Machine learning has the potential to accelerate pain research by transforming the way animal behavioral studies are conducted, a new study has suggested.
Animal behavioral studies are a common avenue of pain research, providing insights into pain responses. However, in addition to the important ethical considerations of such research, these studies can be very time consuming – for each animal it involves painstakingly tracking the position of specific body parts frame by frame in a video.
But in a study published in Neuroscience, researchers at the Indian Institute of Science (IISc) in Bangalore wanted to test the effectiveness of DeepLabCut (DLC) – a machine learning computational tool that tracks the movement of mice – in accurately recording the protective maneuvers, known as ‘nocifensive behaviors’, they make in response to a high surface temperature, such as jumping to escape the heat. They also wanted to see whether DLC could distinguish the effects of chronic and acute pain – an important distinction given the clinical relevance of the research.
“In pain research it has been hard to find effective drugs because the animal behavioral models were not good enough,” says Assistant Professor Arnab Barik at IISc, who led the research. “Most of the responses researchers record are when they give an animal a stimulus and it reacts to it. But that’s only a small part of the pain phenomenon. If you have arthritis, it’s not just hurting when you do strenuous exercise. It’s a chronic pain that stays with you day in, day out.”
Dr. Arnab Barik, the leader of the research
Barik is aware of the ethical considerations of this type of research. “It’s a pain study and it cannot be done without animals because you can’t predict these behaviors in a [petri] dish. It is impossible to perceive and therefore study pain in a model system that does not have a fully functional nervous system and that cannot respond to painful stimuli. As a mouse is a mammal genetically similar to humans, and the circuits of the major brain regions are similar, it’s a good model.” Barik described some of the measures taken to limit the distress of the mice. “We have post experimental and post-surgical care that ensures the mice undergo limited stress, like better housing and they are not experimented on often. After the experiments, we can give them analgesics.”
In the experiments, mice were enclosed in a transparent acrylic chamber with a metal floor whose temperature could be changed. The mice were exposed to 15 different temperatures between 0 and 56oC for 45 seconds each with a break of 15 minutes between each temperature trial. Any changes in their movement were recorded using three webcams providing different perspectives with machine learning tool DLC then analyzing the footage.
Machine learning is a branch of artificial intelligence in which algorithms can be trained to recognize patterns – in this case the appearance of a mouse nose and other body parts in a video frame. In this research, the scientists wanted to investigate the analytical affordances of data DLC captured on the nose of the mice. The specific body part selected was somewhat arbitrary – the researchers just wanted to determine whether tracking one specific body part would be enough to measure how the mouse was responding to the heat.
The researchers found DLC was able to accurately track the noses of mice in different frames and from this they were able to distinguish changes in their behavior at different temperatures. When mice were exposed to low temperatures, they explored the chamber and stood with the support of the chamber walls. At higher temperatures, mice spent time closer to the chamber floor, jumping to try to escape the heat at 56oC.
To measure the effect of acute pain on the response of mice to the thermal plate temperature, allyl-isothiocyanate, the active component of mustard oil, was injected into the left hind paw of some of the mice to stimulate acute pain. Chronic pain was stimulated with the injection of Chronic Freund’s adjuvant into the paw. After this, it was found that DLC was able to discern differences in the behavior of the mice exposed to acute or chronic pain. After the acute pain stimulus, the movement of the mice was reduced across the temperatures of the heat plate, apart from temperatures greater than 48oC, where the mice tried more jumping movements to reach a ledge in the enclosure to escape the heat. Whereas with the chronic pain stimulus, mice were hypersensitive to heat, moving at higher speeds and accelerating more quickly at temperatures above 44oC – they also tried to jump to escape the heat more. “It makes sense, because when you are acutely in pain right now, you feel like you don’t want to move,” says Prannay Reddy, a PhD student at the Champalimaud Foundation in Portugal and formally a Project Assistant at IISc.
The researchers also activated PNBTacr1 neurons which were already known to play an important role in nocifensive responses to painful stimuli in mice, to see whether the response of mice to this could also be measured using DLC. The specific activation of these neurons was driven by chemogenetics or optogenetics stimulation. For this, after the virally-induced expression of particular artificial receptors, such neurons could be activated either chemically, by the administration of the receptor’s ligand, or by blue light. When these neurons were activated, mice were mostly seen to show reduced exploratory behavior across the temperature range and an increased number of escape jumps at high temperatures. “We could clearly see differences using this analytical tool, DLC, by activating these neurons,” says Reddy.
Prannay Reddy – one of the first authors of the article
This research helps to build the case for using DLC with hotplate tests in rodents. “The hotplate test is commonly used in academia and industry by those who study pain,” says Barik. “So having a computational tool that can analyze behavioral responses automatically is useful for everybody. The hotplate test is used widely in drug screening tests in academia and industry. So this can really speed up the process of drug discovery. A major problem with animal behavioral studies is the bias of the experimenter and this alleviates this issue.”
Another area of future research, with the help of DLC technology, will be to disentangle the sensory and emotional responses to pain. “When you are in pain, the sensory component can make you react behaviorally,” says Barik. “But you also become fearful of the stimulus. So now we are starting to think about how much pain-induced behaviors are reversed by anti-anxiety medication.”
It’s the multifaceted aspects of pain that make it so fascinating for Reddy to study. “Pain is so primal and yet so multidimensional,” he says. “There’s the pure reflexive, sensory part of it and the emotional part of it. Yet only now with behavioral neuroscience we are able to illuminate these different aspects and hopefully help people with pain management.”
Barik says the process of publishing with Neuroscience has been a smooth one: “Publishing with Neuroscience has been great. The reviews were timely and constructive.”
This article was written by Dr. Andy Ridgway.
About Neuroscience
Established in 1976, Neuroscience is the flagship journal of IBRO. The journal features papers describing the results of original research on any aspect of the scientific study of the nervous system. Papers of any length are considered for publication provided that they report significant, new, and carefully confirmed findings with full experimental details. Together with IBRO Neuroscience Reports, IBRO’s open access journal, Neuroscience plays a crucial role in supporting the organization’s global neuroscience activities, as proceeds from both journals support more than 90% of IBRO’s initiatives.