Scalable and modular wireless-network infrastructure for large-scale behavioural neuroscience | Nature Biomedical Engineering

Concept and operation principle of WNBN for high-throughput in vivo neuroscience research

Figure 1 illustrates the operational concept of the WNBN ecosystem. The WNBN system integrates wireless neural devices6,7,19,20 and other laboratory equipment with Bluetooth Low Energy (BLE) and Internet technologies to enable semi-automated and fully automated behavioural studies, respectively. The WNBN can either be controlled locally through BLE piconets (ad hoc network that connect the primary node and multiple secondary nodes using Bluetooth technology21) using a smartphone (Fig. 1a), or can be accessed globally over the Internet (Fig. 1b; see Table 1 for detailed features and comparison to existing wireless technologies). Use of BLE for both options enables selective simultaneous control of multiple devices (Supplementary Fig. 1). This integrated approach exploits the unique features of both BLE and Internet protocols, which allow energy efficient control of many different neural devices, behavioural assays or other relevant lab equipment in the vicinity (~100 m) and remote access to the large local networks over the Internet for high-throughput control, respectively. More specifically, the BLE piconet is useful for rapid creation of traditionally small experimental cohorts (~15 devices) with selective device and output controls, particularly when the experimenter is physically present to conduct the experiment. On the other hand, the Internet protocol and webserver on the minicomputer (Raspberry Pi 3 Model B, Raspberry Pi Foundation; Supplementary Fig. 2) allow researchers to control and receive feedback from many more devices in multiple piconet networks both selectively and simultaneously from any remote, Internet-connected part of the world (Fig. 1b–e). Using this scheme, researchers anywhere in the world can securely connect to the WNBN of neural devices, behavioural apparatuses or other lab equipment located inside a minicomputer-equipped lab. Multiple credentialed users with appropriate administrator permissions can log on to a secure private server using a browser on a PC, tablet or smartphone to set up desired experiments. Versatile graphical user interfaces (GUIs) of both the smartphone app19 and the webserver facilitate this process by enabling manipulation of control parameters for multiple simultaneous experiments (Supplementary Figs. 3 and 4, and Videos 1 and 2). The entire system is economical and portable, only requiring commercial hardware in the form of a smartphone for local control or a minicomputer for global remote access.

Fig. 1: Concept and operation principle of WNBN for high-throughput in vivo neuroscience research.

a,b, Remote control modes (local vs global) for WNBN control. For local wireless control (a), a commercial smartphone helps form an individual BLE piconet network, which allows communication with multiple devices simultaneously and/or selectively in its vicinity (<100 m). For global Internet control (b), remote researchers can use any device with an Internet connection to access the custom WNBN server, which is hosted by a local minicomputer inside each laboratory. Multiple users at multiple distinct locations can remotely and simultaneously communicate with various WNBN-enabled devices deployed in multiple laboratories around the globe. Both local (a) and global (b) control schemes can be further extended to conventional tethered equipment by connecting them to off-the-shelf BLE modules. c, Device triggering modes (real-time vs scheduled) for WNBN control. The WNBN Cloud allows multimodal and simultaneous control of neural devices either in real time or scheduled to a specific time in the future. Irrespective of the triggering modes, all controls are selective (controlling a specific group of animals within a large cohort) and simultaneous (control multiple target animals simultaneously). d, Schematic diagram illustrating bidirectional control and sensing capability, which allows not only sending control signals to various neural implants and laboratory tools, but also receiving data from various sensors and feedback systems. Automatic periodic gathering of data from the animal implants and/or laboratory tools can provide insightful information on animal experiments. e, Schematic diagram illustrating capability of selective large-scale control, which allows simultaneous and independent manipulation of various types of neural devices (that is, both emerging wireless and conventional tethered devices) within limited laboratory spaces for high-throughput neuroscience research.

Full size image

Table 1 Comparison of wireless control schemes

Full size table

This WNBN infrastructure allows the user to remotely set up individual, simultaneous or scheduled sequences of commands (Fig. 1c and Supplementary Video 3). This enables experimental parameters to be sent instantly in real time or scheduled at a specific date and time. These commands are received over the Internet by a BLE-enabled minicomputer. Once the received command is processed and the time of triggering is reached, BLE modules in the minicomputer send this command selectively to the large-scale network of wirelessly connected animals implanted with BLE-enabled neural devices or BLE-enabled traditional equipment. In addition, the WNBN infrastructure supports bidirectional control and sensing of information between the remote user and the devices used in various laboratory experiments (Fig. 1d). This capability can be used to check the current status of environmental parameters (such as temperature, pressure and humidity) before remotely triggering a sensitive behavioural experiment. The data-sending and receiving ability can be further extended to form multiple closed-loop control systems, where the WNBN ecosystem, based on received environmental, behavioural or neural data, will send a command signal to a pre-specified set of devices to perform automated actions (for example, photostimulation, drug delivery, etc.). Lastly, the WNBN ecosystem leverages this unique ability to notably enhance experimental throughput by allowing a remote user to have simultaneous selective control over a variety of experiments within a laboratory, as well as across multiple laboratories (Fig. 1e). In summary, all these functions of the WNBN infrastructure – remote control, selective and simultaneous control, bidirectional communication and device-independent scalable manipulation of both wireless and conventional tethered equipment – are powerful features that can address needs within the neuroscience community for efficient, labour-saving technology and can greatly facilitate high-throughput in vivo neuroscience research.

WNBN architecture and signal flow

Figure 2a,b highlights the network architecture of the WNBN controls, compared to conventional Bluetooth-based schemes. Conventional Bluetooth controls rely on one-to-one node control (Fig. 2a(i),b(i)). In this simple architecture, a single wireless transmitter or controller (that is, remote control centre (RCC)) can wirelessly control a single wireless receiver (that is, remote control module (RCM)). Bluetooth single device control technology provides unique advantages, such as a large wireless control range, no line-of-sight and obstacle handicaps, and low cost and setup time for at-large implementation across laboratories. However, the low-throughput efficiency of this single device control scheme makes it unsuitable for large-scale in vivo neuroscience.

Fig. 2: The WNBN architecture and signal flow.

a, Block diagram of network topologies to highlight topological differences and signal flow between (i) conventional Bluetooth single device control and (ii,iii) WNBN multidevice control modes, where RCC, RCM and P stand for remote control centre, remote control module and piconet, respectively; (ii) local piconet WNBN control through a single RCC and (iii) global Internet WNBN control through multiple piconet networks (P1, P2, P3, …, and Pm), all of which are connected to a single RCC. b, Illustration of capabilities and access range of different technologies. (i) Single device control, where 1 RCC can control only 1 RCM at a time in its vicinity using BLE; (ii) local piconet WNBN mode, where a single RCC can selectively and simultaneously control up to 15 independent RCMs within its vicinity using BLE; and (iii) global Internet WNBN mode, where a single remotely located RCC can selectively and simultaneously control multiple devices across the globe at different laboratories (equipped with a minicomputer loaded with a custom WNBN operating system) using a combination of Internet and BLE technologies. The WNBN technology can control not only multimodal neural implants, but also tethered conventional equipment while gathering data from multiple analogue and digital sensors at the same time.

Full size image

To overcome this limitation, we developed the WNBN technology, including operation modes for both local piconet (Fig. 2a(ii),b(ii)) and global Internet control (Fig. 2a(iii),b(iii)) to enable seamless scalability and ease of implementation across laboratories, as well as devices around the globe. In local piconet WNBN control, we employed a single RCC device to simultaneously control multiple wireless RCMs (up to 15 devices at a time) in its vicinity (<100 m). In simple terms, it can be visualized as a star network of multiple RCMs forming multiple single device control connections with a single central RCC (Fig. 2a(ii)). Using BLE wireless protocols, we developed the local control system, which synchronizes multiple data streams by the frequency hopping spread spectrum technique, where it multiplexes the serial data flowing from a single RCC to multiple local RCM receivers. Various receivers in the local piconet WNBN are particularly distinguished from one another through the unique 48-bit Bluetooth device address of each RCM, which is available to RCC upon establishment of a local piconet WNBN. We used this identification technique to enable local piconet mode to control devices simultaneously and selectively with 100% accuracy (Extended Data Fig. 1), in comparison to contemporary technologies such as RF8,9,10,11,13,22,23 or IR technologies7,24, which do not have selective capabilities, leading to potentially limitless inaccurate pairings. Moreover, the communication of RCC with a specific RCM in a local piconet network is completely independent of its communication with other RCMs in the network. We leveraged this ability to enable the user to have complete control of which specific devices to manipulate simultaneously within a group, as well as which target functionalities to implement within each selected device. For example, with this feature, a single RCC can send multiple unique commands to 4 specific RCMs (1, 5, 10, 15) simultaneously within a piconet network of 15 RCMs – for instance, release of a drug using a microfluidic neural device implanted in mouse 1 (RCM 1), optogenetic excitation using a conventional tethered optical fibre implanted in mouse 5 (RCM 5), triggering of a laboratory centrifuge (RCM 10) and simultaneous delivery of light and drug using an optofluidic device implanted in mouse 15 (RCM 15). Hence, the selective and simultaneous control capability of local piconet WNBN technology enables versatile operation not only at device level (devices within a group), but also at functional level for independent control of each individual feature or modality within a device (for control of multimodal devices). The local piconet WNBN can be easily established and controlled using a readily available smartphone and experiments can be rapidly scaled to several devices with minimal setup time and effort.

However, for control over very large-scale experiments across tens or hundreds of animal subjects and equipment or devices within a laboratory, the local piconet control starts reaching its access and scaling limit. To address this limitation, we implemented the scalable Internet control scheme by establishing a star network of multiple local piconet WNBNs, each with multiple wireless receivers (RCMs), connected to a single remotely located transmitter (RCC) (Fig. 2a(iii)). This scheme allows the Internet control to scale across various RCM devices within a laboratory and across multiple laboratories to enable seamless collaboration within or among research groups around the globe (Fig. 2b(iii)). Each RCC in this case can be globally accessed to control multiple piconet networks over the Internet. We used a local minicomputer (Raspberry Pi 3 Model B Quad-Core Broadcom 64 bit ARMv8 1.2 GHz, Raspberry Pi Foundation; Supplementary Fig. 2) with the Internet access as an RCC, which communicates the data to multiple BLE dongles (BLED112, Silicon Labs), each of which then transfers the data wirelessly across local RCMs in the vicinity over BLE. The global Internet WNBN control network demonstrates similar selective and simultaneous capabilities (using BLE protocols) as that of local piconet WNBN. By increasing the number of BLE dongles in the minicomputer, the synchronized hopping and time division multiplexing techniques can be parallelized such that each individual piconet sub-network can wirelessly, selectively and simultaneously control numerous devices at a time. The number of controllable devices scales by increasing the number of connected BLE dongles. Each individual BLE dongle can support up to eight connections. Our custom software driver quickly adapts data multiplexing algorithms and wireless data streaming on the basis of the number of dongles connected for efficient and reliable wireless communication to all end devices. Stimulation parameters for neuromodulation, sensing data obtained from animals or laboratories, or control signals for triggering of conventional tethered equipment such as lasers and imaging probes connected to a BLE module can be communicated over global WNBN control. After developing this WNBN ecosystem, we next sought to test its function in various behavioural neuroscience settings.

Reliability of WNBN systems for in vivo neuroscience

A wireless network ecosystem meant to enable remote neuroscience experimentation that reduces human interaction while maintaining oversight of the experiment must be rapid (millisecond timescale) and reliable. To ensure quality and latency of data communication between the remote user and wireless networks, we validated the reliability of both the local piconet and the global Internet controls for control of both wireless neural devices and conventional neuroscience tools. First, we characterized and compared the WNBN control latencies for selective and simultaneous control of one or more outputs in a real-time or scheduled format. For these studies, we designed and employed a minimalistic, yet highly customizable and rechargeable wireless optogenetic probe (Fig. 3a and Supplementary Fig. 5) as an exemplary RCM device. This BLE-enabled device can be easily controlled using the WNBN ecosystem with minimal cost and effort, while parallel designs can be 3D-printed to further minimize costs25. This device was equipped with bilateral neural probes, where each probe housed two independently controlled microscale inorganic light-emitting diodes (µ-ILEDs) of different wavelengths (470 nm and 589 nm). For wireless control, it was then integrated with a BLE wireless module and a light battery (0.3 g; lithium polymer (LiPo) battery, PowerStream) through modular assembly.

Fig. 3: Implementation of local and global WNBN controls for in vivo neuroscience experimentation.

a, Three major components of the modular and customizable wireless optogenetic system: a rechargeable LiPo battery, a soft µ-ILED probe and a programmable wireless BLE module (RCM). b, Latency of triggering single µ-ILED-based neural probes integrated with wireless RCMs, comparing the responsiveness of local piconet, local and global Internet modes (n = 10). Distance between RCC and RCM is 1 m. c, Latency of simultaneous triggering of dual μ-ILEDs in optogenetic neural probes integrated with wireless RCMs, comparing the responsiveness of local and global Internet modes (n = 10). Orange arrow in b and dark purple arrow in c indicate a global signal propagation delay, which is dependent on the inter-distance between the remote user and RCC (for example, 15 ms delay between Boulder, CO and St. Louis, MO, USA). Note that this propagation delay can be ignored in scheduled controls. Distance between RCC and RCM is 1 m. d, Top: sagittal brain cartoon of implantation of µ-ILED device into the PVH of agrpCre x Ai32 mice. Bottom: representative coronal image showing ChR2-eYFP expression (green) within the PVH. Scale bar, 100 µm. D, Dorsal, V, Ventral, M, Medial, L, Lateral. e, Cartoon depicting ad libitum feeding experiment with either piconet or Internet control. f,g, Amount of food consumed during local piconet-controlled (f) and local Internet-controlled (g) feeding of agrpCre x Ai32 mice, which consumed significantly more food during 20 Hz photostimulation compared with non-stimulation periods (1 h) (repeated measures one-way ANOVA; (f) Hour 1 vs Hour 2, **P = 0.008, Tukey’s multiple comparisons test, n = 7; (g) Hour 1 vs Hour 2, ***P = < 0.001, Tukey’s multiple comparisons test, n = 8). h, Photographs depicting set up of conventional device integration for Internet-controlled conventional optogenetic stimulation. BNC, Bayonet Neill-Concelman connector. i, Cartoon depiction of ad libitum feeding experiment using conventional optogenetic stimulation. j, Amount of food consumed during Internet-controlled conventional laser-coupled optogenetic stimulation of agrpCre x Ai32 mice, which consumed significantly more food during 20 Hz photostimulation compared with non-stimulation periods (1 h) (repeated measures one-way ANOVA, Hour 1 vs Hour 2, *P = 0.0126, Tukey’s multiple comparisons test, n = 9).

Full size image

Figure 3b,c, Extended Data Fig. 1 and Supplementary Videos 1–3 demonstrate the promptness and reliability of local piconet, and local and global Internet control modes for remote control of these wireless optogenetic devices (see Table 2 for summarized key performance metrics). Operation in all control modes is highly stable when a single RCM is paired to RCC, regardless of the distance between RCC and RCM (Extended Data Fig. 1a). These tests had zero failures (that is, 100% of success rate). Transmission was faster in the local piconet compared with the Internet controls (latency ~20 ms for the piconet control versus (~55 + α) ms for the Internet control where α represents a signal propagation delay between the remote user and RCC; Fig. 3b and Extended Data Fig. 1b). This increased latency for Internet control is mainly due to increased RCC–RCM distance, making the local piconet control more suitable for sophisticated experiments requiring higher temporal control resolution. For scheduled controls, RCC uses only the local network, thus the signal propagation delay can be completely eliminated. Therefore, using the scheduled control can make the latency difference between local and global Internet controls negligible. For simultaneous output or device control, the latency of additional devices receiving data increases since the frequency hopping algorithms need to find available frequencies for the additional devices (Fig. 3c). Overall, however, the differences are small and in most cases inconsequential. This is especially true for behavioural experiments with the time course of seconds or longer. Furthermore, the number of paired RCMs does not substantially affect the latency or success rate of BLE signal transmission (Extended Data Fig. 1c,d), except in the case of the local piconet where the number of allowable connections is limited to 15 by a smartphone. These results demonstrate the functionality and scalability of the WNBN control in reliably controlling numerous devices on millisecond timescales.

Table 2 Performance metrics of the WNBN system

Full size table

While benchtop testing provides clear performance hallmarks, integration with real neuroscience experiments requires proof-of-principle in vivo testing. To do so, we used multiple established animal behaviour models to test integration of WNBN infrastructure with in vivo neuroscience experiments. To demonstrate the effectiveness of WNBN-controlled manipulation, we used a well-validated feeding assay in which agouti-related protein (AgRP) expressing neurons were stimulated to drive food consumption in ad libitum fed conditions26,27. First, we implanted BLE-enabled blue µ-ILED devices lateral to the paraventricular nucleus of the hypothalamus (PVH) in agrpCre x Ai32 mice. These mice have Cre-dependent expression of the blue-light sensitive cation channel, channelrhodopsin-2 (ChR2) within AgRP-expressing neurons (Fig. 3d). Following recovery, ad libitum fed mice were given access to a food pellet (~3 g) for 3 h (Fig. 3e) and received 1 h of 20 Hz photostimulation in an off-on-off manner. Blue-light photostimulation wirelessly activated via local piconet (Fig. 3f) or Internet WNBN control (Fig. 3g) significantly increased food consumption above baseline levels, validating the reliability of both WNBN control modes for wireless neuromodulation. Importantly, WNBN control not only works with these specialized BLE-enabled wireless neural devices (Fig. 3a), but can also be used with conventional laboratory equipment (Fig. 3h). Following the same experimental protocol, we used the Internet WNBN control to deliver commands (20 Hz photostimulation) to typical transistor–transistor logic input/output ports on a diode-pumped solid-state (DPSS) laser tethered to fibre-optic-implanted mice, to drive the same increase in feeding behaviour (Fig. 3i,j). These proof-of-principle experiments verify that both control modes (that is, local piconet and Internet controls) of the WNBN system are equally reliable and highly versatile, allowing remote control of both wireless neural devices as well as conventional laboratory tools. This flexibility in downstream RCM targets optimizes laboratory adoptability while minimizing laboratory costs and setup efforts.

Simultaneous and selective control in WNBN systems for high-throughput neuroscience experimentation

To demonstrate the principles of WNBN control systems for multiple simultaneous experiments, we simultaneously targeted BLE-enabled blue µ-ILED devices to the secondary motor cortex (M2) of six mice expressing ChR2 under the Thy1 promoter (Fig. 4a)28. Here, following a baseline exposure to an open arena, smartphone-based local piconet control was used to simultaneously photostimulate (20 Hz) M2 to induce locomotor behaviour (Fig. 4b and Supplementary Video 4). This activation increased total locomotor activity (Fig. 4c), as well as induced rotation behaviour in a time-locked fashion (Fig. 4d). This local piconet-based control is the basis for expansion to scheduled and global control schemes. Using the unique identification of each BLE-enabled device in the WNBN, we can now program more complex interactive experiments, such as those including social interaction and/or those requiring selective control of discrete device features.

Fig. 4: Simultaneous and selective in vivo control in WNBN-controlled systems.

a, Top: sagittal brain cartoon of implantation of µ-ILED device into the M2 of Thy1ChR2 mice. Bottom: representative coronal image showing implant location and ChR2-YFP expression (green) within the cortex. Scale bar, 200 µm. b, Data showing changes in locomotor activity during piconet-controlled stimulation (20 Hz) in Thy1ChR2 mice. c,d, Average distance travelled (c) and average rotations (d) during local piconet-controlled stimulation intervals (c: two-tailed paired t-test (0 vs 20 Hz), *P = 0.0249, n = 6; d: two-tailed paired t-test (0 vs 20 Hz), **P = 0.003, n = 6). e, Photographs and cartoon diagram depicting design of a bipolar multi-µ-ILED optogenetic probe. f, Representative photographs of the wireless optogenetic device highlighting its ability to control individual µ-ILEDs (589 nm, 470 nm and both, respectively). g, Top: sagittal brain cartoon depicting viral injection of AAV1-hSyn1-SIO-stGtACR2-FusionRed, AAV5-Syn-FLEX-Chrimson-tdTomato and bilateral µ-ILED device implantation into the VTA of DATCre mice. Bottom: representative coronal image showing immunohistochemistry for tyrosine hydroxylase (TH) (blue), stGtACR2 (red) and Chrimson (magenta) expression within the VTA. Dashed lines indicate boundaries of separate brain areas; VTA: ventral tegmental area, IPN: interpeduncular nucleus (Scale bar, 100 µm). h, Left: cartoon depicting social interaction test with 20 Hz stimulation at 589 nm for Chrimson-mediated neuron excitation. Right: Chrimson stimulation (20 Hz) significantly increases social interaction in DATCre:Chrimson/GtACR2 mice (two-tailed paired t-test (0 vs 20 Hz), **P = 0.0014, n = 10). i, Left: cartoon depicting social interaction test with 40 Hz stimulation at 470 nm for stGtACR2-mediated neuronal inhibition. Right: this paradigm reduces social interaction in DATCre:Chrimson/GtACR2 mice (two-tailed paired t-test (0 vs 40 Hz), P = 0.086, n = 7). Data are presented as mean ± s.e.m. of biological replicates.

Full size image

To demonstrate this independent selectivity, we used multicolour optogenetics to bidirectionally control the midbrain dopamine system to modulate social interaction29. Here we injected the ventral tegmental area (VTA) of dopamine active transporter Cre mice (DATCre) with viral constructs containing soma-targeted blue-light-sensitive inhibitory opsin stGtACR230 and red-light-sensitive excitatory opsin Chrimson31. We then implanted bilateral, dual µ-ILED (amber/blue) devices (Fig. 4e–g) directed towards the VTA. Following recovery, we selectively activated specific LEDs to target the distinct opsins during a test of social interaction (Fig. 4h, left). Using amber light to activate Chrimson (Fig. 4h, right), we demonstrated that DATCre mice have increased social interaction with non-cagemate ‘stranger’ mice during VTA-dopamine neuron stimulation (paired t-test, P < 0.01). In contrast, blue-light-mediated inhibition of dopamine neuronal activity through stGtACR2 stimulation (Fig. 4i) reduced social interaction, but did not reach significance (P = 0.08). Importantly, this simultaneous selective approach is not limited by device or species. To demonstrate this flexibility, we implanted Sprague-Dawley rats with microfluidic probes19 (Extended Data Fig. 2a) directed at the nucleus accumbens and placed in a palatable food choice consumption assay. In this assay, local piconet WNBN-controlled microinfusion of mu-opioid receptor agonist, D-Ala2, NMe-Phe4, Glyol5-enkephalin (DAMGO), drove binge-like food consumption for palatable high-fat diet as previously described32 (Extended Data Fig. 2b–e).

To demonstrate the WNBN’s high-throughput capability to control systems for simultaneous and parallel experimentation, we conducted two concurrent locomotor and food consumption experiments. Here, Thy1ChR2-YFP mice (n = 8) were targeted with fibre optic implants for BLE-enabled control of conventional DPSS 473 nm laser stimulation of the M2 (Fig. 5a)28. In addition, agrpCre x Ai32 mice (n = 12–13) were targeted with BLE-enabled blue µ-ILED devices for the homecage stimulation of the PVH to drive feeding behaviour (Fig. 5b). Following recovery from surgery, ad libitum fed agrpCre x Ai32 mice were trained to collect pellets from a feeding experimentation device version 3 (FED3)33 overnight for 3 d. FED3 is an open-source homecage feeding device that allows continuous food intake measurements inside a standard mouse homecage with minimal experimenter intervention. Following this training period, all animals from both experiments were placed in separate staging areas within the behavioural testing room – small open fields for the Thy1ChR2-YFP mice and FED3-enabled home cages for the agrpCre x Ai32 mice – 1 h before simultaneous testing. This coordinated experiment began by giving agrpCre x Ai32 mice access to FED3 pellet dispensers (20 mg per pellet) (Fig. 5c). During the first hour of FED3 access, Thy1ChR2-YFP were connected to fibre optic patch cables and allowed to acclimate for 10 min. The experimenter then used the WNBN web interface to schedule a series of ‘ON’ and ‘OFF’ commands to be sent to all 21 BLE-enabled devices to initiate 20 Hz photostimulation in all mice. The first set of 21 ON commands were sent at 60 min. Following 5 min stimulation, 8 OFF commands were sent to stop stimulation for the Thy1ChR2-YFP. Fifty-five minutes later, 8 ON commands were sent to initiate a second Thy1ChR2-YFP stimulation time. Five minutes later, a final set of 21 OFF commands were directed to all BLE-enabled devices to stop all stimulation in both experiments. This resulted in two 5 min stimulation periods for the Thy1ChR2-YFP mice and a single 1 h stimulation period for the agrpCre x Ai32 mice (Fig. 5c). Internet-scheduled photostimulation significantly increased food pellet retrieval in agrpCre x Ai32 (Fig. 5d,g,i) mice while simultaneously increasing locomotion and rotation in Thy1ChR2-YFP mice (Fig. 5d–f,h). Individual heatmaps show simultaneous and distinct behavioural activation during these stimulation periods in each experiment (Fig. 5g,h). Continuous monitoring of pellet retrieval with the FED3 devices shows a leftward shift in the inter-pellet retrieval interval during photostimulation that persists during the hour following photostimulation, despite an overall reduction in pellets obtained during this time (Fig. 5j,k). One week following this test, we coordinated a second multibehaviour experiment to include locomotion and home-cage food consumption without the FED3 devices as described in Extended Data Fig. 3a. Again, scheduled and simultaneous photostimulation increased food consumption in agrpCre x Ai32 mice (Extended Data Fig. 3b,c,g) while concurrently increasing locomotion and rotation in Thy1ChR2-YFP mice (Extended Data Fig. 3d–g). In both feeding experiments, photostimulation increased caloric intake; however, agrpCre x Ai32 mice had higher caloric intake during initial FED3 access (Hour 1) compared with free food access (Extended Data Fig. 3c). In the Thy1ChR2-YFP mice, both days of locomotor stimulation increased overall distance travelled and rotations; however, the second stimulation produced a larger relative response on the second test day, suggesting possible plasticity in response to repeated photostimulation (Extended Data Fig. 3e). Importantly, these simultaneous experiments were all conducted by a single experimenter controlling the experimental parameters through the WNBN web interface. This combination of experiments demonstrates the effectiveness of WNBN in simultaneously controlling specialized BLE-enabled wireless neural devices and conventional laboratory equipment to coordinate high-throughput experimentation.

Fig. 5: Simultaneous WNBN control of high-throughput behavioural experimentation.

a, Left: sagittal brain cartoon of implantation of fibre optic implant M2 on Thy1ChR2-YFP mice. Right: representative coronal image showing implant location and ChR2-YFP expression (green) within the cortex. Scale bar, 250 µm. b, Left: sagittal brain cartoon of implantation of µ-ILED device into the PVH of agrpCre x Ai32 mice. Right: representative coronal image showing ChR2-eYFP expression (green) within the PVH. Scale bar, 100 µm. c, Cartoons and timeline describing locomotion and ad libitum pellet feeding experiment with Internet-scheduled simultaneous control. Blue bar indicates duration of photostimulation. d, Simultaneous behavioural outputs depicting locomotion (m) for Thy1ChR2-YFP mice and pellets retrieved per minute in agrpCre x Ai32 mice in 1 min bins during coordinated behavioural experimentation. e, Data showing changes in locomotor activity during Internet-controlled stimulation (20 Hz) in Thy1ChR2-YFP mice. Average distance travelled during Internet-scheduled stimulation was significantly higher than at pre-stimulation and post-stimulation periods (repeated measures one-way ANOVA, for Stimulation 1: Pre1 vs Stim1, **P = 0.0066 and Stim1 vs Post1, *P = 0.0443, Tukey’s multiple comparisons test, n = 8; for Stimulation 2: Pre2 vs Stim2, **P = 0.0012 and Stim2 vs Post2, *P = 0.0115, Tukey’s multiple comparisons test, n = 8). f, Rotations in Thy1ChR2 mice over 80 min (1 min bins) that includes two stimulation periods. g,h, Heatmaps depicting changes in behavioural activity for agrpCre x Ai32 mice (g) and Thy1ChR2-YFP mice (h) during coordinated scheduled photostimulation. i, Average number of pellets retrieved during 1 h periods. Blue bar indicates the hour of 20 Hz stimulation. agrpCre x Ai32 mice consumed significantly more pellets during 20 Hz photostimulation compared with non-stimulation periods (1 h) (repeated measures one-way ANOVA, Hour 1 vs Hour 2, **P = 0.0048 and Hour 2 vs Hour 3, ****P < 0.0001, Tukey’s multiple comparisons test, n = 12). j,k, Number of pellets (j) and kernel density estimate distributions (k) of inter-pellet intervals during each 1 h period in agrpCre x Ai32 mice. The area under each kernel density estimation curve is 1 and the area under any particular section of the curve estimates the probability of inter-pellet intervals of that duration occurring. Data are presented as mean ± s.e.m. of biological replicates.

Full size image

Scaling wireless networks can introduce issues of network crowding and interference. While these two discrete and independent behavioural experiments were conducted twice within the same animal facility, this facility did not have excessive miscellaneous and unconnected Bluetooth devices within range of the minicomputer. At any given time, there were typically ~10 non-related devices representing BLE-enabled security cameras, smart building infrastructure, smartphones and computer peripherals. Increasing experimental scale to tens or hundreds of devices, however, could be limited due to signal interference among these and unconnected devices. One approach to limiting this interference would be to incorporate advanced Bluetooth mesh technology34 into the WNBN ecosystem. In a proof-of-principle benchtop experiment, we demonstrate this integration by enabling selective device activation of 82 devices over a physical distance of 4 m (Supplementary Video 5 and Fig. 6). This initial test suggests that advanced Bluetooth mesh technology is a feasible approach to limit Bluetooth signal interference in the WNBN ecosystem.

Automated control and data collection to reduce observer effects in behavioural neuroscience

In addition to on-demand device activation, delayed scheduling in the global Internet WNBN enables automated and semi-automated in vivo experimentation. Here, stimulation or monitoring protocols are established for future dates and times. These remote-control capabilities eliminate the need to be physically present in the laboratory, thereby alleviating some elements of an ‘observer effect’1,5. As a versatile demonstration of this scheduled Internet-controlled stimulation, we used this approach to control laser-tethered fibre-optic-implanted mice in a typical optogenetics experiment. Here we tested whether the known arousal modulating properties of the locus coeruleus noradrenergic system11,35 extend to its projections into the medial prefrontal cortex (mPFC). To virally target this system, we used animals that exclusively express Cre recombinase in cells expressing dopamine beta hydroxylase (Dbh), the enzyme necessary for the conversion of dopamine to norepinephrine. DbhCre mice were injected with AAV5-EF1α-DIO-ChR2-eYFP in the locus coeruleus and implanted with a chronic fibre optic36 in the medial prefrontal cortex (Fig. 6a). Following recovery, mice were placed in a behavioural test chamber and allowed to acclimate for 24 h. The experimenter scheduled two global Internet WNBN-controlled stimulations (5 Hz for 15 min) via the web interface to occur following acclimation (Fig. 6b). The next day, early light phase (AM) scheduled stimulation increased locomotor activity in DbhCre animals (Fig. 6c), but late light phase (PM) stimulation had no clear effect (Fig. 6d). This suggests that activation of this pathway promotes general arousal that may be circadian rhythm dependent. Importantly, as these experiments were scheduled to occur automatically and without experimenter intervention, we found a strong negative correlation between baseline locomotion and activity during the photostimulation (Fig. 6e). Notably, the remote access capability of the global Internet WNBN can also be used to log data from selected sensing hardware to obtain vital data on laboratory conditions, such as temperature, humidity, air pressure, or any other sensor with an integrated BLE module (Fig. 6f). Here we used inexpensive off-the-shelf hardware – Simblee breakout board (RFD77101 Breakout Module, Simblee) and atmospheric sensor (Environmental Sensor BME680, SEN-16466, Sparkfun) to detect mouse home-cage temperatures for 24 h (Fig. 6f,g). We remotely detected changes in cage temperature for cages containing standard group-housed mice and empty control cages. Mouse cage temperatures increased during the 12 h ‘night’, while ‘day’ temperatures remained relatively stable (Fig. 6h,i). These data could provide reasonably inexpensive and reliable measures of overall cage activity37,38. Critically, this technology could be expanded to include other behavioural and physiological detection systems as a means to automate, optimize or scale data collection while minimizing experimenter intervention.

Fig. 6: Conventional device integration for Internet-controlled experimentation and data collection.

a, Top: sagittal brain cartoon depicting viral injection of AAV5-ef1α-DIO-ChR2-eYFP into the locus coeruleus (LC) and fibre optic implantation into the mPFC of DbhCre mice. Bottom: representative coronal image showing immunohistochemistry for ChR2-eYFP expression within the mPFC and LC. Images show ChR2-eYFP (yellow) and tyrosine hydroxylase (TH) (pink) or Nissl (blue). Scale bars, 50 µm; 4 V, fourth ventricle. b, Cartoon depicting implementation of scheduled global Internet control in conventional optogenetic stimulation to drive locomotor activity. c, DbhCre:LC-mPFC::ChR2 mice had significantly higher locomotion during the AM 5 Hz stimulation compared with non-stimulation periods (15 min) (repeated measures one-way ANOVA, 5 Hz vs Post, *P = 0.0392, Tukey’s multiple comparisons test, n = 5). d, No significant change was detected during the PM 5 Hz stimulation. e, Baseline movement was negatively correlated with evoked locomotion (r = −0.6398, P = 0.0464, n = 10 pairs). f, Cartoon schematic of Simblee and sensor setup for home-cage temperature detection by remote user. In this case, as depicted in the cartoon, the commands were scheduled in Springfield, MO, USA for detection to occur in St. Louis, MO, USA. g, Photograph of Simblee board, sensor and mouse cage setup for monitoring cage temperature. Inset: photograph of sensor location inside mouse cage. h, Graph depicting average cage temperature collected over a 24 h period. Black circles indicate empty cage temperatures, blue circles indicate mouse cage temperatures. Dark bar indicates dark phase of light cycle (12h :12 h). Data are presented as mean ± s.e.m. of biological replicates. i, Graph depicting temperature change of mouse cage during dark phase of 12h :12 h light cycle (two-tailed paired t test (Day vs Night), **P = 0.0011). Data are presented as individual biological replicates.

Full size image