Fig 1.
(a) Indirect evidence for long-range chemosensory tracking by octopus.
Frame from a video recorded at a depth of 3,238 m near the Davidson Seamount in the Monterey Bay National Marine Sanctuary by the Ocean Exploration Trust vessel EV Nautilus (United States National Ocean and Atmospheric Administration (NOAA), 2019). Numerous Muusoctopus robustus feeding on the carcass of a whale are visible, lower right. At this depth, which is far below the photic zone of the ocean, it is unlikely that octopus can use vision-guided search to arrive at the whale fall, thus chemosensory-guided navigation is the likely search strategy employed. (b) Giant Pacific Octopus (GPO) attached to a spot prawn trap off the coast of Vancouver Island, British Columbia, Canada. Spot prawn traps are baited and placed on the ocean floor, typically at a depth of 130 - 160 m. (c) An octopus eating bait from a baited fish trap. (d) Strategies for Odor Gated Rheotaxis The basic mechanism in odor gated rheotaxis (OGR) is that the animal ‘surges’ upstream (or upwind) on encountering a patch of odorant, and then ‘casts,’ or moves in a zig-zag manner perpendicular to the direction of the mean flow to regain contact with the chemical plume. Several variations on this basic strategy are possible, two of which are shown in the figure. In OGR with counter-turning the animal moves upstream and diagonally across the width of the plume and counter turns into the plume on encountering the plume edge. In OGR with edge-following the animal stays on one side of the plume centerline and makes counter turns to locate a single edge of the plume while also moving upstream. We observed possible evidence of all three mechanisms by octopus during chemosensory tracking. It should be noted that the fine black lines represent a simulated theoretical odor plume and the magenta line is a schematic representation of the animals’ hypothetical trajectory under the three different strategies.
Fig 2.
(a, b) Octopus motion tracking - Principal vectors.
Octopus motion was primarily quantified by two vectors - the body axis vector (blue), and the heading vector (light orange). (b) The angle θ between the body axis and heading vectors is the heading offset angle. A DeepLabCut two-dimensional pose estimation model was trained to locate the eyes of the octopus in the video data (shown in (b) as solid orange dots). The midpoint of the line connecting the two eyes was denoted as the ‘body center’ and was the reference point used to compute kinematic quantities. The body axis vector (blue), is a unit vector that has its base located at the body center, and points in a direction away from the mantle and perpendicular to the line joining the eyes. The heading vector is a unit vector that points in the direction of motion of the body center between two successive video frames. Thus, the heading vector also points in the direction of the velocity of the body center. (c) Flume schematic diagram showing top and side views of the water flume used for the behavioral experiments The flume was 185 cm long and 116 cm wide with a water height between 10 - 12 cm. The water flow was from right to left at a speed of approximately 2 cm/s. Superimposed on the top view is a time-averaged visualization of a simulated chemosensory plume relative to the flume area. (d) Visualization of time-averaged simulated chemosensory plume The heat map shows the time-averaged intensity of light scattered from a simulated chemosensory plume, showing the approximate extent of the plume. The chemosensory plume was simulated by seeding the flow with 9-13 micron diameter glass beads. A syringe pump injected a dilute suspension of the beads approximately iso-kinetically into the flow at the location of the food source. The flow is slightly biased to the bottom of the page because, due to floor drainage and surface ruggedness, it was not possible to level the tank exactly. Typical time traces of the scattered intensity are shown at different locations in the plume demonstrating how chemical concentration fluctuations are likely to vary at increasing distance from the chemosensory source.
Fig 3.
(a, b) Estimating direction of octopus motion relative to arm position.
DeepLabCut predictions of eye coordinates allow for the creation of a body axis vector (blue) that can be used to quantify the orientation of the octopus. The body axis vector has its base at the center of the line joining the two eyes and points perpendicular to this line in a direction pointing away from the mantle. For each octopus, all images were first rotated to a standard orientation with the body axis pointing vertically downwards as shown in (a). A normalized body image was then generated by averaging all images as shown in (b), where the blue arrow shows the stack of images that are added and scaled by the maximum summed pixel intensity to produce the normalized body image (where, r and c represent row and column, respectively, and (r, c)
denotes the normalized intensity). (c) Normalized body images are then used to produce a unique bounding rectangle and bounding circle for each octopus. First, a rectangle that is symmetric about the body axis of the octopus is generated. The extent of the rectangle is such that it contains 95% of all the intensity in the image. Then a circle is drawn with its center at the midpoint between the eyes and its radius equal to the distance to the most distant vertices of the bounding rectangle. Variations in octopus size can now be adjusted for by normalizing the radii of the bounding circles to unity. Bounding rectangles and circles for five different octopuses are shown. (d) Arm positions were estimated by determining the points of intersection of the arms with the bounding circle for that octopus using the NumPy find_peaks algorithm on background subtracted images of the octopus (red dots). The arm that is most closely aligned with the direction of motion, or heading, is termed the heading arm. The angle between the heading arm and the heading vector is called the heading-arm offset angle. (e) Quantifying arm position allows us to track the heading offset angle.
Fig 4.
(a) Octopus can discriminate between baited and unbaited stations in the dark.
Approach paths of the octopus for all approach sequences to the baited stations within the flume. Approaches are backcast ten minutes from the time of eating for each of the three different station configurations (i, ii, iii). The point at which an approach crosses over into a digitally superimposed ‘target box’ is marked by an x on the perimeter of the box. Baited stations are shown in teal/dark green while unbaited stations are shown in orange/brown. Lighter colors denote ‘against flow’ approaches in which the octopus swims upstream to the current and chemosensory plumes could be used for target detection. Darker colors show ‘with flow’ approaches when chemosensory tracking could probably not be used to locate the target. (b) Observed numbers of approaches sorted by their respective location, approach, and station configuration categories. (c) Histogram comparing aggregated approaches sorted by approach condition (with flow or against flow) and target (baited, unbaited/control). (d) Total lengths of target bounding boxes for all four approach and target conditions: 1. With flow - baited target, 2. With flow - unbaited target, 3. Against flow - baited target, and 4. Against flow - unbaited target (colors scheme shown in legend). In the null hypothesis, the number of approaches to a target is proportional to the total external perimeter of the bounding box containing the target (which excludes the common perimeter shared by two target bounding boxes). In other words, the approach probability density, or number of bounding box crossings per unit length is constant. (e) Histograms comparing aggregated number of all approaches to baited and unbaited stations (top), and against flow and with flow approaches (bottom). Statistical comparisons between different approaches were calculated using a binomial test, while comparisons to the null hypothesis (constant probability of approach over the entire target box) were calculated using a chi-squared test. (,
,
,
,
,
, n.s. : not significant).
Fig 5.
Octopus preferentially approached the single baited station from within the plume.
(a) The thick magenta line shows the approximate extent of the plume. This boundary was determined visually using the time-averaged image of the simulated chemosensory plume (Fig 2d). A ‘target circle’ (green) centered on the baited station, and with a radius of 14 cm (the approximate maximum reaching distance for the octopus) was used to quantify the final approach angle of the octopus to the food target. The target circle and the plume boundary were then superimposed on the trajectories of all successful feeding events. The number of approaches to the target from the ‘Inside plume’ region are clearly seen to exceed approaches from the ‘Outside plume’ region. (b) Octopus preferentially approached the target along trajectories that clustered around the centerline of the chemosensory plume rather than from trajectories that lay outside the extent of the plume. (p = , binomial test, expected ratio = 0.17).
Fig 6.
Octopus displayed odor-gated rheotaxis behaviors during chemosensory plume-guided search.
Behaviors included cross-stream redirections, zig-zag casting motions, and surges. Approaches were made in many different body orientations relative to the bait, with minimal body rotation and a complex diversity of arm motions. The baited station location is shown as a green dot, and the approximate extent of the plume is shown in green. Water flow through the flume is from right to left. Octopus images are shown at three-second intervals, and the magenta line shows the trajectory of the body center, which is the midpoint of the line joining the two eyes (Fig 2).
Fig 7.
The distribution of speed and nearest heading arm offset angles is different when octopus are chemosensory tracking.
(a – c) Octopus body center motion trajectories for three different conditions: (i) Inside plume approaches with feeding, (ii) Inside plume trajectories with no feeding, and (iii) No food. Overlaid on the trajectories are yellow ‘hairs’, drawn at three-second intervals, that show the body axis direction of the octopus during the motion. The location of the food is shown by a green dot, and the approximate extent of the chemosensory plume is shown in magenta. (d – f) Polar plots showing time evolution of the body axis direction of the octopus for the three different conditions. (g – l) In panes g - l, actual data points are shown as dots, while the density of dots is represented as a color-coded heatmap. (g – i) Relation between heading offset angle (angle between body-axis vector and velocity vector (shown in inset to pane g)) and speed z-score for all motion trajectories. (j – l) Relation between heading-arm offset angle (angle made by the leading arm with respect to body orientation vector (shown in inset to pane j) and speed z-score for all trajectories.
Fig 8.
Octopus did more fast arm aligned motions during chemosensory tracking.
(a) Relation between speed z-scores and heading-arm offset angles for all motion trajectories. The fast arm-aligned motions (FAAM) are shown in red. (b) Number of FAAM events for three different conditions. (c) Percentage of FAAM for different conditions. (d) Sequences of FAAM motions, shown in red, are superimposed on the ‘in-plume’ chemosensory-tracking motion trajectories. Arrowheads indicate the direction of motion of the octopus. Frames from six examples of FAAM motion sequences are shown in the insets. Octopus images are shown at 0.1 second intervals. The images show the complexity of the body shapes generated during a bout of FAAM. The red arrows point towards the arm most aligned with the velocity vector, which is shown in green. The direction of the body axis is shown with a faint magenta arrow. The quiver plot on the right shows the mean direction vector for all FAAM sequences for the in-plume feeding trajectories. The mean direction vectors are primarily clustered in the direction opposite to the mean flow direction. (e) All motion trajectories for the three different conditions were segmented into 1 second-long lengths and the corresponding speed z-score and heading-arm offset angles for these segments were combined and visualized using UMAP. The top pane shows the behavioral map for all conditions with the FAAM condition superimposed on it. The FAAM sequences are clearly seen to occupy a distinct region in the UMAP visualization.