Fruit flies are attracted to the smell of rotting fruit, and use it to guide them to nearby food sources. However, this task is made more challenging by the fact that the distribution of scent or odor molecules in the air is constantly changing. Fruit flies therefore need to cope with, and exploit, this variation if they are to use odors as cues.

Odor molecules bind to receptors on the surface of nerve cells called olfactory sensory neurons, and trigger nerve impulses that travel along these cells. The olfactory sensory neurons are connected to other cells called projection neurons that in turn relay information to the higher centers of the brain. While many studies have investigated how fruit flies can distinguish between different odors, less is known about how animals can use variation in the strength of an odor to guide them towards its source.

Kim et al. have now addressed this question by devising a method for delivering precise quantities of odors in controlled patterns to fruit flies, and then measuring the responses of olfactory sensory neurons and projection neurons. These experiments revealed that olfactory sensory neurons—which are found mainly in the flies' antennae—responded most strongly whenever an odor changed rapidly in strength, and showed relatively little response to constant odors. An independent study by Schulze, Gomez-Marin et al. found that olfactory sensory neurons in fruit fly larvae also respond in a similar way.

Kim et al. also found that the response of the projection neurons depended on both the rate of nerve impulses in the olfactory sensory neurons and on how quickly this rate was changing. But, unlike the olfactory sensory neurons, projection neurons showed their strongest responses immediately after an odor first appeared.

Thus, in contrast to organisms such as bacteria and worms, which are highly sensitive to the local concentration gradients of odors, fruit flies instead appear to be more responsive to the sudden appearance of an odor in their environment. Kim et al. suggest that this difference may reflect the fact that for ground-based organisms, local gradients are generally reliable predictors of the location of an odor source. However, for flying insects, continually changing air currents mean that predictable local gradients are less common. Therefore, the ability to detect a hint of an odor before the wind changes is a more useful skill.

Odor distribution in nature is intermittent and dynamic (Murlis et al., 1992; Vickers et al., 2001), and animals have evolved the ability to detect and respond to temporal variation of odor stimuli (David et al., 1983; Thesen et al., 1993; Vickers et al., 2001; Porter et al., 2007; Semmelhack and Wang, 2009; Kato et al., 2014). In one of the most sophisticated examples, Drosophila larvae with only a single functional olfactory sensory neuron (OSN) are capable of moving toward a droplet of an attractive odor by actively orienting themselves (Louis et al., 2008). Similarly, adult fruit flies exhibit robust odor-guided behaviors such as turning upwind in flight upon contact with an attractive odor plume (Budick and Dickinson, 2006) and staying within a specific odor zone (Semmelhack and Wang, 2009). In order to enable such odor-guided tasks, it is essential for any olfactory system to process time-varying features of olfactory stimuli and supply behaviorally relevant information to higher brain centers.

Several recent studies have investigated how dynamic olfactory stimuli are processed in insect early olfactory systems (systems consisting principally of OSNs and projection neurons [PNs]) and observed significant temporal processing of odor signals (Bhandawat et al., 2007; Geffen et al., 2009; Kim et al., 2011; Nagel and Wilson, 2011; Martelli et al., 2013). Most of these studies employed a simple odor delivery system that generated step-pulse-like odor stimuli without directly monitoring the actual odor concentration levels. For a rigorous understanding of sensory processing, however, it is essential to precisely measure the input stimuli and systematically explore the input space, as has been successfully done in the field of vision and audition (Wu et al., 2006). Moreover, natural odor plumes are encountered in various spatiotemporal patterns, and their dynamics and statistics can influence the neural encoding mechanism (Brenner et al., 2000; Vickers et al., 2001).

In Drosophila, OSNs expressing the same receptors connect with PNs in one of roughly 50 spherical compartments, termed olfactory glomeruli, which constitute a deutocerebral neuropil called the antennal lobe. PNs subsequently relay olfactory information to higher brain centers such as the mushroom body and the lateral horn (Stocker et al., 1990).

Two recent studies independently reported that Drosophila OSNs encode not only the odor concentration but also its rate of change as a function of time (Kim et al., 2011; Nagel and Wilson, 2011). Building on this recent advance, we asked how PNs further contribute to creating internal representations of dynamic olfactory environments. We tested OSNs and PNs with short plume-like odor stimuli in a variety of settings and analyzed the correlation structure of input/output signals in the odor-OSN-PN pathway. We also constructed a two-dimensional (2D) linear–nonlinear (LN) model of the OSN-to-PN transformation by inducing an ensemble of triangle-shaped OSN spike rates via a systematic design of olfactory stimuli.

We employed a novel odor delivery system that can reliably produce various odor concentration waveforms and provide measurements of the odor concentration with a millisecond resolution on every experiment trial (Figure 1A,B) (Kim et al., 2011). Various odor concentration profiles were designed and tested (Figure 1—figure supplement 1), and the corresponding OSN and PN responses were measured in two separate assays sharing the same odor delivery system (Figure 1A,B). The observed odor concentrations were closely matched between the two assays (Figure 2A–C). We used acetone as the primary odorant because its low ionization potential afforded a high signal-to-noise ratio in our odor concentration measurements. We tested a pair of directly connected OSNs and PNs innervating the DM4 glomerulus with five different acetone concentration waveforms. The dynamics of OSN and PN responses differed significantly from their respective feedforward inputs, and all responses initiated within a few tens of milliseconds of the odor onset (Figure 1C). PNs generally showed a bigger peak spike rate and exhibited more phasic spiking patterns than the presynaptic OSNs. However, the exact functional transformation between OSNs and PNs could not be readily assessed due to the complex dynamics of OSN and PN signals.

Figure 1 with 1 supplement see all

Download asset Open asset

Figure 2 with 1 supplement see all

Download asset Open asset

We therefore designed a set of elementary odor concentration waveforms: a step, a ramp and a parabola (top row of Figure 2A–C). We reasoned that the simple nature of these waveforms would facilitate the analysis of the input/output relationship.

Consistent with previous reports (Kim et al., 2011; Nagel and Wilson, 2011), OSNs responded most strongly to polynomial waveforms when the odor concentration rose rapidly. For the ramp and parabola odor signals, OSN responses were the exact rate-of-change function of their input: a step output to a ramp input (Figure 2B,E) and a ramp output to a parabola input (Figure 2C,F). This pattern of encoding was preserved for different combinations of odorants and OSN-PN pairs (Figure 2—figure supplement 1). To look into this observation formally, we measured the similarity between three different input features—concentration amplitude, rate of change, and acceleration—and the output spike rate of OSNs by performing a set of cross-correlation analyses. The correlation was strong between the OSN output and both odor concentration and its rate of change, but weak between the OSN output and the acceleration of odor concentration (Figure 2J).

Given the very same panel of olfactory stimuli, PN responses exhibited bigger peak amplitude and more stereotyped temporal patterns than OSNs (Figure 2G–I). All PN responses transiently peaked at the time of the odor onset, and their peak times were on average advanced from the OSN peak times. Specifically, in response to ramp-shaped OSN signals, PNs exhibited a rapid step response (Figure 2F,I). To step-like OSN signals, PN showed a phasic onset response, followed by a tonic spiking pattern (Figure 2E,H). These input/output relationships are reminiscent of the rate-of-change encoding in the odor-to-OSN transformation, where the ramp and step odor signals were similarly transformed. Biophysically, this phenomenon suggests a rapid adaptation between OSNs and PNs, which is in agreement with previous reports demonstrating strong short-term depression in synapses between OSNs and PNs (Kazama and Wilson, 2008, 2009; Nagel et al., 2015). From a modeling perspective, this observation predicted a strong correlation between the PN spike rate and the rate of change of the OSN spike rate. A correlation analysis between the PN output and three different input features—OSN spike rate, rate of change, and acceleration—confirmed this prediction (Figure 2K). Together, this supports the hypothesis that PN responses can be modeled as a function of the OSN spike rate and its rate of change, at least for the tested class of odor stimuli.

We therefore propose a simple model of the OSN-to-PN transformation comprising two input blocks: the OSN spike rate and its rate of change, followed by a 2D nonlinearity that maps these inputs into the PN spike rate (Figure 3D). The model has the structure of a classical 2D LN model with its linear blocks postulated as amplitude and rate-of-change filters (Brenner et al., 2000; Geffen et al., 2009). Given the hypothesized input blocks, the model can be fully identified by estimating the nonlinear block from experimental data. The validity of the model can be tested by assessing its ability to predict the PN response to new odor waveforms. In order to estimate the nonlinearity, we first designed OSN spike activity profiles that explore the hypothesized PN input space with an efficient sampling grid. An ensemble of triangle-shaped inputs was shown to be suitable for this purpose, since each up/down ramp stimulus tests the target system with a distinct pair of positive and negative gradients while also sweeping a wide range of input amplitudes (Kim et al., 2011).

Figure 3 with 1 supplement see all

Download asset Open asset

How can we induce an ensemble of triangle signals at the OSN output? We fine-tuned odor concentration signals, while observing the OSN spike rate, until the desired OSN spike rates were produced (Figure 3A). The odor concentration signals inferred from this method are an ensemble of parabolas with systematically varying peak times and amplitudes (Figure 3A). The 2D nonlinearity was estimated subsequently by running a ridge regression analysis on spike rate samples from the OSN and PN signals (Kim et al., 2011) (Figure 3B,C).

The estimated 2D nonlinearity showed a strong dependency on the rate-of-change of OSN spike rate (Figure 3B,C). When the nonlinearity was estimated for samples with a fixed odor concentration value, the PN spike rate rose almost linearly with respect to the rate of change (Figure 3—figure supplement 1). However, the gain of the PN output with respect to the OSN rate of change decreased monotonically with concentration, suggesting that at least a 2D model is required to describe the OSN-to-PN transformation (Figure 3—figure supplement 1). We tested the model by comparing the simulated model output with the experimental PN spike activity (Figure 3C). With step, ramp, and parabola odor inputs, the experimental outputs were closely matched by the model (root-mean-square error = 25 spike/s, Figure 3F). While the odor-evoked responses were well matched by the model, including peak times and amplitudes, the model output often overestimated the actual PN spike rate before and after the stimulus interval. This is because the estimated nonlinearity exhibits a relatively high slope at low input amplitudes (Figure 3—figure supplement 1), which renders the system highly sensitive to the noisy fluctuations at low OSN spike rates (Pahlberg and Sampath, 2011). This problem is thought to be mitigated in olfactory glomeruli by pooling input from a population of OSNs, thereby achieving a higher signal-to-noise ratio than provided by a single OSN (Bhandawat et al., 2007). In summary, the proposed simple 2D model provides a decent approximation of the OSN-to-PN transformation and corroborates the notion that PNs mainly encode OSN spike rate and its rate of change in a nonlinear fashion.

What is the functional consequence of acceleration encoding by antennal lobe PNs? In signal processing theory, the time derivative, or rate-of-change operation, advances the phase of a time signal. For example, cos(t) is the time derivative of sin(t), and its phase is advanced by π/2. Similarly, the rate of change of a sample triangle-shaped signal shows an advancement of a peak time, and its acceleration exhibits a peak time that is further advanced (Figure 4A). Therefore, we reasoned that the computation of the first and second time derivatives (rate of change and acceleration) of odor concentrations by OSN and PNs, respectively, acts to advance peak times of olfactory stimuli. To investigate this idea with a larger set of odor stimuli, we designed a set of triangle-shaped odor waveforms with peak times varying uniformly between 0.6 s and 1.7 s after the stimulus onset (Figure 4B, ‘odor’ curve in Figure 4C). As shown in the previous experiments with triangle-shaped stimuli (Figure 2B), OSNs responded in a pulse-shaped pattern (Figure 4B), with the peak response advanced by 400–1000 ms relative to the stimulus peak (‘OSN’ curve in Figure 4C). In response to the same panel of odor stimuli, PNs consistently produced peak output (Figure 4B) at around 200 ms after the stimulus onset, regardless of the dynamics of the odor/OSN signals (‘PN’ curve in Figure 4C). For the parabola-shaped odor stimuli (Figure 3A), the peak OSN and PN responses were similarly advanced in time relative to the odor and OSN peaks (right plot in Figure 4C) but were more variable. The advancement of a peak time was previously reported for step-pulse odor stimuli (Bhandawat et al., 2007), and our work confirms this result for a larger set of dynamically varying odor stimuli. Together, we hypothesize that acceleration encoding in the Drosophila antennal lobe allows higher olfactory centers to rapidly respond to onsets of slowly rising odor stimuli.

Figure 4

Download asset Open asset

We tested first- and second-order neurons in the Drosophila olfactory system with a wide range of dynamically varying odor concentrations and constructed a simple 2D LN model for the temporal processing of olfactory information. The shape of the 2D nonlinearity and the relatively low prediction error of the model support the hypothesis that the PN spike rate is most strongly dependent on the rate of change of its feedforward OSN input. When combined with the dynamic sensory encoding by OSNs—encoding of odor concentration and its rate of change—PNs signal the rate of change and, most strongly, the acceleration of odor concentration signals to higher brain centers. One remarkable example of the ‘acceleration encoding’ by PNs is found in the OSN/PN responses to the parabolically rising odor input, where the onset is enormously amplified to create step-like responses in PNs (Figure 2C,I).

To the best of our knowledge, the encoding of acceleration by an olfactory sensory system has not been demonstrated elsewhere. This sensory encoding mechanism is distinguished from other chemosensory systems. For example, in bacterial chemotaxis, the rate of change of a stimulus is computed and fed directly into its motor system that controls the flagellar rotation (Bourret et al., 1991). The nematode Caenorhabditis elegans also exhibits a local-gradient-based chemotaxis behavior, and neurons in its sensorimotor pathway are shown to compute the rate of change of concentration gradients (Suzuki et al., 2008).

Why do fruit flies compute acceleration of odor concentrations? How is the acceleration of an olfactory stimulus relevant to a fly's chemotaxis? An efficient chemotaxis strategy should depend on the distribution of odor molecules in the behaviorally relevant range, which can be quantified using a Reynolds number (Weissburg, 2000). Habitats of both nematodes and bacteria have low Reynolds numbers (Re < 1), and thus the odor distribution is primarily determined by a diffusion process. In this environment, the steepest local odor gradient can be directly linked to the location of the odor source, and therefore animals in such environments are expected to encode the rate of change of odor concentrations in order to detect the steepest gradient. In contrast, flying insects experience fluid mechanics that have relatively high Reynolds numbers (Re > 10), with turbulence dictating the distribution of odor molecules over the diffusion process. In this regime, the local odor gradient within an odor plume would be less informative in determining the location of the odor source than the distribution of odor plumes over a macroscopic scale. In fact, it has been well established that insects, including fruit flies, use a strategy of turning upwind during odor plume encounters, a strategy that does not require interpreting local concentration gradients (Budick and Dickinson, 2006).

Furthermore, the relative position of a flying animal and the distribution of odor plumes evolve rapidly over time. It is therefore crucial for flying insects to detect and respond to an encountered odor plume before the fly's position drifts relative to adjacent plumes. We showed that Drosophila's early olfactory system signals the acceleration of odor concentrations to higher brain centers. This encoding mechanism allows an animal to detect the odor onset at a very early phase (Figure 4). Therefore, it is tempting to speculate that the acceleration encoding has evolved in Drosophila to maximize its chance to locate the odor source in a turbulent environment.

It has also been shown that fruit flies can respond to the spatial difference of the odor concentration between two antennae in both walking and flight (Duistermars et al., 2009; Gaudry et al., 2013). However, even with the lack of spatial sampling ability, fruit fly larvae can locate an odor source, based on the temporal sampling of odor gradients (Louis et al., 2008). Therefore, adult fruit flies may rely on both spatial and temporal sampling methods to locate an odor source, although their relative importance may vary depending on the locomotive state, as Reynolds numbers are different between walking and flight.

1. 1. Bhandawat V
   2. Olsen SR
   3. Gouwens NW
   4. Schlief ML
   5. Wilson RI

   (2007) Sensory processing in the Drosophila antennal lobe increases reliability and separability of ensemble odor representations

   Nature Neuroscience 10:1474–1482.

   https://doi.org/10.1038/nn1976

   • Google Scholar
2. 1. Bourret RB
   2. Borkovich KA
   3. Simon MI

   (1991) Signal transduction pathways involving protein phosphorylation in prokaryotes

   Annual Review of Biochemistry 60:401–441.

   https://doi.org/10.1146/annurev.bi.60.070191.002153

   • Google Scholar
3. 1. Brenner N
   2. Bialek W
   3. de Ruyter van Steveninck R

   (2000) Adaptive rescaling maximizes information transmission

   Neuron 26:695–702.

   https://doi.org/10.1016/S0896-6273(00)81205-2

   • Google Scholar
4. 1. Budick SA
   2. Dickinson MH

   (2006) Free-flight responses of Drosophila melanogaster to attractive odors

   The Journal of Experimental Biology 209:3001–3017.

   https://doi.org/10.1242/jeb.02305

   • Google Scholar
5. 1. Datta SR
   2. Vasconcelos ML
   3. Ruta V
   4. Luo S
   5. Wong A
   6. Demir E
   7. Flores J
   8. Balonze K
   9. Dickson BJ
   10. Axel R

   (2008) The Drosophila pheromone cVA activates a sexually dimorphic neural circuit

   Nature 452:473–477.

   https://doi.org/10.1038/nature06808

   • Google Scholar
6. 1. David CT
   2. Kennedy JS
   3. Ludlow AR

   (1983) Finding of a sex pheromone source by gypsy moths released in the field

   Nature 303:804–806.

   https://doi.org/10.1038/303804a0

   • Google Scholar
7. 1. Duistermars BJ
   2. Chow DM
   3. Frye MA

   (2009) Flies require bilateral sensory input to track odor gradients in flight

   Current Biology 19:1301–1307.

   https://doi.org/10.1016/j.cub.2009.06.022

   • Google Scholar
8. 1. Gaudry Q
   2. Hong EJ
   3. Kain J
   4. de Bivort BL
   5. Wilson RI

   (2013) Asymmetric neurotransmitter release enables rapid odour lateralization in Drosophila

   Nature 493:424–428.

   https://doi.org/10.1038/nature11747

   • Google Scholar
9. 1. Geffen MN
   2. Broome BM
   3. Laurent G
   4. Meister M

   (2009) Neural encoding of rapidly fluctuating odors

   Neuron 61:570–586.

   https://doi.org/10.1016/j.neuron.2009.01.021

   • Google Scholar
10. 1. Kato S
    2. Xu Y
    3. Cho CE
    4. Abbott LF
    5. Bargmann CI

    (2014) Temporal responses of C. elegans chemosensory neurons are preserved in behavioral dynamics

    Neuron 81:616–628.

    https://doi.org/10.1016/j.neuron.2013.11.020

    • Google Scholar
11. 1. Kazama H
    2. Wilson RI

    (2008) Homeostatic matching and nonlinear amplification at identified central synapses

    Neuron 58:401–413.

    https://doi.org/10.1016/j.neuron.2008.02.030

    • Google Scholar
12. 1. Kazama H
    2. Wilson RI

    (2009) Origins of correlated activity in an olfactory circuit

    Nature Neuroscience 12:1136–1144.

    https://doi.org/10.1038/nn.2376

    • Google Scholar
13. 1. Kim AJ
    2. Lazar AA
    3. Slutskiy YB

    (2011) System identification of Drosophila olfactory sensory neurons

    Journal of Computational Neuroscience 30:143–161.

    https://doi.org/10.1007/s10827-010-0265-0

    • Google Scholar
14. 1. Louis M
    2. Huber T
    3. Benton R
    4. Sakmar TP
    5. Vosshall LB

    (2008) Bilateral olfactory sensory input enhances chemotaxis behavior

    Nature Neuroscience 11:187–199.

    https://doi.org/10.1038/nn2031

    • Google Scholar
15. 1. Martelli C
    2. Carlson JR
    3. Emonet T

    (2013) Intensity invariant dynamics and odor-specific latencies in olfactory receptor neuron response

    Journal of Neuroscience 33:6285–6297.

    https://doi.org/10.1523/JNEUROSCI.0426-12.2013

    • Google Scholar
16. 1. Murlis J
    2. Elkinton JS
    3. Carde RT

    (1992) Odor plumes and how insects use them

    Annual Review of Entomology 37:505–532.

    https://doi.org/10.1146/annurev.en.37.010192.002445

    • Google Scholar
17. 1. Nagel KI
    2. Hong EJ
    3. Wilson RI

    (2015) Synaptic and circuit mechanisms promoting broadband transmission of olfactory stimulus dynamics

    Nature Neuroscience 18:56–65.

    https://doi.org/10.1038/nn.3895

    • Google Scholar
18. 1. Nagel KI
    2. Wilson RI

    (2011) Biophysical mechanisms underlying olfactory receptor neuron dynamics

    Nature Neuroscience 14:208–216.

    https://doi.org/10.1038/nn.2725

    • Google Scholar
19. 1. Pahlberg J
    2. Sampath AP

    (2011) Visual threshold is set by linear and nonlinear mechanisms in the retina that mitigate noise: how neural circuits in the retina improve the signal-to-noise ratio of the single-photon response

    Bioessays 33:438–447.

    https://doi.org/10.1002/bies.201100014

    • Google Scholar
20. 1. Porter J
    2. Craven B
    3. Khan RM
    4. Chang SJ
    5. Kang I
    6. Judkewitz B
    7. Judkewicz B
    8. Volpe J
    9. Settles G
    10. Sobel N

    (2007) Mechanisms of scent-tracking in humans

    Nature Neuroscience 10:27–29.

    https://doi.org/10.1038/nn1819

    • Google Scholar
21. 1. Semmelhack JL
    2. Wang JW

    (2009) Select Drosophila glomeruli mediate innate olfactory attraction and aversion

    Nature 459:218–223.

    https://doi.org/10.1038/nature07983

    • Google Scholar
22. 1. Stocker RF
    2. Lienhard MC
    3. Borst A
    4. Fischbach KF

    (1990) Neuronal architecture of the antennal lobe in Drosophila melanogaster

    Cell and Tissue Research 262:9–34.

    https://doi.org/10.1007/BF00327741

    • Google Scholar
23. 1. Suzuki H
    2. Thiele TR
    3. Faumont S
    4. Ezcurra M
    5. Lockery SR
    6. Schafer WR

    (2008) Functional asymmetry in Caenorhabditis elegans taste neurons and its computational role in chemotaxis

    Nature 454:114–117.

    https://doi.org/10.1038/nature06927

    • Google Scholar
24. 1. Thesen A
    2. Steen JB
    3. Døving KB

    (1993)

    Behaviour of dogs during olfactory tracking

    The Journal of Experimental Biology 180:247–251.

    • Google Scholar
25. 1. Vickers NJ
    2. Christensen TA
    3. Baker TC
    4. Hildebrand JG

    (2001) Odour-plume dynamics influence the brain's olfactory code

    Nature 410:466–470.

    https://doi.org/10.1038/35068559

    • Google Scholar
26. 1. Wang JW
    2. Wong AM
    3. Flores J
    4. Vosshall LB
    5. Axel R

    (2003) Two-photon calcium imaging reveals an odor-evoked map of activity in the fly brain

    Cell 112:271–282.

    https://doi.org/10.1016/S0092-8674(03)00004-7

    • Google Scholar
27. 1. Weissburg MJ

    (2000) The fluid dynamical context of chemosensory behavior

    The Biological Bulletin 198:188–202.

    https://doi.org/10.2307/1542523

    • Google Scholar
28. 1. Wu MC
    2. David SV
    3. Gallant JL

    (2006) Complete functional characterization of sensory neurons by system identification

    Annual Review of Neuroscience 29:477–505.

    https://doi.org/10.1146/annurev.neuro.29.051605.113024

    • Google Scholar

Author details

1. Anmo J Kim

   Department of Electrical Engineering, Columbia University, New York, United States

   Present address

   The Rockefeller University, New York, United States

   Contribution

   AJK, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article, Contributed unpublished essential data or reagents

   Competing interests

   The authors declare that no competing interests exist.

2. Aurel A Lazar

   Department of Electrical Engineering, Columbia University, New York, United States

   Contribution

   AAL, Conception and design, Analysis and interpretation of data, Drafting or revising the article

   For correspondence

   aurel@ee.columbia.edu

   Competing interests

   The authors declare that no competing interests exist.

3. Yevgeniy B Slutskiy

   Department of Electrical Engineering, Columbia University, New York, United States

   Contribution

   YBS, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article, Contributed unpublished essential data or reagents

   Competing interests

   The authors declare that no competing interests exist.

• 2,920

  views

• 616

  downloads

• 50

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.