Testing spatial working memory in pigs using an automated T-maze

Abstract

Pigs are an important large animal model for translational clinical research but underutilized in behavioral neuroscience. This is due, in part, to a lack of rigorous neurocognitive assessments for pigs. Here, we developed a new automated T-maze for pigs that takes advantage of their natural tendency to alternate. The T-maze has obvious cross-species value having served as a foundation for cognitive theories across species. The maze (17' × 13') was constructed typically and automated with flanking corridors, guillotine doors, cameras, and reward dispensers. We ran nine pigs in (1) a simple alternation task and (2) a delayed spatial alternation task. Our assessment focused on the delayed spatial alternation task which forced pigs to wait for random delays (5, 60, 120, and 240 s) and burdened spatial working memory. We also looked at self-paced trial latencies, error types, and coordinate-based video tracking. We found pigs naturally alternated but performance declined steeply across delays (R² = 0.84). Self-paced delays had no effect on performance suggestive of an active interference model of working memory. Positional and head direction data could differentiate subsequent turns on short but not long delays. Performance levels were stable over weeks in diverse strains and sexes, and thus provide a benchmark for future neurocognitive assessments in pigs.

Keywords: declarative memory; navigation; spatial memory; translational neuroscience; working memory.

Figure 1

Spatial working memory task adapted for pigs using a fully automated T-maze apparatus. A large T-maze was built from PVC sheeting in the Porcine Neuroscience Facility (PNF) at Florida International University (FIU) to test working memory in pigs. (A) A photo of the pig T-maze (angled bird’s eye view, left panel) including one of our young Yorkshire pigs (40–50 lbs.) in the start area about to enter an open guillotine door to the stem. Close-up shots are provided of the guillotine doors (middle), pellet dispenser, and food bowel (right panels). The green and pink markings on the back are spray painted in order to track subjects via cameras and control the maze in a fully automated manner. (B) A simplified diagram of the T-maze showing the location of the corridors, guillotine doors (light blue rectangles), and right and left reward areas (pink boxes). (C) First, a simple alternation task (light blue) was used in which pigs alternated nearly continuously with the exception that the doors cycle open and closed in the start area (5 s delay), until delays (hot pink) were introduced up to 240 s between trials. (D) A close-up picture of a pig next to a task descriptive timeline of part of a delayed alternation session. Alternating responses was scored as correct (providing a food reward in a food bowl). Incorrect turns were unrewarded. After every choice the pig had to walk back to the start area before the next trial

Figure 2

Pigs naturally alternate in the T-maze and acclimate in about a week. (A) Performance plotted for each pig (unique colored dot) showing all pigs were above chance on the first session and all subsequent sessions. (B) The number of trials each pig completed increased over sessions suggesting pigs were becoming more familiar with the apparatus. (C) The start-to-choice latencies decreased over sessions showing pigs were making faster choices over sessions. Note, the pale blue dot was the oldest pig at 6.5 years and was an outlier in the time he took to make choices. However, this did not evidently affect his accuracy. (D) Choice-to-choice latencies decreased across sessions

Figure 3

Imposing delays in the T-maze between trials parametrically taxes spatial working memory in pigs. (A) Performance declined as a result of increased delay times (linear regression with delay duration predicting performance: R² = 0.84, p = 1.078 × 10⁻¹⁷). This demonstrates the main prediction of working memory in the delayed alternation task. (B) The start-to-choice latencies did not change as a function of delay duration. (C) The choice-to-choice latencies did not change as a function of delay duration

Figure 4

Marker-based video tracking rules out many positional confounds for memory. (A) A simplified diagram is shown depicting the areas analyzed to examine if coordinate tracking could reveal any path or location differences in the behavior of the pig that related to subsequent turns, performance and/or delay durations. (B) A picture of one of our Sinclair pigs in the stem and choice area if the maze during a choice. The green and pink circles are superimposed on the images indicating the center of mass of the color spray painted on the dorsal surface of the pig. The blue and pink shading represents tracking zones superimposed on the image for detecting when a pig is in a particular functional location (such as the choice point) to help MATLAB control the guillotine doors at the appropriate time. (C) Two representative pig tracking sessions are shown colored by right and left turns. The thick blue and red lines are the average path for that session and the thin lines are individual trials. The rightmost plot includes all pigs (n = 9) on right and left turn. The thick lines are the average across all pigs, and the thin lines are the average line for individual pigs. (D) A plot of the average difference between right and left turns for each pig. Aggregate right and left trajectories were distinguishable in the choice area, but this was not true in any other location on the stem or when separated by performance or delay duration (one-sample t-tests, df = 8, H0 = 0, ^***, p < 0.0001). (E) A picture of one of our pigs from Mizzou in the start area indicating the tracking locations on the pig and tracking zones as in the stem and choice area. (F) Two representative pig tracking sessions showing subsequent right and left turns colored red and blue. Subsequent turns could be distinguished in the short delays indicating more of a continuous alternation trajectory, but not in the longer delay durations indicating the location of the pig in the start area is not helping with subsequent decisions

Figure 5

DeepLabCut-based analysis rules out many head direction confounds for a memory interpretation. Six-point tracking skeletons were acquired using DeepLabCut (red dot = nose, green dot = right ear, yellow dot = left ear, cyan dot = shoulder center, blue = hip center, hot pink = tail) for all frames. White arrows indicated the head direction calculated for that frame. In all plots and images, 0° is toward the back wall of the start area, 90° is toward the right side of the maze, 180° is toward the choice point, and 270° is toward the left side of the maze. (A) Sample six-point tracking skeleton screenshots for NCLP006 across delays (D5, D6, D120 and D240). B) Polar plots for all head directions calculated in corresponding delays (D5–D240). NCLP006 regularly looked toward the center stem door leading to the choice point during the first 5 s of the delays, but in all longer delays head directions tended to be toward back-wall half of the start area. Head directions for NCLP006 never differentiated subsequent right and left turns (all Watson-Williams tests, p’s > 0.05). (C) Sample six-point tracking skeleton screenshots for NCLP010 across delays (D5, D6, D120 and D240). (D) Polar plots for all head directions calculated in corresponding delays (D5–D240). NCLP010 regularly looked toward the right side before making right turns on 5 s delay trials, and the left side before making left turns on 5 s delays (Watson-Williams test, p < 0.05). However, in all longer delays head directions tended to be in the back right corner of the start area and never differentiated subsequent right and left turns (all Watson-Williams tests, p’s > 0.05)