The primary goal of this study was to establish an Arduino-based behavioral platform, which can be easily utilized for a variety of behavioral tasks with head-fixed rodents. The basic components of the platform for an operant conditioning behavioral task include: (1) a delivery system for water reward; (2) a lick detection sensor; and (3) a sensory stimulation apparatus. In our system (Figure 1), the Arduino microprocessor is the central hub which operates these components. The Arduino repeatedly monitors the signals from the lick sensor and controls the timing of water and stimulus delivery. As required, the Arduino sends a TTL trigger to other external devices to synchronize recordings and the animals’ behavior. A host computer regularly communicates with the Arduino via a serial connection to update the behavioral settings and readout the behavioral data logged in the Arduino. The source code for this system can be found online at https://github.com/palmerlab/behaviour_box, as well as additional documentation at https://palmerlab.github.io

During operant conditioning, a subject’s operant behavior (in this case, lick response to a sensory stimulus) is reinforced by a reward. For head-fixed animals, the water reward is delivered through a waterspout located within reach of the animal’s mouth using a gravity flow water system. The reward delivery is controlled by a solenoid pinch valve (12 V DC, Takasago Fluidic Systems, Nagoya, Japan) and the volume delivered is determined by the duration of the valve opening and the height of the gravity water system. In our behavioral paradigm described below, mice receive approximately 10 μl water reward each correct trial.

Licking frequency is monitored using a custom-made piezo-based lick sensor. This consists of a piezoelectric wafer (0.6 mm Range Piezo Bender Actuator, PiezoDrive Pty Ltd, Callaghan, NSW, Australia) glued along the shaft of the waterspout (blunted 18-Gauge syringe needle). A tongue lick to the spout induces a small deformation of the wafer, which generates a voltage change inside the piezoelectric element. This signal is then amplified using a simple operational amplifier circuit (LM358N, Texas Instruments) to ensure the Ardunio microcontroller detects the lick-evoked voltage changes.

The Arduino Uno offers 13 digital input/output pins, as well as 6 analog to digital inputs with 4.8 mV resolution (0–5 V, 1024 bits; Figure 2). To operate the operant conditioning Go/No-Go task, the following analog and digital inputs/outputs are necessary (see Figure 2 for a schematic of the wiring diagram). Analog inputs: the amplified signal from the lick sensor is sent to an analog input pin. Digital outputs: four digital output pins are connected to: (1) sensory stimulator; (2) punishment consisting of a TTL-triggered valve gating a pressurized air line; (3) water valve; and (4) recording trigger.

The operant conditioning program running on the Arduino continuously reads the analog signal and detects individual lick events when the reported piezoelectric voltage signal crosses a given threshold. This sensing function runs in a loop, adding to a counting variable each time a lick is detected. The Arduino tracks the time and initiates the scheduled events for the task (e.g., sensory stimulation, opening the water valve, etc.).

The Arduino supports two-way serial port communications via an USB interface to a host computer. A Python script, (Python version 2.7.10) written around the pyserial library, is used to handle the sending and receiving of messages. The status of each experiment is continuously monitored and communicated between the Arduino serial port and the host computer. This requires the digital pins 0 and 1 to be unassigned in the Arduino code as these carry the relevant signals. The full set of global variables (listed in Table 1), excluding pin out assignments, is available for updating. In addition, during a trial, the status of each trial event is reported to the host computer. Successfully updating a variable results in the Arduino sending a message echoing the new variable and value to the serial port. In this way the system creates a running log of all settings and changes as they occur. All communications from the Arduino conform to yaml¹ specification, with debug messages about trial events commented out. Yaml is a useful format for serialized data which emphasizes human readability. This self-documenting system was implemented because the stimulus timing and reward condition is updated on a trial-by-trial basis.

The Python script is used to initiate each Go/No-Go task and therefore controls the timing of individual trials. After initialization, the Arduino defaults to a listening mode where it parses any variables that are transmitted. It will remain in this mode indefinitely until it receives the message “START” from the host computer, which triggers the initiation of a trial. This has the advantage of allowing the experimenter to pause the trials and make quick changes to the parameters without the need for touching the Arduino code, in turn avoiding recompiling.

The Arduino microcontroller is loaded with a custom written program² which consists of two main functions; a basic Habituation mode, and a slightly more complex Operant mode. Here we discuss how to run these different behavioral training modes using the system described above (Figure 3). All procedures were approved by the Florey Institute of Neuroscience and Mental Health Animal Care and Ethics Committee and followed the guidelines of the Australian Code of Practice for the Care and Use of Animals for Scientific Purposes.

The behavioral platform described here can be easily combined with various existing techniques/devices for recording neural activity as the Arduino processor can be used to trigger other software/hardware which support external trigger modes (EXT Trigger pin, Figure 2). Here, the EXT trigger sends a HIGH TTL signal at the start of a trial and remains on for the entire duration. The rising edge of this signal can be used to trigger episodic recordings, or the full signal may be used for devices that support variable length recordings. To prove the utility of this system, we performed two-photon calcium imaging from cortical pyramidal neuron dendrites while the animal was engaged in a reward-based sensory task. Mice expressing a genetically-encoded calcium indicator, GCaMP6f (AAV1.Syn.GCaMP6f.WPRE.SV40), in layer 2/3 pyramidal neurons in the primary somatosensory cortex were implanted with a chronic window. In brief, animals were anesthetized with isoflurane vapor (0.5% in 0.5 mL/min O₂) during all surgery procedures and administered with lidocaine and meloxicam for local and general analgesia respectively. A small incision was made in the scalp, and using the stereotactic coordinates as a guide, a small craniotomy was made above the somatosensory cortex. The virus was delivered via a glass pipette, which had been backfilled with the virus aliquot and silicon hydraulic oil. A hydraulic piston was used to precisely deliver 100 nL of virus to the target site, 200–300 μm below the dura. After an incubation time of 7–10 days a head post and chronic window surgery was performed. Under isofluorane anesthesia (0.5% in 0.5 mL/min O₂), the soft tissue of the scalp was removed. A small aluminum post was glued to the skull, using cyanoacrylate glue (Loctite 401, Henkel Australia Pty. Ltd., Sydney, VIC, Australia), and fixed firmly in place with C&B metabond® (Parkell Inc, Brentwood, NY, USA). A 3 mm circular craniotomy was made around the injection site, and a glass coverslip was fixed in place. The edges were sealed with cyanoacrylate glue, and dental cement (Jet Denture Repair, Lang Dental, Wheeling, IL, USA) was used to make a small well for the water-immersion lens (see Holtmaat et al., 2009).

After a recovery period of at least 3 days, the mice were acclimatized to head-fixation through several sessions of gradually increasing head restraint; beginning with momentary catch-and-release, and leading to 5–10 min restraints. During this time, and throughout the period of the experiment, animals had their access to water restricted to 1 mL/day. Their weight was monitored daily to ensure they did not drop below 80% of their pre-restriction mass. The animals were then introduced to the experimental rig, and trained to associate licking behavior with a reward, and stimulus, through the “habituation” mode described previously. Subsequently, they were trained to perform the Go/No-Go behavioral task as described above, in the “operant” mode. Here, the sensory stimulation was delivered to the contralateral forepaw using a small button (linear resonance actuator, precision microdrives, London, UK) and the ability of the mouse to detect the forepaw stimulation was tested by using stimulation of different durations (0, 5, 25, 50, 100, 150, 200 and 300 ms). Mice were able to reliably detect and report forepaw stimulation durations greater than 100 ms (Figure 4A). The great advantage of the Go/No Go behavioral paradigm over and above other complex reward-based behaviors is that it is rapidly leant. Mice typically reached a criterion of successful behavior (usually >80% “hit” and <20% false alarm rate, d′ > 1.5) after four sessions (Figure 4B). Using a two-photon microscope, we imaged the dendritic calcium activity of GCaMP6f-expressing layer 2/3 neuron dendrites in the forepaw area of the somatosensory cortex, while the animal was performing the Go/No-Go task (Figure 5A). Large Ca²⁺ transients occurred both spontaneously and during the Go/No-Go task in layer 2/3 pyramidal neuron dendrites (Figure 5B), with a greater number of large Ca²⁺ transients occurring after the initial stimulus delivery (Figure 5C). Using the behavioral platform to drive the sensory stimulation and behavioral recording enables direct correlation and association between neural activity and behavior, and opens the door to address many unanswered questions about the neural basis of behavior.