Study animals
Research occurred in the Neotropical lowland rainforest at La Selva Biological Station, Costa Rica (10° 26′ N, 84° 00′ W, 50 m elevation), from June to August 2015. We studied a currently undescribed species of Prionostemma Pocock harvestmen (Opiliones: Sclerosomatidae, referred to as P. sp1. in Refs.35,41,42,43, which roosts during daytime in tree trunks, buttresses, and palm leaves44. At nighttime, they actively forage on the ground and foliage41,42. Harvestmen face a wide diversity of predators, including small mammals, lizards, frogs, spiders, centipedes, and insects45. We collected 135 eight-legged individuals and placed them in clear plastic deli containers (15 × 12 × 10 cm) 24 h before trials. Individuals were fed cucumber, apple, and cat food in captivity every 2 days.
Experimental setup
We recorded harvestmen moving during daytime across a horizontal arena (Supplementary Video S1) in the lab, using the same procedure as Ref.35. To simulate predation attempts, we grabbed individuals by their hind legs and then released them. Hence, we interpreted the animal’s subsequent movements as escape behaviors46. Trials were video recorded with a GoPro HERO4 camera (GoPro, San Mateo, CA, USA) at 120 frames/s. A mirror at 45° perpendicular to the ground on the opposite side of the arena to the camera allowed recording lateral and dorsal views.
The accurate 3D body’s location in the videos was obtained as follows. The focal length, optical center, and radial and tangential lens distortions (i.e. intrinsic camera parameters) for the GoPro camera and lens were obtained using the built-in checkerboard calibration app in MATLAB vR2016a (The Mathworks, Natick, MA, USA). Then, the translation and orientation of the lateral and dorsal views relative to the track (i.e. the extrinsic camera parameters) were estimated using an M-estimator sample consensus algorithm.
Experimental treatments and trials
We induced autotomy by gently grasping the femur of the target leg with forceps as Refs.20,31,33, which immediately resulted in the release of that leg at the coxa-trochanter joint. We measured locomotion for each animal at seven different times: once prior to autotomy, once immediately after, and then 2 h, 6 h, 24 h, and 2 days after autotomy.
The experimental treatments used here varied in the number and type of legs missing (Fig. 1) to reflect the natural occurrence of autotomy. The treatments are named with a number representing the number of legs lost, and a letter representing the type of loss. Treatments were (1) C: intact control individuals with 8 legs. Legs were grabbed with forceps but without inducing autotomy, (2) 1L: missing the right locomotor leg I, (3) 2L: missing both of the locomotor legs I, (4) 2S: missing both sensory legs (legs II), (5) 2A: asymmetrical loss, missing legs I and III from the same side, and (6) 3L: missing three legs (both locomotor legs I and one leg II). Individuals were randomly assigned one treatment (n = 21–24 per treatment). We induced autotomy of hind legs, given that our field survey showed that missing hind legs was 53% more likely than missing rear legs.
Experimental comparisons. Body diagrams in the center column represent Prionostemma sp.1 harvestmen seen in dorsal view. The main prediction for each comparison is described in the right column. Figure was created in Microsoft Power Point version 16.39 (URL: https://www.microsoft.com/en-us/microsoft-365/microsoft-office).
Video analyses
We digitized videos using the custom scripts developed in Mathematica 10.4 (Wolfram Research, Inc., Champaign, IL, USA) described in Ref.35. Briefly, scripts automatically tracked the position of the animal’s body across each view. We then reconstructed its three-dimensional trajectory over time using built-in functions (i.e. estimateFundamentalMatrix and triangulate) and tools developed by Ref.47 for MATLAB. Using the XYZ trajectory of body position, we calculated the kinematics of the animal’s center of mass (CoM). Automatically tracking the movement of each leg was not possible, unfortunately. Harvestmen legs are very thin (Supplementary Video S1), which prevents a good contrast with the background.
We calculated nine performance, postural, or stride variables to describe harvestmen locomotion as in Escalante et al.35. For performance metrics, we used the XYZ positions over time to calculate (1) the average horizontal velocity, hereafter referred to as ‘velocity’, calculated as v_h = (x_final − x_initial)/(t_final − t_initial), where x_initial and x_final are the (x, y) coordinates of the body at the start and end of the trial, respectively. (2) Maximal horizontal acceleration, calculated as a_hmax = max_{t in trial} a_h(t), where a_h(t) is the horizontal component of the acceleration calculated from a quintic smoothing spline fit to 3D position over time. We consider these variables reflect biologically relevant performance. We assumed that harvestmen would aim to sustain fast speed to avoid being captured (velocity), as well as a fast burst of speed to quickly move away from a potential predator (acceleration).
For postural variables, we calculated (3) the three-dimensional sinuosity normalized by time. This unitless measurement is the total path length of the trajectory divided by the linear distance between the endpoints and quantifies the lateral and vertical deviations from a straight path48. We also measured (4) the minimal and (5) the maximal height of the CoM.
For stride kinematics, we visually followed the movement of a focal leg (left leg I). We noted the time when each leg was on the ground (stance phase), and when it was lifted (aerial phase), which together represent one stride. For 2L and 3L treatments we followed the third left leg as the focal leg. We followed three strides to calculate (6) the average duty factor, the proportion of time during each stride that the focal leg was on the ground, (7) average stride frequency, the number of complete strides per second, (8) average stride period, the time to complete one stride, and (9) average stride length, the maximal distance along the x-axis the leg moved during one stride.
To investigate patterns of leg use, we visually followed all legs during three strides and constructed gait diagrams. We did this for five individuals in each experimental treatment, before autotomy, immediately after, as well as 2 days after autotomy. Finally, we visually scored each video based on the type of gait performed. We grouped the type of gaits into “fast gaits” (running and stotting) and “slow gaits” (bobbing and walking). We grouped gaits this way because performance (velocity and acceleration) is similar within gait groups35. Additionally, our focus here was on understanding the consequences of autotomy regardless of gait type.
Body measurements
We measured the length of the left leg IV for each individual to the nearest 0.05 mm using digital calipers. Leg IV length (see Supplementary Table S2) is a good proxy of body size since leg IV was never autotomized, and it correlated with leg I and III lengths (r = 0.39, 0.49, respectively. P < 0.02 for both. See Supplementary Fig. S1). Leg morphology is not sexually dimorphic40 and thus we included both adult females and males in this study. Voucher specimens of all individuals are preserved in 70% ethanol in the Essig Museum of Entomology, University of California Berkeley.
Data analyses
To test the influence of autotomy on the locomotor performance we performed generalized linear mixed models (GLMMs) using velocity or acceleration as the response variable. Predictor variables included as fixed effects were treatment, time since autotomy (treated as categorical, hereafter referred to as ‘time’), gait group (fast or slow), and all possible interactions (Table 1). Individual identity was included as a random effect to account for repeated measurements. Preliminary analyses revealed that neither the leg length nor sex affected locomotor variables (GLM: leg length, sex, and leg length × sex interaction all P > 0.42). Hence, these variables were excluded from the final models. None of the nine kinematic variables were normally distributed (Shapiro tests P < 0.05). However, GLMMs and GLMs are robust to deviations from normality, so they were useful for between treatment and time comparisons.
To identify the variables (and interactions) that affected locomotor performance we ran likelihood ratio tests49 for each model. Three sets of GLMMs (Fig. 1, Table 1) compared for the effect of (1) the number of missing legs (comparing C, 1L, 2L, and 3L treatments), (2) the type of missing leg (C, 2L and 2S), and (3) the side of the body where harvestmen lost legs (C, 2L, and 2A). Post hoc Tukey comparisons examined for differences within treatments over time, as well as differences between treatments at a given time after autotomy. We defined locomotor recovery as the first time point at which the mean locomotor performance was statistically indistinguishable from pre-autotomy levels following experimental leg loss.
Lastly, to examine potential changes in the gait use after autotomy, we compared the number of individuals in each treatment that performed each gait type before and immediately after autotomy using an independence chi-square. To examine gait type changes over time we compared the numbers before and two days after autotomy with an independence chi-square for each treatment. Statistical analyses were run in R (R Development Core Team 2018). A small portion of this dataset (the before autotomy trials of all individuals) were collected as part of another study describing gait kinematics35. Hence, information collected from some of the same individuals in Ref.35 were included in the current dataset, along with five other time points.
Source: Ecology - nature.com
