the proposed kinds of circuitry may provide a useful scaffold for such an endeavour.

The authors thank V. M. Saxon, K. Tschirner and K. Gerber for help with the experiments, and M. Heisenberg, B. Michels, T. Saumweber and M. Schleyer for discussions and support.