Elemental and configural olfactory coding by antennal lobe neurons of the honeybee (Apis mellifera) Anneke Meyer' C. Giovanni Galizia Abstract When smelling an odorant mixture, olfactory systems can be analytical (i.e. extract information about the mixture elements) or synthetic (i .e. creating a configural percept of the mixture). Here, we studied elemental and configural mixture coding in olfactory neurons of the honeybee antennal lobe, local neurons in particular. We conducted intracellular recordings and stimulated with monomolecular odorants and their coherent or incoherent binary mixtures to reproduce a temporally dynamic envi­ ronment. We found that about half of the neurons respon­ ded as 'elemental neurons ', i.e. responses evoked by mixtures reflected the underlying feature information from one of the components. The other half responded as 'con­ figural neurons', i.e. responses to mixtures were clearly different from responses to their single components. Ele­ mental neurons divided in late responders (above 60 ms) ~nd early responder neurons (below 60 ms), whereas responses of configural coding neurons concentrated in-between these divisions. Latencies of neurons with configural responses express a tendency to be faster for coherent stimuli which implies employment in different processing circui ts . Keywords Hetero-glomerular local interneuron . E lectroantennogram . Intracellular recording· Morphology . Mixture interaction A. Meyer· C. G. Gali zia (181) Department of Biology, University of Konstanz, Constance, Germany e- mail: giovanni.galizia @uni -konstanz.coll1 Abbreviations AL Antennal lobe EAG Electroantennogram LN Local neuron PN Projection neuron ORN Olfactory receptor neuron Introduction Olfactory systems need to extract the biologically relevant information from chemically and temporally complex stimuli in a turbulent environment. This task is further complicated by the fact that in some incidences the important information might be carried by single com­ pounds and in others by mixtures (for review see Lei and Vickers 2008). Accordingly, the olfactory system has to process configural information of the mixture as a whole (sometimes referred to as synthetic coding), but at the same time preserve elemental information about single com­ pounds (sometimes referred to as analytical coding). Olfactory guided behavior, which relies on effi cient olfactory coding, is prominent amongst insects. Their comparably simple nervous system makes them excell ent models to study mechanisms of the evolutionarily preserved olfactory system (Ache and Young 2005; Sato and Touhara 2009). The honeybee, Apis me llifera, has the ability to differentiate many odors, and is therefore well suited to study coding of complex odorant compositions (Menzel et al. 1996; Galizia and Menzel 2001). Importantly, in behavioral experiments bees are able to recognize a mixture of two odorants as a new odor (,configural coding'), but also to extract the information about the two odor components (,elemental coding') (Deisig et al. 2001, 2002, 2003). Thus, http://www.springer.com/life+sciences/animal+sciences/journal/359 http://nbn-resolving.de/urn:nbn:de:bsz:352-217979 160 the olfactory system of the bee should allow for both, configural and elemental elements to be processed. Configural coding is achieved in the primary olfactory area, the antennal lobe (AL). The AL consists of functional subunits with high synaptic density, the glomeruli. In each glomerulus three classes of neurons synapse onto each other: olfactory receptor neurons (ORN), projection neu­ rons (PNs), and local interneurons (LNs). The AL network, build by these three types of neurons, is thought to reformat the input signal such that discernibility of stimuli is increased (Galizia 2008). ORNs detect odorant molecules at the antenna and form the input level of the AL. Each glomerulus receives sensory input from one type of ORN, which in honeybees branch throughout the superficial layer of the glomerular 'cap' (Pareto 1972). Optophysiological measurements of AL input in different insects indicate that it follows rules of elemental summation (Tabor et al. 2004; Deisig et al. 2006; Si lbering and Galizia 2007). PNs form the AL output and send their axons from the AL to higher processing areas (Mobbs 1982). In compari­ son to the input signal, representation of mixtures is more configural in PNs (Tabor et al. 2004; Si lbering and Galizia 2007; Silbering et al. 2008; Deisig et al. 2010) . LNs branch exclusively within the AL and interconnect glomeru li . They are suggested mediators of linear and non­ linear transformations between AL input and output (Sun et al. 1993; Ng et al. 2002; Sachse and Galizia 2003; Sachse et al. 2006; Bhandawat et al. 2007; Olsen and Wilson 2008). In Hymenoptera, the honeybee in particular, two main morphological groups are distinguished: homo LNs uni­ formly innervate many glomeruli , and hetero-LNs innervate one glomerulus densely and several sparsely. Neurons of both groups can interconnect the cap and the central 'core' within one glomerulus (Fonta et al. 1993). Simi lar to findings in moth, LN response latencies in the honeybee were found to be shorter than those of PNs suggesting that signal transfer from ORNs to PNs is mediated via LNs (Christensen et al. 1993; Krofczik et al. 2009). Furthermore, the presence of reciprocal synapses (Gascuel and Masson 1991) suggests a complex synaptic layout including both, LN mediated and direct ORN-PN signal transduction . In the present study, we investigated in how far elemental odorant information is preserved in individual neurons. Further, we wondered if configural and elemental coding strategies can be attributed to morphologically different neurons. We approached these questions conducting intra­ cellular recordings and morphological reconstructions from single AL neurons (LNs, as wel l as PNs) of the honeybee. Single cell recordings were combined with electroantenno­ gram (EAG) recordings, in order to establish a precise tem­ poral reference frame . To simulate properties of a turbulent environment, we created coherent or incoherent binary mixtures. In a coherent mixture, odorant components are delivered in synchrony, while in an incoherent mixture the components are delivered asynchronously. Materials and methods Animal preparation Worker honeybees (Apis mellifera) were caught at the entrance of the hive or at a feeder, immobilized by cooling, and mounted in custom-made Plexiglas holders. The bees were allowed to acclimate to the new environment for 1-6 h before the experiment started. Antennae were immobilized with Eicosane (melting point 37°C; Sigma-Aldrich Chemie GmbH, Germany) whilst head and mandibles were immobilized with Deiberit 502 (melting point 60°C; Boehme-Schoeps, Germany). To reduce brain movements the esophagus was detached from its muscles. The head capsule was opened between the median ocellus and the base of the antennae. G lands and tracheal sheaths were removed carefully . The exposed brain was kept moist during recordings by dribbling saline onto it if necessary (in mM: 130 NaCl, 6 KCI, 4 MgCI2, 5 CaCI2 , 10 Hepes, 25 D-Glucose, 160 sucrose, pH 6.7, 500 mosm/L). Odorants Odorant identity (Krofczik et al. 2009) as well as con­ centration (Christensen et al. 1993; Stopfer et al. 2003) impact the response onset measured in electroantennogram recordings (EAG) as well as the latency of individual neurons (Junek et al. 2010). We chose two monomolecular odorants (I-octanol and 2-heptanone) that naturally occur in the honeybee's environment as both, components in floral mixtures (Ornata et al. 1990; Tollsten and Knudsen 1992; Baraldi et al. 1999) and pheromones (Balderrama et al. 1996,2002). We presented the single components alone, their coherent mixture (both components with syn­ chronized odorant onset), and their incoherent mixtures (both components with odorant onsets shifted with respect to each other). In doing so, we created a controlled recre­ ation of the elements in a dynamic odorant environment. In addition, comparing absolute response latencies can then be used to identify the odorant to which a particular neuron responds in a mixture, and thus to differentiate between 'elemental' and 'configuraI' coding (see below). Stimulation paradigm A custom-built olfactometer, similar to a previously pub­ lished model (Galizia et al. 1997) was used for stimulation. Stimulus delivery was controlled by TTL pulses triggered Block Stimulus Duration f-----> on off on off on off i"" i"" i"" i"" i"" i"" i"" i" i' i"" i" Time a 1 2 3 4 5 6 7 8 9 [sec] Fig. 1 Stimulus protocol. Stimulus duration was 800 ms. The inter­ tri al interval was 1,800 ms. Each stimulus repetition was considered as one trial. One block consisted of three trials with identical st imulation. The onset de lay for odorants in incoherent mixture blocks was 50 ms by the recording software, Clampex (Axonlnstruments Inc., USA) . Odorant flow through individual channels of the olfactometer was calibrated using a photo ionization detector (Aurora Scientific Inc. , Canada). The primary odorants I-octanol and 2-heptanone, diluted in mineral oi l were used for stimulation. Odorant concentration was I :200 for I-octanol and I: 100 for 2-heptanone. The effect of odorant concentration on response latency was tested with the following dilutions for I-octanol and 2-heptanone, respectively: I :200 and I: 100, I: 100 and I :200, and I: 100 and I: 100. Airborne stimuli were delivered in a constant stream of clean air (1.2 m/s) that was directed to both antennae. Stimulus duration was 800 ms at an inter-trial interval of 1,800 ms. Three trials using identical stimul i followed each other in immediate succession and consti­ tuted one stimulus block. An interval of 5,000 ms sepa­ rated two blocks from each other (Fig. I). A completed recording consisted of the presentation of five stimulus blocks and two control blocks: each of the two primary odorants, their coherent mixture, their two incoherent mixtures , a control of mineral oi l and a control of pure air. To an incoherent mixture the onset of the second odorant was delayed by 50 ms with respect to the first odorant. The sequence between presentations of different blocks was pseudo-randomized. Electrophysiology Three types of electrophysiological experiments were performed: electroantennogram (EAG) recordings, intra­ cellular recordings from single AL neurons, and parallel recordings of both of these. EAG recordings served to validate the stimulus apparatus and to analyze odorant concentration-spec ific effects in ORNs. Intracellul ar recordings were used to investigate the role of single AL neurons within the AL network. Parallel recordi ngs with both methods a llowed measuring the response onset of sing le AL neurons with respect to a sensory time reference. 16 1 Glass electrodes were pulled from borosi licate capillar­ ies (GCI50F- 10, Clark electronic instruments, UK) using a horizontal Pu ller (P-97, Sutter Instruments Co.,USA). Sharp electrodes ( 100- 250 MQ) used for intrace'tlular recordings were tip-filled with fixable fluorescent dye (4% Alexa 488 hydrazid in 0.2 M KCI, 4% Micro Ruby in 0 .2 M K-Acetate or 3% Lucifer Yellow in 0.1 % LiCI). Blunt e lectrodes (5 - 20 MQ) used for EAG recordings were filled with 0.2 M NaCI. The sharp electrodes were placed on the AL and grad­ ually advanced employing a micro-manipulator (Kleindiek Nanotechnik, Germany) until a cell was impaled. For EAG recordings, a blunt electrode was placed on the antenna tip using a second micro-manipulator (Brinkmann Instrumen­ tenbau, Germany). In cases where both signals were recorded in parallel , the EAG was always taken from the antenna ipsilateral to the AL recorded from. A common reference electrode was placed through a small incision in between the lower ocelli . Recordings were performed in current-clamp mode, using an Axoclamp 2B Amplifier (gain 10, Axonlnstru­ ments Inc. , SA). EAG signals were additionally amplifi ed by means of a custom-built external amplifier (gain 10). A 50-Hz filter (Hum-Bug, Quest Scientific, Canada) removed line hum . Data were di gitized using the Axon Interface, DigiData 1200B (Axoninstruments Inc. , USA) and stored on hard-drive using Clampex 8.2 (Axonlnstruments Inc., USA). Morphology After an intracellular recording was fini shed successfully , the dye loaded in the tip of the sharp electrode (see above) was iontophoretically expelled, with the polarity of the current pulses (0.2 s width, 2 Hz, 1-4 nA) chosen according to the dye ' s charge. Subsequently, in order to visualize all glomeruli we counterstained the sensory tracts of the penetrated AL with Neurobiotin (Vector Laboratories Inc. , USA) . For this purpose, the cuticle previously removed from the head capsu le was now repositioned and closed carefully with eicosan. The ipsil ateral antenna was brought in an upri ght position and surrounded by a basin made from vaseline that was filled with 2% neurobiotin (in aq.dest.). The antenna was cut at the scapus and the neurobiotin was given 2- 3 h to be taken up by the antennal nerve stump. For morphological preparations, brains were removed and fixed in 4% paraformaldehyde for 3 h at RT, or overnight at 4°C. Subsequently, preparations were washed 3 times for 10, 30, and 45 min in phosphate buffered so lution and incubated in 0.5% avidin-coupled fluorescent dye to visualize the neurobiotin in the ORNs (either AMCA-avidin , or cy3, depending on the single-cell 162 marker) for at least 5 h. Brains were then washed again 3 times for 15, 30 and 45 min, dehydrated in an ascending ethanol series, cleared for 20 min in xylol and finally embedded in DPX mounting medium (Fluka, Sigma­ Aldrich Chemie GmbH). To visualize staining results, confocal image stacks were taken with a Zeiss LSM 510 Meta confocal microscope (Carl Zeiss AG, Germany) . Data analysis Sharp recordings were filtered off-line (10kHz lowpass cutoff). Spikes were detected, using custom-written rou­ tines based on the open source R packages SpikeOMatic (Pouzat et al. 2004) and STAR (Pouzat and Chaffiol 2009). To determine firing rate and response latencies, algorithms provided by the open source Matlab toolbox FIND (http: // find.bccn.uni-freiburg.de) were employed. Analysis of single cell data was chosen so as to maximize compara­ bil ity to related work (Krofczik et al. 2009). Image pro­ cessing of confocal stacks and reconstruction of cell morphology were achieved using AMIRA 5.1 software (Mercury Computer Systems, Germany) . For analysis of EAG recordings, custom-written routines in R (http: //www. R-project.org) were used. a excited sing le spike Temporal electroantennogram analysis EAG recordings were filtered off-line (\ 00 Hz low-pass), and averaged over repeated trials. Response onset was defined as the relative maximum preceding the steepest negative slope of the potential drop which dema rcated an odorant response. This point was found to be least affected by temporal displacement attributable to response ampli­ tude and, hence, evaluated as most reliable. Response latency analysis Using sharp electrodes intracellularly, we recorded from a total of 21 cells in the antennal lobe. Neuron s differed greatly in their physiological properties: some had low background activity, and responded to odors with single or few spikes (Fig. 2b, right), some responded with clearly increased firing rate (excited, Fig. 2b, left), others with a drop in spike rate (inhibited, Fig. 2b, middle). In neurons that yielded a good morphological staining it was possible to identify their morphology as PNs or LNs, in all other cases this was not possible, due to the 'blind ' nature of our recording: we refer to the presumably mixed population of recorded LNs and PNs as AL neurons. We assume that no JIYV'-""" "'-'~"""'''''''''''-''''''~'lW I'''~OW'''''''''_~''''''''';'''''',r ... ¥''rItA''''~~ 1000 ms b excited inhibited sing le spike 3 iii 2 .;: l- S ...... 5 ..!!! 4 :!:. 3 oS CII 2 a: I I II II I I I I 0 500 o 500 Time [ms] o 500 Fig. 2 Estimation methods of response latency was chosen with respect to response patterns. a Exemplary traces of neurons with different response patterns: excitation (top), inhibition (middle), or few spikes on a depolarization (bottom). b Estimation of response latency. Spiking activ ities shown for three repeated trial s under identica l stimulation. Superimposed red traces indicate the response rate function. Blue lines mark the est imated response onset and blue bars indicate their across trial variability. Response onset for exc itation is defined as the point of steepest rising in the rate function after stimulus onset (leJi). Response onset for inhibition is defined as the point of steepest fall ing (middle) . Response onset for neurons with cont inuously sparse firing is not we ll captured by a rate function and best characterized by the peak time of a s ingle spike after depolarization onset (right). Grey bars indicate st imulus delivery time ORNs were stained, because the small axon terminals of these neurons make it unlikely that we impaled them with our sharp electrode. In our preparations we never found stainings of neurites directed towards the antennal nerve. We judged a cell as responding to a stimulus, if repeated stimulation elicited similar modulation in the neuron's firing pattern . Absolute latency, that is the mean latency across trials, and relative latencies, that is trial-to-tri al differences in latency, were calcul ated with one of three methods (1 -3). The method was chosen based on the respective firing pattern (Fig. 2). I. Latencies of cells that responded to stimulation with an increased firing-rate were estimated based on the derivative of the trial-al igned firin g rate as described elsewhere (Meier et al. 2008). This method processes the data in four success ive steps. First, the derivative of each single trial spike train of a given cell was estimated by convolving with an asymmetric Savitzky-Golay filter (Savitzky and Golay 1964) (polynomial order 2, 300 ms width , Welch windowed) . Second, all si ngle trial-derivatives were optimall y aligned, finding the greatest possible pair-wise cross correlation (Nawrot et al. 2003) . The resulting time­ shifts correspond to each tri al's relative latency. Their standard deviation (J gives a measure for the across­ trial latency variability. Third , the single trial spike trains are temporally aligned by shifting each by its individual relative latency. Fourth, the aligned spike trains were merged into one train , representing the cell s activity pooled over trial s. The convolution of this merged spike trai n with the same asymmetric Savitzky-Golay filter gave an estimate about the derivative of the cell 's absolute firing rate, based on which the absolute latency within a given block of stimulation was determined. The stimulus-specific absolute latency was defined as that point in time where the slope of the firing rate is steepest, that is the derivative's max imum . 2. Latencies of cell s that responded to stimulation with a decrease in firing rate were estimated with an approach nearly identical to I), but instead of the steepest rising slope, the steepest fa lling slope of the absolute firin g rate was defined as response onset. 3. Latencies of cell s that had very low spontaneous activity and responded to stimulation with a membrane depolarization ridden by one or few single spikes were estimated based on spike peak time rather than rate. The membrane depolari zation in these cell s was taken as indicative for an apparent response. The response latency was defi ned as the peak time of the first spike riding such a depolari zation. 163 Single cell morphology Confocal image stacks were processed using A mira 5. 1 software (Visage Imaging GmbH, Germany). All prepara­ tions were carefully inspected for complete stainings: spec­ imens with. irregular labeling, abruptly ending neurites or strong fluorescent background were excluded from further morphological analysis. However we did include prepara­ tions in which the neuron was presumably incompletely filled but sufficient to transfer valuable information. Incomplete stainings can never be excluded in in tracellu lar dye fill s, for example due to extremely small neurite diam­ eter and consequently compromised distribution of fluores­ cent dye. We searched very carefully for stained axons leaving the AL in order to differenti ate between LNs and PNs. Neurons were reconstructed using the filament editor without further estimation of neurite diameter. Location and size of single glomeru li were registered by interactive seg­ mentation based on OSN mass fill s. Glomerulus identi ty was determined by visual inspection and comparison with the morphological atlas of the honeybee (Gali zia et al. 1999, http://neuro .uni-konstanz.delhoneybeeALatlas). To com­ pare innervation patterns with spatial patterns of AL activity in response to the stimuli applied, the physiological atlas of the honeybee (http://neuro.uni-konstanz.de) was consulted. Results Latency shifts in receptor neurons are inherited by antennal lobe neurons We chose two odorants and analyzed latency difference between them at two different concentrations. At the level of the compound olfactory receptor neuron response as measured with EAGs, we found that response latency decreases with increasing concentration (two-way Anova, F collcelltratioll < 0.000; Tukey test, P CO llcclltrat ion < 0.000; mean difference = 22.646 ms) irrespective of the odorant used (Podorant = 0.7; Pinteraction = 0.9). In order to investigate how ORN latency shift transfers to the AL network we conducted simultaneous intracellular and EAG recordings (Fig. 3a) . The odorant at lower concentra­ tion evoked a smaller EAG amplitude (Fig. 3b) as well as a weaker firing rate in phasic-toni c AL neurons (Fig. 3c). As for EAGs, the latency in AL neurons decreased with increas ing odorant concentration (one-sided, paired ttest, p = 0.048). Indeed , the di fference between mean response onsets of EAGs and AL neurons was identical for both stimuli (37 ms; Fig. 3d). Thus, concentration-dependent latency shifts originate in ORN act ivity and transfer direct ly to AL 164 a EAG I AL neuron: bEAG 1 "10·2Hept 500 ms C AL neuron 0.S"10·' Oct! 1 "10" Hept ·500 o 500 Time [ms] Fig. 3 AL neurons inherit a concentration-dependent latency shift from peripheral neurons. a Exempl ary traces of simultaneous EAG and single-cell recording. Grey bars indicate stimulus deli very time. Dolled lili es indicate the time of stimulus detection on the antennae, measured by EAG (upper trace). b EAG amplitude differs for different stimuli . EAG in response to lower concen trated I-octanol is depicted in red. EAG in response to higher concentrated 2-heptanone is depicted in blue. Traces shown are averaged (II = II ). cAL neurons neurons. Accordingly, odorant concentration can be used to create stimuli with di stinct, characteristic latency shifts. This phenomenon offers a tool to identify a response evoking component from a mixture, within single AL neurons. Simu ltaneous recording of AL neurons and EAG also allowed us to report the response latency of single neuro ns more accurately . Timing of olfactory stimuli is always constrained by laboratory conditi ons, like air turbulences or distance between odorant source and receiver (Vetter et al. 2006). While easy to control within one lab, these constraints are difficult to standardize between different laboratories. Comparability of data between laboratories is increased when latencies are estimated with respect to a reference derived from the animal rather than the stimulation mach inery. The EAG re li ably reflects stimulus arrival at the antennae. Hence we used the mean EAG response onset as a reference time point to estimate single AL neuron latencies. Both, elemental and confi gural coding occur in antennal lobe neurons In a natural environment, components of an odorant mi x­ ture mi ght impinge on the antenn a simultaneously (as in a 90 80 70 60 ~ 50 Q) 40 .... III a: 30 20 10 1000 EAG dill ____ ---'I ~ _--::-::::-::-_...JI ~ 1000 ms 61 +/· 12 ms 51 +/· 14 ms 37 37 120 Vi' 100 oS >- 80 g 60 .!! (II 40 ...J c: 20 :ll o :E OCII Hepl Oell Hepl EAG AL neuron mean fir ing rate differs for different stimuli . Color code as in b. Rate functions are aligned and averaged over all neurons which responded with a phasic-tonic pattern (/I = 12). (d) Response latencies of EAG and AL neuron recordings differ significantly for different stimuli , but keep the same relative time shift (37 ms). Color code as in b, c. Bars illustrating mean latency in EAGs are stripy. Bars illustrating mean latencies from single ce ll measurements are colored solid. Same animals as in b, c; error bars give standard deviations mixture from one source), or separate in time (as when they originate from di ffe rent sources). An olfactory system that can differentiate between these two conditions mi ght be able to select whether to process a mixture in a configural way (i .e., the mi xture being more than the summation of its components), or in an e lemental way (i.e., with access to the components' identity). To study thi s question , we presented single odorants and their coherent and incoherent mixtures. We asked whether single AL neurons can be attributed to elemental or configural processing, and whe­ ther their coding strategies are related to the cell 's own response latencies. About half of all recorded AL neurons responded to the mi xture in the same way as to one of its components (n = 12, 'elemental neurons'). The remaining cell s (n = 9, 'configural neurons') responded to the mixtures as to a di ffe rent odorant. 'E lemental neurons' fe ll in two selectivity groups: e lement selective and element dominant. The first group (n = 7, element selective) responded to the mixture as to one of its components, its dominant (D) odorant. Response latency to the coherent mixture, and the incoherent mixture in which odorant D was presented 165 a s lilllJlllJjlJJjjjW1Jm~llil cS~ e 12 'E 8 :::l o llilllJJll~~l/IIIWWIJ.W SD_ S-D~ D-S llJllLWllLl~ D~ so~ s.o~ D-S~I~ o () 4 o o 50 100 150 Latency [ms] f Ui' 100 .§. ~ 500 ms -'5""0""0-m-s-' ~" •• o early early late dominant selective b d 9 S 111[1111[[1111[[[11[11[1[11[[[11[[11[[1[[1[11[1[[[[1 o 1[111111[111 [[111.111[[1111[11[111[111 so 1[111111[11[1[1.11[1.[.11111 [[[ [ SoD [[1[[ 111[[11[[[11[[1[[.11111.1[11[1[11111 D-S 1[[1[[1111[11[111[1 ••• [1[1111[1111111[1[1[[1 S 1[[[[[ 1111[[[11[111[[[[[ 11[[ 111[[1 [1111 [1 [[mlll[ o 1[[ II 1I [1I 11 1 [[111II11I1~[[ III [ [1 [i[11I1I1I so [ [ [[II [[ [[[[111[1 [rullli II I 1 [[ [[[1[[[ SoD [[11 1[[1[[[[[[11111 11[11[1 1I[1I11!1[[lIImlllll [[[1I [I D-S [[ [1[[l m[[[I[[I[ [11II11I 1Im[[II[[II[II[II [II [[[1 mean [ms] sd [ms] early early late dominant selective selective 31 28 93 28 20 16 -500 o 500 1000 Time [ms] -500 o 500 Time [ms] 1000 Fig. 4 Grouping 'elemental neurons', according to element selectivity and mean cell latency. a Example of an element-selective neuron . The dominant odorant (D) evokes responses both as si ngle stimulus and in mixtures (S O = coherent mixture, SoD/DoS = incoherent mixtures) . The subordinate odorant (S) does not contribute to the response. Grey bars indicate stimulus deli very time. b Merged spike trains comprising all element-selective 'elemental neurons' (II = 5). Note the lack of response to the subordinate odorant (S), as compared to the other conditions. c Example of an element-dominant neuron. Both odorants evoke responses, but in incoherent mi xtures a latency shift indicates first, corresponded to the response latency to odorant D alone. Responses to the incoherent mixture in which odorant D was presented as the second component were shifted by about 50 ms, corresponding to the delay of this component. Hence we conclude that these neurons responded to odorant D alone, and were not influenced by the presence of the other, subordinate (S) odorant (Fig. 4a, b) . The second group of 'elemental neurons' had a different coding characteristic (n = 5, element dominant) . These neurons responded to both of the single components. However, in the mi xture only one odorant (D) contributed to the response, whi le the other (S) did not have any apparent impact on response pattern or latency (Fig. 4c, d). Further, 'elemental neurons' fe ll in two latency clusters: short latencies (early responders) below 60 ms and long latencies (late responders) above 60 ms (Fig. 4e) . All ele­ ment-dominant neurons fe ll in to the early responders preference of odorant D. d Merged spike trains compnsmg all element-dominant neurons (II = 7). Note that subordinate (S) as well as dominant (D) odorant evoke a response. c Irrespect ive of selectivity, neurons' latencies cluster into an early (g reell ) and a late (magellta ) group. r Based on the combination of latency and selectivity, three neuron groups become apparent. Element-dominant neurons always respond early (lighf greell ), element-selective neurons distribute in an early (dark greell ) and a late group (magenfa), with significant different latencies (paired I test, one-s ided, p = 0.00 I). g Table of mean cell latencies sorted according to group cluster, while element-selective neurons di stributed into both clusters . Based on the combination of selectivity and latency, we end up with three subgroups of 'elemental neurons' : e lement-dominant with early responses (n = 5), element-selective with early responses (n = 4) and ele ment­ selective with late responses (n = 3) (Fig. 4f, g). Responses of 'configural neurons' were more diverse, both in terms of selectivity and latency . One cell responded on ly to coherent stimuli (Fig . 5a), some to mixtures on ly (1'1 = 2), but most cell s responded to mixtures as well as single compounds (n = 6, Fig. 5b) . Response latencies of 'configu ral neurons' scattered broadly around 60 ms and thus concentrated exactly between the groups of 'elemental neurons' with early and late responses (Fig . 5c, d) . Unlike for 'elemental neurons', latencies within one 'configural neuron ' cou ld be short for one stimu lus and long for another one. Fastest responses were, in the mean, evoked by single compounds (62 ± 26 ms) and slowest by 166 C 12 • ea rly elementa l a A illlWillJJJJJJJ B~~ AB~ b A_.JJJ~ .l .. .... I: :;, 0 0 8 4 • configura l . Iate ele menta l A-B~ B-A I I1IIIII1IIII j I 11111 ~-,-::=_-,15mv 500ms Fig. 5 'Configural neurons' and distribut ion of latencies with respect to 'elemental neurons'. a Responses to coherent stimuli only fLatencies (ms): B 94, AB 132). Grey bars indicate stimulus delivery time. b Example of an AL neuron which responded to single compounds and coded mi xtures configurally In = 6, latencies of the individual neuron (ms): A 62, B 49, ab 61, A-B 67, B-A 871. a 80 : ~~~~ e~~~enta l 60 . Iate elementa l ~ .!!! ~ Q) 40 .... CIS a: 20 a -500 0 500 1000 Time [ms] Fig. 6 Early responding AL neurons may impact late responses in AL neurons. Group averaged rate functions of early responding 'elemental neurons' (green trace, II = 9), late responding ' elemental neurons' (magel/ta, 1/ = 3) and 'configural neurons' (b lue trace, II = 6). Grey shades indicate stimulus delivery time. The odorant-selective, late group comprised responses of exc itation (a) and inhibit ion incoherent mixtures (84 ± 49 ms, paired t test, p < 0.05; Fig.5e). Antennal lobe neurons are active sequenti ally We analyzed response latencies in more detail by calcu­ lating population rate functions for each of the subgroups (Fig. 6). Early responses of 'elemental neurons' were either excited, or single spikes. We pooled these and estimated a common rate function. Late responses of 'elemental neu­ rons' were either spike excited or inhibited responses, 0 0 50 100 150 200 d Latency [ms] early- late - elemental con Ii ural elementa l mean [msl 29 62 93 sd Ims) 22 36 16 e 1 : l ~~~==~~~~~~ o 50 100 150 200 Latency [ms] C Histogram of latencies from neurons with configural responses (blue), early (greel/) and late (magenta) neurons with elemental responses (green and magenta as in Fig. 4e). d Table of mean cell latencies sorted according to group. e Histogram resolv ing response latencies of 'configural neurons' to sing le compounds (white), coherent mixtures (black) and incoherent mixtures (grey) b 80 : ~~~h e~~~enta l 60 . Iate e~ementa l ~ ~ Q) 40 .... CIS a: 20 0 500 1000 Time [msj (b) which are treated separate ly . a Rate peaks of early neurons precede pos iti ve rate peaks of excited late AL neurons. b Rate peaks of early neurons also precede negative rate peaks of inhibited late neurons. Note the initi al small activity boost which precedes the inhibited period which we separated in two rate functions. Amongst the 'configural neurons' all types of responses were repre­ sented. We pooled all non-inhibited responses in a common rate function. Superimposition of these rate functions ill ustrates that activity peaks of the three groups follow each other in immediate succession , suggesting that the corresponding cell s might be arranged sequentially in a functional net­ work. Early responding elemental AL neurons (green) precede the positive rate peak of excited (magenta; Fig. 6a) , as well as the negative rate peak of inhibited late responding 'elemental neurons' (magenta ; Fig. 6b) . Inter­ estingly, the rate function of inhibited late responding neurons appears to have a small activity boost, which relapses just before early responding neurons reach their maximal response frequency. While 'elemental neurons' have rather distinct activity peaks, 'configural neurons' appear more diverse, reflecting the scatter in latencies within and across this neuron group. Hetero local neurons are involved in configural as well as elemental processing What is the relationship between elemental and configural responses on one side, and the cell's morphology on the other? Using intracellular staining we obtained the mor­ phology of one PN and three LNs. The PN fell into the element-selective, late response group. The LNs all had early mean responses. Two of the LN stainings were of sufficient quality to be reconstructed into a 3D skeleton model (Fig. 7). Both neurons responded to 2-heptanone and to its binary mixtures (Fig. 7a, h) and were hetero-LNs, i.e. with one densely innervated glomerulus ('main glo­ merulus') and several sparsely innervated glomeruli. One was an odorant-selective 'elemental neuron' (Fig. 7, top row) and the other a 'configural neuron' (Fig. 7, middle and bottom row). We asked whether their glomerular innervation pattern could explain their different response profiles. We identified the innervated glomeruli and com­ pared these with the AL's spatial activity pattern evoked by 2-heptanone as published in the physiological atlas of the honeybee (http://neuro.uni-konstanz.de; c.p.: Fig. 7f). The densely innervated glomerulus of the 'elemental neuron' was one of the 2-heptanone responsive glomeruli (TI-29, Fig. 7b, c) . The neurites branched within the core of the glomerulus and reached out into an intermediate layer between cap and core. Counterstaining of ORNs showed that LN branches and ORN axons overlapped suggesting that a direct input from ORNs is possible (Fig. 7d, white arrows). The main glomerulus of the 'configural neurons', however, innervated a glomerulus that is not responsive to 2-heptanone (TI-19, c.p.: Fig. 7f, g). This neuron inner­ vated several glomeruli sparsely, among which at least three that are weakly responsive to 2-heptanone (T3- 18, T3- 31, T3- 52; Fig. 7i, j; see also movies in supplemental material). Sparse arborization fibers did not reach into the glomerular cap (Fig. 7kl, magenta arrows), which would suggest that this neuron did not receive input from ORNs. However, a careful reconstruction of the neuron and the glomerular cap based on counterstained ORNs showed that the sparsely arborizing neurites in fact distributed just between cap and core (Fig. 7k2, magenta-white arrows) . Hence, from our data we cannot decide whether hetero­ LNs have, in their sparsely innervated glomeruli , direct, 167 monosynaptic ORN input from the cap, poly-synaptic input through LNs and PNs from the core, or both. While the response latency of the 'elemental neuron' to the dominant odorant and its mixtures was similar (36 ± 2 vs. 38 ± 4 ms), the 'configural neuron' clearly responded faster to the single compound than to the mix­ tures (18 ± I vs. 49 ± II ms). This change in latency indicates the occurrence of both, mono- and poly-synaptic input to the configural processing LN. We hypothesize that this neuron was embedded in two different processing circuits that were differentially activated depending on the sensory stimulus delivered. Discussion In the present study, we investigated how single neurons in the honeybee antennal lobe (AL) process odorant mixtures. More particularly, we wanted to know if it is possible to differentiate between 'elemental' and 'configural' neurons. We found that indeed, there are 'elemental' and 'configu­ ral' neurons, though the same neuron may fall into either category depending on the particular stimulus used. Fur­ thermore, the fastest neurons have elemental responses, while neurons with configural response properties respond later (Fig. 6), suggesting that computing configural responses is more demanding to the network than com­ puting elemental responses. At the same time, configural coding is not the end result of AL processing, as shown by even later responses of elemental neurons. It should be noted that intracellular recordings in the honeybee antennal lobe only allow for short measurements (less than 10 min), and therefore the sample size of stained neurons is small. Thus, some of the observations that we draw, in particular about the relationship between mor­ phology and functional properties, need to be taken more as hypotheses than as proven facts. However, together with published reports from our colleagues (Flanagan and Mercer 1989; Fonta et al. 1993; Galizia and Kimmerle 2004; Krofczik et al. 2009), these data help understanding the intricate coding networks in the insect antennal lobe, in particular with respect to mixture coding. In what situation would elemental coding on the one, and configural coding on the other hand be relevant for an organism? Natural odorants as they are produced by, e.g. flowers are usually complex blends of many chemical compounds. A pollinating bee will perceive this bouquet as an individual odor, much as we do the smell of coffee. In an other instance, the bouquet of two different flowers might mix in the air. Here, the olfactory system has the task to separate odor componl'lnts, i.e. to analyze the elements of the mixture: 'it smells two different flowers' or 'it smells coffee and fresh baked bread ' . In a turbulent environment 168 a - j .1 .oJ Octl Hept I • iJba l iii 'il I! Mix Octl - Hept N 0 I III 7o==H: Hept - Octl SOOms e T3-5 h Ocd Hept Mix T1 -29 I IWlWI II lUI , aw I , JUM III Octl - H'!II'~""'p"t--I---'" Q\ ~",'.'.'l".hI."oI.l _~~ Hept - Octl SOOms b \ I d m+1 v f Fig, 7 Glomerular innervation patterns of hete ro-LNs responding to 2-heptanone. a Electrophysiologica l recording from a hetero-LN exhibiting elemental coding of 2-heptanone. b Frontal view of a reconstruction of the neuron corresponding to the traces in a . The highlighred glomerulus (TI -29) is known to be responsive to 2-heptanone and dense ly innervated. c Parasagitta l view of the same hetero-LN as in b. d Confocal image illustrating the type of dense innervation observed in the neuron in b, c. ORN innervation is given in green , LN innervation in magell.fa. Note the overlapping innerva­ tion area (white arrows). e Schematic drawing of the AL illustrating in co lor code the invo lvement of sing le glomeruli in the response to 2-heptanone, as determined by ca lci um imaging with bath applied dye (cp. http://neuro .uni-konstanz.de/honeybeeALatlas for the physio­ logica l atlas of the honeybee). Arrows indicate glomeruli which are innervated by the neurons presented. f Color coded 2-heptanone response as determined in ca lci um imaging experiments, projected on a reconstruction of a hetero-LN exhibi ting config ural coding of odorants fro m diffe rent sources frequently mi x and we assume that the 'coffee' mi xture will generally travel as a coherent mixture, while 'coffee and bread ' will more likely generate incoherent mixtures. This hypothesis has been tested using moth sexual pheromone components: when the components were released from the same locati on, male moths were attracted by the mi xture ('coherent' case, the moth recogni zed the mi xture as an odor), while spati all y separated locations did not attract the males (, incoherent' case, the moth did not behave as to a mi xture, suggesting elemental coding (Andersson et al. 2011). In our study of mi xture process ing at the level of single neurons in the AL of the honeybee, we have therefore used two kind of mixture stimul i: coherent and incoherent mi xtures. " v+d a V+d a 2-heptanone. Clearly the main glomerulus (TI - 19) is no t part of the stereotypic 2-heptanone response pattern . g Parasag ittal view of f. h Electrophysio logical recording from the hetero-LN in f, g exhibiting configura I coding of 2-heptanone. i Frontal view of the same neuron as in f- h . The highlighred glomeruli (T3-52 , T3-18, T3- 3 1) are sparsely innervated by the depicted neuron and known to be responsive to 2-heptanone. Bille dotted line ind icates the location of the densely innervated glomerulus (TI - 19). j Parasagitta l view o( i. kl Confocal image illustrating the sparse innervation in i , j of glomeruli in the core region. Sparse arbors seem not to overl ap with ORNs (magenta arrows). k2 Reconstruction of glomerular cap and core as well as sparse arbors from k, i. Note that nelll'ites di stribute just between cap and core (lI1agellw-while arrows). d dorsa l, v ventral, III medial, I lateral, p posteri or, CI anteri or. f , g, i , j are ava ilable as movies in supplemental materi al for better visua li zation. Arrows in b, c, f, g, i , j indicate ce ll body When looking at the response of a single neuron to the mixture AB of two odorants A and B, it is not always a trivi a l task to characterize whether the neuron responded to A or to B , or to the mixture. A neuron that does not respond to either A or B, but responds to AB , is clearly a 'configural neuron ' , selective to the mi xture stimulus. But how about a neuron that responds both to A and to B, and also to the mi xture AB? Does it respond to a single one of the com­ ponents, to both individually, or to the mi xture? We made use of concentrati on-dependent latency shifts that are generated in the receptor neurons to address thi s question. We recorded EAG potenti als in paralle l to intracellular recordings, and could determine response latencies with great precision. We also titered odorant concentration in a way to generate distinct latencies for the two odorants A and B. Using the distinct latencies as a marker for stimulus identity, it was possible to infer for each neuron if a response was evoked by a mixture as a whole, or by one of the mixture compounds. We found that half of the neurons responded to one of the compounds rather than the mixture: 'elemental neu­ rons' . These neurons could be subdivided into element­ selective neurons with early responses, element-dominant neurons with late responses, and element-selective neurons with late responses. Honeybee LNs have been shown to respond faster to odorant stimulation than PNs (Krofczik et al. 2009), which would suggest that the early responding neurons are LNs, the late responding PNs. Since our sample of stained neurons is small , in this work we do not differentiate among them, and lump them all as AL neurons. The other half of the neurons gave 'configuraI' respon­ ses, i.e. their response patterns to the single components and to the mixtures differed, and in particular mi xture responses did not reflect feature information of single compounds. Response latencies of 'confi gural neurons' vari ed considerably, but responses were often faster for coherent than for incoherent stimuli . This observation suggests that these neurons may be embedded in different processing circuits, depending on the stimulus components. This property may be used in context-dependent coding, whether the context is another stimulus or the physiological state of the animal. E lectrophysiological recordings have the advantage of hi gh temporal resolution and thus allow to detect minute delays of responses within and between neurons. Previous studies have shown that LNs respond prior to PNs (Krofczik et al. 2009) and recordings in moth revealed LN subgroups with different latencies that allow fo r LN- LN interaction (Christensen et al. 1993). These findin gs suggest the ex is­ tence of successively active neuron populations. In agreement with previous data from the honeybee (Flanagan and Mercer 1989; Sun et al. 1993), we could not find latency clusters comparable to those shown for the moth . Our finding that 'configural neurons' had a tendency to respond later to incoherent (delayed stimulus onset of two compounds) than to coherent stimuli (single com­ pounds and their temporally coherent mi xture) suggests a less well ordered, but functionally more fl ex ible AL­ archi tecture: these neurons may be recruited by di fferent functional networks, depending on the network activity elic ited by the stimulus. We found another observation hinting towards non-lin­ ear process ing mechani sms in the AL, in neurons with late elemental responses: inhibi ted late responding neurons had a small activity boost, re lapsing just before early responding neurons reached their max imal response 169 frequency. It is inspiring to think that thi s findin g could be explained by recurrent inhibition, as is known between granule and mitral cells in the mammalian olfactory bulb (for review see Urban and Arevian 2009) . Taken together, the data shown here and in other studies show that the AL is ?ot a simple feed-forward relay station, but a dense and multi -layered neural network with recur­ rent connectivity . Local neurons with asymmetric shape have been fou nd in many insect species (Flanagan and Mercer 1989; Stocker et al. 1990; Christensen et al. 1993; Sun et al. 1993; Seki and Kanzaki 2008; Chou et al. 2010). Hetero-LNs as found in honeybees, with a single, densely innervated g lomerulus and a limited number of sparsely innervated other glome­ ruli , have so far only been found in bees, and occasionally in other hymenoptera (Dacks et al. 2010) . What is the spec ific function of these neurons? Based on their mor­ phology, hetero-LNs have been proposed to be function ally polarized, with the main glomerulus being dendritic, and the sparsely innervated glomeruli being axonal (Galizia and Kimmerle 2004). This 'central-input ' polarity is useful to shape inter-glomerular inhibition and elemental mixture processing, in particular given that inter-glomerular inhi­ bition is dependent on the functional similarity of glome­ ruli (Linster et al. 2005) . The neuron shown in Fig. 7a- c shows all the features of such a neuron , in particular given its innervation of glomerulus T 1-29, which receives OSN input responding to 2-heptanone, an odorant that elici ts response in this neuron . However, the neuron shown in Fig. 7d- k does not fit this pattern . This suggests that some hetero-LNs may have another polarity ('central -output') : dendritic innervation in the sparse glomeruli , and axonal in the dense glomerulus, as suggested from developmental studi es where pruning in a dendrite-like fashion was observed in sparse but not dense arbors (Devaud and Masson 1999). A central-output hetero-LN may be suited to create configural response patterns, when its focal inhibitory action on a particu lar glomerulus is based on a di stributed input from a defined glomerular assembly . However, the data also allows a third possibility. Hetero­ LNs mi ght not have a fixed polarity (and form two groups, i.e . 'central-input' and 'central-output' ), but they might in fact ac t ei ther way depending on the activity pattern in the antennal lobe. Dendrodendritic interactions have been shown in olfactory granule cell s in the vertebrate olfactory bulb (Shepherd et al. 2007), and rec iprocal synapses abound in insect anten nal lobes, as shown in cockroach (Malun 1991 ; Distler et al. 1998) and bees (Gascuel and Masson, 1991) . Multiple spike heights have often been recorded from LNs (Flanagan and Mercer 1989; Chris tensen et al. 1993 , 2001 ; Sun et al. 1993; Galizia and Kimmerle 2004; Krofczik et al. 2009). If these are not due to mUltiple neuron recordings or to gap-junctions between neurons, mUltiple 170 spike heights might reflect distributed active membranes, resulting in multiple spike initiation zones. Whether hetero­ LNs are capable of performing both, elemental and confi­ gural odorant processing, in an odor-context-dependent manner, remains to be shown in future studies. If this hypothesis were confirmed, this could explain their high number in the honeybee AL. Honeybees rely to a greater degree on flexible odorant coding than most species, given that they are flower-constant polylectic pollen and nectar foragers. Thus, their olfactory system must be capable of processing and memorizing many odorant mixtures as unique flower identifiers. In addition, social insects have a complex communication system relying on many odors that function as pheromones, for which so far no dedicated subgroup of glomeruli forming a labeled line has been found. Thus, the same combinatorial logic may be used for floral odorants and for pheromones. In this situation, it is an advantage for the bee to identify both the odorant elements in a mixture, and the uniqueness of the mixture itself. The 'elemental' and 'configuraI' neurons presented here are likely to form an important role in the neural networks performing this task. Acknowledgments We thank two anonymous referees for their constructive and helpful comments. We thank Martin P. Nawrot for providing data analysis scripts, practical advice and comments on the manuscript. We are gratefu l to David Gustav and Birgit Rapp for beekeeping, Jacob Stierle, Christoph Kleineidam, Henning Proske, Cyrille Girardin and Sabine Kreissl for fruitfu l discussions, Christine Dittrich and Tobias MUller for technical assistance, Jago Wallenschus and Corinna Geiss for identification of glomeru li and Marit Stranden for help with neuron reconstructions. This work was supported by the Landesgraduiertenfoderung Baden-WUrtemberg, the Deutsche Aka­ demische Austausch Dienst, and the DFG. Open Access This article is distributed under the terms of the Creat ive Commons Attribution Noncommercial License which per­ mits any noncommercial use, distribution , and reproduction in any medium, provided the original author(s) and source are credited. References Ache BW, Young JM (200S) Olfaction: diverse species, conserved principles. Neuron 48(3):4 17-430. doi: I O. 1016/j.neuron.200S. 10.022 Andersson MN, Binyameen M, Sadek MM, Schlyter F (20 11 ) Attraction modulated by spacing of pheromone components and anti -attractants in a ' bark beetle and a moth. J Chem Ecol 37(8):899-9 11. doi: 10. lOO7/s I0886-0 11 -999S-3 Balderrama N, iiez IN, Giurfa M, Torrealba J, Albornoz ED, Almeida LO (1996) A deterrent response in honeybee (Apis mellifera) foragers: dependence on disturbance and season. J Insect Physiol 42:463-470 Balderrama N, iiez IN, Guerrieri F, Giurfa M (2002) Different func tions of two alarm substances in the honeybee. J Comp Physiol A Neuroethol Sens Neura l Behav Physiol 188(6): 48S-49 1. doi: 10.1 007/s003S9-002-0321-y Baraldi R, Rapparani F, Rossi F, Latella A, Ciccio li P (1999) Volatile organic compound emissions from flowers of the most occurring and economically important species of fruit trees . Phys Cem Earth 24:729-732 Bhandawat V, Olsen SR, Gouwens NW, Schlief ML, Wilson RI (2007) Sensory processing in the drosophila an tenna l lobe increases reliability and separabi lity of ensemble odor represen­ tations. Nat Neurosci 10(11 ): 1474- 1482. doi: IO.1038/nnI976 Chou YH, Spletter ML, Yaksi E, Leong JCS, Wilson RI , Luo L (20 I 0) Diversity and wiring variability of olfactory local interneurons in the drosophila antenna l lobe. Nat Neurosci: doi: 1O.1038/nn.2489 (journal club IS.02.20 10) Christensen TA, Waldrop BR, Harrow ID, Hildebrand JG (1993) Local interneurons and information processing in the olfactory g lomeru li of the moth Mallduca sexta. J Comp Physiol [A] 173(4):38S-399 Christensen TA, D' Alessandro G, Lega J, Hildebrand JG (200 1) Morphometric modeling of olfactory circuits in the insect antennal lobe: I. Simulations of spiking local interneurons. Biosystems 61(2-3): 143- IS3 Dacks AM , Reisenman CE, Paulk AC, Nighorn AJ (20 10) Histamine­ immunoreactive local neurons in the antennal lobes of the hymenoptera. J Comp Neurol SI8(1S):2917-2933. doi: 10.10021 cne.2237I Deisig N, Lachnit H, Giurfa M, Hellstern F (2001) Configural olfactory learning in honeybees: negative and positive patterning discrimination. Learn Mem 8(2):70-78. doi: 10.110 Ilm.3830 I Deisig N, Lachnit H, Giurfa M (2002) The effect of similarity between e lementa l stimuli and compounds in olfactory pattern­ ing discriminations. Learn Mem 9(3): 112-121. doi: 10.11011 Im.41002 Deisig N, Lachnit H, Sandoz JC, Lober K, Giurfa M (2003) A modified version of the unique cue theory accounts for olfactory compound processing in honeybees. Learn Mem 10(3): 199-208. doi:l 0.110 Illm.55803 Deisig N, Giurfa M, Lachnit H, Sandoz JC (2006) Neural representation of olfactory mixtures in the honeybee antennallobe. Eur J Neurosci 24(4):1161 - 1174. doi: IO.1111/j .1460-9568.2006.04959.x Deisig N, Giurfa M, Sandoz JC (2010) Antennal lobe processing increases separabi lity of odor mixture representations in the honeybee. J Neurophysiol 1 03(4):2185-2194. doi: 10.1 I 52/jn. 00342.2009 Devaud JM, Masson C (1999) Dendritic pattern development of the honeybee antenna 1 lobe neurons: a laser scanning confocal microscopic study. J Neurobiol 39(4):461-474 Distler PG, Gruber C, Boeckh J (1998) Synaptic connections between GAB A-immunoreactive neurons and uniglomerular projection neurons within the antennal lobe of the cockroach, Periplalleta americalla. Synapse 29( I): 1- 13 Flanagan D, Mercer AR (1989) Morphology and response character­ istics of neurones in the deutocerebrum of the brain in the honeybee Apis mellifera. J Comp Physiol 164:483-494 Fonta C, Sun XJ, Masson C ( 1993) Morphology and spatial distribution of bee antenna l lobe interneurones to odours. Chem Senses 18(2): 101 - 119, doi: 10.1093/chemse/18.2.10I, http:// chemse.oxfordjournals.org/cgi/contentlabstractlI81211 0 I, http:// chemse.oxfordjournals.org/cgi/reprintlI8/211 0 I .pdf Ga li zia CG, Menzel R (2001) The role of glomeruli in the neural representation of odours: results from optica l record ing studies. J Insect Physiol 47(2): II S- 130 Gali zia CG (2008) Insect olfaction. In: The senses: a comprehensive reference. pp 72S-769 Galizia CG, Kimmerle B (2004) Physiological and morphological characterization of honeybee olfactory neurons combining electrophysiology, calcium imaging and confocal microscopy. J Comp Physiol A Neuroethol Sens Neural Behav Physiol 190( I ):21 - 38. doi: 1O.1007/s00359-003-0469-0 Galizia CG, Joerges J, KUttner A, Faber T, Menzel R (1997) A semi­ in-vivo preparation for optical recording of the insect brain. J Neurosci Methods 76(1):61 - 69 Ga lizia CG, Mcllwrath SL, Menzel R ( 1999) A digilal three­ dimensional alias of the honeybee antennallobe based on optical sections acqui red by confocal microscopy. Cell Tissue Res 295(3):383-394 Gascuel J, Masson C ( 199 1) A quantitative ultrastructural study of the honeybee antennal lobe. Tissue Cell 23(3):34 1-355 Junek S, Kludt E, Wolf F, Schild D (20 10) Olfactory coding with patterns of response latencies. Neuron 67:872- 884 Krofczik S, Menzel R, Nawrot MP (2009) Rapid odor processing in the honeybee antennal lobe network. Front Comput Neurosci 2: 1- 13 Lei H, Vickers N (2008) Centra l processing of natural odor mixtures in insects. J Chem Ecol 34(7):9 15-927. doi: 1O.1oo7/sI0886- 008-9487-2 Linster C, Sachse S, Ga lizia CG (2005) Computational modeling suggests that response properties rather than spatial position determine connectivity between olfactory glomeruli . J Neuro­ physiol 93(6):3410- 34 17. doi : 10.1 152/jn.01285.2oo4 Malun D ( 199 1) Inventory and distribution of .synapses of identified uniglomerular projection neurons in the antenna l lobe of Periplaneta americana. J Comp Neurol 305(2):348- 360. doi: 10.1002/cne.903050215 Meier R, Egert U, Aertsen A, Nawrot MP (2008) FIND- a unified framework for neural data analysis. Neural Netw 2 1 (8): 1085- 1093. doi: 10.1016/j .neunet.2008.06.019 Menzel R, Hammer M, MUller U, Rosenboom H (1996) Behavioral , neural and cellular components undei'lying olfactory learning in the honeybee. J Physiol Paris 90(5-6):395-398 Mobbs PG (1982) The brain of the honeybee Apis melli/era I. the connections and spatial organization of the mushroom bodies. Philos Trans R Soc Lond B 298:309- 354 Nawrot MP, Aertsen A, Rotter S (2003) Elimination of response latency variability in neuronal spike trains. Bioi Cybern 88(5):32 1- 334. doi: 10.1 007/s00422-002-039 1-5 Ng M, Roorda RD, Lima SQ, Zemel man BV , Morcillo P, Miesenbck G (2002) Transmission of olfactory information between three populations of neurons in the an tennal lobe of the fly . Neuron 36(3):463-474 Olsen SR, Wilson RI (2008) Lateral presynaptic inhibition mediates gain control in an olfactory circuit. Nature 452(7 190):956- 960. doi: 1O.1038/nature06864 Omata A, Yomogida K, Nakamura S (1990) Volatile components of apple flowers. Flavour Fragr J 5: 19-22 Pareto A ( 1972) Spatial distribution of sensory antenna l fibres in the central nervous system of worker bees. Z Zellforsch Mikrosk Anat 13 1 ( I): 109- 140 Pouzat C, Chaffio l A (2009) Automatic spike train analysis and report generation. An implementat ion with r, r2html and star. J Neurosci Methods 18 1( 1): 11 9- 144. doi: 10.1016/j.jneumeth.2009.01.037 17 1 Pouzat C, Delescluse M, Viot P, Diebolt J (2004) Impro ved spike­ sorting by modeling firing statistics and burst-dependent spike amplitude attenuation: a markov chain monte carlo approach . J Neurophysiol 9 1 (6):29 1 0- 2928. doi:10.1152/jn.002 27.2oo3 Sachse S, Ga lizia CG (2003) The coding of odour-intensity in the honeybee antennal lobe: local computation optimizes odour representation. Eur J Neurosci 18(8):2 11 9- 2 132 Sachse S, Peele P, Silbering AF, Ghmann M, Galizia CG ( 2006) Role of histamine as a putative inhibitory transmitter in the honeybee antennal lobe. Front Zool 3:22. doi: 10.1186/1742-9994-3-22 Sato K, Touhara K (2009) Insect 01 faction: recept ors, signal transduction, and behavior. Results Probl Cell Differ 47: 12 1- 138 Savitzky A, Golay M (1964) Smoothing and differentiation of data by simplified least squares procedures. Anal Chem 36: 1627- 1639 Seki Y, Kanzaki R (2008) Comprehensive morphological identifka­ tion and gab a immunocytochemistry of antennal lobe local interneurons in bombyx mori. J Comp Neurol 506( 1):93-107. doi: 1O.1oo2/cne.21528 Shepherd GM, Chen WR, Willhite D, Migliore M, Greer CA (2007) The olfactory granule cell: from classical enigma to central role in olfactory processing. Brain Res Rev 55(2):373- 382. doi: 10.10 16/j .brainresrev.2oo7 .03 .005 Silbering AF, Galizia CG (2007) Processing of odor mix.tures in the drosophila antennal lobe reveals both global inh ib ition and glomerulus-specific interactions. J Neurosci 27(44): 11 ,966- I 1,977. doi: I 0.1523/JNEUROSCI.3099-07 .2007 Silbering AF, Okada R, Ito K, Galizia CG (2008) Olfactory infonna­ tion processing in the drosophila antennal lobe: anything goes. J Neurosci 28(49): 13075- 13087. doi: 10. 1523/JNEUROSCI. 2973-08.2008 Stocker RF, Lienhard MC, Borst A, Fischbach KF (1990) Neuronal architecture of the antennal lobe in Drosophila me lanogaster. Cell Tissue Res 262(1 ):9-34 Stopfer M, Jayaraman V, Laurent G (2003) Intensity versus identity coding in an olfactory system. Neuron 39(6):99 1- 1004 Sun XJ , Fonta C, Masson C (1993) Odour quality processing by bee antennal lobe interneurones. Chem Senses 18(4):355- 377. doi: 10.1 093/chemse/18.4.355 (http://chemse.oxfordjournals.org/cgi/ conten t! abs tract! 18/4/355 , http://chemse.oxford journals .org/cgil reprint!18/4/355 .pdf) Tabor R, Yaksi E, Weislogel JM, Friedrich RW (2004) Processing of odor mixtures in the zebraflsh olfactory bulb. J Neurosci 24(29):6611 - 6620. doi: 10. 1523/JNEUROSCJ. I 834-04.2004 Tollsten L, Knudsen JT (1992) Scent in dioecious Salix (sal i­ caceaae)- a cue determining the pollination system ? Plant Syst Evol 182:229- 237 Urban NN, Arevian AC (2Q09) Computing with dendrodendritic synapses in the olfactory bulb. Ann NY Acad Sci 11 70:264- 269. doi:J0. llll/j .1749-6632.2009.03899.x VelteI' RS, Sage AE, Justus KA, Card RT, Gal izia CG (2006) Temporal integrity of an airborne odor stimulus is greatly affected by physical aspects of the odor delivery system. Chem Senses 31 (4):359- 369. doi: 10.1 093/chemse/bjj040 Text1: First publ. in: Journal of Comparative Physiology A ; 198 (2012), 2. - S. 159-171 http://dx.doi.org/10.1007/s00359-011-0696-8 Text2: Konstanzer Online-Publikations-System (KOPS) URN: http://nbn-resolving.de/urn:nbn:de:bsz:352-217979