Swimming direction of the glass catfish is responsive to magnetic stimulation

Several marine species have developed a magnetic perception that is essential for navigation and detection of prey and predators. One of these species is the transparent glass catfish that contains an ampullary organ dedicated to sense magnetic fields. Here we examine the behavior of the glass catfish in response to static magnetic fields which will provide valuable insight on function of this magnetic response. By utilizing state of the art animal tracking software and artificial intelligence approaches, we quantified the effects of magnetic fields on the swimming direction of glass catfish. The results demonstrate that glass catfish placed in a radial arm maze, consistently swim away from magnetic fields over 20 μT and show adaptability to changing magnetic field direction and location.

The glass catfish (Kryptopterus vitreolus) is found in slow moving fresh water in Southeast Asia, ranges from 31.4-64.6 mm in length and is transparent except for their organ packed head [15]. The glass catfish has historically been of interest to a wide range of scientific disciplines, including circulation [16], cell line establishment [17] and electroreception [18,19]. The interest in this species' magnetoreception has recently been reignited due to its potential to be part of a synthetic system that will allow remote, magnetic control, of neural function [20]. New advances in molecular biology have brought new tools to identify the mechanisms by which this species respond to electromagnetic fields. Recently, we have discovered a gene (electromagnetic perceptive gene (EPG)) that is expressed in the glass catfish's ampullary organ and is specifically activated in response to magnetic stimuli. This genetic-based magnetoreception has a great potential as a neuromodulation technology and as a valuable tool to study neural behavior from the molecular to network levels [20][21][22]. However, the mechanism by which magnetoreception manifests and functions is not well understood [23][24][25][26][27][28][29].
This work was designed to characterize the natural behavior of glass catfish in response to magnetic fields. This understanding may lead to improved engineering of magnetic-receptive modulation and sets a foundation for a new magnetically sensitive animal model. We capitalized on new concepts of artificial intelligence as well as traditional video tracking algorithms to quantify how glass catfish respond to magnetic stimulation with high spatial and temporal resolution.

Methods
All animal procedures were conducted in accordance with the NIH Guide for the Care and Use of Laboratory Animals and approved by the Michigan State University Institutional Animal Care and Use Committee.
Thirteen Glass catfish of undetermined sex and age [30] were imported from Thailand to the United States and housed in a standard 30-gallon fish tank with a 12-hour day/night cycle with water maintained at 27 degrees and provided with an enriched environment. After arriving from Thailand, fish were acclimatized for 6 weeks before starting experiments. Water quality was checked daily for ammonium, nitrite, and nitrate. Fish were fed a diet of fresh hatched brine shrimp twice per day. All experiments were conducted between the hours of 11 am and 4 pm to eliminate behavioral changes due to feeding [31] and light cycle.
The Y-maze's arms were 60 cm long and 10 cm wide with a central area of 10x10x10 cm (AnyMaze, San Diego Instruments, CA). One week prior to starting experiments all fish were transitioned from their tanks to the radial Y-maze where they lived for the duration of each experiment (2 weeks). Enrichment materials were carried over from permanent housing to Ymaze, rearranged daily and removed prior to starting experiments. A five-gallon water change was done weekly in the Y-maze with mature water to control water quality and reduce the impact of any unknown pheromones. The same fish were used for all experiments, thirteen fish were tested in constant location and twelve in alternating location experiment. The magnetic stimulus and the sham stimulus were placed 10 cm from the end of an arm, inside of the maze. Each trial was recorded for 30 minutes by overhead cameras while the experimenter was out of the room. Each trial was repeated four times for each condition for a total of 24 trials. In order to negate the effects of Earth's intrinsic magnetic field, the location of the magnet was rotated during the changing location experiment: Arm 1 was oriented in the south west direction, Arm 2 in the northern direction and Arm 3 in the south east direction.
A permanent Neodymium Rare Earth Magnet (11.5 cm x 3.18 cm x 2.2 cm) with a horizontal magnetic flux of 577 mT at the magnet's surface was placed 10 cm from the end of one of the Y-maze arms. The strength of the magnetic field induced by the magnet was calculated by COMSOL (Fig 1). A sham stimulus was made from plastic and aluminum foil, with similar dimensions to the magnet. All recordings were analyzed by a radial-maze tracking software written in Matlab by Delcourt et al. [32]. Delcourt et al. software is available on github:: https://github.com/sjmgarnier/projectRadial. Videos were taken originally in AnyMaze format and converted to.mp4. The Matlab program then created a background image by taking an average of 100 frames. Fish location was determined by subtracting the background image from each frame, remaining pixels with a grey scale value higher than threshold were given a value of 1, continuous pixels with a value of 1 were labeled as a fish. The spatial resolution of all videos was 3.57 ± .52 pixels/cm and were recorded at 30 frames per second. The number of fish in each arm was reported every second, however fish located in the center of the Y-maze were not reported. The average of the 4 trials for the arm of interest (Magnet/Sham) and the pooled average of the two empty arms (No Magnet/Sham) is reported in Figs 2 and 3.
Data is reported as median ± 95% confidence intervals (CI). Data was analyzed in R Studio, using the bootstrap method to calculate CI's and Mann-Whitney U Test to compare the arm of interest with the pooled average of the two empty arms across the 4 trials. Shapiro-Wilke tests and Q-Q plot analysis were used to determine distribution of the data which was found to be not normally distributed.
In a separate set of experiments, a single fish was selected from the school and placed in the center of the Y-maze. Two trials of magnet and sham conditions were conducted: one set of trials was used for computer training, and the other for analysis using the trained software. DeepLabCut [33,34] was used to track the location of a single fish in the Y-maze. DeepLabCut code is available at the following address: https://github.com/DeepLabCut/DeepLabCut.git. The program was initially trained on 20 frames and ran through 1,300,000 iterations. After initial training, outlier frames were extracted and re-labeled. The program was re-trained with the addition of the outlier frames for 100,000 iterations. Retraining was done three times until visual inspection and expected error were satisfactory (±15 pixels). DeepLabCut results were extracted using an output CSV file from the code, which provides the x,y coordinates of the tracked fish for every frame of the video. This file was exported to R-Studio and plotted as a standard scatter plot and overlaid on a Y-maze diagram. Tracked videos are two minutes and fifty seconds long (5,100 frames).

Constant location of stimulus
We characterized fish behavior as a response to a magnetic stimulation that was consistently present at the same location. In these sets of experiments the magnet was always placed in Arm 1 of the Y-maze. The initial location of the fish school (n = 13) was changed randomly to one of the Y-maze's three arms. Each trial was repeated four times for each arm for a total of 24 trials. The number of fish present in Arm 1 (Arm with magnet) was significantly lower than the number of fish in the other two arms in the first minute (Arm 1-Magnet, 7.69% ± 8.4%; Arm 2 & 3-No Magnet, 38.46% ± 8.84%; P = < .0005, U = 147424, Mann-Whitney U Test), after 5 minutes (Arm 1-Magnet, 7.69% ± 2.3%; Arm 2 & 3-No Magnet, 38.46% ± 6.41%; P = < .0005, U = 3227686, Mann-Whitney U Test), and over the entire recording that lasted 30 minutes (Arm 1-Magnet, 7.69% ± 1%; Arm 2 & 3-No Magnet, 38.46% ± 3%; P = < .0005, U = 175417129, Mann-Whitney U Test) (Fig 2). When the initial fish location was also in Arm 1, the school immediately swam away from that arm and stayed away. In contrast, when the sham stimulus was placed in Arm 1, the fish exhibited less preference for the no-sham arms than they did in the previous experiment, after 1  Between experiments it was also observed that there is significantly less fish in Arm 1 when the magnet is present compared to when the sham is present at all time points (p < .0005, Mann-Whitney U Test). These experiments showed that glass catfish prefer to avoid swimming in water with a magnetic strength over 20 μT (Figs 1c and 2). Results indicate that regardless of the initial location of the fish, they tend to avoid Arm 1 more when magnet was present compared to sham at all time points (p < .0005,Mann-Whitney U Test).

Changing location of stimulus
We then sought to determine if fish behavior changed with the location of the magnetic stimulation. In this set of experiments the fish school (n = 12) was barricaded in the middle of the Ymaze, and the magnetic or sham stimuli were placed randomly in one of the arms (Fig 1b). After barricade removal the fish swam away from the magnet and explored the two other maze-arms. In line with previous experiments, there are significantly less fish present in the arm on average when the magnet was present compared to when the sham stimulus was present (Arm 1-Magnet, 8.33% ± 8.25%, Arm 1-Sham, 25% ± 15.68%, U = 908254; Arm 2-Magnet, 11.53% ± 9.71%, Arm 2-Sham, 16.67% ± 23.4%, U = 890468; Arm 3-Magnet, 41.67% ± 22.6%; Arm 3-Sham, 50% ± 5.11%, U = 983677; P = < .0005 Mann-Whitney U Test for all tests). For the purpose of individual swim pattern analysis, one fish was placed in the middle of the Y-maze. Once the barricade had been removed the fish swam across two arms where no magnet was present but exhibited a clear avoidance from the arm containing the magnet. We used a state-of-the-art artificial intelligence (AI) approach, DeepLabCut [33,34] to track the fish's swimming path. DeepLabCut was successfully trained on a single fish with an error of

Discussion
Several marine species have developed a magnetic perception that is useful in navigation and the detection of prey and predators [6,35], for review see [36][37][38]; animals such as sharks and platypus, use magnetoreception for prey detection [39][40][41][42][43]. Others, like ants [44], use this sense for predator avoidance and the nematode, Caenorhabditis elegance uses magnetoreception for vertical navigation in soil [45]. Even cattle have been shown to align themselves with electromagnetic pulses [46]. Recently, evidence of magnetoreception has been unveiled in fish species previously thought to me non-magnetic, such as zebrafish [47]. The glass catfish is a transparent fish found in slow moving rivers of Southeast Asia where visibility is low [15,17]. Until recently, the transparent glass catfish was commonly identified as Kryptopterus bicirrhis but is now known to be Kryptopterus vitreolus [15]. It appears plausible that in these types of conditions, magnetic perception is an advantageous trait to conserve. However, the mechanisms allowing magnetic sensation remain largely unknown [26,38]. We have used state-ofthe-art software based on artificial intelligence object tracking algorithms to characterize glass catfish behavioral response to magnetic fields. The results indicate that glass catfish consistently swim away from magnetic fields over 20 μT and show adaptability to changing magnetic field direction and location. In addition, our results show that this magnetic avoidance behavior is not influenced by school behavior. It is important to note that there is significantly less fish in the Sham arm compared to No-Sham arm in the constant location stimulus experiment. We hypothesize that this is because the fish developed an avoidance for Arm 1 due to repeatedly exposed to the magnet. In the alternating location experiment we see no avoidance of the sham object in all three arms.
We have previously demonstrated that the modulatory effects of magnetic stimulation on mammalian cells transfected with EPG was induced by magnetic fields of 50 mT [21]. However, in this experiment we see that fish are significantly more sensitive to magnetic fields than transfected cells, with fish starting to exhibit avoidance behavior at~20 μT (Figs 2 and 3). While the pathway by which EPG modulates calcium channels is unknown [20,21] it is possible that there are accessory proteins that are currently remain unknown, which amplify magnetic sensitivity in glass catfish. It could also be possible that different cellular mechanisms occur over the time when EPG is exposed to magnetic fields. Currently, we can only evoke a cellular response by using strong magnetic fields in culture (> 50mT). However, the Earth's magnetic field is only 30 μT-60 μT, and yet, is readily detected by glass catfish. One of the major challenges in EPG's development as a neuromodulatory technology is the attenuation of magnetic fields over distance. If the biological amplification properties of the glass catfish are uncovered, this technology could be used to treat deep brain afflictions without the need for surgery.
Using AI such as DeepLabCut can be transformative to animal behavior studies. Using this method, we could follow the swimming pattern of an individual fish, over thousands of frames with high spatial and temporal resolution. Another advantage is the machine learning components of AI. The more trials run through DeepLabCut, the more efficient and accurate it becomes at tracing animals in similar situations. However, the transparency of glass catfish caused detection difficulties with DeepLabCut and recording hardware when rapid movement caused insufficient contrast between the fish and Y-maze. In Fig 4B, during the sham stimulus the fish swam at a gradual pace throughout the maze. However, during magnetic stimulus the fish tend to stay in one area then dash to the end of an arm and back. During these rapids movements our recording software experienced a drop in frame rate Once the fish slowed down and the contrast was restored, the tracking became accurate. While the transparent glass catfish poise unique challenges for AI tracing due to its transparent body, this AI tracing approach can be beneficial to track movement of other marine species.
We have established that the glass catfish has unique magnetic field sensing capabilities that position it as a valuable model to study magnetoreception in animal species. The cellular mechanisms allowing this capability remains to be determined. We have already identified and cloned the EPG from glass catfish. But is this the only magnetic-sensitive protein? Does it work with other proteins to amplify and modulate its activity? Do other animal species that have been shown to be sensitive to magnetic fields have similar proteins? This animal model can provide key information to address these questions. By characterizing the behavior of glass catfish, we are now working towards developing a fish with a knock-out in the EPG gene. This will elucidate if there are additional genes associated with magnetic responses and will facilitate the development of the next generation of additional magnetic sensing molecular tools.