Odor-guided navigation is common across the animal kingdom (Baker et al., 2018). Olfactory cues inform an animal of its location in a natural environment (Boie et al., 2018), and allow it to adjust its locomotion to navigate an odor landscape in a goal-directed manner (Bargmann and Horvitz, 1991; Berg and Brown, 1972; Aceves-Piña and Quinn, 1979). Odor-guided navigation is an ethologically relevant task that is important for the animal’s survival, and it has been a useful framework with which to study genes and circuits underlying sensory-motor transformations (Calhoun and Murthy, 2017; Clark et al., 2013). We focus in particular on the navigation of small model organisms in continuous, non-turbulent gradients established by the spread of odorants from their sources due to diffusion and drift. How animals interpret these gradients and use them to inform their actions remains an active and productive area of research especially in genetic model systems like Caenorhabditis elegans and Drosophila melanogaster (Bargmann and Horvitz, 1991; Aceves-Piña and Quinn, 1979; Levy and Bargmann, 2020; Mattingly et al., 2021; Gomez-Marin et al., 2011; Gepner et al., 2015; Tastekin et al., 2018).

A major challenge is to quantitatively relate the animal’s behavior to the precise olfactory cue that the animal experiences moment-by-moment. Therefore, it is critical to precisely control the odor environment and record the sensory cues experienced by these animals. The need to control and measure odorants still pose a formidable challenge. While many techniques exist to present odor stimuli (Gorur-Shandilya et al., 2019; Kadakia et al., 2022), measure odor concentration (Tariq et al., 2021; Celani et al., 2014), or develop virtual odor environments in a lab (Yamada et al., 2021; Radvansky and Dombeck, 2018), it remains challenging for precise control and continuous monitoring of an odor landscape.

All approaches to generate odor landscapes in a lab environment must contend with the odor’s diffusivity and interaction with other substrates. Early approaches to control odor concentration relied on passive diffusion to construct a quasi-stationary spatial odor gradient, for example by adding a droplet of odorant in a Petri dish in a ‘droplet assay’ (Monte et al., 1989; Pierce-Shimomura et al., 1999; Louis et al., 2008; Iino and Yoshida, 2009). Diffusion places severe limits on the space of possible landscapes that can be created and on the timescales over which they are stable, and the created odor profile is sensitive to adsorption of odor to surfaces, absorption into the substrate, temperature gradients, and air currents, all parameters that are difficult to measure, model, or control. Microfluidics allow water-soluble odors to be continuously delivered to a chamber in order to provide spatiotemporal control (Chronis et al., 2007; Albrecht and Bargmann, 2011; Lockery et al., 2008). Microfluidics devices, however, are limited in extent, require water-soluble odors, and must be tailor-designed to the specific attributes of the animal’s size and locomotion. A microfluidic post-array has been shown to support C. elegans locomotion, but no similar microfluidic device has been demonstrated to support olfactory navigation of Drosophila larvae, for example.

We previously reported a macroscopic gas-based active flow cell that uses parallel flow paths to construct temporally stable odor profiles (Gershow et al., 2012). That approach allows for finer spatiotemporal control of the odor gradient, is compatible with Drosophila larvae, and works with volatile airborne-based odor cues. This device used an array of solenoid valves to generate programmable odor profiles, but perhaps because of its complexity has not been widely adopted.

Most methods to create an odor landscape do not provide a means for knowing or specifying the spatiotemporal odor concentration. In other words, while an experimenter may know that some regions of an area have higher odor concentrations, they cannot quantify the animal’s behavior given a precise concentration of the odor. This limits the ability to quantitatively characterize sensorimotor processing. To address this shortcoming, various methods have been proposed to measure odor concentration across space. For example, gas samples at specific locations have been taken and measured offline (Yamazoe-Umemoto et al., 2018).

In one of the most comprehensive measurements to date, Louis et al., 2008 and Tadres et al., 2022, used infrared spectroscopy to measure the spatial profile of a droplet-based odor gradient. They used this approach to study bilateral olfactory sensing and olfactory encoding in Drosophila larvae, clearly demonstrating the value of measuring odor landscapes. That work measured odor concentration at a relatively low temporal resolution ($\sim 1$ sample/min), only when animals weren’t present, and relied on a relatively expensive Fourier transform infrared spectroscopy machine.

In all of these previous works, measurements were performed offline, not during animal behavior, and the odor concentration was assumed to be the same across repeats of the same experiment, or when animals were present. But even a nominally stable odor landscape is subject to subtle but significant disruptions over time from small changes in airflow, from temperature variation, and from the odor’s interaction with the substrate, which can include absorption, adhesion, and reemission (Gorur-Shandilya et al., 2019; Yamazoe-Umemoto et al., 2018; Tanimoto et al., 2017; Yamazoe-Umemoto et al., 2015). This is challenging to account for and control within a single behavior experiment, and is even more difficult to account for across multiple instances of such experiments. Additional variability may also arise across experiments as a result of the introduction of animals, changes to agar substrates, and alteration in humidity or other environmental conditions. To recover the odor concentration that an animal experienced, there is a need to monitor odor concentration and animals’ behaviors concurrently.

Our previously reported flow cell used a photo-ionization detector (PID) sensor moved across the lid before behavioral experiments to measure the odor concentration across space at a single point in time (Gershow et al., 2012). During experiments, the total concentration of odor in the chamber was monitored concurrently with measurements of behavior. While this provided some assurances that the overall odor concentration was relatively stable, it did not provide any spatial information concurrently with behavior measurements.

Metal-oxide sensors are small and scalable digital gas sensors that have recently been used to monitor odor concentration in studies of olfaction (Schmuker et al., 2016; Drix and Schmuker, 2021; Tariq et al., 2021) especially in the context of turbulent odor plumes. Arrays of metal-oxide sensors have been used to measure spatial odor landscapes by sampling across space (Burgués et al., 2020). Here, we leverage these approaches to present a new multi-sensor odor array with metal-oxide sensors combined with a flow chamber in order to measure the odor landscape that an animal crawls in with high spatial and temporal resolution. The array of sensors can be used in two ways: the full array can be used to measure the generated gradient throughout the extent of the chamber, or parts of the array can be used on the borders to monitor, during behavioral experiments, the odor profile along the boundaries of the chamber. By varying flow rates and the sites of odor introduction, we show a variety of odor landscapes can be generated and stabilized.

To demonstrate the utility of the apparatus, we applied this instrument to quantitatively characterize the sensorimotor transformation underlying navigational strategies used by C. elegans and D. melanogaster larva to climb up a butanone odor gradient. Butanone is a water-soluble odorant found naturally in food sources (Worthy et al., 2018) that is often used in odor-guided navigation studies and is of particular scientific interest for studies of olfactory learning (Bargmann et al., 1993; Levy and Bargmann, 2020; Cho et al., 2016; Torayama et al., 2007). We show that the agar gel used during behavioral experiments greatly disrupts an applied butanone gradient, and we demonstrate a pre-equilibration protocol allowing generation of stable gradients taking into consideration the effects of agar. Moreover, we monitor these gradients during ongoing behavior measurements via continuous measurements of the odor profile along the boundaries of the arena.

Using these stable and continuously measured butanone gradients, we measure odor-guided navigation in animals by tracking their posture and locomotion as they navigate the odor landscape. We record chemotaxis behavior and identify navigation strategies in response to the changing odor concentration they experience. In C. elegans, we observe navigation to the odorant butanone employs similar strategies to those reported for other sensory-guided navigation conditions, such as salt chemotaxis (Iino and Yoshida, 2009; Dahlberg and Izquierdo, 2020; Luo et al., 2014), namely a biased random walk strategy that adjusts teh frequency of turns, known as pirouettes (Pierce-Shimomura et al., 1999), and a gradual veering strategy, known as weathervaning (Iino and Yoshida, 2009; Izquierdo et al., 2015). We also measure navigation strategies of a mutant C. elegans and implicate the interneuron AIB mutant as being important for olfactory-guided weathervaning. In D. melanogaster larvae, we identify runs followed by directed turns (Gershow et al., 2012; Louis et al., 2008; Gomez-Marin et al., 2011). By using concurrent measurements of behavior and odor gradient, we characterize olfactory navigation in these small animals on agar with known butanone odor concentrations, which for C. elegans has not been reported before.

We developed new methods for both generating and measuring odor gradients which we describe here. The systems are modular, scalable, and flexible. The components, which can be used independently of each other, can be fabricated directly from provided files using online machining services.

Flow chamber for generating spatiotemporal patterns of airborne odors

We first sought to develop a method of creating odor gradients that satisfied the following criteria:

1. The spatial odor profile should be controllable. Varying control parameters (e.g. flow rates, tubing connections) should result in predictable changes to the resulting odor landscape.

2. The odor profile should be stable and verifiable. The same spatial profile should be maintained over the course of an experiment lasting up to an hour, and this should be verifiable via concurrent measurements during behavior experiments.

3. The apparatus should be straightforward to construct and to use, and flexible to adapt to various experimental configurations, including using with either C. elegans or Drosophila larva, and with agar arenas of various sizes.

We constructed a flow chamber to control odor airflow across an arena (Figure 1a). Odor and humidified air are sourced from two bubblers, one containing pure water and the other an aqueous solution of odorant and water. Flow rates are controlled by separate mass flow controllers (MFCs) upstream of the bubblers. Downstream of the bubblers, the odor and air streams are divided into parallel sections of equal lengthed tubing. Each tube is connected to one input port of the flow chamber. The pattern of connections and the flow rates set by the two MFCs determines the shape of the produced odor profile. For instance, if odor is provided at a single central inlet, the resulting profile is a ‘cone’ (Figure 1d–g) whose peak concentration and divergence are controlled with the MFCs (e.g. speeding up the odor flow while slowing down the airflow broadens the cone). Temporal gradients can be achieved by varying the odor flow in time, subject to constraints imposed by the odor’s absorption into the agar gel.

Figure 1 with 6 supplements see all

Download asset Open asset

The outflux from the flow chamber is connected to a flow meter and a photo-ionization detector (PID) to monitor the overall flow rate and odor concentration respectively. The geometry of this flow chamber is shown in Figure 1b, where parallel tubings are connected from the side and the chamber is vacuum sealed with a piece of acrylic on top during experiments. The chamber is designed for use with ∼100 mm square agar plates. The extra width (2.5 cm on either side) diminishes the influence of the chamber boundary on the odor profile over the arena. Interchangeable inserts allow for different agar substrates (e.g. circular plates) or for full calibration by odor sensor arrays (OSAs) (Figure 1c), discussed in the next section. Metal components are designed for low-cost fabrication by automated mechanisms (either laser cut-able or three-axis CNC machinable). The fabrication plans for the flow chamber and inserts (Data sharing), list of required components (Table 1, Supplementary file 2), and a detailed tutorial (Supplementary file 1) are all publicly available.

Table 1

Item	Manufacturer	Part number	Price estimate ($/unit)
Mass flow controller	Aalborg	GFCS-010654	$900
Photo-ionization detector	Ametek	piD-TECH	$500
Metal-oxide sensors	Sensirion	SGP30	$10
PCB board	MacroFab and OSH Park	Customized	$400
Flow chamber	eMachineShop	Customized	$250
Flow meter	Aalborg	Model P flow tube meters	$160
Teensy	PJRC	Teensy 4.0	$25
Camera	Basler	acA4112-30um	$3000
Bubbler	Sigma-Aldrich	Duran GL 45 connection	$200
Tubing connection	idex-upchurch	Fitting, manifold, and tubings	∼$1000

Measuring the spatiotemporal odor distributions

A central difficulty in measuring animals’ responses to olfactory cues is quantifying airborne odor concentrations that vary in space and time (Vickers and Baker, 1994; Tariq et al., 2021; Kadakia et al., 2022; Yamazoe-Umemoto et al., 2018). This difficulty is exacerbated in turbulent environments where odor plumes carry abrupt spatial and temporal jumps in concentration far from the source with fundamentally unpredictable dynamics (Celani et al., 2014; Dennler et al., 2022). But even in laminar flows, like those considered here, boundary conditions, slight changes in temperature, and the presence of absorbing substrates like agar make this challenging.

There is therefore a need, even for quasi-stationary gradients, to characterize the odor profiles in situ and to monitor these profiles during experiments. Various optical techniques, like laser-induced fluorescence or optical absorption (Louis et al., 2008; Tadres et al., 2022; Demir et al., 2020), exist to monitor concentration across planar arenas, but these have not been demonstrated for behavior experiments, are expensive to construct, require specially designed arenas, or some combination of these disadvantages. Electronic chemical sensors can reveal the time-varying concentration at a particular point in space. A tiled array of these sensors acts as a ‘camera’ forming a two-dimensional (2D) spatiotemporal reading of the concentration. The gold standard for measurement of odor concentration is the PID, but even the smallest versions of these sensors are both too large (∼2 cm in all dimensions) and too expensive (∼$500 each) to make an array. Metal-oxide odor sensors, designed to be used in commercial air quality sensors, are available in inexpensive and compact integrated circuit packages (Schmuker et al., 2016; Burgués et al., 2020). However, in general, commercial metal-oxide sensors are not designed for precision work - they tend to drift due to variations in heater temperature, humidity, adsorption of chemicals, and aging effects (Dennler et al., 2022). Most such sensors are designed to detect the presence of gas above a particular concentration but not to precisely measure the absolute concentration. We became aware of a newer metal-oxide sensor, the Sensirion SGP30 that was designed for long-term stability and concentration measurement; we wondered if such a sensor could be calibrated for use in an OSA.

To calibrate the sensor, we created a controllable concentration source by bubbling air through butanone. The odor reservoir contains butanone dissolved in water and is kept below the saturation concentration (11 mM or 110 mM odor sources). We then mixed this odorized airflow into a carrier stream of pure air. We kept the carrier airflow rate constant ($\sim 400$ mL/min) and varied the flow rate through the odor source ($0-50$ mL/min); the odor flow rate was slow enough that the vapor remained saturated, so the concentration of butanone in the mixed stream was proportional to the flow rate through the butanone bubbler, as directly measured with a PID (Figure 1—figure supplement 2). We typically calibrated concentration with continuously ramped flow rate in triangle waves with a 500 s period for 2–3 cycles. We performed calibrations across a range of odor concentrations and against two PIDs, each with a different dynamic range (0–200 ppm, Figure 1—figure supplement 1, 0–1000 ppm, Figure 1—figure supplement 3), and also compared against the PIDs to characterize the metal-oxide sensors’ temporal response (Figure 1—figure supplement 2). We found a one-to-one correspondence between the odor sensor reading and the PID reading that persisted over time and showed no obvious hysteresis. These metal-oxide sensors’ temporal response and stability appear to be well matched for capturing the slow fluctuations in our stable and non-turbulent odor landscapes. We also characterized the heating effects of these sensors in the arena (Figure 1—figure supplement 4), and conducted measurements to show that powering the sensors does not noticeably change the concentration of odorant in the chamber (Figure 1—figure supplement 3). We reasoned that after calibrating the metal-oxide sensors to a PID, we could use the array to measure spatiotemporal odor concentration distributions.

We constructed a sensor array from ‘odor sensor bars’ (OSBs), printed circuit boards each containing 16 sensors in two staggered rows of 8. Each OSB also contained 8 temperature and humidity sensors to allow compensation of the odor sensor readings. The OSBs are mounted orthogonal to the direction of airflow; 7 OSBs fit inside our flow chamber (112 sensors total) allowing a full measurement of the odor profile. Taken together these 112 sensors formed an OSA, capable of measuring odor concentrations with ∼1 cm spatial resolution sampled at 1 Hz. Prior to all experiments we calibrated the OSA in situ by varying the butanone concentration across the entire anticipated range of measurement while simultaneously recording the odor sensor and PID readings.

To verify the ability of the OSA to measure concentration gradients, we created an artificially simple steady-state odor landscape and compared it to our predictions based on a convection-diffusion flow model. We flowed in odorized air ($\sim 30$ mL/min) through a central tube and clean air through the others ($\sim 400$ mL/min distributed into 14 surrounding tubings) in an environment without agar and without animals. This results in an airflow velocity $\sim 5$ mm/s in the flow chamber. The odor concentration profile was established in a few minutes and remained stable over the timescale of 30 min behavioral experiments (Figure 1—figure supplement 5). As the flow rates and concentrations are all known and the flow is non-turbulent, and there is no agar or animals present, the concentration across the chamber should match a convection-diffusion model. After establishing the gradient, we recorded from the discrete odor sensors on the array (Figure 1d) and estimated the values in between sensors using spline interpolation with length scale equal to the inter-sensor distances (Figure 1e). We compared this stationary profile with a two-parameter convection-diffusion flow model (Figure 1f and g) fit to the data and described in the Materials and methods section. The measured concentrations in this artificially simplistic odor gradient show good agreement with the fit convection-diffusion model, especially in the central region where experiments are to be conducted, leading us to conclude that the OSA can accurately report the odor concentration. Note that the model serves as an independent validation of the sensors, and is not required to perform the measurement. We proceed to consider more complex odor environments.

We demonstrate several examples of flexible control over the odor profile. In the configuration shown in Figure 1, the center-most input provides odorized air and all surrounding inputs provide moisturized clean air to form a cone-shape stationary odor pattern. A narrower cone can be created by increasing the airflow (600 mL/min) of the surrounding inputs relative to the middle odorized flow (Figure 2a). The cone can be inverted by placing odorized air in the two most distal inputs, and clean air in all middle inputs (Figure 2b). This inverse cone has lower concentration in the middle and higher on the sides. In later animal experiments, we restrict odorized air to one side to form a biased-cone odor landscapes, resulting in a cone with an offset from the middle line of the arena. Many more configurations are possible, demonstrating that the odor flow chamber enables the flexible control of airborne odor landscapes that are much more complex than a single odor point source.

Figure 2 with 1 supplement see all

Download asset Open asset

Figure 2—source data 1

Physical properties of odorants.

Table odor_physics.xlsx in https://doi.org/10.6084/m9.figshare.21737303.

https://cdn.elifesciences.org/articles/85910/elife-85910-fig2-data1-v2.xlsx

Download elife-85910-fig2-data1-v2.xlsx

To demonstrate that the flow control and measurement methods are not restricted to any single odor molecule, we create and measure various stationary odor landscapes with benzaldehyde, ethanol, and ethyl butyrate odorants that, together with butanone, span a range of different vapor pressure and water solubility (Figure 2, Figure 2—source data 1). For each molecule, the metal-oxide odor sensors are separately calibrated against a PID (Figure 2—figure supplement 1). To our knowledge, most volatile organic compounds that can be sensed and calibrated by a PID should be compatible with our sensor array system. Depending on the vapor pressure, molecular weight of different molecules, and the geometry of our flow chamber, the cone-shape odor profiles can be skewed or less smooth. The skewed profile is possibly a result of the inhomogeneity in the pocket created by the odor sensor hub. Nonetheless, we empirically verified that these odor profiles are stationary and reproducible.

Odor-agar interactions dominate classical butanone droplet assays

Classic chemotaxis experiments in small animals commonly construct odor environments with odor droplets in a Petri dish, usually with a substrate such as agar. Our odor delivery instrument is designed to be compatible with a similar environment. To first better understand classical chemotaxis experiments with butanone, we sought to characterize the spatiotemporal odor profile from an odor droplet point source using our OSA. We first considered the case without agar. In that case the odor concentration should be governed entirely by gas phase diffusion. We placed a 2 µL droplet of 10% butanone in water on the lid of our instrument centered in the arena above the full OSA and without any airflow (Figure 3a and b). Butanone was observed to diffuse across the arena in the first 3 min (Supplementary Video; Figure 3—video 1) resulting in an equilibrium concentration that was close to uniform across the odor sensors. We note that the final concentration of roughly 100 ppm and the equilibration timescale both match what we would expect from first principals for $\sim {10}^{-6}$ mol of butanone in an ${\displaystyle \sim 225\mathrm{m}\mathrm{L}}$ arena that have a diffusion rate of ${\displaystyle \sim 0.08{cm}^2/\mathrm{s}}$ in air. A uniform odor landscape is not helpful for studying odor-guided navigation, but most behavioral experiments are not conducted in a bare flow chamber but contain a biologically compatible substrate, such as an agar gel, as is typically used in droplet assays. We therefore sought to investigate the role that agar plays in sculpting the odor landscape in butanone droplet experiments.

Figure 3 with 2 supplements see all

Download asset Open asset

We introduced agar into the droplet assay by removing two sensor bars and replacing them with agar. We placed a butanone droplet directly on the agar or on the lid, as done classically in worm and larval assays respectively, and measured the odor landscape over time (Figure 3b and c). The odor concentration measured with agar is dramatically different from that measured without agar. Instead of quickly equilibrating to a uniform concentration, in the presence of agar there was instead a local maximum of butanone surrounding the droplet that persists even after 3 min. This difference in airborne odor concentration with and without agar persists after experimental perturbation such as removing and replacing the lid over the chamber (Figure 3—figure supplement 1). This convinced us of the need to account for odor-agar interactions when working with water-soluble odorants such as butane.

Agar poses challenges for establishing stable odor landscapes

By placing the agar on the underside of the lid hanging above the odor sensors, we were able to simultaneously record spatial odor concentrations generated by our flow chamber opposite the surface of the agar during animal behavior (Figure 1—figure supplement 6). Unfortunately, when imaging in this configuration, the odor sensors in the background obstruct the animals from being tracked over the extent of the arena. We therefore sought an alternative strategy: we chose to measure the odor profile along only the boundary of the arena in order to infer the odor concentration inside the arena in such a way as to also take into account the odor-agar interactions. This approach relies on establishing temporally stable odor landscapes, even in the presence of agar, which we sought to accomplish by leveraging our ability to control the flow of odor.

We first sought to measure how the presence of agar altered the odor profiles that would have been established by flow in a bare chamber lacking agar (Figure 1, Figure 2). As in Figure 3c and d, we replaced two OSBs with a rectangular strip of agar gel (‘w/ agar’) or a metal plate as a control (‘w/o agar’), and then measured airborne odor concentration upstream and downstream of the agar under odorized airflow that would normally produce a cone profile (Figure 4a–b). While the agar had little effect on the odor landscape upstream of the agar, it drastically altered the downstream odor landscape (Figure 4b), suggesting that the agar absorbs the airborne butanone molecules. This finding is consistent with the odor droplet experiments (Figure 3), repeated with a full-sized 96 mm square agar plate intended for use with animals (Figure 4c), and to be expected since butanone is highly soluble in water (275 g/L). Pulse-chase style experiments confirm that agar does indeed absorb and reemit butanone (Figure 4—figure supplement 1b). To accommodate the full-sized agar plate, we measured only the 1D odor profiles upstream and downstream of the agar (Figure 4d). Taken together, these experiments suggest that agar-butanone interaction presents a challenge for setting up and maintaining stable butanone odor landscapes.

Figure 4 with 1 supplement see all

Download asset Open asset

Monitoring the boundary determines the odor landscape at quasi-equilibrium

Pre-equilibration allows us to quickly establish a temporally stable odor landscape even in the presence of agar, and stability is indicated by monitoring the odor sensors along the boundary. Now we seek an approach to infer the odor concentration over the agar from the measurements along the boundary. In this section we will show that the odor concentration above the agar after pre-equilibration should be the same as the odor concentration when no agar is present and we will then validate this equivalence empirically (to within 10%). We will then use this knowledge to infer the odor concentration above the agar at quasi-equilibrium using the measurements along the boundary. If the airborne odor concentration along the boundaries upstream and downstream of the agar matches the profile in the absence of agar, one can infer that the airborne odor concentration landscape above the agar is also the same. We validated this approach in three ways: via analytical argument, via numeric simulation, and via direct empirical measurement.

The physical intuition is as follows: in the absence of sources or sinks, the fact that two odor concentration distributions obey the same differential equations and share the same conditions on all boundaries means that the distributions are identical throughout the interior. Agar adds complexity because it can act as a source or sink by absorbing and reemiting the odor, but even then, knowledge that the odor along the boundary is the same with and without agar places strong constraints on the interior odor landscape. The only way the interior could differ would be if sources and sinks of odor in the agar were precisely arranged so that all excess odor emitted from one point is exactly reabsorbed somewhere else before reaching the downstream boundary. Not only is such an arrangement unlikely, it is inherently temporally unstable and should tend to eliminate the excess sources and sinks over time. A mathematical version of this argument is presented in Appendix 1. Therefore, when we measure steady odor concentration along the boundary with agar that matches the concentration without, we are comfortable concluding that the odor landscape with agar is the same as without.

A numerical reaction-convection-diffusion model also shows that after pre-equilibration, the with-agar odor landscape matches the without-agar odor landscape (Figure 5). The model requires reasonable assumptions about the absorption rate, reemission rate, and capacity of the agar. The numerical simulation provides qualitative agreement to our empirical observations in Figure 4.

Figure 5

Download asset Open asset

We experimentally verified that the airborne odor concentration above the agar matched our expectations based on measurements made along the boundary, by performing direct measurements of the airborne odor opposite the agar. We used an upside-down configuration as in Figure 1—figure supplement 6 and Figure 6: we poured agar onto the lid and measured the odor concentration below it with the full sensor array underneath (Figure 6, Figure 1—figure supplement 6).

Figure 6 with 1 supplement see all

Download asset Open asset

After the pre-equilibrium protocol, measured airborne concentration in the arena with agar was similar to measurements made without agar to within an average of less than 10% fractional difference in the area that the animal would crawl (Figure 6b). The agreement of odor landscape measured with and without agar at quasi-equilibrium agrees with our modeling and indicates that the odor landscape is stable and identifiable in our setup by monitoring the odor profile along the boundary. Therefore, an experimenter could establish a desired odor gradient without agar, and then, even in the presence of agar, once the boundary measurements matched, the experimenter would be confident that the odor landscapes also matched. Note that our hypothetical experimenter would not be required to perform any simulations or calculations.

The presence of worms did not noticeably alter the odor concentration (Figure 6—figure supplement 1). Our approach relies on reaching quasi-steady state. In measurements made without the pre-equilibration protocol, and therefore presumably before quasi-steady state, the interior odor concentration with agar no longer matches the without agar, as expected (Figure 6c).

Butanone chemotaxis in C. elegans

We sought to use our experimental approach to directly quantify C. elegans’ navigation strategies for airborne butanone. C. elegans are known to climb gradients toward butanone (Cho et al., 2016; Levy and Bargmann, 2020). Microfluidic environments suggest that they use a biased random walk strategy to navigate in a liquid butanone environment (Levy and Bargmann, 2020; Albrecht and Bargmann, 2011). Worms are also known to use weathervaning to navigate airborne gradients of other odors (Iino and Yoshida, 2009; Kunitomo et al., 2013) although to our knowledge this has not been specifically investigated for butanone.

Worms were imaged crawling on agar in the flow chamber under an airborne butanone odor landscape illuminated by infrared light. Here, six recording assays were presented, with approximately 50-100 animals per assay, and two different odor landscapes were used. C. elegans navigated up the odor gradient toward higher concentrations of butanone, as expected (Figure 7a and b). Importantly, the odor concentration experienced by the animal at every point in time was inferred from concurrent measurements of the odor profile along the boundary of the agar (Figure 7c). On average, animals were more likely to travel in a direction up the local gradient than away from the local gradient, as expected for chemo-attraction (Figure 7d). We use the term ‘bearing to local gradient’ to describe the animal’s direction of travel with respect to the local odor gradient that it experiences.

Figure 7 with 2 supplements see all

Download asset Open asset

We find quantitative evidence that the worm exhibits both biased random walk and weathervaning strategies. To investigate biased random walks, we measured the animal’s probability of turning (pirouette) depending on its bearing with respect to the local airborne butanone gradient (Figure 7e). We find that the animal is least likely to turn when it navigates up the local gradient and most likely to turn when it navigates down the gradient, a key signature of the biased random walk strategy (Berg, 2018; Mattingly et al., 2021).

To test for weathervaning, we measured how the curvature of the animal’s trajectory depended on its bearing with respect to the local airborne butanone gradient (Figure 7f). When the animal’s navigation aligns to the axis of the butanone gradient, the distribution of the curvature of its trajectory was roughly symmetric and centered around 0 (straight line trajectory). By contrast, when the animal navigated more perpendicular to the gradient, the distribution of the curvature of its trajectories was skewed. The skew was such that it enriched for cases where the animal curved its trajectories toward the local gradient. We further calculate the average curvature as a function of bearing to local gradients following methods in Iino and Yoshida, 2009; Figure 7—figure supplement 2a. The result shows that the worm on average turns toward the gradient direction, which is the key signature of weathervaning strategy. Both the biased random walk and weathervaning behavior was absent in control experiments with flow but not odor, and we observed no evidence of anemotaxis at the $\sim 5$ mm/s air velocities encountered by the animals (Figure 7—figure supplement 1, Figure 7—figure supplement 2c). We conclude that C. elegans utilize both biased random walk and weathervaning strategies to navigate butanone airborne odor landscapes (Iino and Yoshida, 2009; Yoshida et al., 2012). Both strategies had been observed for salt chemotaxis before, but to our knowledge had not previously been quantitatively characterized for airborne butanone. More broadly, previous measurements of navigation in response to airborne odors for C. elegans have not had access to the underlying odor concentration and therefore could only use proxies, such as the bearing angle or distance with respect to a point odor source. Our approach allows a more direct description of the navigation strategy because we characterize turn probabilities and other features with respect to the local gradient directly (Figure 7), as opposed to a proxy.

To identify potential neurons involved in mediating the navigation strategy, we measured butanone-guided navigation of mutant animals in which the interneuron AIB was chronically inhibited. Disruptions to AIB had previously been shown to either increase weathervaning (AIB ablation) (Iino and Yoshida, 2009) or decrease biased random walks (Luo et al., 2014) (AIB inhibition) when navigating in a salt gradient. But to our knowledge, AIB’s role in influencing the animal’s butanone-guided navigation strategy had not previously been measured. We observed that AIB-inhibited animals still climbed butanone gradients, but the worm relied on different behavioral strategies (Figure 8). The turn probability was still biased by the concentration gradient suggesting that the animal still used a biased random walk strategy, although turning rate was overall lower. Most dramatically, the animal showed down-regulated weathervaning. In wild-type (N2) animals the distribution of the curvature of the animal’s trajectory depends on its orientation to the local odor gradient (Figure 7f, Figure 7—figure supplement 2a) indicating weathervaning, but in AIB-inhibited animals the distribution of trajectory curvature is much less dependent on orientation to the local gradient (Figure 8d, Figure 7—figure supplement 2b). Together this suggests that AIB is important for supporting turning rate and the weathervaning strategy, and demonstrates the potential for this approach to characterize mutants or investigate neural circuits underlying navigation.

Figure 8

Download asset Open asset

Having access to quantitative information about the odor the animal experiences provides an opportunity to characterize the animal’s overall gradient climbing performance through observations of its local movements on a local gradient (Mattingly et al., 2021). We compute the animal’s drift velocity as a function of local gradient steepness (Figure 9). This captures the animal’s overall gradient climbing performance as a result of all the navigational strategies it uses, including the biased random walk and weathervaning. The analysis provides information relevant to a chemotaxis index, but it has the additional benefit of describing quantitatively how the animal responds to a local gradient and on what timescale. This analysis is only possible with a knowledge of the odor concentration experienced by the animal, and is naturally accessible in our instrument.

Figure 9

Download asset Open asset

Butanone chemotaxis in Drosophila larvae

To further evaluate the utility of the flow chamber for the study of small animal navigation, we investigated how larval Drosophila navigate butanone. Although butanone is not as commonly used as a stimulus with Drosophila as with C. elegans, butanone is known to be attractive to larval flies (Dubin et al., 1995; Dubin et al., 1998) and has been variously reported to be attractive (Park et al., 2002; Jung et al., 2015) and aversive (Israel et al., 2022; Lerner et al., 2020) to adult flies. To investigate the larva’s navigational strategy in a butanone gradient, we created a ‘cone’-shaped butanone gradient over the agar substrate using the pre-equilibration protocol, as before, and we confirmed the presence and stability of the gradient by continuously measuring the spatial distribution of butanone upstream and downstream of the agar arena. We monitored the orientation and movement of 59 larvae over six separate 10 min experiments (∼ 10 larvae per experiment) with an average observation time of 7 min per larva (Figure 10).

Figure 10

Download asset Open asset

Larvae moved toward higher concentration of butanone (Figure 10a). To analyze the strategy by which they achieved this, we first constructed a coordinate system in which 0° was in the direction of the odor gradient (toward higher concentration) and 180° was directly down-gradient; angles increased counterclockwise when viewed from above. We found that larvae initiated turns at a higher rate when headed down-gradient ($\pm {180}^{\circ }$ bearing with respect to the local gradient) than up-gradient ($0^{\circ }$) (Figure 10b). When larvae turned, their reorientations tended to orient up-gradient (negative angle changes from $+{90}^{\circ }$ bearing with respect to the local gradient, and positive angle changes from $-{90}^{\circ }$) (Figure 10c). Bias in turning rate and reorientations in larvae are known navigation strategies in larvae (Gershow et al., 2012; Gomez-Marin et al., 2011; Gomez-Marin and Louis, 2014), but here it is to our knowledge the first characterization for butanone. With modulation of turning rate and orientation in the same odor environment, we conclude that Drosophila larvae use similar navigational strategies to C. elegans to move toward butanone. We note that the chemotaxis efficiency is much higher in larvae compared to worms given the order of magnitude difference in run velocity (∼3 mm/s in larvae and ∼0.1 mm/s in worms).

We present a custom-designed flow chamber and OSA that enables us to measure navigation strategies of worms and fly larvae within the context of a controlled and measured odor environment. The key features of this odor delivery system are that (1) the odor concentration profile through space is controlled in the flow chamber, (2) the OSA provides a spatial readout to calibrate and measure the profile, and (3) the odor concentration profile is monitored during animal experiments. This last feature, the ability to monitor the spatial profile of odor concentration on the boundary during experiments, to our knowledge sets this method apart from previous approaches.

The odor landscape experienced by the animal can be recovered by one of two ways, with different tradeoffs. Typically, for a given set of flow parameters, we first quantify the odor gradient using the whole sensor array when agar is absent, then during behavioral experiments, we confirm the desired odor profile is achieved using sensor bars mounted along the boundary upstream and downstream of agar. During experiments with agar, the real-time monitoring of the odor profile at the edges of the arena guards against a range of possible errors or experimental imperfections. For example, exhaustion of the odor source, blocked connections, imperfect seals, and so on will all be detected by a change in the odor profile along the boundary, thereby avoiding silent changes to the odor landscape that may skew behavioral results. Indeed, it was this real-time monitoring that first alerted us to the fact that absorption of odor into the agar substrate caused the odor landscape to differ from our naive expectations.

If real-time monitoring of the entire arena during behavioral experiments is desired, the odor array can instead be mounted opposite the agar surface (Figure 6). We show that this introduces complications in imaging and makes tracking animals challenging (Figure 1—figure supplement 6). Future work could address these challenges through alternate lighting schemes (e.g. frustrated TIR in Risse et al., 2013) or by masking the sensor array with a darker material. For most applications, however, we expect that confirming the profile via sensors along the boundary will be sufficient because we have demonstrated that the odor landscape established by flow at quasi-equilibrium is stable and repeatable in the absence of experimenter or instrument errors (and detectable when not).

The ability to monitor spatial profile during experiments, along with a quantitative understanding of odor-agar interactions, provides confident knowledge of the odor experienced by the animal over time. This is especially important when we analyze time-varying navigation behavior in an odor environment that potentially changes in time too (Yamazoe-Umemoto et al., 2018; Yoshida et al., 2012; Levy and Bargmann, 2020). Past work focusing on neural circuit mechanisms for odor navigation in worms have often relied on coarser readout such as the chemotaxis index or using the distance to odor source as a proxy for the inverse of odor concentration. However, it has been shown that the exact odor concentration alters the animal’s behavioral response to odor (Yoshida et al., 2012; Levy and Bargmann, 2020). Quantifying tuning curves with respect to more precise odor measurements may better constrain the investigations into neural mechanisms driving the sensorimotor transformations underlying navigation algorithms (Clark et al., 2013; Liu et al., 2020).

In contrast to liquid delivery of odor gradients via microfluidic chips (Albrecht and Bargmann, 2011; Larsch et al., 2015), our method allows worms to crawl freely on an agar surface. This allows our behavior measurements to be directly compared against classical chemotaxis assays (Louis et al., 2008; Pierce-Shimomura et al., 1999). Additionally, the macroscopic odor airflow chamber makes it straightforward to flexibly adjust the spatial pattern between experiments without the need to redesign the chamber. A detailed tutorial for assembling the instrument and the protocol for experiments is provided in Supplementary file 1.

Our setup uses low flow rates corresponding to low wind speed velocities (5 mm/s) to avoid anemotaxis. Larger organisms, including adult flies, navigate toward odor sources by combining odor and wind flow measurements (Vergassola et al., 2007; Matheson et al., 2022). Fly larvae exhibit negative anemotaxis at wind speeds 200–1000 times higher than those used here (Jovanic et al., 2019), but previous work showed that they do not exhibit anemotaxis at lower wind speeds like the ones used here (Gershow et al., 2012). For example, they do not exhibit anemotaxis at 12 mm/s which is still higher than the velocities they experience here (Gershow et al., 2012). Therefore, we do not expect Drosophila larvae to exhibit anemotaxis under our flow conditions. C. elegans are not thought to respond to airflow. In experiments in aqueous microfluidic chips under flow, C. elegans move toward higher concentrations of attractant and do not respond to the flow of the liquid (Albrecht and Bargmann, 2011). In agreement, we do not observe any evidence of C. elegans anemotaxis in our chamber in response to control experiments without odor and with wind speeds of 5 mm/s (Figure 7—figure supplement 1).

We focused on the odor butanone because it is important for a prominent associative learning assay (Torayama et al., 2007; Kauffman et al., 2011). Butanone has high vapor pressure and is highly soluble in water and therefore it interacts strongly with agar. In this work we show that this odor-agar interaction makes it difficult to a priori infer an odor landscape experienced by the animal when agar is present, but that continuously monitoring the odor profile on the boundary overcomes this challenge.

Odorants with less solubility and lower vapor pressure may not require pre-equilibration to reach quasi-equilibrium quickly, but instead may be able to stabilize across agar on a shorter timescale. For example, ethyl butyrate is less water soluble (∼0.586 g/100 mL in water, compared to 27.5 g/100 mL for butanone), has lower pressure (11.3 mmHg compared to 78 mmHg for butanone), and has a smaller gas phase diffusion coefficient (5.6 × ${10}^{-3}$ cm²/s compared to ∼69 cm²/s for butanone). A droplet of ethyl butyrate has been shown to form a relatively stable odor landscape over a 5 min timescale in a dish on agar without flow (Tadres et al., 2022; Louis et al., 2008). Here, we demonstrate stable odor landscapes over longer timescales of at least 30 min with a more volatile odor, butanone, by using our flow-based assay (Figure 1—figure supplement 5). A strength of the flow-based approach with monitoring is that it should, in principle, allow for the use of a wider range of odors. Crucially, the continuous monitoring makes this approach more robust against other unforeseen odor interactions, such as interactions between an odorant and glass, aluminum or plastic.

Here, we have addressed the problem of creating non-turbulent airborne odor landscapes. The biophysical processes governing odor sensing in small animals such as C. elegans are not fully understood. The worm carries a thin layer of moisture around its body as it moves on the agar substrate (Bargmann, 2006) and it is unclear to what extent the worm pays attention to the concentration of an odorant in the agar below it vs the air above it. Our reaction-convection-diffusion model suggests that at the quasi-equilibrium conditions used in our experiments the odor concentration in agar is related to the airborne odor concentration directly above it up to a scalar that we predict to be constant across the agar. Although we have not measured this empirically, this suggests that even in the extreme case that the animal only senses odor molecules in the agar, the odor concentration experienced by the animal in our experiments should differ by no more than a scaling factor compared to our estimates based on the airborne odor concentration.

Here, we also showed the possibility of and challenges associated with directly measuring the full spatial odor profile (instead of using the boundary) during simultaneous behavior tracking (Figure 1—figure supplement 6). Future work is needed to improve concurrent animal tracking methods, to design more miniaturized and scalable sensors, and to generalize to more complex or 3D environments.

Knowing the concentration experienced by the animal is not only useful for measuring navigational strategies more precisely than in classical assays, like the droplet chemotaxis assays. It will also be crucial for studying changes in navigational strategy, such as those in the context of associative learning (Cho et al., 2016; Torayama et al., 2007), sensory adaptation (Levy and Bargmann, 2020; Itskovits et al., 2018), and long timescale behavioral states (Calhoun et al., 2014; Gomez-Marin et al., 2011; Klein et al., 2017). In all those cases, it will be critical to disambiguate slight changes to the odor landscape from gradual changes in the navigational strategies. Continuously monitoring the odor landscape during behavior will remove this ambiguity.

Recordings for odor flow control, concentration measurements, and behavioral tracking data are publicly available: https://doi.org/10.6084/m9.figshare.21737303. Compressed 'zip' archive Chen_flow_20222.zip contains raw data. Compressed 'zip' archive flow_chamber_machined_components.zip contains CAD and other design files related to the hardware.

1. 1. Chen K
   2. Gershow M
   3. Leifer A

   (2023) figshare

   Recordings for odor flow control, concentration measurements, and behavioral tracking data.

   https://doi.org/10.6084/m9.figshare.21737303

1. 1. Aceves-Piña EO
   2. Quinn WG

   (1979) Learning in normal and mutant Drosophila larvae

   Science 206:93–96.

   https://doi.org/10.1126/science.206.4414.93

   • PubMed
   • Google Scholar
2. 1. Albrecht DR
   2. Bargmann CI

   (2011) High-content behavioral analysis of Caenorhabditis elegans in precise spatiotemporal chemical environments

   Nature Methods 8:599–605.

   https://doi.org/10.1038/nmeth.1630

   • PubMed
   • Google Scholar
3. 1. Baker KL
   2. Dickinson M
   3. Findley TM
   4. Gire DH
   5. Louis M
   6. Suver MP
   7. Verhagen JV
   8. Nagel KI
   9. Smear MC

   (2018) Algorithms for olfactory search across species

   The Journal of Neuroscience 38:9383–9389.

   https://doi.org/10.1523/JNEUROSCI.1668-18.2018

   • PubMed
   • Google Scholar
4. 1. Bargmann CI
   2. Horvitz HR

   (1991) Chemosensory neurons with overlapping functions direct chemotaxis to multiple chemicals in C. elegans

   Neuron 7:729–742.

   https://doi.org/10.1016/0896-6273(91)90276-6

   • PubMed
   • Google Scholar
5. 1. Bargmann CI
   2. Hartwieg E
   3. Horvitz HR

   (1993) Odorant-selective genes and neurons mediate olfaction in C. elegans

   Cell 74:515–527.

   https://doi.org/10.1016/0092-8674(93)80053-h

   • PubMed
   • Google Scholar
6. 1. Bargmann CI

   (2006) Chemosensation in C. elegans

   WormBook 1:1–29.

   https://doi.org/10.1895/wormbook.1.123.1

   • PubMed
   • Google Scholar
7. 1. Berg HC
   2. Brown DA

   (1972) Chemotaxis in Escherichia coli analysed by Three-dimensional Tracking

   Nature 239:500–504.

   https://doi.org/10.1038/239500a0

   • Google Scholar
8. Book

   1. Berg HC

   (2018) Random Walks in Biology

   Princeton University Press.

   https://doi.org/10.2307/j.ctv7r40w6

   • Google Scholar
9. 1. Boie SD
   2. Connor EG
   3. McHugh M
   4. Nagel KI
   5. Ermentrout GB
   6. Crimaldi JP
   7. Victor JD

   (2018) Information-theoretic analysis of realistic odor plumes: What cues are useful for determining location?

   PLOS Computational Biology 14:e1006275.

   https://doi.org/10.1371/journal.pcbi.1006275

   • PubMed
   • Google Scholar
10. 1. Burgués J
    2. Hernández V
    3. Lilienthal AJ
    4. Marco S

    (2020) Gas distribution mapping and source localization using a 3D grid of metal oxide semiconductor sensors

    Sensors and Actuators B 304:127309.

    https://doi.org/10.1016/j.snb.2019.127309

    • Google Scholar
11. 1. Calhoun AJ
    2. Chalasani SH
    3. Sharpee TO

    (2014) Maximally informative foraging by Caenorhabditis elegans

    eLife 3:e04220.

    https://doi.org/10.7554/eLife.04220

    • PubMed
    • Google Scholar
12. 1. Calhoun AJ
    2. Murthy M

    (2017) Quantifying behavior to solve sensorimotor transformations: advances from worms and flies

    Current Opinion in Neurobiology 46:90–98.

    https://doi.org/10.1016/j.conb.2017.08.006

    • PubMed
    • Google Scholar
13. 1. Celani A
    2. Villermaux E
    3. Vergassola M

    (2014) Odor landscapes in turbulent environments

    Physical Review X 4:041015.

    https://doi.org/10.1103/PhysRevX.4.041015

    • Google Scholar
14. 1. Cho CE
    2. Brueggemann C
    3. L’Etoile ND
    4. Bargmann CI

    (2016) Parallel encoding of sensory history and behavioral preference during Caenorhabditis elegans olfactory learning

    eLife 5:e14000.

    https://doi.org/10.7554/eLife.14000

    • PubMed
    • Google Scholar
15. 1. Chronis N
    2. Zimmer M
    3. Bargmann CI

    (2007) Microfluidics for in vivo imaging of neuronal and behavioral activity in Caenorhabditis elegans

    Nature Methods 4:727–731.

    https://doi.org/10.1038/nmeth1075

    • PubMed
    • Google Scholar
16. 1. Clark DA
    2. Freifeld L
    3. Clandinin TR

    (2013) Mapping and cracking sensorimotor circuits in genetic model organisms

    Neuron 78:583–595.

    https://doi.org/10.1016/j.neuron.2013.05.006

    • PubMed
    • Google Scholar
17. 1. Coen P
    2. Clemens J
    3. Weinstein AJ
    4. Pacheco DA
    5. Deng Y
    6. Murthy M

    (2014) Dynamic sensory cues shape song structure in Drosophila

    Nature 507:233–237.

    https://doi.org/10.1038/nature13131

    • PubMed
    • Google Scholar
18. Conference

    1. Dahlberg BA
    2. Izquierdo EJ

    (2020)

    Contributions from parallel strategies for spatial orientation

    The 2020 Conference on Artificial Life.

    • Google Scholar
19. 1. Demir M
    2. Kadakia N
    3. Anderson HD
    4. Clark DA
    5. Emonet T

    (2020) Walking Drosophila navigate complex plumes using stochastic decisions biased by the timing of odor encounters

    eLife 9:e57524.

    https://doi.org/10.7554/eLife.57524

    • PubMed
    • Google Scholar
20. 1. Dennler N
    2. Rastogi S
    3. Fonollosa J
    4. van Schaik A
    5. Schmuker M

    (2022) Drift in a popular metal oxide sensor dataset reveals limitations for gas classification benchmarks

    Sensors and Actuators B 361:131668.

    https://doi.org/10.1016/j.snb.2022.131668

    • Google Scholar
21. 1. Drix D
    2. Schmuker M

    (2021) Resolving fast gas transients with metal oxide sensors

    ACS Sensors 6:688–692.

    https://doi.org/10.1021/acssensors.0c02006

    • PubMed
    • Google Scholar
22. 1. Dubin AE
    2. Heald NL
    3. Cleveland B
    4. Carlson JR
    5. Harris GL

    (1995) Scutoid mutation of Drosophila melanogaster specifically decreases olfactory responses to short-chain acetate esters and ketones

    Journal of Neurobiology 28:214–233.

    https://doi.org/10.1002/neu.480280208

    • PubMed
    • Google Scholar
23. 1. Dubin A.E
    2. Liles MM
    3. Seligman F
    4. Le T
    5. Tolli J
    6. Harris GL

    (1998) Involvement of genes encoding a K+ channel (ether a go-go) and a Na+ channel (smellblind) in Drosophila olfaction

    Annals of the New York Academy of Sciences 855:212–222.

    https://doi.org/10.1111/j.1749-6632.1998.tb10569.x

    • PubMed
    • Google Scholar
24. 1. Evans TC

    (2006) Transformation and microinjection

    WormBook 10:450.

    https://doi.org/10.1895/wormbook.1.108.1

    • Google Scholar
25. 1. Gepner R
    2. Mihovilovic Skanata M
    3. Bernat NM
    4. Kaplow M
    5. Gershow M

    (2015) Computations underlying Drosophila photo-taxis, odor-taxis, and multi-sensory integration

    eLife 4:e06229.

    https://doi.org/10.7554/eLife.06229

    • PubMed
    • Google Scholar
26. 1. Gershow M
    2. Berck M
    3. Mathew D
    4. Luo L
    5. Kane EA
    6. Carlson JR
    7. Samuel ADT

    (2012) Controlling airborne cues to study small animal navigation

    Nature Methods 9:290–296.

    https://doi.org/10.1038/nmeth.1853

    • Google Scholar
27. Software

    1. GershowLab

    (2023a) OdorSensorArray, version swh:1:rev:d290dc2f878919c4a13e3e4528ff50c6de34563d

    Github.

    https://archive.softwareheritage.org/swh:1:dir:ff0935296df626ff9698f3852829751ea5bd1b4c;origin=https://github.com/GershowLab/OdorSensorArray;visit=swh:1:snp:b0195c9db5fbb265a93d2c2f86a62ca6d22fed38;anchor=swh:1:rev:d290dc2f878919c4a13e3e4528ff50c6de34563d
28. Software

    1. GershowLab

    (2023b) Chen-Wu-eLife-Drosophila-image-capture, version swh:1:rev:9e3a0e7b341243b0c32c4f352bae681aa30966b4

    Software Heritage.

    https://archive.softwareheritage.org/swh:1:dir:bb4b2b8d00ab4ed0ce5a3457b4b22a633a6c6f75;origin=https://github.com/GershowLab/Chen-Wu-eLife-Drosophila-Image-Capture;visit=swh:1:snp:c38fc090078907f9256a01b5d4a0e90136e1614b;anchor=swh:1:rev:9e3a0e7b341243b0c32c4f352bae681aa30966b4
29. 1. Gomez-Marin A
    2. Stephens GJ
    3. Louis M

    (2011) Active sampling and decision making in Drosophila chemotaxis

    Nature Communications 2:441.

    https://doi.org/10.1038/ncomms1455

    • PubMed
    • Google Scholar
30. 1. Gomez-Marin A
    2. Louis M

    (2014) Multilevel control of run orientation in Drosophila larval chemotaxis

    Frontiers in Behavioral Neuroscience 8:38.

    https://doi.org/10.3389/fnbeh.2014.00038

    • PubMed
    • Google Scholar
31. 1. Gorur-Shandilya S
    2. Martelli C
    3. Demir M
    4. Emonet T

    (2019) Controlling and measuring dynamic odorant stimuli in the laboratory

    The Journal of Experimental Biology 222:jeb207787.

    https://doi.org/10.1242/jeb.207787

    • PubMed
    • Google Scholar
32. 1. Iino Y
    2. Yoshida K

    (2009) Parallel use of two behavioral mechanisms for chemotaxis in Caenorhabditis elegans

    The Journal of Neuroscience 29:5370–5380.

    https://doi.org/10.1523/JNEUROSCI.3633-08.2009

    • PubMed
    • Google Scholar
33. 1. Israel S
    2. Rozenfeld E
    3. Weber D
    4. Huetteroth W
    5. Parnas M

    (2022) Olfactory stimuli and moonwalker SEZ neurons can drive backward locomotion in Drosophila

    Current Biology 32:1131–1149.

    https://doi.org/10.1016/j.cub.2022.01.035

    • PubMed
    • Google Scholar
34. 1. Itskovits E
    2. Ruach R
    3. Kazakov A
    4. Zaslaver A

    (2018) Concerted pulsatile and graded neural dynamics enables efficient chemotaxis in C. elegans

    Nature Communications 9:2866.

    https://doi.org/10.1038/s41467-018-05151-2

    • PubMed
    • Google Scholar
35. 1. Izquierdo EJ
    2. Williams PL
    3. Beer RD

    (2015) Information Flow through a Model of the C. elegans Klinotaxis Circuit

    PLOS ONE 10:e0140397.

    https://doi.org/10.1371/journal.pone.0140397

    • PubMed
    • Google Scholar
36. 1. Jovanic T
    2. Winding M
    3. Cardona A
    4. Truman JW
    5. Gershow M
    6. Zlatic M

    (2019) Neural Substrates of Drosophila Larval Anemotaxis

    Current Biology 29:554–566.

    https://doi.org/10.1016/j.cub.2019.01.009

    • PubMed
    • Google Scholar
37. 1. Jung SH
    2. Hueston C
    3. Bhandawat V

    (2015) Odor-identity dependent motor programs underlie behavioral responses to odors

    eLife 4:e11092.

    https://doi.org/10.7554/eLife.11092

    • PubMed
    • Google Scholar
38. 1. Kadakia N
    2. Demir M
    3. Michaelis BT
    4. DeAngelis BD
    5. Reidenbach MA
    6. Clark DA
    7. Emonet T

    (2022) Odour motion sensing enhances navigation of complex plumes

    Nature 611:754–761.

    https://doi.org/10.1038/s41586-022-05423-4

    • PubMed
    • Google Scholar
39. 1. Kauffman A
    2. Parsons L
    3. Stein G
    4. Wills A
    5. Kaletsky R
    6. Murphy C

    (2011) C. elegans positive butanone learning, short-term, and long-term associative memory assays

    Journal of Visualized Experiments 11:2490.

    https://doi.org/10.3791/2490

    • PubMed
    • Google Scholar
40. Software

    1. Kevin-SC

    (2023) Leifer-behavior-triggered-averaging-Tracker-new, version swh:1:rev:f7e33d743b02d085e3f2413d0e3d024ae8fb5626

    Github.

    https://archive.softwareheritage.org/swh:1:dir:53ccf21365e6957660aeac39e8115f969356fa15;origin=https://github.com/Kevin-Sean-Chen/leifer-Behavior-Triggered-Averaging-Tracker-new;visit=swh:1:snp:169a87d71d227b3e12683f35261a0677c198c191;anchor=swh:1:rev:f7e33d743b02d085e3f2413d0e3d024ae8fb5626
41. 1. Klein M
    2. Krivov SV
    3. Ferrer AJ
    4. Luo L
    5. Samuel AD
    6. Karplus M

    (2017) Exploratory search during directed navigation in C. elegans and Drosophila larva

    eLife 6:e30503.

    https://doi.org/10.7554/eLife.30503

    • Google Scholar
42. 1. Kunitomo H
    2. Sato H
    3. Iwata R
    4. Satoh Y
    5. Ohno H
    6. Yamada K
    7. Iino Y

    (2013) Concentration memory-dependent synaptic plasticity of a taste circuit regulates salt concentration chemotaxis in Caenorhabditis elegans

    Nature Communications 4:1–11.

    https://doi.org/10.1038/ncomms3210

    • PubMed
    • Google Scholar
43. 1. Larsch J
    2. Flavell SW
    3. Liu Q
    4. Gordus A
    5. Albrecht DR
    6. Bargmann CI

    (2015) A Circuit for Gradient Climbing in C. elegans Chemotaxis

    Cell Reports 12:1748–1760.

    https://doi.org/10.1016/j.celrep.2015.08.032

    • PubMed
    • Google Scholar
44. 1. Lerner H
    2. Rozenfeld E
    3. Rozenman B
    4. Huetteroth W
    5. Parnas M

    (2020) Differential Role for a Defined Lateral Horn Neuron Subset in Naïve Odor Valence in Drosophila

    Scientific Reports 10:6147.

    https://doi.org/10.1038/s41598-020-63169-3

    • PubMed
    • Google Scholar
45. 1. Levy S
    2. Bargmann CI

    (2020) An adaptive-threshold mechanism for odor sensation and animal navigation

    Neuron 105:534–548.

    https://doi.org/10.1016/j.neuron.2019.10.034

    • PubMed
    • Google Scholar
46. 1. Liu M
    2. Sharma AK
    3. Shaevitz JW
    4. Leifer AM

    (2018) Temporal processing and context dependency in Caenorhabditis elegans response to mechanosensation

    eLife 7:e36419.

    https://doi.org/10.7554/eLife.36419

    • PubMed
    • Google Scholar
47. 1. Liu A
    2. Papale AE
    3. Hengenius J
    4. Patel K
    5. Ermentrout B
    6. Urban NN

    (2020) Mouse navigation strategies for odor source localization

    Frontiers in Neuroscience 14:218.

    https://doi.org/10.3389/fnins.2020.00218

    • PubMed
    • Google Scholar
48. 1. Lockery SR
    2. Lawton KJ
    3. Doll JC
    4. Faumont S
    5. Coulthard SM
    6. Thiele TR
    7. Chronis N
    8. McCormick KE
    9. Goodman MB
    10. Pruitt BL

    (2008) Artificial dirt: Microfluidic substrates for nematode neurobiology and behavior

    Journal of Neurophysiology 99:3136–3143.

    https://doi.org/10.1152/jn.91327.2007

    • Google Scholar
49. 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

    • PubMed
    • Google Scholar
50. 1. Luo L
    2. Wen Q
    3. Ren J
    4. Hendricks M
    5. Gershow M
    6. Qin Y
    7. Greenwood J
    8. Soucy ER
    9. Klein M
    10. Smith-Parker HK
    11. Calvo AC
    12. Colón-Ramos DA
    13. Samuel ADT
    14. Zhang Y

    (2014) Dynamic encoding of perception, memory, and movement in a C. elegans chemotaxis circuit

    Neuron 82:1115–1128.

    https://doi.org/10.1016/j.neuron.2014.05.010

    • Google Scholar
51. 1. Matheson AMM
    2. Lanz AJ
    3. Medina AM
    4. Licata AM
    5. Currier TA
    6. Syed MH
    7. Nagel KI

    (2022) A neural circuit for wind-guided olfactory navigation

    Nature Communications 13:4613.

    https://doi.org/10.1038/s41467-022-32247-7

    • PubMed
    • Google Scholar
52. 1. Mattingly HH
    2. Kamino K
    3. Machta BB
    4. Emonet T

    (2021) Escherichia coli chemotaxis is information limited

    Nature Physics 17:1426–1431.

    https://doi.org/10.1038/s41567-021-01380-3

    • PubMed
    • Google Scholar
53. 1. Monte P
    2. Woodard C
    3. Ayer R
    4. Lilly M
    5. Sun H
    6. Carlson J

    (1989) Characterization of the larval olfactory response in Drosophila and its genetic basis

    Behavior Genetics 19:267–283.

    https://doi.org/10.1007/BF01065910

    • PubMed
    • Google Scholar
54. 1. Park SK
    2. Shanbhag SR
    3. Dubin AE
    4. de Bruyne M
    5. Wang Q
    6. Yu P
    7. Shimoni N
    8. D’Mello S
    9. Carlson JR
    10. Harris GL
    11. Steinbrecht RA
    12. Pikielny CW

    (2002) Inactivation of olfactory sensilla of a single morphological type differentially affects the response of Drosophila to odors

    Journal of Neurobiology 51:248–260.

    https://doi.org/10.1002/neu.10057

    • PubMed
    • Google Scholar
55. 1. Pierce-Shimomura JT
    2. Morse TM
    3. Lockery SR

    (1999) The fundamental role of pirouettes in Caenorhabditis elegans chemotaxis

    The Journal of Neuroscience 19:9557–9569.

    https://doi.org/10.1523/JNEUROSCI.19-21-09557.1999

    • PubMed
    • Google Scholar
56. 1. Radvansky BA
    2. Dombeck DA

    (2018) An olfactory virtual reality system for mice

    Nature Communications 9:839.

    https://doi.org/10.1038/s41467-018-03262-4

    • PubMed
    • Google Scholar
57. 1. Risse B
    2. Thomas S
    3. Otto N
    4. Löpmeier T
    5. Valkov D
    6. Jiang X
    7. Klämbt C

    (2013) FIM, a novel FTIR-based imaging method for high throughput locomotion analysis

    PLOS ONE 8:e53963.

    https://doi.org/10.1371/journal.pone.0053963

    • PubMed
    • Google Scholar
58. 1. Schmuker M
    2. Bahr V
    3. Huerta R

    (2016) Exploiting plume structure to decode gas source distance using metal-oxide gas sensors

    Sensors and Actuators B 235:636–646.

    https://doi.org/10.1016/j.snb.2016.05.098

    • Google Scholar
59. 1. Tadres D
    2. Wong PH
    3. To T
    4. Moehlis J
    5. Louis M

    (2022) Depolarization block in olfactory sensory neurons expands the dimensionality of odor encoding

    Science Advances 8:eade7209.

    https://doi.org/10.1126/sciadv.ade7209

    • PubMed
    • Google Scholar
60. 1. Tanimoto Y
    2. Yamazoe-Umemoto A
    3. Fujita K
    4. Kawazoe Y
    5. Miyanishi Y
    6. Yamazaki SJ
    7. Fei X
    8. Busch KE
    9. Gengyo-Ando K
    10. Nakai J
    11. Iino Y
    12. Iwasaki Y
    13. Hashimoto K
    14. Kimura KD

    (2017) Calcium dynamics regulating the timing of decision-making in C. elegans

    eLife 6:e21629.

    https://doi.org/10.7554/eLife.21629

    • PubMed
    • Google Scholar
61. 1. Tariq MF
    2. Lewis SM
    3. Lowell A
    4. Moore S
    5. Miles JT
    6. Perkel DJ
    7. Gire DH

    (2021) Using head-mounted ethanol sensors to monitor olfactory information and determine behavioral changes associated with ethanol-plume contact during mouse odor-guided navigation

    eNeuro 8:ENEURO.0285-20.2020.

    https://doi.org/10.1523/ENEURO.0285-20.2020

    • PubMed
    • Google Scholar
62. 1. Tastekin I
    2. Khandelwal A
    3. Tadres D
    4. Fessner ND
    5. Truman JW
    6. Zlatic M
    7. Cardona A
    8. Louis M

    (2018) Sensorimotor pathway controlling stopping behavior during chemotaxis in the Drosophila melanogaster larva

    eLife 7:e38740.

    https://doi.org/10.7554/eLife.38740

    • PubMed
    • Google Scholar
63. 1. Torayama I
    2. Ishihara T
    3. Katsura I

    (2007) Caenorhabditis elegans integrates the signals of butanone and food to enhance chemotaxis to butanone

    The Journal of Neuroscience 27:741–750.

    https://doi.org/10.1523/JNEUROSCI.4312-06.2007

    • PubMed
    • Google Scholar
64. 1. Vergassola M
    2. Villermaux E
    3. Shraiman BI

    (2007) “Infotaxis” as a strategy for searching without gradients

    Nature 445:406–409.

    https://doi.org/10.1038/nature05464

    • PubMed
    • Google Scholar
65. 1. Vickers NJ
    2. Baker TC

    (1994) Reiterative responses to single strands of odor promote sustained upwind flight and odor source location by moths

    PNAS 91:5756–5760.

    https://doi.org/10.1073/pnas.91.13.5756

    • PubMed
    • Google Scholar
66. 1. Worthy SE
    2. Haynes L
    3. Chambers M
    4. Bethune D
    5. Kan E
    6. Chung K
    7. Ota R
    8. Taylor CJ
    9. Glater EE

    (2018) Identification of attractive odorants released by preferred bacterial food found in the natural habitats of C. elegans

    PLOS ONE 13:e0201158.

    https://doi.org/10.1371/journal.pone.0201158

    • PubMed
    • Google Scholar
67. 1. Yamada M
    2. Ohashi H
    3. Hosoda K
    4. Kurabayashi D
    5. Shigaki S

    (2021) Multisensory-motor integration in olfactory navigation of silkmoth, Bombyx mori, using virtual reality system

    eLife 10:e72001.

    https://doi.org/10.7554/eLife.72001

    • PubMed
    • Google Scholar
68. 1. Yamazoe-Umemoto A
    2. Fujita K
    3. Iino Y
    4. Iwasaki Y
    5. Kimura KD

    (2015) Modulation of different behavioral components by neuropeptide and dopamine signalings in non-associative odor learning of Caenorhabditis elegans

    Neuroscience Research 99:22–33.

    https://doi.org/10.1016/j.neures.2015.05.009

    • PubMed
    • Google Scholar
69. 1. Yamazoe-Umemoto A
    2. Iwasaki Y
    3. Kimura KD

    (2018) Measuring spatiotemporal dynamics of odor gradient for small animals by gas chromatography

    Bio-Protocol 8:e2797.

    https://doi.org/10.21769/BioProtoc.2797

    • PubMed
    • Google Scholar
70. 1. Yoshida K
    2. Hirotsu T
    3. Tagawa T
    4. Oda S
    5. Wakabayashi T
    6. Iino Y
    7. Ishihara T

    (2012) Odour concentration-dependent olfactory preference change in C. elegans

    Nature Communications 3:739.

    https://doi.org/10.1038/ncomms1750

    • PubMed
    • Google Scholar

Decision letter after peer review:

Thank you for submitting your article "Continuous odor profile monitoring to study olfactory navigation in small animals" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by Piali Sengupta as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

As described in their reports, the reviewers appreciate the merits of the sophisticated methodology and assays introduced in your manuscript to create controlled odorant landscapes. The reviewers raised a series of concerns related to the methodology and its use. We ask that you address these concerns in a revised manuscript.

Methodological concerns:

– The authors should experimentally characterize the sensitivity range (log concentrations over which detection is reliable), temporal resolution, and stability (potential existence of drifts) of their sensor array in response to more rapid changes in odor concentration. They should discuss their results in the context of published studies where metal oxide sensors have previously been employed to measure odor dynamics in turbulent airborne environments.

– It would be important to rule out the heat effects of the metal oxide sensor arrays on the behavior of C. elegans and Drosophila larvae. The authors should discuss the existence of such heat effects. They should also discuss whether physical interactions with the sensors themselves could affect the odor stimulus.

– The authors must generalize their conclusions to other odors that are commonly used to study olfactory behavior in Drosophila and C. elegans, and that have lower water solubility. More generally, they should discuss the type of odors that are compatible with their methodology.

– The authors should include a more thoughtful comparison of the pros and cons of their methodology with respect to the spectroscopic approach presented in Louis et al. 2008. In Tadres et al. 2022, numerical simulations of odor gradients were constrained by spectroscopy data. These simulations showed that gradients produced by a single source of ethyl butyrate are relatively stable, which is consistent with the behavior of larvae. While the reviewers don't dispute the fact that the more complex assay introduced by the authors might enhance the stability of odor gradients, it is probably misleading to conclude that all gradients produced by a single odor source are unstable.

– As supporting material, it would be important to list cost estimates to build the assay and provide general information related to the sensitivity, odor selectivity, and temporal resolution of the methodology. Given that the methodology is considered a Tools and Resources, it would also be extremely valuable to provide more information (possibly a tutorial) to help other labs adopt the methodology to characterize odor landscapes in new assays.

Additional request:

The methodology and experiments presented in the manuscript offer the opportunity to compare odor-evoked behaviors in two species in response to the same environment. To illustrate the potential of the methodology, the authors are strongly encouraged to analyze the behavior of both species using the same procedure and metrics to report similarities and differences in behavioral algorithms. This addition should help the reader define whether the methodology enables to draw new conclusions or if it recapitulates previous observations. If the authors decide that they won't proceed with a deeper behavioral analysis, they should tone down their claims about algorithmic findings in their abstract and the rest of the manuscript.

Reviewer #1 (Recommendations for the authors):

Measuring airborne odorant gradient is notoriously challenging. The goals of this manuscript are twofold: (1) to develop a technique to characterize the spatial properties and temporal stability of airborne odorant landscapes; (2) to produce stationary odorant landscapes with controlled and predictable geometries that are suitable to refine the study of navigation behavior (chemotaxis) in small animals such as C. elegans and the Drosophila larva. Tackling these two problems should improve future correlations between sensation and behavior to study the neural bases of sensory encoding and action selection.

To achieve their first goal, the authors propose to use a distributed array of digital gas sensors. The resolution of the measurements reported in the manuscript is impressive: they go beyond the state of the art. Although these measurements remain discrete in space, they provide valuable data to fit a biophysical model of the diffusion of the odor and its reaction with the arena where the odorant landscape is created. While the technique allows monitoring odorant landscapes in space and time, it also has limitations.

First, the measurements require modifying the experimental conditions where the behavior is tested. The landscape cannot be directly measured while the behavior is taking place. Second, the dimensions of the digital gas sensors (~1 cm) are relatively large compared to the size of the tested animals (~1 mm). This limits the resolution of the spatial measurements of odorant landscapes with small dimensions where even smaller animals are studied. In particular, the system might not be able to detect differences in concentrations between the individual inlets of the odor delivery system, which reduces the accuracy of the reconstruction of a landscape and the evaluation of its variability.

The second goal of the study is to produce odorant landscapes with enhanced stability compared to existing systems. The airflow-based olfactometer features spatially patterned odorized air. While this system represents an improvement with respect to published assays in C. elegans, its contribution to the larva is less obvious. The geometry of the landscapes resulting from the linear array of odorized air flows is essentially uni-dimensional. This produces geometries that are fundamentally different from the radially symmetric gradients emitted by odor cups that have been used by other labs for many years. The authors conclude that gradients resulting from the diffusion of a single source are unstable due to the absorption of the odor by the agar. This conclusion was drawn for odorant molecules that are water-soluble (or miscible in water like ethanol). Whether the same conclusion holds for odorant molecules with low or negligible water solubility is unknown, which precludes a generalization of the results to odors that have been more commonly used in the field.

The manuscript would benefit from additional work to demonstrate how the tool can advance the understanding of chemotaxis in C. elegans and the Drosophila larva. Besides the fact that the authors chose to characterize larval chemotaxis with an odor that has not been used in previous work, the behavioral characterization is limited given the potential capabilities of the new tool. Presently, it falls short of reporting "chemotaxis strategies for C. elegans and D. melanogaster larvae populations under different spatial odor landscapes" as stated in the abstract. This incompleteness is a missed opportunity to validate the merit of the methodology.

I praise the authors for the technical achievement presented in the manuscript. Combining a spatial array of digital gas sensors with computational simulations of the odor diffusion process is ingenious. However, several aspects of the methodology require further characterization and more careful validation to be helpful as a tool that can be widely adopted by other labs. Below is a list of shortcomings that should be addressed:

1. Although the multi-odorous-air-flow delivery system is simplified compared to the original design published by Gershow et al. in 2012, it remains complex. It would be important to test the reproducibility of landscapes produced by this system. In addition, it would be useful to demonstrate the flexibility with which landscapes with different geometries can be created. One notable limitation of the system is that gradients can only be designed in one direction. This precludes the creation of radially symmetric gradients produced by a point source, such as those that have been typically used in C. elegans (e.g. Pierce-Shimomura & Lockery 1999) or in the Drosophila larva (learning assay developed by Gerber and coworkers). Given the tiled arrangement of the airflow inlets, is the gradient smooth at the scale of a worm or "bumps" expected in the odor profile over the y-axis? This question is likely to be unresolved by the odor measurements since there is a higher density of tubes compared to sensors along the y-axis.

2. The manuscript reports that the high absorption of butanone by the agar significantly affects the properties of the resulting odorant landscape. As a result, odorant landscapes produced by a single odor droplet can flatten over time in such a way that the gradient might be nearly inexistent after a couple of minutes. This conclusion is significant since larval olfaction has been frequently studied with a single odor source (a cup or a droplet). As indicated by the authors in lines 270-271, butanone is a water-soluble odorant molecule. By comparison, most odors that have been used to study olfactory behaviors in the Drosophila larva have a low solubility in water. For instance, 1-octanol (<1g/L), benzaldehyde (<10g/L), pentyl acetate (<2g/L) and geranyl acetate (<1 g/L) are all far less soluble than butanone (275 g/L). This difference in water solubility is likely to affect the stability of the odorant gradient. Were odor measurements not combined with computational modeling of the odor diffusion-reaction process in Tadres et al. 2022 (Science Advances) to establish that gradients of ethyl butyrate and ethyl acetate are relatively stable for 5 min?

3. Technical question related to Figure 3: in this figure, the butanone droplet appears to be placed on the lid without agar (line 24) whereas it is placed directly on the agar gel in the condition with agar (line 251). The placement of the butanone droplet on the lid versus on the agar is an important distinction besides just the presence or absence of agar. This point is not addressed in the discussion of Figure 3. The authors should use the same type of odor source to make comparisons across conditions.

4. The behavioral quantification of the chemotactic behavior of C. elegans reproduces a series of results that have already been published. Weathervaning appears to be reproduced, but its strength might be lower than that reported by IIno and Yoshida 2009. It would be helpful if the authors could present a graph with the average curvature rate as a function of the bearing of the gradient to permit a comparison with Iino and Yoshida 2009.

In contrast with the behavioral quantification of C. elegans, the behavioral quantification of larvae is minimal. Essentially, the authors report that larvae turn less while moving upgradient and turn more while moving downgradient. This quantification does not justify the highly-quantitative assay introduced in the manuscript. In addition, could the authors show that larvae are capable of "weathervaning"? Another interesting application of the assay would be to establish how the navigational performances of larvae change with the steepness (or other geometric features) of stationary odorant gradients. This would also show that larvae are capable of precise chemotaxis in gradients that are essentially 1D compared to 2D gradients resulting from point sources.

5. The introduction is slightly misleading. One motivation of the work is to design a setup where measurements of odor concentrations characterizing an odorant landscape can be done during behavioral experiments. One would expect that these measurements would be done in 2D across the assay. Instead, the measurements are made on the border of the arena, which partly contradicts the original goal of the methodology. I appreciate that the presence of the sensors in the arena is not compatible with behavioral experiments, but the implications of this limitation should be more thoroughly discussed in the manuscript. It also reinforces the need to establish the level of fluctuations in odor concentrations inside the arena in real-life conditions.

More generally, how do the authors envision the use of their new methodology by regular labs working on olfaction? If, on the one hand, their primary goal is to propose a technique that will be adopted by many experimental labs, one should acknowledge that most labs might find it challenging to conduct odor measurements, optimize the convection-reaction-diffusion model and simulate new odorant landscapes. Going through the pipeline would require a "tutorial". If, on the other hand, the goal of the methodology is to draw attention on potential artifacts associated with the instability of odorant gradients in published assays, the authors should conduct a more thorough analysis of the stability of representative experimental conditions.

6. Appendix 1 argues that the convection-diffusion model and the reaction-convection-diffusion model are equivalent if the odor flow between the air and agar are in equilibrium. How is this equilibrium defined quantitatively?

Reviewer #2 (Recommendations for the authors):

This manuscript by Chen et al. describes an apparatus for measuring odor-evoked navigation behavior in C. elegans and Drosophila larvae. The major advance is using an array of metal oxide sensors to measure odor gradients. The authors apply this tool to generate stable gradients in an agar environment. They then measure odor-evoked behavior in both worms and fly larvae, demonstrating the ability to recover previously described stimulus-behavior associations such as biased random walking and weathervaning.

The problem of controlling and measuring odor dynamics is a challenge for all studies of odor-guided navigation and new approaches to these problems are welcome. While the present study shows the potential of the metal oxide sensor array approach, several considerations are missing that would help other researchers to evaluate whether this approach would be useful, and are important for evaluating claims made with this device.

1) Metal oxide sensors have previously been employed to measure odor dynamics in turbulent airborne environments (e.g. Schmuker et al. 2016, Tariq et al. 2021, Dinnler et al. 2022). Two major issues raised by these previous studies are that (1) metal oxide sensors have a fast onset but very slow offset which complicates inference of odor dynamics, and (2) sensor drift can complicate absolute concentration measurements. The authors should discuss these previous studies and show data on the temporal resolution and stability of their sensor array in response to more rapid changes in odor concentration.

2) Previous studies have used a spectroscopic approach to quantify odor gradients in agar (Louis et al. 2008). The authors state that this approach is not compatible with simultaneous behavioral measurements, while the metal oxide sensors are, however, it is not totally clear to me why this should be. While a direct comparison of these two methods would be ideal, a more thoughtful comparison of the pros and cons of the two methods would be most helpful to other researchers.

3) A major issue for using the metal oxide sensor arrays during behavior would seem to be the heat generated by the arrays. This should be discussed and any heat effects on behavior should be described, as thermotaxis behavior has been described in both species studied here.

4) It is not clear that this device allows for "precise" measurements at the location of the animal as claimed in the abstract. First, the measurements are in the air and are predicted to be related to the concentration in agar through a scalar. Second, measurements are made at the edges of the agar plate and internal concentrations are inferred. Although these appear to be stable in the absence of animals, local fluctuations due to animal movement cannot be measured.

Suggestions:

The focus of this study is on animals that move in a viscose substrate such as worms and larvae, where odors form stable gradients. While this is implied by the term "small model organisms" the authors should distinguish this from animals other than small model animals (adult flies, larval fish) that move in turbulent environments where the temporal resolution of the system would need to be much higher.

The ability to measure odor-evoked behavior in two species in response to the same environment seems like a bit of a missed opportunity here. Can the behavior of both species be analyzed using the same methods and similarities and differences in behavioral algorithms described?

Overall I think this is an interesting approach but I think there are a large number of specific claims that need to be softened or toned down:

line 18: "Crucially and unlike previous methods, our method allows continuous monitoring of the odor profile during behavior" I don't think this is true. For example, the Vickers and Baker 1994 study used an extra moth antenna to measure odor plume fluctuations in flying moths, and the Tariq 2021 study measured odor at the location of a navigating mouse using metal oxide sensors.

line 23-24: "accurately inferred" "precise odor concentration" not sure these are true for the reasons listed in 4 above.

line 35-36: "small mode organisms" I guess this means worms and larvae but I think it would be helpful to specifically say animals that move in a substrate or mostly navigate in gradient (as opposed to turbulent) environments.

line 46: "no technique currently exists for precise control and continuous monitoring of an odor landscape." This is not true either. Many published approaches here include optogenetics to create virtual environments, precise generation of odor waveforms, controlled flow chambers, etc.

line 97: "odor profile in the chamber" I think the abstract implies that you can make precise measurements at the location of the animal but it is clearer later on that these are inferred measurements across the arena.

line 160: "quantifying airborne odor concentrations" There is a large literature on this in the turbulent navigation field that is not cited or discussed in this paper.

line 175: "metal oxide sensors" can you give estimates of cost, sensitivity, odor selectivity, and temporal resolution? I would expect this to be in a supplement.

line 204: 1-second temporal resolution. What is the evidence for this? Is this a sampling rate or does it take into account the dynamics of the sensor as shown in the Tariq paper?

line 251: "remove two sensor bars and replace them with agar" I think it needs to be clearer upfront that the sensors have to be removed to do agar measurements.

line 371: butanone chemotaxis in Drosophila larvae. should cite the Jung and Bhandawat 2015 paper here that closely examines butanone-evoked navigation in adult flies and compares it to vinegar-evoked navigation.

line 399: "This last feature…sets this method apart from previous approaches" It is not clear to me why the spectroscopic approach could not in principle be used during behavior.

line 405: "In the future, such tuning curves may form the basis of investigations into neural mechanisms driving the sensorimotor transformations underlying navigation" Lots of this has been done! Seems weird to say this as a future thing and not cite the many many circuit cracking papers in worms and flies that have been pursued with other apparatus.

lines 438-440: "at the quasi-equilibrium conditions used in our experiments the odor concentration in agar is related to the airborne odor concentration directly above it up to a scalar that we predict to be constant across the agar." But is this actually true? This is quite far from the claim in the abstract that you can precisely measure the concentration at the location of the animal.

Reviewer #3 (Recommendations for the authors):

In this paper, Chen et al. propose a new method to measure odor stimuli in space and time. Measuring the odor stimulus is a key step in interpreting odor-driven behavior and understanding the neural mechanisms that mediate it, but this task still challenges every experimenter in the field. As described in the introduction, there is basically not a single method that is good for most behavioral assays, even when these involve small animals such as C. elegans and D. melanogaster. Previous approaches are either invasive, very expensive, or limited to very small behavioral arenas. The strength of the proposed method is to be cheap and to have a reasonable spatial (1 cm) and temporal (1 s) resolution. The full sensor array introduced here cannot be located throughout the behavioral arena, but a mathematical model shows that it is sufficient to measure the odor stimulus at certain specific positions to reconstruct the full spatial profile. In other assays, such mathematical considerations might not be possible and physical constraints might make it impossible to use the sensor during behavior, however, it can still be used to measure the odor landscape in a separate experiment, and, with a certain degree of reproducibility (which will depend on the specific delivery system), this is still better than no measure. It should be noted that the method is limited to 2D measurements, which is sufficient for the walking or crawling behavior of small animals in non-turbulent conditions, but it cannot be extended to 3D assays. It remains moreover unclear what is the sensitivity range of the sensor for the odors used in the paper (butanone and ethanol) and what is expected for other odorants (which compounds are detectable?).

As proof of principle, the authors use this new method to characterize the behavior of C. elegans and D. melanogaster as a function of the concentration gradient encountered along their moving trajectories. In this respect, it remains unclear whether the method allows new conclusions or simply recapitulates previous observations.

Sensors:

It would be important to know more about the sensitivity range of the sensors. It seems that the concentration used here does not saturate the sensors: what is the full scale of sensitivity for the x-axis in Figure 1- S1? Is the calibration curve similar for ethanol? And what kinds of odors are expected to be detectable?

Another point that is not mentioned is whether the sensor itself affects the odor stimulus: for example, through an absorption/release mechanism similar to what happens with the agar: I guess one should compare Figure 2a to Figure 4b? No interference with the stimulus would be a clear advantage of this approach over the PID and should be stated.

Behavioral analysis:

Regarding the paragraphs on C. elegans and D. melanogaster, I would suggest that the authors clarify what is a new finding vs what is already known, and in which cases the measurement of the odor gradient is critical.

Clearly, Figure 7 requires such measurements; however, the significance of the result is somehow obscure: what do we expect for the relationship between drift velocity and gradient? Also, there seems to be a very small and possibly not significant (I do not find a statistical test) positive drift for the high range of the tested gradients: could these gradients be too shallow?

Is the same analysis of drift vs gradient not possible with D. melanogaster? Moreover, I wonder why the turn rate for -90 and 90 degrees do not have similar values: both directions are perpendicular to the gradient, wouldn't one expect the same behavior? The heading change, in that concern, is as expected, with similar absolute value and opposite direction. Also unclear why the heading change is similar for 180 and 0 degrees.

Essential revisions:

As described in their reports, the reviewers appreciate the merits of the sophisticated methodology and assays introduced in your manuscript to create controlled odorant landscapes. The reviewers raised a series of concerns related to the methodology and its use. We ask that you address these concerns in a revised manuscript.

We thank the reviewers and editors for this thoughtful, detailed and constructive review.

Methodological concerns:

– The authors should experimentally characterize the sensitivity range (log concentrations over which detection is reliable), temporal resolution, and stability (potential existence of drifts) of their sensor array in response to more rapid changes in odor concentration. They should discuss their results in the context of published studies where metal oxide sensors have previously been employed to measure odor dynamics in turbulent airborne environments.

We performed new experiments (Figure 1 – Figure Supp 2; Figure 1 – Figure Supp 3) to characterize sensitivity range, temporal resolution and stability in response to changing odor concentrations. This is in addition to the previous experiments presented in (Figure 1 – Figure Supp 1).

Sensitivity range: Previously we had made paired recordings of the metal oxide sensor (OS) and the photoionization detector (PID) over the range of odors used in the manuscript, ~0-200 ppm (Figure 1 – Figure supp 1). Now we have added new experiments with a higher range PID (rated to 2,000 ppm) to evaluate OS responses to odors up to 1,000 ppm (Figure 1 – Figure supp 3), roughly an order of magnitude beyond the range of odors we use with animals. The OS shows non-saturating one-to-one mapping to the PID across this range.

Temporal resolution: To evaluate temporal resolution of our OS we conducted measurements of the OS in response to steps of odor concentration (Figure 1 – Figure Supp 2). We observe that both the OS and PID take tens of seconds to respond to steps in odor concentration induced by our changes to odor flow. Since our odor landscapes have nominally constant odor flow over the 30 minute timescale, we are confident that this is sufficient to capture the slow odor concentration changes that we might expect to observe.

Stability The relationship between OS and PID is stable over the course (>1 h) of our new odor step response measurements (Figure 1 – Figure supp 2), and consistent with our previous triangle wave measurements (Figure 1 – Figure Supp 1).

Thank you for pointing us to additional literature about the use of metal oxide sensors for the study of olfaction. We have now added additional references throughout:

(from Introduction):

“Metal-oxide sensors are small and scalable digital gas sensors that have recently been used to monitor odor concentration in studies of olfaction (Schmuker et al. 2016, Drix et al. 2021, Tariq et al. 2021) especially in the context of turbulent odor plumes.

Arrays of metal-oxide sensors have been used to measure spatial odor landscapes by sampling across space (Burgues et al. 2020).

Here we leverage these approaches…”

(from results):

“Metal-oxide odor sensors, designed to be used in commercial air quality sensors, are available in inexpensive and compact integrated circuit packages (Schmuker et al. 2016, Burgues et al. 2020). However, in general, commercial metal-oxide sensors are not designed for precision work – they tend to drift due to variations in heater temperature, humidity, adsorption of chemicals and ageing effects (Dennler et al. 2022). Most such sensors are designed to detect the presence of gas above a particular concentration but not to precisely measure the absolute concentration. We became aware of a newer metal-oxide sensor, the Sensirion SGP30 that was designed for long-term stability and concentration measurement; we wondered if such a sensor could be calibrated for use in an odor sensor array.”

– It would be important to rule out the heat effects of the metal oxide sensor arrays on the behavior of C. elegans and Drosophila larvae. The authors should discuss the existence of such heat effects. They should also discuss whether physical interactions with the sensors themselves could affect the odor stimulus.

We have now performed new analysis (Figure 1 – Figure supp 4) of temperature measurements to characterize heating effects of the metal oxide sensors. The existing spatial array of thermo-hydro sensors on the sensor bars already recorded the local temperature in the arena during measurements. We now present those measurements in Figure 1 – Figure Supp 4. Our measurements in an arena without agar when all 112 odor sensors are producing heat show that the temperature stabilizes after ~20 mins, a similar time frame to our pre-equilibration protocol. The spatial temperature fluctuations are small (~0.5 C) across the region where an animal would be present. In separate experiments with agar present we used a thermocouple to directly measure temperature on the agar surface and find the temperature differences to be similarly small:

“We experimentally validate heat management when the full sensor array is active for odor measurements. The temperature increases in the first 5 minutes of recording and later stabilizes (Figure 1 – Figure supp 4). After stabilizing, the temperature readout in space shows modest spatial fluctuations on the scale of 0.5 C. In separate experiments with agar, we measured similar spatial fluctuations on the agar surface.”

As we described in the caption of Figure 1 – Figure supp 4:

“We expect the heating shown here to represent an upper bound for agar experiments with animals, because here all 112 odor sensors were active and generating heat, while with agar experiments only 32 odor sensors would be active. Consistent with this, during separate experiments with agar, the temperature measured via a thermocouple at the four corners of the agar surface showed a difference in temperature within 0.4 C at four corners that are 10 cm apart.”

In the methods we describe thermal management strategies we have used to minimize the effects of heating:

“To avoid generating thermal gradients, the system has been designed to dissipate heat to the optics table. Each metal-oxide odor sensor contains a micro hotplate which consumes 86 mW power during readings. To dissipate this heat the aluminum insert inside the flow chamber serves as a heat sink. Odor sensor bars are connected to the insert using heat conductive tape and thermal paste. The insert and chamber are in turn in direct thermal contact with the optics table. Temperature and humidity is constantly monitored at 8 locations per OSB via the on-board temperature and humidity sensors during experiments to confirm that there is minimal thermal or moisture gradient created in the environment.”

Importantly, we observed little or no bias in the animal’s trajectories in control experiments that lacked odor (mock odor landscape) but still had animals, air flow and powered odor sensors, suggesting that the animals are not undergoing thermotaxis (Figure 7 – Figure Supp 1).

To test whether the odor sensor interacts with and/or depletes the odor, we conducted new experiments where we turned off the odor sensors part way through the experiment, but still recorded downstream overall odor concentration with the PID (Figure 1-Figure supp 3b). Our measurements indicate no discernible change in downstream odor concentration between when the metal-oxide odor sensors are on or off, suggesting that the metal-oxide sensors do not deplete the odor. Reviewer 3 points out that this is an expected advantage of metal-oxide sensors over PID sensors because PID sensors are expected to deplete the odor molecules that they measure.

– The authors must generalize their conclusions to other odors that are commonly used to study olfactory behavior in Drosophila and C. elegans, and that have lower water solubility. More generally, they should discuss the type of odors that are compatible with their methodology.

Thank you for the suggestion. In addition to ethanol and butanone, we now include new measurements with benzaldehyde and ethyl butyrate, two additional odor molecules that are common in larvae and worm chemotaxis studies (Figure 2 c and e). These molecules have different vapor pressure and water solubility compared to butanone (Author response table 1, Figure 2 – Figure Supp 2), but can all be measured and controlled with our apparatus and calibration protocol (new measurements in Figure 2 – Figure Supp 1).

According to these additional measurements and the specifications provided by the manufacturer, the sensors can generally sense most volatile organic compounds of interest.

– The authors should include a more thoughtful comparison of the pros and cons of their methodology with respect to the spectroscopic approach presented in Louis et al. 2008. In Tadres et al. 2022, numerical simulations of odor gradients were constrained by spectroscopy data. These simulations showed that gradients produced by a single source of ethyl butyrate are relatively stable, which is consistent with the behavior of larvae. While the reviewers don't dispute the fact that the more complex assay introduced by the authors might enhance the stability of odor gradients, it is probably misleading to conclude that all gradients produced by a single odor source are unstable.

We have updated the text to emphasize that measurements with our sensor array do not cast doubt on the stable odor landscapes reported in Louis et al. 2008 and Tadres et al. 2022, but that our approach can also create stable landscapes with molecules that have stronger odor-agar interactions (such as butanone) or other oderants that may also interact with their environment.

“Odorants with less solubility and lower vapor pressure may not require pre-equilibration to reach quasi-equilibrium quickly, but instead may be able to stabilize across agar on a shorter time scale.

For example, ethyl butyrate is less water soluble (~0.586 g/100 ml in water, compared to 27.5 g/100 ml for butanone), has lower vapor pressure (11.3 mmHg compared to 78 mmHg for butanone) and has a smaller gas phase diffusion coefficient (5.6 x 10^-3 compared to ~69 cm^2/s for butanone). A droplet of ethyl butyrate has been shown to form a relatively stable odor landscape over a 5 minute timescale in a dish on agar without flow (Tadres et al. 2022, Louis et al. 2008). Here we demonstrate stable odor landscapes over longer timescales of at least 30 mins with a more volatile odor, butanone, by using our flow-based assay (Figure 1 – Figure Supp 5). A strength of the flow based approach with monitoring is that it should, in principle, allow for the use of a wider range of odors. Crucially, the continuous monitoring makes this approach more robust against other unforeseen odor interactions, such as interactions between an odorant and glass, aluminum or plastic.“

To facilitate direct comparison with Figure 1 in Louis et al., 2008 and Figure S10 in Tadres et al. 2022, we have added a similar analysis for odor landscape stability in Figure 1 – Figure Supp 5. We also added more explicit details about the Louis 2008 approach and how it inspires and motivates our own work.

“In one of the most comprehensive measurements to date, Louis and colleagues (Louis et al. 2008, Tadres et al. 2022) used infrared spectroscopy to measure the spatial profile of a droplet based odor gradient. They used this approach to study bilateral olfactory sensing in Drosophila larvae, clearly demonstrating the value of measuring odor landscapes. That work measured odor concentration at a relatively low temporal resolution (~1 sample / min), only when animals weren't present, and relied on a relatively expensive Fourier-transform infrared spectroscopy machine.”

And we draw contrasts with it and all previous works here:

“In all of these previous works, measurements were performed offline, not during animal behavior, and the odor concentration was assumed to be the same across repeats of the same experiment, and when animals are present.”

We now also discuss more explicitly a caveat with our approach: namely that it is suboptimal for simultaneous direct measurements of the full spatial gradient and animal behavior. We conducted new experiments where we demonstrate that the instrument can directly measure the odor landscape (including on the agar) simultaneously with animal behavior but that tracking then becomes a challenge (Figure 1 – Figure Supp 6). We describe this in more detail in response to Reviewer #1 and mention it in the Results section:

“Ideally, one would like to directly measure odor opposite the surface of the agar, such that odor-agar-interactions are fully accounted for. By placing the agar on the lid hanging above the odor sensors, we were able to simultaneously record spatial odor concentrations generated by our flow chamber opposite the surface of the agar during animal behavior (Figure 1 – Figure Supp 6). Unfortunately, when imaging in this configuration, the odor sensors in the background obstruct the animals from being tracked over the extent of the arena. We therefore sought an alternative strategy: we chose to measure the odor profile along only the boundary of the arena in order to infer the odor concentration inside the arena in such a way as to also take into account the odor-agar-interactions…”

And in the discussion:

“Here we also showed the possibility of and challenges associated with directly measuring the full spatial odor profile (instead of using the boundary) during simultaneous behavior tracking (Figure 1 – Figure Supp 6). Future work is needed to improve concurrent animal tracking methods or to design more miniaturized and scalable sensors.”

– As supporting material, it would be important to list cost estimates to build the assay and provide general information related to the sensitivity, odor selectivity, and temporal resolution of the methodology. Given that the methodology is considered a Tools and Resources, it would also be extremely valuable to provide more information (possibly a tutorial) to help other labs adopt the methodology to characterize odor landscapes in new assays.

We thank the reviewer for this suggestion. We now include cost breakdowns in Table 1 in the methods section. We continue to provide comprehensive parts and catalog numbers (PartsList.xlsx), now as Supplementary File 2. And we provide a new detailed step-by-step tutorial (Tutorial.pdf, Supplementary File 1) with photographs and instructions to assemble and operate the instrument. Both Supplementary Files are also available in a FigShare repository, http://doi.org/10.6084/m9.figshare.21737303.

Additional request:

The methodology and experiments presented in the manuscript offer the opportunity to compare odor-evoked behaviors in two species in response to the same environment. To illustrate the potential of the methodology, the authors are strongly encouraged to analyze the behavior of both species using the same procedure and metrics to report similarities and differences in behavioral algorithms. This addition should help the reader define whether the methodology enables to draw new conclusions or if it recapitulates previous observations. If the authors decide that they won't proceed with a deeper behavioral analysis, they should tone down their claims about algorithmic findings in their abstract and the rest of the manuscript.

We have now added new experiments and a new maintext figure characterizing chemotaxis of a mutant worm with ectopic expression of an activated potassium channel (twk-18(gf)) expressed in the interneuron AIB (Figure 8). We compare their navigation to wild type and show that these mutant animals still climb gradients but they have lower turning rate and rely less on weather-vaning. This comparative analysis demonstrates how our setup can provide insight into potential neural circuit mechanisms that contribute to navigation behavior in worms. We separately have added additional language comparing worms and Drosophila larvae:

“We note that the chemotaxis efficiency is much higher in larvae compared to worms given the order of magnitude difference in run velocity (~ 3 mm/s in larvae and ~0.1 mm/s in worms).”

Reviewer #1 (Recommendations for the authors):

Measuring airborne odorant gradient is notoriously challenging. The goals of this manuscript are twofold: (1) to develop a technique to characterize the spatial properties and temporal stability of airborne odorant landscapes; (2) to produce stationary odorant landscapes with controlled and predictable geometries that are suitable to refine the study of navigation behavior (chemotaxis) in small animals such as C. elegans and the Drosophila larva. Tackling these two problems should improve future correlations between sensation and behavior to study the neural bases of sensory encoding and action selection.

To achieve their first goal, the authors propose to use a distributed array of digital gas sensors. The resolution of the measurements reported in the manuscript is impressive: they go beyond the state of the art. Although these measurements remain discrete in space, they provide valuable data to fit a biophysical model of the diffusion of the odor and its reaction with the arena where the odorant landscape is created. While the technique allows monitoring odorant landscapes in space and time, it also has limitations.

First, the measurements require modifying the experimental conditions where the behavior is tested. The landscape cannot be directly measured while the behavior is taking place. Second, the dimensions of the digital gas sensors (~1 cm) are relatively large compared to the size of the tested animals (~1 mm). This limits the resolution of the spatial measurements of odorant landscapes with small dimensions where even smaller animals are studied. In particular, the system might not be able to detect differences in concentrations between the individual inlets of the odor delivery system, which reduces the accuracy of the reconstruction of a landscape and the evaluation of its variability.

We thank the reviewer for this thoughtful critique. We now demonstrate direct recording of the entire extent of the odor landscapes when agar is present in the flow chamber (Figure 6) and when animals are also present on agar (Figure 1 – Figure Supp 6). We show that the current system, as designed, has challenges tracking animals when the sensors are present, and we now explain how this motivated us to pursue a strategy of measuring only on the boundary when animals are present.

We analyzed the subset of animals that we are able to track simultaneously with odor measurements and show that chemotaxis strategies qualitatively agree with the findings presented in Figure 7.

Regarding the spatial resolution of our odor measurements, we are confident that our sensors are sufficient to capture spatial variation for the geometries and flows used throughout this manuscript which are all far from the turbulent regime. Inlets are spaced 1 cm apart while our sensors are spaced 0.75 cm apart vertically along the sensor bar. We now clarify in the revised methods section:

“The 16 odor sensors are arranged in two interleaved columns spaced 1 cm apart. The vertical distance between sensors in each column is 1.5 cm, which results in 0.75 cm vertical sensor resolution.”

Crucially, our model-based fitting indicates that the flow patterns we use are governed by a non-turbulent convection-diffusion process, which has a spatial scale of ~1 cm (Figure 1 g). Therefore we are confident that we have sufficient spatial resolution. We also are reassured that we see stability in the time domain. We present new analysis (Figure 1 – Figure Supp 5) to show that our sensor size and spacing is sufficient to capture the spatial concentration changes we expect in the portion of the arena that we measure over time. Our measurements show that there is little temporal fluctuation throughout the time scale of behavioral recordings (30 minutes) and that the spatial profile is smooth and robust in its shape. We therefore conclude that the current setup should be sufficient to characterize the spatially patterned odor landscape.

The second goal of the study is to produce odorant landscapes with enhanced stability compared to existing systems. The airflow-based olfactometer features spatially patterned odorized air. While this system represents an improvement with respect to published assays in C. elegans, its contribution to the larva is less obvious. The geometry of the landscapes resulting from the linear array of odorized air flows is essentially uni-dimensional. This produces geometries that are fundamentally different from the radially symmetric gradients emitted by odor cups that have been used by other labs for many years. The authors conclude that gradients resulting from the diffusion of a single source are unstable due to the absorption of the odor by the agar. This conclusion was drawn for odorant molecules that are water-soluble (or miscible in water like ethanol). Whether the same conclusion holds for odorant molecules with low or negligible water solubility is unknown, which precludes a generalization of the results to odors that have been more commonly used in the field.

We apologize for the confusion. We did not mean to imply that single source droplets are unstable for all odor molecules or all cases. We also address this point in more detail in response to the Essential Revisions above. We now clarify in the text:

“We focused on the odor butanone because it is important for a prominent associative learning assay (Torayama et al. 2007, Kauffman et al.2011). Butanone has high vapor pressure and is highly soluble in water and therefore it interacts strongly with agar. In this work we show that this odor-agar interaction makes it difficult to a priori infer an odor landscape experienced by the animal when agar is present, but that continuously monitoring the odor profile on the boundary overcomes this challenge.”

…

“Odorants with less solubility and lower vapor pressure may not require pre-equilibration to reach quasi-equilibrium quickly, but instead may be able to stabilize across agar on a shorter time scale.

For example, ethyl butyrate is less water soluble (~0.586 g/100 ml in water, compared to 27.5 g/100 ml for butanone), has lower pressure (11.3 mmHg compared to 78 mmHg for butanone) and has a smaller gas phase diffusion coefficient (5.6 x 10^-3 cm^2/s compared to ~69 cm^2/s for butanone). A droplet of ethyl butyrate has been shown to form a relatively stable odor landscape over a 5 minute timescale in a dish on agar without flow (Tadres et al. 2022, Louis et al. 2008).”

And we have clarified in the relevant subheading (emphasis added): “Odor-agar interactions dominate classical butanone droplet assays”

We now also conducted additional experiments, in new Figure 2c and e, to show that our olfactometer is capable of making gradients of three other odors with other properties: benzaldehyde, ethanol and ethyl butyrate. In principle, molecules with lower water solubility should be easier to work with because they interact less with agar. A strength of our approach is that it works with a range of odor molecules including ones like butanone that we are interested in for scientific reasons but that have high interactions with agar.

The manuscript would benefit from additional work to demonstrate how the tool can advance the understanding of chemotaxis in C. elegans and the Drosophila larva. Besides the fact that the authors chose to characterize larval chemotaxis with an odor that has not been used in previous work, the behavioral characterization is limited given the potential capabilities of the new tool. Presently, it falls short of reporting "chemotaxis strategies for C. elegans and D. melanogaster larvae populations under different spatial odor landscapes" as stated in the abstract. This incompleteness is a missed opportunity to validate the merit of the methodology.

We have now removed the offending language from the abstract. Importantly, we have added new experiments to demonstrate how the tool can advance our understanding of chemotaxis. In particular we now perform new experiments comparing a mutant worm that has defects in a key interneuron AIB to a wild-type animal (Figure 8). These measurements show that defects in AIB result in a down-regulation of weathervaning, and lower overall turning rate, but that the bias in turning rate is preserved and the animal still climbs odor gradients. Experiments like the one with the AIB mutant show that our instrument will enable new advances. We think the work is particularly well suited to its “Tools and Resources” format, as described in the eLife guide for authors:

“Tools and Resources articles do not have to report major new biological insights or mechanisms, but it must be clear that they will enable such advances to take place”.

I praise the authors for the technical achievement presented in the manuscript. Combining a spatial array of digital gas sensors with computational simulations of the odor diffusion process is ingenious. However, several aspects of the methodology require further characterization and more careful validation to be helpful as a tool that can be widely adopted by other labs. Below is a list of shortcomings that should be addressed:

1. Although the multi-odorous-air-flow delivery system is simplified compared to the original design published by Gershow et al. in 2012, it remains complex. It would be important to test the reproducibility of landscapes produced by this system. In addition, it would be useful to demonstrate the flexibility with which landscapes with different geometries can be created. One notable limitation of the system is that gradients can only be designed in one direction. This precludes the creation of radially symmetric gradients produced by a point source, such as those that have been typically used in C. elegans (e.g. Pierce-Shimomura & Lockery 1999) or in the Drosophila larva (learning assay developed by Gerber and coworkers). Given the tiled arrangement of the airflow inlets, is the gradient smooth at the scale of a worm or "bumps" expected in the odor profile over the y-axis? This question is likely to be unresolved by the odor measurements since there is a higher density of tubes compared to sensors along the y-axis.

The reviewer raises several related concerns about (1) reproducibility (2) limitations of possible geometries and (3) spatial resolution.

• We demonstrate several aspects of the instrument's reproducibility. Perhaps the most dramatic form of reproducibility is that we show in Figure 6 that the system reproduces a desired odor landscape using the same steady-state flow inputs even when a large agar odor sink is present that would otherwise dramatically perturb the landscape. Specifically Figure 6b generates the same odor landscape as in Figure 6a to within 10% fractional difference even though the agar pad is present. We also show Figure 1 – Figure Supp 1 that the odor sensors themselves are reproducible across time. And in a new analysis in Figure 1 – Figure Supp 5 we show that the odor gradient itself is stable over time.

• Regarding concerns about possible geometry: We argue that the droplet assay is more limited than the geometries allowed by our instrument because the droplet assay allows only a single odor landscape geometry and the experimenter has limited control over its properties. In contrast, our instrument can support and measure a droplet based gradient, as we show in Figure 3. So if a droplet is desired, our instrument can measure from that too. But importantly our instrument can also use flow to generate other odor landscapes such as a cone with a different slope or inverse cone as shown in Figure 2. As an aside, in the C. elegans literature and in some early fly larva experiments the radial symmetry of the droplet assay is often not used because chemotaxis arrays are conducted when the droplets are positioned on the far side of a plate (Bargmann 2006, Cho et al. 2016, Aceves-Piña et al. 1979).

• We address concerns about spatial resolution of our odor sensors in more detail in response to Reviewer #1’s Public Review, above. The inter-channel spacing (1 cm) is in fact larger than the inter-sensor spacing (0.75 cm vertical sensor resolution). Fine structure due to odor plumes is ruled out by the Reynold’s number (<10), which is too low to support turbulent flow, and by quantitative agreement of the measurements with the convective-diffusive model lacking turbulence. This model is characterized by a length scale larger than the inter-sensor spacing. Therefore we are confident that we have the spatial resolution to characterize the relevant changes of the landscape.

2. The manuscript reports that the high absorption of butanone by the agar significantly affects the properties of the resulting odorant landscape. As a result, odorant landscapes produced by a single odor droplet can flatten over time in such a way that the gradient might be nearly inexistent after a couple of minutes. This conclusion is significant since larval olfaction has been frequently studied with a single odor source (a cup or a droplet). As indicated by the authors in lines 270-271, butanone is a water-soluble odorant molecule. By comparison, most odors that have been used to study olfactory behaviors in the Drosophila larva have a low solubility in water. For instance, 1-octanol (<1g/L), benzaldehyde (<10g/L), pentyl acetate (<2g/L) and geranyl acetate (<1 g/L) are all far less soluble than butanone (275 g/L). This difference in water solubility is likely to affect the stability of the odorant gradient. Were odor measurements not combined with computational modeling of the odor diffusion-reaction process in Tadres et al. 2022 (Science Advances) to establish that gradients of ethyl butyrate and ethyl acetate are relatively stable for 5 min?

We agree that water solubility plays an important role. We now clarify in the main text (copied below) that our claims about the instability of butanone droplets apply to butanone and not necessarily any other odorant. We also address this concern in response to Essential Revisions (third bullet point) and in response to Reviewer #1’s Public Review.

“Odorants with less solubility and lower vapor pressure may not require pre-equilibration to reach quasi-equilibrium quickly, but instead may be able to stabilize across agar on a shorter time scale. For example, ethyl butyrate is less water soluble (~0.586 g/100 ml in water, compared to 27.5 g/100 ml for butanone), has lower pressure (11.3 mmHg compared to 78 mmHg for butanone) and has a smaller gas phase diffusion coefficient (5.6 x 10^-3 cm^2/s compared to ~69 cm^2/s for butanone). A droplet of ethyl butyrate has been shown to form a relatively stable odor landscape over a 5 minute timescale in a dish on agar without flow (Tadres et al. 2022, Louis et al. 2008). Here we demonstrate stable odor landscapes over longer timescales of at least 30 mins with a more volatile odor, butanone, by using our flow-based assay (Figure 1 – Figure Supp 5).”

We are not casting doubt on the claims made in Tadres et al. 2022. It is interesting to compare the stability of the odor landscapes in Tadres et al. to the ones shown here. For the sake of comparison we copy Figure S10 from Tadres et al. and show it next to a panel from our new Figure 1 – Figure Supp 5. Even with a much less stable odorant our flow based approach demonstrates stable odor landscapes that persist five-times longer.

This prolonged stability would be important for research in learning, sensory adaptation, or any chemotaxis experiments in worms, which take longer than for flies.

3. Technical question related to Figure 3: in this figure, the butanone droplet appears to be placed on the lid without agar (line 24) whereas it is placed directly on the agar gel in the condition with agar (line 251). The placement of the butanone droplet on the lid versus on the agar is an important distinction besides just the presence or absence of agar. This point is not addressed in the discussion of Figure 3. The authors should use the same type of odor source to make comparisons across conditions.

Thank you for the suggestion. We conducted the requested experiments. We now show both droplets on the lid and droplet on agar in revised Figure 3. It's perhaps an interesting historical tradition that odor droplets are normally placed in agar for worm chemotaxis (Bargemann 2006; Cho et al. 2016) and on the lid for larval chemotaxis experiments (Tadres et al. 2022 and Louis et al. 2008).

4. The behavioral quantification of the chemotactic behavior of C. elegans reproduces a series of results that have already been published. Weathervaning appears to be reproduced, but its strength might be lower than that reported by IIno and Yoshida 2009. It would be helpful if the authors could present a graph with the average curvature rate as a function of the bearing of the gradient to permit a comparison with Iino and Yoshida 2009.

Thank you for the suggestion. We generate a new Figure 7–Figure Supp 2 to more directly compare weathervaning in butanone chemotaxis to the salt chemotaxis reported by Iino and Yoshida 2009. As shown in Iino 2009 figure 4B, their results with isoamyl alcohol show average curving rate on a similar scale (they do not have direct odor concentration measurements). Our result not only demonstrates weathervaning strategy in butanone chemotaxis but also indicates that the tuning curve depends on the magnitude of the gradient (as opposed to just the bearing angle).

In contrast with the behavioral quantification of C. elegans, the behavioral quantification of larvae is minimal. Essentially, the authors report that larvae turn less while moving upgradient and turn more while moving downgradient. This quantification does not justify the highly-quantitative assay introduced in the manuscript. In addition, could the authors show that larvae are capable of "weathervaning"? Another interesting application of the assay would be to establish how the navigational performances of larvae change with the steepness (or other geometric features) of stationary odorant gradients. This would also show that larvae are capable of precise chemotaxis in gradients that are essentially 1D compared to 2D gradients resulting from point sources.

These are excellent suggestions for a future larger scale investigation. In this work we demonstrate that the instrument will enable future discoveries for both worms and fly larvae. For example we show the diversity of odor landscapes (e.g. Figure 2) that one could use to address the scientific questions like those posed by the reviewer. We demonstrate that larvae navigate in the new apparatus as expected and as previously reported, supporting its use in the sorts of future experiments described by the reviewer.

5. The introduction is slightly misleading. One motivation of the work is to design a setup where measurements of odor concentrations characterizing an odorant landscape can be done during behavioral experiments. One would expect that these measurements would be done in 2D across the assay. Instead, the measurements are made on the border of the arena, which partly contradicts the original goal of the methodology. I appreciate that the presence of the sensors in the arena is not compatible with behavioral experiments, but the implications of this limitation should be more thoroughly discussed in the manuscript. It also reinforces the need to establish the level of fluctuations in odor concentrations inside the arena in real-life conditions.

We now include new measurements of the full sensor array concurrently with agar gel (Figure 6) and worm behavior (Figure 1 – Figure Supp 6). The new measurements improve the manuscript in several ways: First, they show that simultaneous direct measurements are possible, although suboptimal because they create challenges for behavioral tracking shown in Figure 1 – Figure Supp 6. Second, they provide strong motivation for moving to a boundary based approach, which we now explain more clearly to the reader. And third, the agar measurements in Figure 6 provide additional empirical evidence that our model (Figure 5 and Appendix) and boundary conditions (Figure 4) are effective for inferring concentration across the 2D space during behavioral recordings. And the measurements in Figure 6 provide direct evidence for a stable, controlled odor landscape in our system.

We describe the textual changes we have made in more detail in our response to Reviewer #1’s Public Review, and in our response to the requested Essential Revisions.

More generally, how do the authors envision the use of their new methodology by regular labs working on olfaction? If, on the one hand, their primary goal is to propose a technique that will be adopted by many experimental labs, one should acknowledge that most labs might find it challenging to conduct odor measurements, optimize the convection-reaction-diffusion model and simulate new odorant landscapes. Going through the pipeline would require a "tutorial". If, on the other hand, the goal of the methodology is to draw attention on potential artifacts associated with the instability of odorant gradients in published assays, the authors should conduct a more thorough analysis of the stability of representative experimental conditions.

Our goal is to provide a resource to the field. Our goal is not to systematically explore potential artifacts in other systems. We chose to include measurements of the butanone odor droplet (Figure 3) in order to motivate the need for new methods to better deal with the odor butanone, which we are interested in using for studying learned olfactory navigation.

We have made several changes throughout the text to make clear that we are not casting doubt on the previous measurements, such as those in Tadres et al. 2022 and Louis et al. 2008. We describe these textual changes in more detail in response to requested Essential Revisions, Reviewer #1’s public review, and Reviewer #1 point 2, above.

There may be a misunderstanding about the convection-reaction-diffusion model. That model is to verify our protocol to a reader, but is NOT required for the experiment. No modeling is needed for a new lab to adopt this method and build their own custom stationary odor gradients, as we now state:

“Note our hypothetical experimenter would not be required to perform any simulations or calculations.”

A lab needs only to construct stationary landscapes without agar, and then ensure that the odor measured along the boundary is similar when an agar plate is introduced. We now state this explicitly in the main text to avoid confusion:

“Note that the experimental procedure does not require one to conduct numerical simulations repetitively. According to the empirical (Figure 6) and theoretical (Appendix) evidence, one can confidently infer the odor landscape from boundary conditions when the system reaches quasi-equilibrium.”

…

“A detailed tutorial for assembling the instrument and the protocol for experiments is provided in Supplementary File 1.”

We also now include a step-by-step tutorial, with photographs, to help other labs perform experiments with the instrument. The tutorial is included as Supplementary File 1, and crossposted on FigShare https://doi.org/10.6084/m9.figshare.21737303.

6. Appendix 1 argues that the convection-diffusion model and the reaction-convection-diffusion model are equivalent if the odor flow between the air and agar are in equilibrium. How is this equilibrium defined quantitatively?

We now clarify in the appendix:

“According to first-order kinetics, we expect the concentration change \Δ C(x,y) to be proportional to the concentration at the point in space C(x,y). Therefore in practice we monitor the fractional difference as shown in (Figure 6). Quantitatively when the concentration is within 10% fractional difference from the steady-state profile without agar, we categorize it as the quasi-equilibrium regime we conduct animal experiments with.”

We define equilibrium between air and agar when the concentration difference in two substrates has little change in time, namely k1\deltaC ~ k2\epsilon. Empirically and quantitatively, when the outflux odor concentration under conditions with or without agar is close (within a given fraction that the experimentalist chooses), then we conduct animal experiments in this quasi-steady-state regime. In our revised manuscript, we have new measurements for odor concentration across the agar region (Figure 6). The fractional difference is below 10% at equilibrium in our system.

Reviewer #2 (Recommendations for the authors):

This manuscript by Chen et al. describes an apparatus for measuring odor-evoked navigation behavior in C. elegans and Drosophila larvae. The major advance is using an array of metal oxide sensors to measure odor gradients. The authors apply this tool to generate stable gradients in an agar environment. They then measure odor-evoked behavior in both worms and fly larvae, demonstrating the ability to recover previously described stimulus-behavior associations such as biased random walking and weathervaning.

The problem of controlling and measuring odor dynamics is a challenge for all studies of odor-guided navigation and new approaches to these problems are welcome. While the present study shows the potential of the metal oxide sensor array approach, several considerations are missing that would help other researchers to evaluate whether this approach would be useful, and are important for evaluating claims made with this device.

1) Metal oxide sensors have previously been employed to measure odor dynamics in turbulent airborne environments (e.g. Schmuker et al. 2016, Tariq et al. 2021, Dinnler et al. 2022). Two major issues raised by these previous studies are that (1) metal oxide sensors have a fast onset but very slow offset which complicates inference of odor dynamics, and (2) sensor drift can complicate absolute concentration measurements. The authors should discuss these previous studies and show data on the temporal resolution and stability of their sensor array in response to more rapid changes in odor concentration.

We thank the reviewer for pointing us to this literature, which we now cite at several points in the manuscript, as described in response to the first requested Essential Revision. To characterize temporal dynamics, we now present new measurements in response to step-wise odor concentration changes (Figure 1 – Figure Supp 2). We find the response timescale of 10’s-of-seconds to be sufficient for our odor landscapes which are nominally static over the timescale of our recordings (30 mins). We also characterize temporal drift across 90 minutes in Figure 1 – Figure Supp 1 by comparing the metal oxide sensors to a PID. We see no obvious systematic drift on this timescale. Some of those earlier metal-oxide studies involve detecting odor in the turbulent regime which is much more demanding than the odor landscapes used here

We now highlight these measurements and comparisons to previous uses of metal oxide sensors in the text:

“Metal-oxide sensors are small and scalable digital gas sensors that have recently been used to monitor odor concentration in studies of olfaction (Schmuker et al. 2016, Drix et al. 2021, Tariq et al. 2021) especially in the context of turbulent odor plumes. Arrays of metal-oxide sensors have been used to measure spatial odor landscapes by sampling across space (Burgues et al. 2020). Here we leverage these approaches to present a new multi-sensor odor array with metal-oxide sensors combined with a flow chamber”.

…

“However, in general, commercial metal-oxide sensors are not designed for precision work – they tend to drift due to variations in heater temperature, humidity, adsorption of chemicals and ageing effects (Dennler et al. 2022).”

2) Previous studies have used a spectroscopic approach to quantify odor gradients in agar (Louis et al. 2008). The authors state that this approach is not compatible with simultaneous behavioral measurements, while the metal oxide sensors are, however, it is not totally clear to me why this should be. While a direct comparison of these two methods would be ideal, a more thoughtful comparison of the pros and cons of the two methods would be most helpful to other researchers.

We do not claim that the FTIR method is fundamentally incompatible with simultaneous behavioral measurements, merely that it has not been demonstrated before. We have now provided a more detailed comparison in the introduction:

“In one of the most comprehensive measurements to date, Louis and colleagues (Louis et al. 2008, Tadres et al. 2022) used infrared spectroscopy to measure the spatial profile of a droplet based odor gradient. They used this approach to study bilateral olfactory sensing and olfactory encoding in Drosophila larvae, clearly demonstrating the value of measuring odor landscapes. That work measured odor concentration at a relatively low temporal resolution (~1 sample / min), only when animals weren't present, and relied on a relatively expensive Fourier-transform infrared spectroscopy machine.”

…

“In all of these previous works, measurements were performed offline, not during animal behavior, and the odor concentration was assumed to be the same across repeats of the same experiment, and when animals are present.”

3) A major issue for using the metal oxide sensor arrays during behavior would seem to be the heat generated by the arrays. This should be discussed and any heat effects on behavior should be described, as thermotaxis behavior has been described in both species studied here.

We thank the reviewer for pointing out the importance of heat management. We address this in more detail in response to the second point in “Essential revisions.” Briefly, we added new analysis for heating (Figure 1 – Figure Supp 4) and more detailed description in the caption and method sections. Specifically, we leverage existing spatial temperature measurements made by thermo-hydro sensors located across the odor sensor array. The metal-oxide sensors produce heat during the experiment but we dissipate much of this heat through thermal coupling to the optics table. We also observe no obvious signs of thermotaxis or biased trajectories in control experiments where we would expect any heating to be present (Figure 7 – Figure Supp 1). We therefore conclude that the heating effects are minor.

4) It is not clear that this device allows for "precise" measurements at the location of the animal as claimed in the abstract. First, the measurements are in the air and are predicted to be related to the concentration in agar through a scalar. Second, measurements are made at the edges of the agar plate and internal concentrations are inferred. Although these appear to be stable in the absence of animals, local fluctuations due to animal movement cannot be measured.

We have removed the offending language from the text and added new experiments that directly measure odor concentration above the agar (Figure 6) even when animals are present (Figure 1 – Figure Supp 6). This direct measurement validates our approach of inferring internal concentrations from the boundary. We describe this in more detail in response to Reviewer #1’s public review, and in response to the requested Essential Revisions. We now write in the text:

“Ideally, one would like to directly measure odor just opposite the surface of the agar, such that odor-agar-interactions are fully accounted for. By placing the agar on the underside of the lid hanging above the odor sensors, we were able to simultaneously record spatial odor concentrations generated by our flow chamber opposite the surface of the agar during animal behavior (Figure 1 – Figure Supp 6). Unfortunately, when imaging in this configuration, the odor sensors in the background obstruct the animals from being tracked over the extent of the arena. We therefore sought an alternative strategy: we chose to measure the odor profile along only the boundary of the arena in order to infer the odor concentration inside the arena in such a way as to also take into account the odor-agar-interactions.”

In experiments, such as new measurements in Figure 6 – Figure Supp 1, we see no obvious evidence to suggest that the movement of animals significantly alters the odor landscape:

"After the pre-equilibrium protocol, measured airborne concentration in the arena with agar was similar to measurements made without agar to within an average of less than 10% fractional difference between the two conditions in the area that the animal would crawl, FIGURE 6b. The agreement of odor landscape measured with and without agar at quasi-equilibrium agrees with our modeling and indicates that the odor landscape is stable and identifiable in our setup by monitoring the odor profile along the boundary…

The presence of worms did not noticeably alter the odor concentration (Figure 6 – Figure Supp 1)."

Suggestions:

The focus of this study is on animals that move in a viscose substrate such as worms and larvae, where odors form stable gradients. While this is implied by the term "small model organisms" the authors should distinguish this from animals other than small model animals (adult flies, larval fish) that move in turbulent environments where the temporal resolution of the system would need to be much higher.

Thank you for this suggestion. To further clarify that the method is developed for stable gradients, we now write:

“Here we have addressed the problem of creating non-turbulent airborne odor landscapes.”

In the abstract we make clear we discuss worms and fly larvae:

“For small model organisms like C. elegans and larval D. melanogaster…”

and that we deal with stationary landscapes:

“We construct stationary chemical landscapes in an odor flow chamber…”

The ability to measure odor-evoked behavior in two species in response to the same environment seems like a bit of a missed opportunity here. Can the behavior of both species be analyzed using the same methods and similarities and differences in behavioral algorithms described?

We have now expanded the language comparing the navigational strategy of these two organisms in the same environment:

“With modulation of turning rate and orientation in the same odor environment, we conclude that Drosophila larvae use similar navigational strategies to C. elegans to move towards butanone. We note that the chemotaxis efficiency is much higher in larvae compared to worms given the order of magnitude difference in run velocity (~3 mm/s in larvae and ~ 0.1 mm/s in worms).”

Overall I think this is an interesting approach but I think there are a large number of specific claims that need to be softened or toned down:

line 18: "Crucially and unlike previous methods, our method allows continuous monitoring of the odor profile during behavior" I don't think this is true. For example, the Vickers and Baker 1994 study used an extra moth antenna to measure odor plume fluctuations in flying moths, and the Tariq 2021 study measured odor at the location of a navigating mouse using metal oxide sensors.

Thank you for pointing us to this literature. We removed the offending phrase from the abstract and now highlight the prior work:

“A central difficulty in measuring animals' responses to olfactory cues is quantifying airborne odor concentrations that vary in space and time (Vickers et al. 1994, Tariq et al. 2021, Kadakia et al. 2022, Yamazoe-Umemoto et al. 2018).”

line 23-24: "accurately inferred" "precise odor concentration" not sure these are true for the reasons listed in 4 above.

We now show more direct evidence to support our claims that our inference is accurate to within 10% (Figure 6), and we have also removed the offending language. We discuss this in more detail in response to Reviewer #2’s Public Review point #4, above.

line 35-36: "small mode organisms" I guess this means worms and larvae but I think it would be helpful to specifically say animals that move in a substrate or mostly navigate in gradient (as opposed to turbulent) environments.

We now clarify in the introduction:

“We focus in particular on the navigation of small model organisms in continuous, non-turbulent, gradients established by the spread of odorants from their sources due to diffusion and drift.”

line 46: "no technique currently exists for precise control and continuous monitoring of an odor landscape." This is not true either. Many published approaches here include optogenetics to create virtual environments, precise generation of odor waveforms, controlled flow chambers, etc.

We removed the original claim and revised the introduction section to include methods used in past studies, including virtual environments and controlled odor presentation:

“While many techniques exist to present odor stimuli (Gorur-Shandilya et al. 2019, Kadakia et al. 2022), measure odor concentration (Tariq et al. 2021, Celani et al. 2014), or develop virtual odor environments in a lab (Yamada et al. 2021, Radvansky et al. 2018), it remains challenging for precise control and continuous monitoring of an odor landscape.”

Optogenetic virtual environments are not a direct substitute for precise odor environments because optogenetics bypasses the biophysics that govern receptor binding and may or may not evoke sensory adaptation in a physiological way.

line 97: "odor profile in the chamber" I think the abstract implies that you can make precise measurements at the location of the animal but it is clearer later on that these are inferred measurements across the arena.

We have reworded the abstract to further clarify:

“Careful placement of the sensors allows the odor concentration across the arena to be continuously inferred in space and monitored through time.”

Moreover, we now show direct measurements of the odor profile in the chamber in new experiments in Figure 6 and Figure 1 – Figure Supp 6.

line 160: "quantifying airborne odor concentrations" There is a large literature on this in the turbulent navigation field that is not cited or discussed in this paper.

We thank the reviewer for pointing us to these references. We now discuss them:

“A central difficulty in measuring animals' responses to olfactory cues is quantifying airborne odor concentrations that vary in space and time (Vckers et al. 1994, Tariq et al. 2021, Kadakia et al. 2022, Yamazoe-Umemoto et al. 2018). This difficulty is exacerbated in turbulent environments where odor plumes carry abrupt spatial and temporal jumps in concentration far from the source with fundamentally unpredictable dynamics (Celani et al. 2014, Dennler et al. 2022)”

line 175: "metal oxide sensors" can you give estimates of cost, sensitivity, odor selectivity, and temporal resolution? I would expect this to be in a supplement.

Table 1 now lists the cost of key components. At the time of purchase, the Metal Oxide Sensors were approximately $10 each. We also provide a manifest of detailed parts numbers (Supplementary File 2, PartsList.xlsx, cross listed on FigShare 10.6084/m9.figshare.21737303).

We address reviewer concerns about sensitivity, selectivity and temporal resolution in more detail in our response to the first requested “Essential revision”. Briefly, we conducted new measurements to characterize their responses and stability presented in new figure panels (Figure 1 – Figure Supp 2; Figure 1 – Figure Supp 3; and Figure 2):

Sensitivity: We added calibration with odor sensor at higher concentration range up to 1,000 ppm (Figure 1 – Figure Supp 3) and characterized response to step functions (Figure 1 -Figure Supp 2). These measurements show that the array of metal oxide sensors is sensitive across a large range of odorant concentrations, significantly larger than the 0-200 ppm we use in behavior experiments.

Selectivity: We added measurements with other molecules commonly used in worm and larval chemotaxis studies (Figure 2 c,e), specifically ethyl butyrate and benzaldehyde, in addition to butanone and ethanol and compare the metal oxide sensors to readings with a PID for each.

Temporal resolution: We added new figures showing the odor sensor’s temporal response to step-wise modulated odor flow (Figure 1 – Figure Supp 3). This result indicates that the sensors act on a time scale that is on the order of ~10 seconds and is comparable to the downstream PID readout. This temporal response is sufficient for the slow changes we expect in our stable odor gradients.

line 204: 1-second temporal resolution. What is the evidence for this? Is this a sampling rate or does it take into account the dynamics of the sensor as shown in the Tariq paper?

We apologize for the confusion. We meant to refer to sampling rate, not the temporal response. We have rewritten as “1 Hz sampling rate.” As described in response to the previous comment, and in response to the requested Essential Revisions, we have added new measurements, analysis, and new language to address the temporal response of the sensors.

line 251: "remove two sensor bars and replace them with agar" I think it needs to be clearer upfront that the sensors have to be removed to do agar measurements.

We now show in Figure 6 and Figure 1 – Supp 6 that sensor bars do not strictly have to be removed, but that tracking is suboptimal when all sensor bars are in place. We now better emphasize the two “modes” of experimental measurements early on in the manuscript, in the introduction section, that do or do not require removing the sensor bars:

“The array of sensors can be used two ways: the full array can be used to measure the generated gradient throughout the extent of the chamber, or parts of the array can be used on the borders to monitor, during behavioral experiments, the boundary odor profile in the chamber. By varying flow rates and the sites of odor introduction, we show a variety of odor profiles can be generated and stabilized.”

line 371: butanone chemotaxis in Drosophila larvae. should cite the Jung and Bhandawat 2015 paper here that closely examines butanone-evoked navigation in adult flies and compares it to vinegar-evoked navigation.

Added. We thank the reviewer for suggesting this reference:

“Although butanone is not as commonly used as a stimulus with Drosophila as with C. elegans, butanone is known to be attractive to larval flies (Dubin et al. 1995, Dubin et al. 1998) and has been variously reported to be attractive (Park et al. 2002, Jung et al. 2015) and aversive (Israel et al. 2022, Lerner et al. 2020) to adult flies. ”

line 399: "This last feature…sets this method apart from previous approaches" It is not clear to me why the spectroscopic approach could not in principle be used during behavior.

To our knowledge this last feature has not been demonstrated with a spectroscopic approach. In principle it may be possible, but one would have to consider technical challenges such as: (1) design the chamber to be compatible with simultaneous optical access for both infrared imaging and imaging of the animals; (2) confirm that the spectroscopic measurements are not detected by the animal (thermally or visually), do not interfere with animal behavior, and do not optically conflict with the illumination needed to visualize the animals; and (3) be fast enough to meaningfully scan through the environment within the time course of a single animal experiment (Louis et al. 2008 reported 1 minute per scan at one cross section).

line 405: "In the future, such tuning curves may form the basis of investigations into neural mechanisms driving the sensorimotor transformations underlying navigation" Lots of this has been done! Seems weird to say this as a future thing and not cite the many many circuit cracking papers in worms and flies that have been pursued with other apparatus.

We apologize for the confusion. We were trying to point out that we have a new ability to measure tuning curves with respect to local odor gradients in more complex odor landscapes (not radially symmetric). To avoid confusion we have removed the offending text.

We now reference additional sensorimotor literature:

“Past work focusing on neural circuit mechanisms for odor navigation in worms have often relied on coarser readout such as the chemotaxis index or using the distance to odor source as a proxy for the inverse of odor concentration. However, it has been shown that the exact odor concentration alters the animal's behavioral response to odor (Yoshida et al. 2012, Levy et al. 2020). Quantifying tuning curves with respect to more precise odor measurements may better constrain the investigations into neural mechanisms driving the sensorimotor transformations underlying navigation algorithms (Clark et al. 2013, Liu et al. 2020).”

lines 438-440: "at the quasi-equilibrium conditions used in our experiments the odor concentration in agar is related to the airborne odor concentration directly above it up to a scalar that we predict to be constant across the agar." But is this actually true? This is quite far from the claim in the abstract that you can precisely measure the concentration at the location of the animal.

We now directly measure the odor concentration immediately opposite the surface of agar, as shown in new Figure 6 and new Figure 1 – Supp 6. As we discuss in response to comments from Reviewers #1 and #2 above, we show that the odor landscape inferred from measurements on the boundary is within some fractional error (10%) from the quasi-equilibrium profile without agar. In other words, the inference for odor concentration is supported both experimentally (direct measurement) and theoretically (numerical example and derivation in the Appendix).

As mentioned in response to Reviewer #2 Public Review #4, we have removed the offending word “precise.”

We are careful not to claim that this technique measures concentration ‘in’ agar. But in the Appendix we argue that it is reasonable to assume that the odor concentration that the worm likely experiences at the gas-liquid interface is a scalar multiple of the odor concentration in air. It is also mentioned in the Discussion section:

“Our reaction-convection-diffusion model suggests that at the quasi-equilibrium conditions used in our experiments the odor concentration in agar is related to the airborne odor concentration directly above it up to a scalar that we predict to be constant across the agar.”

Reviewer #3 (Recommendations for the authors):

In this paper, Chen et al. propose a new method to measure odor stimuli in space and time. Measuring the odor stimulus is a key step in interpreting odor-driven behavior and understanding the neural mechanisms that mediate it, but this task still challenges every experimenter in the field. As described in the introduction, there is basically not a single method that is good for most behavioral assays, even when these involve small animals such as C. elegans and D. melanogaster. Previous approaches are either invasive, very expensive, or limited to very small behavioral arenas. The strength of the proposed method is to be cheap and to have a reasonable spatial (1 cm) and temporal (1 s) resolution. The full sensor array introduced here cannot be located throughout the behavioral arena, but a mathematical model shows that it is sufficient to measure the odor stimulus at certain specific positions to reconstruct the full spatial profile. In other assays, such mathematical considerations might not be possible and physical constraints might make it impossible to use the sensor during behavior, however, it can still be used to measure the odor landscape in a separate experiment, and, with a certain degree of reproducibility (which will depend on the specific delivery system), this is still better than no measure. It should be noted that the method is limited to 2D measurements, which is sufficient for the walking or crawling behavior of small animals in non-turbulent conditions, but it cannot be extended to 3D assays. It remains moreover unclear what is the sensitivity range of the sensor for the odors used in the paper (butanone and ethanol) and what is expected for other odorants (which compounds are detectable?).

We thank the reviewer for positive comments and suggestions. We now clarify in the text that the method is restricted to studying crawling behavior in 2D environments:

“Future work is needed to improve concurrent animal tracking methods, to design more miniaturized and scalable sensors, and to generalize to more complex or three-dimensional environments.”

We clarify a potential misunderstanding: in all experiments with agar we measure odor sensors concurrent with behavior, but these measurements typically occur on the boundary around the agar. The exception is Figure 6 and Figure 1 – Supp 6 which measure directly opposite the surface of the agar, including when animals are present (Figure 1 – Figure Supp 6). We show how this creates difficulty for tracking animal behavior, which motivates our preferred approach to measure along the boundary during behavior.

In new experiments we further characterize the sensitivity range of the sensors across a 0-1000 ppm range (Figure 1 – Figure Supp 3). And we conduct new experiments to demonstrate use with oderants benzaldehyde and ethyl butyrate (Figure 2 c and e) in addition to the butanone and ethanol shown previously. We describe this in more detail in our response to requested Essential Revisions.

As proof of principle, the authors use this new method to characterize the behavior of C. elegans and D. melanogaster as a function of the concentration gradient encountered along their moving trajectories. In this respect, it remains unclear whether the method allows new conclusions or simply recapitulates previous observations.

Some of our conclusions are new: for example, while the parallel use of both weathervaning and biased random walk strategies had been previously well characterized for other forms of chemotaxis, such as salt chemotaxis (Iino et al. 2009, Luo et al. 2014), to our knowledge we are the first to characterize navigation in airborne butanone. We now state:

“We conclude that C. elegans utilize both biased random walk and weathervaning strategies to navigate butanone airborne odor landscapes (Iino et al. 2009, Yoshida et al. 2012). Both strategies had been observed for salt chemotaxis before, but to our knowledge had not previously been quantitatively characterized for airborne butanone. More broadly, previous measurements of navigation in response to airborne odors for C. elegans have not had access to the underlying odor concentration and therefore could only use proxies, such as the bearing angle or distance with respect to a point odor source. Our approach allows a more direct description of the navigation strategy because we characterize turn probabilities and other features with respect to the local gradient directly (FIGURE 7), as opposed to a proxy.”

We also added new experiments that for the first time characterize the changes to butanone navigational strategy induced by defects in interneuron AIB. AIB’s role in weathervaning had previously been studied in the context of salt chemotaxis, but not in the context of butanone olfactory guided navigation. We now clarify:

"To identify potential neurons involved in mediating the navigation strategy, we measured butanone guided navigation of mutant animals in which the interneuron AIB was chronically inhibited. Disruptions to AIB had previously been shown to either decrease weathervaning (AIB ablation) (Iino et al. 2009) or decrease biased random walks (Luo et al. 2014) (AIB inhibition) when navigating in a salt gradient. But to our knowledge, AIB's role in influencing the animal's butanone-guided navigation strategy had not previously been measured. We observed that AIB-inhibited animals still climbed butanone gradients, but the worm relied on different behavioral strategies (FIGURE 8). The turn was still biased suggesting that the animal still used a biased random-walk strategy, although turning rate was overall lower. The animal showed down-regulated weathervaning. In wild-type (N2) animals the distribution of the curvature of the animal's trajectory depends on its orientation to the local odor gradient (FIGURE 7f, Figure 7 -Figure Supp 2a) indicating weathervaning, but in AIB-inhibited animals the distribution of trajectory curvature is much less dependant on orientation to the local gradient (FIGURE 8d, Figure 7 – Figure Supp 2b). Together this suggests that AIB is important for supporting turning rate and the weathervaning strategy, and demonstrates the potential for this approach to characterize mutants or investigate neural circuits underlying navigation."

We agree with the reviewer that an important aspect of this work has been to recapitulate previous findings using our new approach. This is intentional and consistent with its submission as a “Tools and Resources” article, which explicitly “do not have to report major new biological insights or mechanisms.” We address this further in response to the 6th point in the requested Essential Revisions.

Sensors:

It would be important to know more about the sensitivity range of the sensors. It seems that the concentration used here does not saturate the sensors: what is the full scale of sensitivity for the x-axis in Figure 1- S1? Is the calibration curve similar for ethanol? And what kinds of odors are expected to be detectable?

Another point that is not mentioned is whether the sensor itself affects the odor stimulus: for example, through an absorption/release mechanism similar to what happens with the agar: I guess one should compare Figure 2a to Figure 4b? No interference with the stimulus would be a clear advantage of this approach over the PID and should be stated.

As described in response to requested “Essential Revisions'', we now included new measurements to characterize the metal-oxide based odor sensors across a range of odor concentrations up to 1,000 ppm (Figure 1 – Figure Supp 3). We also characterize their temporal response profile and their stability (Figure 1 – Figure Supp 2). We now perform new measurements to compare the metal-oxide sensors to PIDs separately for four odorants that span different water solubilities and vapor pressures (Figure 2 – Figure Supp 1): butanone, benzaldehyde, ethyl butyrate, and ethanol, as described in response to the third requested Essential Revision. We now clarify:

“To our knowledge, most volatile organic compounds that can be sensed and calibrated by a PID should be compatible with our sensor array system.”

In new measurements shown in Figure 1 – Figure Supp 3b we observe that the downstream odor concentration measured by the PID is not affected by powering on or off the metal oxide odor sensors. This suggests that the metal oxide sensors are not depleting the odorant during measurements. We agree that metal oxide sensor’s lack of interference with the stimulus is an advantage over the PID.

Behavioral analysis:

Regarding the paragraphs on C. elegans and D. melanogaster, I would suggest that the authors clarify what is a new finding vs what is already known, and in which cases the measurement of the odor gradient is critical.

Clearly, Figure 7 requires such measurements; however, the significance of the result is somehow obscure: what do we expect for the relationship between drift velocity and gradient? Also, there seems to be a very small and possibly not significant (I do not find a statistical test) positive drift for the high range of the tested gradients: could these gradients be too shallow?

We now clarify what is a new finding and what has been known for worm chemotaxis:

“We conclude that C. elegans utilize both biased random walk and weathervaning strategies to navigate butanone airborne odor landscapes (Iino et al. 2009, Yoshida et al. 2012). Both strategies had been observed for salt chemotaxis before, but to our knowledge had not previously been quantitatively characterized for airborne butanone. More broadly, previous measurements of navigation in response to airborne odors for C. elegans have not had access to the underlying odor concentration and therefore could only use proxies, such as the bearing angle or distance with respect to a point odor source. Our approach allows a more direct description of the navigation strategy because we characterize turn probabilities and other features with respect to the local gradient directly (FIG7), as opposed to a proxy.”

We have also added a new experiment studying navigational strategy of an AIB defective mutant resulting in another new finding:

“To identify potential neurons involved in mediating the navigation strategy, we measured butanone guided navigation of mutant animals in which the interneuron AIB was chronically inhibited. Disruptions to AIB had previously been shown to either decrease weathervaning (AIB ablation) (Iino et al. 2009) or decrease biased random walks (Luo et al. 2014) (AIB inhibition) when navigating in a salt gradient. But to our knowledge, AIB's role in influencing the animal's butanone-guided navigation strategy had not previously been measured….”

We have added more statistical rigor in figure 9 (the original figure 7). We now perform t-tests for each bin against the null-hypothesis (zero drift velocity) and found significant differences for every bin other than the first. Given the non-Gaussian distribution, we also performed two-sample Kolmogorov-Smirnov between bins and found that they are all significantly different from the lowest drift velocity bin. These statistical tests are incorporated in the revised figure 9 caption:

“Other than the left-most bin with the lowest gradient, all other bins show significant difference from zero drift velocity using t-test (p<0.001, ***). Both two-sample t-test and Kolmogorov-Smirnov test (given the non-Gaussian distribution) show significant differences between bins and the first bin with the lowest concentration gradient (p<0.001).”

We do not find the amplitude of the drift velocity to be shallow compared to our expectations of common worm chemotaxis kinetics: The average arrival time from the center of the plate to droplet source 5 cm away is normally on the order of ~10 minutes. Our measured mean drift velocity reported in Figure 9 is ~0.1 mm/s which corresponds to ~ 8 minutes and is roughly consistent with what we would expect. We think that this measurement has a wide spread due to the stochastic behavior in worm chemotaxis, and can be compared to similar saturating drift velocity in bacterial chemotaxis (Mattingly et al. 2021). It is interesting to consider whether the drift velocity tuning to local gradients might be dependent on the gradient distribution of the environment. Investigating how behavioral performance changes as a function of the gradient statistics would be an interesting future direction.

Is the same analysis of drift vs gradient not possible with D. melanogaster? Moreover, I wonder why the turn rate for -90 and 90 degrees do not have similar values: both directions are perpendicular to the gradient, wouldn't one expect the same behavior? The heading change, in that concern, is as expected, with similar absolute value and opposite direction. Also unclear why the heading change is similar for 180 and 0 degrees.

We hesitate to apply the drift vs gradient analysis to Drosophila larvae because their runs have a much longer persistence length than C. elegans which could confound the interpretation given the relative small size of our arena. We chose to optimize the arena size for C. elegans. We added Drosophila experiments later to demonstrate compatibility. If one was interested in a dedicated instrument for Drosophila, one could consider fabricating a larger arena (~30cm, as we used previously in Gershow et al., 2012) with identical design principles and larger odor sensor bars.

We also expected the turn rates of Drosophila larvae to be the same for -90 and 90 degrees; that they differ likely reflects the limited size of the data set used in this analysis. The average heading change is signed, so a non-zero average represents a directional bias (e.g. positive average heading change represents a bias towards the counter-clockwise or left-hand direction). We do not expect a directional bias for the 0 or 180 degree directions, and therefore expect both to be nearly 0, in agreement with the data presented in Figure 10b.

Author details

1. Kevin S Chen

   Princeton Neuroscience Institute, Princeton University, Princeton, United States

   Contribution

   Software, Formal analysis, Investigation, Visualization, Methodology, Writing – original draft

   Contributed equally with

   Rui Wu

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-8792-4625

2. Rui Wu

   Department of Physics, New York University, New York, United States

   Contribution

   Resources, Software, Validation, Methodology, Writing – review and editing

   Contributed equally with

   Kevin S Chen

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0009-0008-8707-4075

3. Marc H Gershow

   1. Department of Physics, New York University, New York, United States
   2. Center for Neural Science, New York University, New York, United States

   Contribution

   Conceptualization, Resources, Formal analysis, Supervision, Funding acquisition, Visualization, Methodology, Project administration, Writing – review and editing

   For correspondence

   mhg4@nyu.edu

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-7528-6101

4. Andrew M Leifer

   1. Princeton Neuroscience Institute, Princeton University, Princeton, United States
   2. Department of Physics, Princeton University, Princeton, United States

   Contribution

   Conceptualization, Supervision, Funding acquisition, Writing – original draft, Project administration, Writing – review and editing

   For correspondence

   leifer@princeton.edu

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-5362-5093

• 391

  Page views

• 87

  Downloads

• 0

  Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.