A microfluidic device to study neuronal and motor responses to acute chemical stimuli in zebrafish

Zebrafish larva is a unique model for whole-brain functional imaging and to study sensory-motor integration in the vertebrate brain. To take full advantage of this system, one needs to design sensory environments that can mimic the complex spatiotemporal stimulus patterns experienced by the animal in natural conditions. We report on a novel open-ended microfluidic device that delivers pulses of chemical stimuli to agarose-restrained larvae with near-millisecond switching rate and unprecedented spatial and concentration accuracy and reproducibility. In combination with two-photon calcium imaging and recordings of tail movements, we found that stimuli of opposite hedonic values induced different circuit activity patterns. Moreover, by precisely controlling the duration of the stimulus (50–500 ms), we found that the probability of generating a gustatory-induced behavior is encoded by the number of neurons activated. This device may open new ways to dissect the neural-circuit principles underlying chemosensory perception.


Supplementary information
Fluid dynamics of chemical delivering Stationary flows Our microfluidic system has two stationary states, namely resting and injection, depending on the pressure applied to the solution reservoirs. In this section, we derive analytical expressions for pressure and volumic flow rates in all channels for both states, which provides a complete description of the flows inside the chip.
The hydrodynamics in the chamber channels and the delivery circuit are rather decoupled, such that these circuits can be separated. The microfluidic delivery circuit can be synthesized by the following diagram, where P , q and R stand for pressure (all expressed as offsets to the atmospheric pressure), volumic flow rate and hydraulic resistance:

Injection channels
Delivery channel

Stimulus circulation channels
Supplementary diagram 1. Equivalent diagram of the delivery circuit. Pressures are defined at the nodes and each channel is described by its hydraulic resistance R and flow rate q. Arrows indicate the direction of positive flow rates, defined as the direction of fluid in the resting state.
The pressures of the stimuli reservoirs p A and p B and of the waste p w are directly controlled by the operator, while the pressure at the end of the delivery channel p D is very close to atmospheric pressure 1 . The pressures p a , p b and p y are set internally. For each of the seven segments the Hagen-Poiseuille relation applies : where the hydraulic resistance R of a rectangle channel of width w, height h and length L is defined by : with µ = 10 −3 P a.s the dynamic viscosity of water. In units of 10 8 mbar.s.m −3 , the hydraulic resistances are R 1 = 170, R 2 = 132, R 3 = 320 and R D = 1.3. We obtain three more equations with flow conservation: 1 p D is controlled by the flow rate in the chamber such that p D = −ρwq 2 ch /2S 2 ch , with ρw the volumic mass of water, q ch = 2µL.s −1 the flow rate in the chamber and S ch = 4.10 −8 m 2 the surface of the cross-section of the chamber channel, which gives p D = −0.05 mbar.
The system of ten linear equations in eq. 1 and eq. 3 can be solved. Introducing the reduced quantities α = R3 R D , γ 1 = R2R3 R1R2+R2R3+R1R3 , γ 2 = R1R3 R1R2+R2R3+R1R3 and γ 3 = R1R2 R1R2+R2R3+R1R3 , the internal pressures write: The pressures and flow rates in the chamber circuit can be calculated in a similar way. All pressures and flow rates are displayed on Supplementary Figure 6, in both the resting and injection states.
The flow rate at the outlet of the delivery channel q D follows an affine dependence with the pressure applied in the stimulus reservoir : q D = a.p A + b, with a = −1.10 × 10 −11 m 3 .mbar −1 .s −1 and b = 1.42 × 10 −9 m 3 .s −1 . At the typical pressure at which recordings have been performed (p A = 500 mbar) we found q D = −4.08 µL.s −1 . Conversely, the reservoir pressure at which the outlet flow becomes negative, i.e. the minimal pressure for which the system is injecting stimulus in the chamber, is at p min A = 128 mbar. This is coherent with the minimal pressure found experimentally, namely 125 mbar.

Comparison with standard injection methods
In this section, we discuss how the dynamical aspects of our system compares with methods that have been most commonly used for delivering chemicals to aquatic species.

Methods with uncontrolled advection/diffusion
A class of methods is based on local concentration increase, either via a solid media (e.g. soaked cotton pellets [1,2], freeze-dried pellets [3], starch gel [4], agar gel [5]) or by directly releasing drops of the product in the sample tank [6].
The transport of chemical compounds involves two distinct mechanisms, advection and diffusion [7]. In most cases, advection is a much faster transporting process, and diffusion can be neglected. In the previously cited methods, advection processes are not controlled (flows created in the fluid during product release, movements of the animal, thermal convection, etc.) which introduce a huge variability in the front propagation speed and important concentration heterogeneity. These methods are viable in studies where temporal aspects or control of the concentration are not important, but a precise understanding of the gustatory neuronal processes requires a better level of control. Hence, to generate sharp and reproducible chemical stimulations, it is crucial to control the advection processes around the specimen.
In addition, in advection-based delivery systems where the chemical is stored very close to the sample, care should be taken to eliminate diffusion-based cross-pollution. Indeed, diffusion should be avoided for two reasons: (i ) since the propagation front spreads with time, the concentration at a given point slowly increases which makes it difficult to define an onset time of stimulation, and (ii) since diffusion is an irreversible process, without proper cleaning of the chemoreceptors the delivery cannot be repeated on the same animal, which hampers multitrials averaging. Diffusion is a rather slow process 2 , but can appear during the course of a experiment. In our microfluidic device, the distance between the stimuli circulation channels and the larva's mouth is ≈ 375 µm, so a front of stimulus could reach the sample in a few minutes. Our system has thus been designed such that in the resting state a permanent flow is maintained in the injection and delivery channels to suck water from the chamber (q x , q y and q D are positive), which completely eliminates diffusion-based cross-pollution, while the flow around the sample is unchanged between the resting and injection states.

Valve-driven injection
Common advection-based methods involve a switching valve system (see e.g. [9, 10] for the gustatory system or [11,12,13] for the olfactory system). In these devices a continuous flow is set into an injection tube, and the fluid (buffer or stimulus) is switched at the inlet of the injection tube. The valve prevents cross-diffusion between the buffer and the stimulus solution and ensures a constant flow rate around the specimen. In this section, we compute the evolution of the concentration at the outlet of such an injection tube and show that it has a slow evolution, which we compares to measurements in our device.
Let us consider a cylindrical 3 tube of length L and internal radius ρ in which a continuous flow is imposed at a constant flow rate q (see Supplementary diagram 2). We also assume that at time t = 0, solution A 2 For instance, citric acid has a diffusion coefficient of D 0.65 × 10 −9 m 2 .s −1 in water [8], which gives a propagation of the diffusion front in ≈ 15 s for a distance of 100 µm, and ≈ 385 s for a distance of 500 µm. 3 Similar results are found with rectangular channels, though the calculus is more complicated. See [14] for details.
where v 0 is the maximum velocity, at the center of the tube. With this expression the flow rate can be derived as a function of v 0 : At t = τ , the front tip reaches the outlet of the injection tube. This time is defined by : Let us note C A and C B the concentrations of solutions A and B. We aim at computing the time-evolution of the average concentrations c A (t) and c B (t) at the tube outlet : For t < τ , we have : c A (t) = C A and c B (t) = 0.
For t ≥ τ , the concentrations are given by : where r f (t) is the radius of the cross section of the front at the outlet of the tube. The radius r f (t) corresponds to a velocity of exactly v(r f (t)) = L/t, which gives, by using the expression of the Poiseuille profile in eq. 5 : and, finally : The evolution of the concentrations at the outlet in t −1 imposes a slow-exchange dynamics, and the system continues to deliver a significant concentration of the initial compound A for a long time after offset. Given the sensitivity of gustatory chemoreceptors 4 , this implies that sensory activation can still occur even at a very long time after the valve switch have replaced the stimulus with the buffer. For instance, with a tube of length 10 cm, inner radius 1 mm and a flow rate of 10 µL.s −1 , we get τ = 15.7 s and after 315 s the liquid presented to the specimen would still contain 5% of the original solution. To significantly decrease the characteristic time τ , it is necessary to reduce the size of the injection channel. Reducing the channel radius ρ has a strong limitation, since at a given flowrate the bulk velocity v 0 will increase quadratically and rapidly exceed the limit of physiologically relevant values (≈ 50mm.s −1 ). So if ρ decreases, q has to be decreased accordingly, such that there is no net effect on the timescale τ . The only acceptable strategy is therefore to reduce the length of the delivery channel L. By using microfluidic devices, typical values of L decrease from ≈ 10 cm to ≈ 100 µm, yielding three orders of magnitude in τ . As shown in Supplementary Figure 7-b, in our system the onset of stimulus presentation follows the slow evolution of eq. 9 with a typical time τ ≈ 10 ms.
Finally, one important asset of our microfluidic device is that it completely eliminates the stimulus in a few tens of milliseconds at injection offset, as shown in Supplementary Figure 7-b. This is due to the inversion of flow direction in the injection and delivery channels, which stops any arrival of the stimulus in the chamber and eliminates the stagnant volume at the outlet of the delivery channel, combined with the continuous washing of the specimen with clean buffer. Altogether, our system allows for the delivery of pseudo-square pulses at an almost constant concentration with unprecedented steep-rising and falling edges and it allows for a complete cleaning of the specimen in a few tens of milliseconds after stimulus presentation. The high-level of control over advection offered by microfluidic systems is used here to (i ) deliver several pulses of stimuli in a very reproducible fashion to the same animal, which greatly facilitates response-averaging across trials, and (ii ) deliver pulses as short as 10 ms, which is impossible with standard techniques. Finally, as the delivery channel has an extremely short length (100 µm) it is cleaned very fast and be used for the delivery of another chemical in a very short delay following the first stimulus.