Evaluating Skilled Prehension in Mice Using an Auto-Trainer

Our auto-trainer evaluates forelimb reach-to-grasp (prehension) in an automated manner. To achieve this endpoint, many of the parameters designed for the mouse prehension task, including pellet placement, pellet size, and training criteria, have been iterated over several years and adapted from prior protocols^2,5,6. The advancement here is the automation of the task using a robot that allows home-cage housing. Home-cage housing allows the mice to remain calm and perform the task with less anxiety. Non-home-cage training is associated with increased stress which can lead to the increased time and decreased precision^7,8,9. We demonstrate here precision resembling our own results with manual home-cage training^5,7,8. Although home-cage training exists for the rat¹⁰, to our knowledge, this is the first auto-trainer that takes advantage of training mice in their home-cage.

Our auto-trainer includes an adjustable platform on which a slotted cage rests and may be lowered or raised to the appropriate height for alignment with a food pellet holder (also referred to as a diving board). A pellet-dispensing system places the food pellet on the diving board holder. The food pellet holder has a bait pellet sensor assembly consisting of a reflective object sensor to detect whether a food pellet is present or not on the diving board holder. Due to the light sensitivity issues, the reflective object sensor may be calibrated at installation to suit the lighting environment of the laboratory. Each mouse's cage is placed on the auto-trainer such that the pellet's inner edge is in line with the outer edge of the cage gate's slot, corresponding to step 2.3.4 of the manual procedure detailed above. Two lost pellet sensors oriented in opposite directions in a funnel beneath the diving board pellet holder detect falling pellets. One benefit of employing two lost pellet sensors is that it ensures high detection accuracy for various food pellets of different sizes and shapes. Both lost pellet sensors consist of a standard transmissive photo-interrupter with a through-hole design, to sense the motion of a falling pellet without requiring contact.

The software consists of a program that runs the auto-trainer and collects data on successes and failures. User input consists of the file location in which the data is recorded, how many pellets are dispensed in one training session, an option to pause the training session after dispensing a particular pellet, a field to record the increased distance (if any) across which the mouse must reach, and a field to control the array size used in the program's calculations (which may be ignored during normal use). Further, the software enables the user to tune the diving board's reflective object sensor in order to recalibrate the light sensitivity as necessary. An output of each trial is displayed to the user as well as recorded and saved in a log file for later retrieval.

A single trial consists of a single food pellet's time spent on the diving board until it is removed by action from the mouse. If a pellet leaves the diving board as determined by the bait pellet sensor and the pellet is detected falling through the funnel shortly afterward by either of the lost pellet sensors, it is recorded as a failed trial by the software. If the bait pellet sensor determines that the pellet leaves the diving board, but no falling object is detected by either of the lost pellet sensors, it is assumed to have been pulled into the cage by the mouse and is counted as a successful trial.

This formulation is used because it is useful to design a behavioral assay in which the task to perform directly, rather than indirectly, provides the reward. In this way, there is no ambiguity on the animal's part on what the task is (e.g., be hungry, locate food, get food, eat food). Among the many tasks that utilize such a paradigm, the prehension task has become quite popular for such assessments. The task merely requires that an animal use a single limb to reach for and grasp a single food item, which the animal subsequently brings to its mouth for consumption. The prehension task assesses a behavior that is very similar to an everyday behavior used by many mammals. Most importantly, the prehension task resembles human motor behavior⁴. This generalizability enhances the expectation that principles derived from the preclinical assessment of the behavior are clinically applicable in disease states. For example, impairments in skilled forelimb and hand use are seen in stroke, Huntington's disease, Parkinson's disease, and multiple sclerosis. Thus, modeling behavioral deficits and subsequent recovery in mice is invaluable to understanding human recovery and how it might be encouraged^2,11,12,13.

Many aspects of the auto-trainer proposed herein greatly benefit the research process. First, most behavioral assays require an experimenter to closely train and monitor daily sessions, which can be costly, labor intensive and prohibitively time consuming. Our auto-trainer allows for behavioral data to be collected independently of an experimenter. Second, our auto-trainer can be replicated to allow multiple mice to be trained and evaluated objectively, efficiently and concurrently, thus minimizing time and effort. Third, the low-cost of the auto-trainer allows replication and use of multiple auto-trainers concurrently for large scale and efficient testing.

It should be noted that the critical point that requires careful supervision is during the shaping phase of the training. Notably, this protocol's chief weakness is the risk of bad use becoming fixed in some mice¹¹. The protocol aims to mimic tests like the ladder rung test in that succeeding in the task provides the reward. However, the task itself still must be taught to the mice in step 2.3 of the protocol, unlike the ladder rung test. The concept most likely to cause a mouse to stumble in learning this task is from extending a paw out of the cage to using the paw to actually grasp the pellet. In the first session of step 2.3.1, mice should be rewarded simply for extending a paw out of their cage. However, over the following few days, investigators should take care to reward mice less for just extending the paw, and more for extending the paw and touching the pellet, as we describe in step 2.3.3.

Please note that approximately 5% of mice will fail to progress past this stage, typically because of limited extension of their digits to pull in the food pellet. Such mice will fail with one or both paws with little consideration of the actual location of the pellet, providing little or no useful data. To minimize a mouse's potential for failure at this stage, caution is strongly recommended when pulling away the pellet with the tweezers during the learning process. In particular, the mouse should be rewarded with food not only when it extends a paw, but also when that paw grips the pellet and applies enough force sufficient to investigator's satisfaction. A similar risk to potential failure at this stage is posed by mice using their tongue to lick pellets towards them. When training mice that tend to lick, place the pellet further laterally away from the slot. Mice will find it difficult to reach with the tongue over a greater sideways distance, but the range of motion of the arm and paw are more capable of closing the distance.

Our described protocol is readily extended to different lab environments or different methods of data collection. The auto-trainer, for example, is very useful as a labor-saving device, but is not strictly required for data collection, as pellets can be provided and successes/failures can be recorded by hand. Individual reaches can also be categorized based on more detailed information than simply success/failure, for example by considering the angle of approach of each mouse, the number of reaching attempts that do not touch the pellet, or the mechanics of the retrieval motion, which has received more attention in recent years¹⁴. The ability of the animal to retrieve a pellet is but one measure. Using additional hardware, we will also be able to measure the velocity, angle, and trajectory of the animal's limb movements. This kinematics is an important aspect of motor learning both before and after a neurological injury. To this end, we are currently incorporating various novel means of analyzing the locomotion and kinematics of the mouse's grasping action. We are exploring using high-speed cameras to obtain kinematic measurements of the grasp and attaching pressure transducers and accelerometers to the food pellet holder to measure force and mass data associated with the grasp. These new features will enhance the functionality of the auto-trainer to collect significant data passed a simple pass or fail trial and help illustrate the gait of the mouse's grasp through disease progression. In the future, we will be using the robot assisted prehension task as a platform to evaluate type, dose, and timing of rehabilitation after neurological injury. Moving forward, we will continue to improve the task, with refinements to help lessen incorrect behaviors and improve task acquisition rate and training time.

In summary, we have developed a new auto-trainer for assessing upper forelimb prehension skill in mice. The task requires mice to reach their paw through a slit, grasp a small food pellet, and pull the pellet in the direction of their body so that they may eat the pellet. The task setup is mechanically constrained to assure dominant paw usage. Mice can be trained quickly and simultaneously, with only the shaping process requiring manual input. The test can be administered efficiently and analyzed automatically. This high throughput behavioral assay quantifies success rate and is easily modified for future analysis of kinematics and force dynamics.