|Home | About | Journals | Submit | Contact Us | Français|
In the Drosophila antennal lobe, excitation can spread between glomerular processing channels. In this study, we investigated the mechanism of lateral excitation. Dual recordings from excitatory local neurons (eLNs) and projection neurons (PNs) showed that eLN-to-PN synapses transmit both hyperpolarization and depolarization, are not diminished by blocking chemical neurotransmission, and are abolished by a gap junction mutation. This mutation eliminates odor-evoked lateral excitation in PNs and diminishes some PN odor responses. This implies that lateral excitation is mediated by electrical synapses from eLNs onto PNs. In addition, eLNs form synapses onto inhibitory LNs. Eliminating these synapses boosts some PN odor responses and reduces the disinhibitory effect of GABA receptor antagonists on PNs. Thus, eLNs have two opposing effects on PNs, driving both direct excitation and indirect inhibition. We propose that when stimuli are weak, lateral excitation promotes sensitivity, whereas when stimuli are strong, lateral excitation helps recruit inhibitory gain control.
Sensory neurons are generally selective for particular stimulus features. Neurons at the same level of sensory processing that are tuned to different features can be thought of as representing different “processing channels” A fundamental question in sensory neuroscience is to understand the mechanisms and functions of cross-talk between such channels.
The notion of a sensory channel is particularly well-defined in early olfactory processing. This is because each glomerulus in the olfactory bulb or antennal lobe defines both an anatomical module and a discrete feedforward circuit. Each olfactory receptor neuron (ORN) is presynaptic to a single glomerulus, and each second-order neuron is postsynaptic to a single glomerulus (Bargmann, 2006). Functional connections between processing channels were until recently thought to be mainly inhibitory, with little or no spread of excitation between principal neurons in different glomeruli (but see Laurent et al., 2001; Lledo et al., 2005; Schoppa and Urban, 2003). Recently, however, several studies in the Drosophila antennal lobe demonstrated the existence of excitatory connections between second-order neurons in different glomeruli (Olsen et al., 2007; Root et al., 2007; Shang et al., 2007). These studies found that when the ORN inputs to a second-order neuron were silenced, that neuron still received indirect odor-evoked excitation from other glomeruli (which we define here as “lateral excitation”). These studies proposed that lateral excitation was mediated by local neurons (LNs) that extend dendrites into many glomeruli and form dendrodendritic synapses with second-order neurons (termed projection neurons, or PNs).
Both the mechanism and function of lateral excitation are uncertain. It has been suggested that its function might be to boost responses to weak stimuli (Olsen et al., 2007; Shang et al., 2007), but this could not be directly tested because there was no known way to abolish lateral excitation. It was also proposed that lateral excitation is mediated by the release of acetylcholine from LNs (Shang et al., 2007). However, this could not be directly tested because cholinergic antagonists block the transmission of all olfactory signals to the brain; this is due to the fact that ORNs are themselves cholinergic (Kazama and Wilson, 2008). Moreover, lateral excitation is recruited very rapidly after ORN signals reach the brain, with a delay almost too short for a disynaptic connection (1.5 msec; Kazama and Wilson, 2008). This rapid recruitment of lateral excitation suggests that the underlying mechanism might be unusual.
In this study, we had two broad aims. Our first aim was to determine the synaptic mechanisms responsible for the spread of excitation between glomeruli. Our second aim was to discover how eliminating these mechanisms alters the output of the antennal lobe in response to olfactory stimuli.
When direct ORN input to a PN is silenced, olfactory stimuli can still elicit an excitatory response (Olsen et al., 2007; Root et al., 2007; Shang et al., 2007). This is thought to reflect the action of excitatory LNs (eLNs). However, it has not been directly demonstrated that any LNs actually have excitatory effects on other neurons. Several studies have noted the existence of GABA-immunonegative LNs (Chou et al., 2010; Shang et al., 2007; Wilson and Laurent, 2005) which are potential candidates for excitatory LNs. Among the Gal4 lines that reportedly drive expression in LNs, none is specific to GABA-negative LNs, but about half of the LNs labeled by krasavietz-Gal4 are GABA-negative (58–61%; Chou et al., 2010; Shang et al., 2007), making it a useful starting point. Shang et. al. (2007) reported that GABA-negative krasavietz LNs are immunopositive for choline acetyltransferase (Cha), although a later study reported that not all are Cha-positive (Chou et al., 2010), casting some doubt on their function.
We therefore began by asking whether krasavietz LNs can excite PNs. We used krasavietz-Gal4 to drive expression of both a GFP reporter and a light-activated cation channel (channelrhodopsin-2 or ChR2; Boyden et al., 2005). Targeted whole-cell in vivo recordings from GFP-labeled LNs confirmed that blue light depolarized these neurons and elicited a train of spikes (Figure 1A,B).
We then made whole-cell in vivo recordings from PNs in these flies. Optogenetic stimulation of krasavietz LNs evoked both excitation and inhibition in PNs (Figure 1C,D). If the PN response were due to the combined action of cholinergic and GABAergic LNs, it should be completely blocked by Cd2+, a broad-spectrum antagonist of voltage-dependent calcium channels and thus a blocker of chemical synaptic transmission. However, in the presence of Cd2+ the excitatory component was actually increased, and only the inhibitory component was abolished (Figure 1D). Subtraction of the traces recorded before and after adding the drug revealed that the Cd2+-sensitive component is hyperpolarizing and slow, whereas the Cd2+-insensitive component is depolarizing and fast (Figure 1E). As a negative control, we also recorded from PNs in flies that lacked the krasavietz-Gal4 driver. Light evoked almost no response, confirming that most of the PN response in flies harboring the krasavietz-Gal4 driver was due to ChR2-mediated currents in Gal4-expressing cells (data not shown). Taken together, these results imply that the krasavietz population includes eLNs, but eLNs do not excite PNs through chemical synapses.
As a next step, we aimed to locate the GABA-negative LNs within the krasavietz population. We labeled krasavietz LNs with CD8:GFP and performed dual immunofluorescence confocal microscopy using anti-GABA and anti-CD8 antibodies. We observed that GABA-negative krasavietz LN somata were located just ventro-lateral to the antennal lobe neuropil (Figure 2A). By comparison, most of the GABA-positive labeled somata were dorso-lateral to the antennal lobe, fewer located ventrally. This suggested that we could preferentially study eLNs by targeting the ventral region.
To test this idea, we performed dual whole-cell recordings from PNs and krasavietz LNs. We injected depolarizing current into each recorded LN to produce a depolarization and a train of spikes (Figure 2B,C). This could evoke either depolarization (Figure 2B) or hyperpolarization (Figure 2C) in the simultaneously-recorded PN. In some cases, there was no PN response. Consistent with the immunostaining results, we only observed depolarization in the PN when we were recording from LNs located in the ventral region. When we observed hyperpolarization in the PN, it was generally when we were recording from LNs located dorsally. In total, we performed 74 dual recordings from PNs and krasavietz LNs, of which 37 showed an excitatory LN-PN connection, 17 showed an inhibitory LN-PN connection, and the remainder showed no connection.
In these experiments, we also noticed that the LNs that depolarized PNs had distinctive electrophysiological properties. Specifically, these cells were always barraged by spontaneous inhibitory postsynaptic potentials (IPSPs; Figure 2D). In these cells we also typically saw small events resembling attenuated action potentials (~10 mV amplitude; Figure 2D) in addition to full-sized spikes (~40 mV amplitude). In our paired recordings, every LN that made made an excitatory connection with the simultaneously-recorded PN also had these distinctive electrophysiological properties. The converse was also true: every LN with these properties also made an excitatory connection with the PN, implying that each eLN is connected to most or all PNs. By contrast, we never saw these properties in LNs that hyperpolarized PNs (Figure 2E). In the experiments that follow, we use the presence of spontaneous IPSPs, together with soma location and GFP expression, as our diagnostic method of identifying eLNs.
If eLNs mediate lateral excitation, then they should be excited by odors. In addition, their odor selectivity should correlate with the selectivity of lateral excitation. In order to test these predictions, we made recordings from eLNs while presenting a chemically diverse panel of odors. We selected odors that would elicit a broad range of total activity levels in the ORN population (Hallem and Carlson, 2006). We verified this using local field potential recordings from the antenna, which provide a rough estimate of the total amount of ORN activity (Olsen et al., 2010). These recordings confirmed that this odor panel elicits a wide range of total ORN activity levels (Figure 3A,B).
Next, we recorded from eLNs to determine how they respond to odors. Interestingly, all the odors in our test panel elicited similar eLN responses, regardless of their chemical structure or the total amount of ORN activity they elicited (Figure 3C,D). Every eLN we recorded from was broadly tuned to odors and was sensitive to even weak ORN input. Finally, we asked how the responses of eLNs compare with the properties of lateral excitation. We removed the feedforward inputs to a subset of PNs by bilaterally removing the olfactory organ housing their cognate ORNs (either the antennae for PNs in an antennal glomerulus, or the palps for PNs in a palp glomerulus). We labeled deafferented PNs with GFP and recorded specifically from these cells while stimulating ORN input to other glomeruli using our panel of test odors. Again, we observed that all the odors in the panel elicited similar lateral excitation in these PNs (Figure 3E,F). Qualitatively similar results were observed for PNs in three different glomeruli (VC1, VC2, DM1).
These results are consistent with the hypothesis that eLNs are the neural substrate of lateral excitation. The sensitivity and broad tuning of lateral excitation have been noted previously (Olsen et al., 2007; Shang et al., 2007). Our results imply that these properties reflect the odor response characteristics of eLNs themselves.
In contrast to a previous study that proposed that eLNs excite PNs by releasing acetylcholine (Shang et al., 2007), our optogenetic stimulation experiments suggest that eLNs do not excite PNs through chemical synapses. However, because this technique activates a mixed population of LNs, the interpretation of this result is complicated by co-activating excitatory and inhibitory inputs. We therefore turned to dual intracellular recordings to examine the properties of eLN-PN connections in a more selective manner. We found that depolarizing the eLN elicited a depolarization in the PN, whereas hyperpolarizing the eLN elicited a hyperpolarization in the PN (Figure 4A,B). Blocking chemical synaptic transmission with Cd2+ did not weaken the PN responses (Figure 4B,C). Similarly, the nicotinic acetylcholine receptor antagonist mecamylamine had no effect, although two other nicotinic antagonists had weak effects in some experiments (Figure S1). It is difficult to determine whether there is a small chemical component to these synapses, and if so, whether this represents spillover or conventional synaptic transmission (see Discussion). What is clear is that eLNs are electrically coupled to PNs, and this represents the main mechanism by which eLNs depolarize PNs.
Two points are worth noting about the values of the coupling coefficients that we measured in these experiments (Figure 4C). First, although these coefficients are small, they almost certainly underestimate the strength of the connection at the synapse. This is because both electrodes are located at the soma, and the soma can be electronically distant from synaptic sites (Gouwens and Wilson, 2009). Thus, voltages will decay substantially while traveling from the presynaptic electrode to the synaptic site, and from the synaptic site to the postsynaptic electrode. Second, depolarizing signals were transmitted more effectively across this electrical connection than were hyperpolarizing signals (Figure 4B,C). This could reflect better propagation of depolarizing voltages to the site of the gap junction, and/or electrical rectification at the junction (Phelan et al., 2008).
We next asked whether we could genetically disrupt the connections from eLNs onto PNs. The Drosophila genome contains multiple genes coding for gap junction subunits (Phelan et al., 1998). Among these, shaking-B (shakB) is a good candidate. The shakB locus produces alternative transcripts, a set of which (shakB.neural) are expressed in the adult central nervous system (Sun and Wyman, 1996; Zhang et al., 1999). The shakB2 allele produces a complete elimination of shakB.neural proteins. This mutation disrupts electrical connections in the optic lobe and in the giant fiber escape pathway (Curtin et al., 2002; Phelan et al., 1996; Sun and Wyman, 1996; Thomas and Wyman, 1984), and it produces defects in visual escape behaviors and increased seizure susceptibility (Kuebler and Tanouye, 2000; Thomas and Wyman, 1984).
We began by asking whether shakB is expressed by antennal lobe neurons. We used patch electrodes to collect the somata of individual GFP-labeled PNs, pooled these samples, and performed RT-PCR with nested primers designed to detect shakB transcripts. We detected a clear band at the predicted size (Figure S2), meaning that this gap junction subunit is likely expressed in antennal lobe PNs.
We then tested whether the shakB2 mutation eliminates electrical synapses from eLNs onto PNs. Dual recordings showed that connections from eLNs onto PNs were completely abolished in mutant flies (Figure 4B–C). These results are consistent with the idea that eLN-to-PN synapses are electrical connections.
Dual recordings also gave us the opportunity to examine synaptic transmission in the reverse direction, from PNs onto eLNs (Figure 4D). In every pair we recorded, stimulating the PN depolarized the eLN (Figure 4E). This implies that each eLN receives excitation from most or all PNs. PN-to-eLN synapses transmitted both depolarizing and hyperpolarizing signals (Figure 4E,F). Cd2+ had no effect on transmission of hyperpolarizing steps, but did significantly reduce transmission of depolarizing steps (Figure 4E,F). A nicotinic antagonist had a similar effect (Figure S1). These results imply that PN-to-eLN synapses are mixed chemical-electrical synapses. Consistent with this conclusion, we found that the shakB2 mutation abolishes the transmission of hyperpolarizing steps but not depolarizing steps (Figure 4E,F). Together, these results show that PNs release acetylcholine onto eLNs, in addition to coupling electrically to eLNs.
Our results suggest that eLNs do not release acetylcholine onto PNs. However, some GABA-negative krasavietz LNs are Cha-positive (Chou et al., 2010; Shang et al., 2007), implying that these cells do synthesize acetylcholine. This raises the question of whether eLNs release acetylcholine onto cells other than PNs. In particular, we wondered whether eLNs might make cholinergic synapses onto iLNs.
To investigate this, we performed dual recordings between eLNs and iLNs (Figure 5A). In many of these pairs, depolarizing the eLN produced a depolarizing response in the iLN (Figure 5B). This was substantially reduced by Cd2+ (Figure 5B,C) and by a nicotinic antagonist (Figure S1). This implies that eLN-to-iLN synapses are largely cholinergic, whereas eLN-to-PN synapses are mainly or purely electrical. (It is possible that some eLN-to-iLN connections are polysynaptic, but the fact that some of these connections were relatively strong makes it unlikely that they are all polysynaptic.)
These synapses also likely have an electrical component, because hyperpolarizing the eLN generally hyperpolarized the iLN (Figure 5B,C). We therefore tested whether the shakB2 mutation alters these connections. Surprisingly, connections from eLNs onto iLNs were completely gone in mutant flies (Figure 5B,C). This is unexpected because the chemical component of these connections should not necessarily depend on the electrical component. This result implies that the electrical component of this synapse is required for the proper development of the chemical component.
We also examined synaptic transmission in the reverse direction, from iLNs onto eLNs (Figure 5D). In some cases, we saw the signature of an electrical connection—namely, weak transmission of both depolarizing and hyperpolarizing pulses from the iLN to the eLN (Figure 5E). In other cases, depolarizing the iLN strongly hyperpolarized the eLN (Figure 5E), suggesting that iLNs can release GABA onto eLNs.
Together, these results demonstrate that excitatory and inhibitory LNs are interconnected. This in turn suggests that eLNs play a role in the recruitment of GABAergic inhibition.
These recordings revealed that shakB is required for the proper development of the chemical component of eLN-to-iLN synapses. Given this, we wondered whether shakB is also required for the chemical component of yet another chemical-electrical synapse—namely, the reciprocal synapse between PNs in the same glomerulus. This synapse transmits both depolarizing and hyperpolarizing steps, and the transmission of depolarizing steps is partially blocked by Cd2+ (Kazama and Wilson, 2009).
In order to record from pairs of such “sister” PNs simultaneously, we took advantage of a Gal4 line which labels seven PNs in glomerulus DA1 (Berdnik et al., 2006; Jefferis et al., 2004). We recorded from pairs of GFP-positive DA1 PNs and probed their reciprocal connections (Figure 6A). In control flies, we found that DA1 PNs were always reciprocally connected, and both depolarizing and hyperpolarizing steps were transmitted across these connections. In the shakB2 mutant, we found that these connections were completely abolished (Figure 6B,C).
We interpret this as evidence that sister PNs are normally coupled by mixed chemical-electrical connections, but that the electrical component is required for the proper development of the chemical component. Interestingly, this result has a direct precedent in the mouse olfactory bulb, where electrical synapses between sister mitral cells are required for the development of chemical synapses between these cells (Maher et al., 2009). We cannot exclude the idea that sister PNs couple indirectly by synapsing onto the same eLN, but this seems unlikely given that PN-PN interactions are relatively strong.
We have shown that the shakB2 mutation eliminates eLN-to-PN synapses. Therefore, if eLNs mediate odor-evoked lateral excitation, then the shakB2 mutation should eliminate this phenomenon. To test this idea, we focused on PNs in three different glomeruli: VC1, VC2, and DM1. We labeled these PNs with GFP to target them specifically with our electrodes, and we removed their feedforward inputs by removing the olfactory organ housing their cognate ORNs. For example, VC1 ORNs are housed in the maxillary palp, and so we removed the palp when we recorded from these PNs so that we could observe purely lateral signals from antennal glomeruli. In this type of recording configuration, we found that all test odors elicited reliable lateral excitation in wild-type PNs, but no odors elicited any lateral excitation in shakB2 mutant PNs (Figure 6). This result supports the conclusion that eLNs mediate odor-evoked lateral excitation. (Note that in the mutant, weak odor-evoked lateral inhibition was observed; this is consistent with a previous report that there is a small amount of postsynaptic lateral inhibition in this circuit (Olsen and Wilson, 2008), although most lateral inhibition is presynaptic.)
For all three PN types we recorded from, the shakB2 mutation had the same effect, arguing that the mechanism of lateral excitation is not glomerulus-specific. This is an important result because different PNs can receive either strong or weak lateral excitation depending on the glomerulus they innervate (Olsen et al., 2007). In these experiments, we noticed that wild-type VC1 and VC2 PNs showed lateral excitation of a size that was typical of most other glomeruli (Olsen et al., 2007), whereas wild-type DM1 PNs consistently showed smaller lateral excitation (Figure 6). Because the shakB2 mutation abolishes lateral excitation in all three cases, we would interpret this heterogeneity as reflecting stronger electrical coupling with the eLN network in some glomeruli, and weaker coupling in other glomeruli.
As a control, we verified that expressing a shakB.neural transgene under Gal4/UAS control rescues odor-evoked lateral excitation (Figure S3). This result shows that the phenotype is due to loss of shakB, and is not an artifact of the genetic background. We also performed a series of control experiments to verify the specificity of the mutant phenotype. First, we confirmed that PN morphology and glomerular compartmentalization is normal in the mutant (Figure S4). Then, we verified that ORN odor responses are also normal (Figure S4). Finally, we checked that iLN-PN connections are normal (Figure S4). Taken together, these results show that the antennal lobe is essentially normal in the shakB2 mutant, except for three types of synapses which have an electrical component and which are abolished: eLN-to-PN synapses, eLN-to-iLN synapses, and PN-PN reciprocal synapses.
Given that the shakB2 mutation eliminates three specific types of synapses in the antennal lobe, we then asked whether this mutation alters PN spiking responses to odors in an otherwise intact circuit. Because shakB2 eliminates synapses from eLNs onto PNs, we might predict that it reduces some PN odor responses. Furthermore, shakB2 also eliminates PN-PN reciprocal synapses, and this is another reason why we would predict that it reduces some PN odor responses. However, because this mutation also eliminates eLN-to-iLN synapses, it might reduce the recruitment of inhibition, and so we might predict that it actually increases some PN odor responses.
To investigate this, we compared PN odor responses in control flies and shakB2 mutants. We used GFP labeling to target our electrodes to PNs in four glomeruli: VC1, DA1, VC2, and DM1. We exploited differences in the way these four types of PNs couple to other glomeruli and to sister PNs in order to disambiguate changes in lateral excitation, lateral inhibition, and PN-PN synapses.
We began with glomerulus VC1 (Figure 8A) because only one PN is known to be present in this glomerulus (Tanaka et al., 2004; see also Experimental Procedures). If only one PN is present, this would simplify the situation because it would mean that there are no PN-PN synapses. In VC1 PNs, we found that responses to all the test odors were weaker in shakB2 flies as compared to controls, and for many odors, this difference was statistically significant (Figure 8B,C).
If this phenotype reflects a loss of lateral excitation, it should disappear for a stimulus which is specific to VC1 ORNs. Fortunately, these ORNs are the only neurons in the palp that respond to the odor fenchone (Goldman et al., 2005). Some antennal ORNs also respond to this odor (data not shown), but we made fenchone a “private” odor for VC1 by removing the antennae. When we recorded from VC1 PNs in flies with intact antennae, fenchone elicited an excitatory response which was significantly smaller in shakB2 mutant flies (Figure 8A,B), implying that this odor elicits lateral excitation onto VC1 from antennal glomeruli. By contrast, when we recorded from VC1 PNs in flies with antennae removed, there was no difference between control and shakB2 responses (Figure 8D,E). This supports the idea that the phenotype is due to the loss of lateral excitation, at least for this PN and these odors.
Next, we examined a second glomerulus, DA1. This glomerulus is notable for containing an unusually large number of sister PNs (seven in total; Berdnik et al., 2006). The only known ligand for DA1 ORNs is cis-vaccenyl acetate, which is also relatively selective for these ORNs (Clyne et al., 1997; Schlief and Wilson, 2007; van der Goes van Naters and Carlson, 2007). Thus, cis-vaccenyl acetate should elicit reciprocal excitation among sister DA1 PNs, but little or no lateral input to these PNs, providing us with an opportunity to look specifically at the role of sister PN interactions in shaping PN odor responses.
We confirmed that cis-vaccenyl acetate elicits an excitatory response in DA1 PNs (Figure 8F,G), as previously reported (Schlief and Wilson, 2007). In shakB2 mutant flies, we found that this response was significantly smaller than in control flies (Figure 8G). Because shakB2 eliminates reciprocal synapses between DA1 PNs (Figure 6), and because cis-vaccenyl acetate is a relatively “private” odor for DA1 ORNs, this result implies that PN-PN synapses can amplify odor responses.
We next turned to glomerulus VC2. In this glomerulus, as in VC1, only one PN may be present (Tanaka et al., 2004). As in VC1, we found that some VC2 PN spiking responses to odors were significantly diminished in the shakB2 mutant (Figure 9A–C). However, other responses were significantly larger in the mutant (Figure 9B,C). Given that this mutation eliminates eLN-to-iLN synapses, this phenotype might reflect a defect in the recruitment of GABAergic inhibition. To test this, we compared the effect of adding GABA receptor antagonists (5 μM picrotoxin and 20 μM CGP54626) in control versus mutant PNs. If shakB2 reduces the amount of GABAergic inhibition recruited by these stimuli in VC2 PNs, then the antagonists should have a smaller effect on mutant VC2 PN odor responses. Indeed, we found that the disinhibitory effect of adding the antagonists was significantly smaller in the mutant than in controls (Figure 9D,E). This is consistent with the conclusion that eLNs are involved in recruiting GABAergic inhibition, presumably via their excitatory synapses onto iLNs.
Finally, we also examined glomerulus DM1. As in VC1 and VC2, only one PN is known to reside in this glomerulus. Unlike VC1 and VC2 PNs, DM1 PNs receive only weak lateral excitation, suggesting weak coupling to the eLN network (Figure 7). Perhaps not surprisingly, the shakB2 mutation had no significant effect on DM1 PN odor responses (Figure S5). This negative result is consistent with our conclusion that the gross anatomy of the antennal lobe is normal in shakB2 mutants, as are ORN odor responses (Figure S4).
Our findings directly demonstrate that LNs can excite PNs. Putative excitatory local neurons have been identified in the olfactory bulb (Aungst et al., 2003), but they have not been shown to have excitatory effects on principal neurons. In this study, we show directly for the first time that local neurons can excite principal neurons and thereby spread activity between glomeruli.
A previous study proposed that excitatory LNs depolarize PNs by releasing acetylcholine (Shang et al., 2007), based on the finding that some krasavietz LNs are immunopositive for Cha. However, using an optogenetic approach, we find that selectively stimulating krasavietz eLNs produces an excitatory response which is insensitive to blocking synaptic vesicle exocytosis, suggesting primarily electrical rather than chemical coupling from eLNs onto PNs. Moreover, dozens of dual recordings from krasavietz eLNs and PNs revealed clear evidence of electrical connections. Together, these results are strong evidence that krasavietz LNs couple to PNs electrically. We found it was difficult to determine conclusively whether eLN-to-PN synapses have a small cholinergic component. On one hand, we found that, on average, neither Cd2+ nor mecamylamine nor D-tubocurarine had any effect, and this argues against a cholinergic component. On the other hand, α-bungarotoxin slightly inhibited coupling (Figure S1). In any event, it is clear that this synapse is largely (if not purely) electrical.
Given these results, it is interesting that eLNs synapses onto iLNs clearly have a strong chemical component. Unlike eLN-to-PN synapses, eLN-to-iLN synapses were almost completely blocked by either Cd2+ or mecamylamine (Figure S1). This implies that the properties of eLN output synapses are target-cell specific: synapses onto PNs are largely or purely electrical, whereas synapses onto iLNs are largely chemical with a smaller electrical component. There are several examples in the literature of a neuron forming synapses with different properties onto different types of target cells (Pelkey and McBain, 2007), but this example of target-cell specialization seems particularly striking.
What is the functional relevance of our finding that eLNs form electrical synapses onto PNs? One distinctive property of electrical connections is their speed. This helps explain why lateral excitation is recruited so rapidly. Indeed, electrical stimulation of the antennal nerve elicits depolarization in maxillary palp glomeruli only about 1.5 msec after onset of depolarization in an antennal glomerulus (Kazama and Wilson, 2008). This suggests that the eLN network contributes to the earliest time points in the PN odor response. Consistent with this idea, we often saw differences between control and shakB2 mutant PNs during the earliest epoch of PN responses (e.g., Figure 8B). Thus, lateral excitation may be preferentially involved in the rising phase of PN odor responses, whereas inhibition seems to be recruited more slowly (Figure 1E). Preferential transmission of the rising phase of an odor pulse may speed odor detection and promote resolution of odor rapid fluctuations (Bhandawat et al., 2007).
Another characteristic feature of electrical synapses is that they are less noisy than chemical synapses (Connors and Long, 2004). A previous study showed that whereas noise in sister PNs is highly correlated, noise is almost entirely uncorrelated in PNs innervating different glomeruli (Kazama and Wilson, 2009). That result implied that LNs contribute relatively little correlated noise to PNs. Our finding that eLN-PN connections are electrical rather than chemical may help explain why that is so. It also suggests that the eLN network is unlikely to add substantial noise to PN odor responses, contrary to a previous suggestion that this is the major function of eLNs (Shang et al., 2007).
A further characteristic property of electrical connections is that they can alter the way signals propagate through a cell. This is because an electrical connection acts as a shunt which diminishes the effect of a synaptic current on a cell’s membrane potential. Thus, eliminating electrical connections can make neurons more electrotonically compact (Bennett and Zukin, 2004). Indeed, we observed that the shakB2 mutation significantly increased PN input resistance in PNs that normally receive relatively strong lateral excitation (p<0.05 for VC1 and VC2, Mann-Whitney U-tests; not significant in DA1 and DM1). This might be expected to increase PN responses to their direct ORN inputs, but this is not what we observed: when VC1 PNs were disconnected from lateral input (by removing the antennae), the shakB2 mutation did not increase the spiking responses of these PNs to their direct ORN inputs, despite the fact that PN input resistance was increased. The change in PN input resistance may be too small to have an effect on PN spike rate; alternatively, changes in PN input resistance may trigger compensatory changes in the strength of ORN-to-PN synapses, as has been shown previously (Kazama and Wilson, 2008).
Previous studies have noted two curious features of odor-evoked lateral excitation in the antennal lobe. First, even weak levels of ORN activity are sufficient to recruit lateral excitation onto PNs. Second, odor stimuli that differ in chemical structure and/or concentration elicit somewhat similar levels of lateral excitation (Olsen et al., 2007; Shang et al., 2007).
Our results imply that these features reflect properties of eLNs themselves. Namely, eLNs respond robustly to even weak levels of ORN activity, and they are relatively indiscriminate in their responses. Each eLN extends its neurites into most or all glomeruli (data not shown), and so it may pool excitatory input from most or all glomeruli. Moreover, we found that the probability of finding a connection from a randomly-chosen PN onto an eLN was 100%. This suggests that all PNs converge onto each eLN. If so, this would help explain why eLNs are sensitive to even weak odors and are excited by all chemical classes of odors.
What is the function of spreading excitation between glomeruli? We suggest that eLNs may function to slightly and transiently increase the excitability of all PNs when any ORN channel is activated. In addition, eLNs might have a role in promoting synchrony among PN spikes. As a result, the eLN network could improve odor detection when stimuli are weak. An obvious potential problem with this network is that spreading excitation between glomeruli could destroy PN odor selectivity. However, this is evidently not a problem in practice because PNs are in fact odor selective (Bhandawat et al., 2007). In this study, we found that increasing level of ORN activity does not substantially increase the strength of eLN odor responses. This ceiling on eLN activity might be useful in preventing lateral excitation from becoming too powerful when odors are strong. In other words, the tendency for eLN odor responses to saturate should help preserve PN odor selectivity.
Thus far, thinking about the functional relevance of eLNs has considered only their role in connecting PNs in different glomeruli. In this study, we discovered that eLNs provide excitatory input not only to PNs, but also to iLNs. Indeed, eLN synapses onto iLNs are stronger than their synapses onto PNs. This implies that a major function of eLNs is to recruit GABAergic inhibition.
Consistent with this, we found that some PN odor responses are actually potentiated by the shakB2 mutation, which suggests a loss of inhibition which is large enough to outweigh the loss of lateral excitation. Moreover, whereas in control flies some odor responses were substantially disinhibited by GABA receptor antagonists, in shakB2 flies these responses were much less disinhibited when GABA receptors were blocked. This result supports the idea that eLNs provide an important source of excitatory drive to iLNs, although iLNs also receive excitatory input from PNs (Figure S4; see also Wilson et al., 2004).
The idea that interneurons can excite other interneurons—thereby modulating inhibition of principal cells—has a precedent in other neural circuits. For example, the vertebrate retina contains two layers of electrically-coupled inhibitory interneurons: horizontal cells in the outer retina, and amacrine cells in the inner retina. Because these retinal networks are purely electrical, they are not thought to boost the overall level excitation in the interneuron network; rather, they are thought to simply average signals across neighboring cells, thus creating a more uniform inhibitory surround (Bloomfield and Volgyi, 2009). In the antennal lobe, electrical coupling between eLNs and iLNs may serve an analogous “smoothing” function. But because eLNs also excite iLNs through chemical synapses, eLNs are likely to also boost the overall level of excitation in iLNs, thereby boosting inhibition of PNs.
Why might it be useful for eLNs to drive both direct excitation of PNs and indirect inhibition of PNs? We propose that the relative importance of these two effects depends on odor intensity. The excitatory drive relayed by eLNs onto PNs is probably most important when odor stimuli are weak. When stimuli are strong, the excitation that eLNs provide to iLNs may be relatively more important. Although we found that increasing stimulus intensity does not substantially increase eLN activity, PN input to iLNs is likely growing as ORN activity rises. It may be the combined excitatory drive from eLNs and PNs onto iLNs that causes GABAergic inhibition to grow with rising ORN activity (Olsen et al., 2010; Olsen and Wilson, 2008). Increasing inhibition helps prevent PN activity from saturating, and may promote odor discrimination (Asahina et al., 2009; Olsen et al., 2010; Root et al., 2008; Sachse and Galizia, 2003).
Electrical networks are pervasive in both vertebrates and invertebrates (Bennett and Zukin, 2004; Connors and Long, 2004; Phelan, 2005). Thus, understanding signal propagation in electrical networks has fundamental importance for understanding how neural circuits function. Several new genetic approaches to control neural circuits in vivo involve disrupting synaptic vesicle release (Luo et al., 2008; Simpson, 2009). However, electrical synapses will be unaffected by these perturbations. Other approaches involve introducing channels that are controlled by light or unnatural ligands (Luo et al., 2008; Simpson, 2009). However, the effects of opening a channel in specific cell populations can differ for electrical versus chemical networks, because electrical connections are bidirectional whereas chemical connections are not. These considerations should inspire caution in interpreting experiments using these approaches, and also emphasize the need for new tools to specifically perturb electrical networks.
Flies were raised at 25°C on conventional cornmeal agar medium under a 12/12 hour light/dark cycle. All experiments were performed on adult female flies, 1–3 days post-eclosion (except where otherwise noted). Gal4 lines were previously described as follows: krasavietz-Gal4 on chromosome 3 (drives Gal4 expression in both GABA+ LNs and GABA- LNs; Chou et al., 2010; drives Gal4 expression in both GABA+ LNs and GABA- LNs; Shang et al., 2007); GH146-Gal4 on chromosome 2 (drives Gal4 expression in a large fraction of PNs; Stocker et al., 1997); NP5221-Gal4 on chromosome 2 (drives Gal4 expression in 1 VC1 PN, 1 VC2 PN, and 1 DM1 PN; Tanaka et al., 2004); Mz19-Gal4 on chromosome 2 (drives Gal4 expression in 7 DA1 PNs; Berdnik et al., 2006; drives Gal4 expression in 7 DA1 PNs; Jefferis et al., 2004). The krasavietz-Gal4 line drives Gal4 expression in at least three eLNs per antennal lobe, based on the fact that we have recorded sequentially from three GFP-positive eLNs in the same antennal lobe when GFP was expressed under the control of this driver. While we assume that the NP5221-Gal4 line drives Gal4 expression all the PNs in glomeruli VC1, VC2, and DM1 (i.e., 1 PN in each of these three glomeruli), we cannot exclude the idea that there are other PNs in these three glomeruli that do not express Gal4. However, in experiments where we labeled PNs in specific glomeruli with photo-activatible GFP (expressed under the Cha promoter), we have found that the total number of PNs in most individual glomeruli is generally the same as the number labeled by enhancer-trap Gal4 lines in our laboratory (W.W. Liu and R.I. Wilson, unpublished observations), so we think this assumption is reasonable. Lines with UAS-linked transgenes were previously described as follows: UAS-CD8:GFP on chromosomes 2 and 3 (Lee and Luo, 1999); UAS-ChR2:YFP on chromosomes 2 and 3 (Hwang et al., 2007; lines “C” and “B”); UAS-shakB.neural on chromosome 2 (Curtin et al., 2002). The shakB2 mutation has been previously characterized (Baird et al., 1990; Homyk et al., 1980).
In vivo whole cell patch clamp recordings from the somata of PNs and LNs were performed as described previously (Wilson and Laurent, 2005). The external saline solution bathing the brain contained (in mM): 103 NaCl, 3 KCl, 5 N-Tris(hydroxymethyl)methyl-2- aminoethane-sulfonic acid, 8 trehalose, 10 glucose, 26 NaHCO3, 1 NaH2PO4, 1.5 CaCl2, and 4 MgCl2 (osmolarity adjusted to 270–275 mOsm). The saline was bubbled with 95% O2/5% CO2 and the pH equilibrated at 7.3. Patch-clamp electrodes were filled with an internal solution consisting of the following (in mM): 140 potassium aspartate, 10 HEPES, 1 EGTA, 4 MgATP, 0.5 Na3GTP, 1 KCl, and 13 biocytin hydrazide. The pH of the internal solution was adjusted to 7.3 and the osmolarity was adjusted to ~ 265 mOsm. Local field potential recordings in the antenna (Figure 3) and single-sensillum recordings of ORN spikes (Figure S4) were performed essentially as described previously (Bhandawat et al., 2007; Olsen et al., 2007). Recordings were performed in current clamp mode using an Axopatch 200B amplifier (Axon Instruments). Recorded voltages were low-pass filtered at 2 kHz and digitized at 10 kHz. Data acquisition and all the analyses were performed in MATLAB (MathWorks) using custom software. All antagonists were prepared as concentrated stock solutions in water and then added to the saline perfusate to achieve the stated final concentration.
Identified PNs were filled with biocytin during the recording, and the morphology of the recorded neurons were visualized post hoc after fixation and after incubation with 1:1000 streptavidin:Alexa Fluor 568 (Invitrogen), as described previously (Bhandawat et al., 2007). The antennal lobe neuropil was visualized with a primary incubation of 1:10 mouse anti-nc82 antibody (Developmental Studies Hybridoma Bank, University of Iowa, Iowa City, IA) and a secondary incubation of 1:250 anti-mouse:Alexa Fluor 633 (Invitrogen), as described previously (Bhandawat et al., 2007). Images were acquired on a Zeiss LSM 510 confocal microscope with a 40× oil-immersion objective.
Odors were diluted in paraffin oil at the concentrations specified, except 4-methylphenol which was diluted in water. Dilutions were freshly prepared every 5 days. Details on the odor sources are posted at http://wilson.med.harvard.edu/odors.html. Odors were delivered with a custom-built olfactometer as described previously (Bhandawat et al., 2007) that dilutes the headspace of the odor vial a further ~10-fold in clean air before it reaches the fly. The flow rate of the odor delivery stream was 2.2 L/min. Odor stimuli were applied for 500 msec every 45 sec for 6 trials per odor per cell.
Newly eclosed flies were cultured for 2 days in the dark on conventional medium supplemented with potato flakes rehydrated in an aqueous solution of all-trans-retinal. All-trans-retinal was prepared as a stock solution in ethanol (35 mM) and diluted 20-fold in water just before mixing with the potato flakes. Blue light was delivered using a 100-W Hg arc lamp (Olympus) attenuated with a 25% neutral density filter. Pulses of light (500 msec) were delivered every 5 sec using a shutter (Uniblitz) controlled by a TTL pulse. As a negative control, we recorded from PNs in flies lacking the Gal4 driver (genotype UAS-Chr2:YFP; UAS-ChR2:YFP) and confirmed that light elicited only a very small depolarization in these PNs (mean ± SEM = 0.2 ± 0.2 mV, n=5).
In the dual recordings involving eLNs, we expressed GFP under the control of the krasavietz Gal4 line to label eLNs, and we recorded from randomly-selected PNs or iLNs. Although PNs were not GFP-labeled in these recordings, they are identifiable based on their small-amplitude action potentials (Wilson et al., 2004; <12 mV). Cells identified as PNs in this way always formed excitatory connections onto eLNs. Similarly, although iLNs were not GFP-labeled in these experiments, they are identifiable based on their large-amplitude action potentials (Wilson et al., 2004; >25 mV) and lack of IPSPs (which distinguishes them from eLNs; see Figure 2). Cells identified as iLNs in this way never formed excitatory chemical synaptic connections onto other cells, but sometimes formed inhibitory chemical synaptic connections onto other cells which were blocked by GABA receptor antagonists (data not shown). In all dual recording experiments, the antennae were removed just before the experiment in order to minimize spontaneous activity. The intensity of current injection was adjusted in each experiment to achieve voltage deflections of approximately ± 40 mV in the cell into which current was injected. Current injections (500 msec duration) were repeated every 5 sec for 40–50 trials. The response of the unstimulated cell was low-pass filtered (50 Hz cutoff) to remove any spikes, and was averaged over all trials. The coupling coefficient was computed as the average change in membrane potential of postsynaptic neuron divided by that of the presynaptic neuron.
Spikes were detected using custom software. The coupling coefficient was calculated by dividing the trial-averaged membrane potential change in the postsynaptic cell by the change in the presynaptic cell. Peri-stimulus time histograms were generated by calculating the firing rate in 50-msec bins that overlapped by 25 msec. Mean PN spiking responses were quantified as the average spike rate during a 500-msec window beginning 100 msec after nominal stimulus onset, averaged across all 6 trials with a given stimulus (Figures 8 and and9).9). Lateral excitation in PNs was calculated as the average odor-evoked change in membrane potential (versus the pre-odor baseline membrane potential in each trial) during a 200-msec time window beginning 100 msec after nominal stimulus onset, averaged across all 6 trials (Figure 7 and Figure S3). All error bars/bands represent ± SEM values.
We are grateful to L.C. Griffith, K. Ito, L. Luo, G. Miesenböck, M.A. Tanouye, W.D. Tracey, and R.J. Wyman for gifts of fly stocks. D. Schmucker, N. Obholzer, and N. Jurish-Yaksi provided advice on RT-PCR experiments, and K.I. Nagel helped with the writing of the data acquisition software. We thank members of the Wilson lab for their feedback on the manuscript. This work was funded by a long-term fellowship from the Human Frontiers Science Program (to E.Y.), a grant from the NIH (R01DC008174), a McKnight Scholar Award, and Beckman Young Investigator Award (to R.I.W.). R.I.W. is an HHMI Early Career Scientist.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.