兽研摘要
捕食可强烈影响种群结果，而运动性能在追逐-逃避互动中可能尤为重要。我们假设基于性能的逃避能力差异可能促成猎物种群的丰度和分布。我们比较了在澳大利亚大陆持续存活的北褐袋狸（Isoodon macrourus）和在大陆已衰退的北袋鼬（Dasyurus hallucatus）的运动性能，并以家犬（Canis familiaris；多品种，3.4–44 kg）作为犬科性能的近似参照。从俯视视频影像中提取速度、加速度和航向变化，用于估计相对逃避性能。犬达到最高的平均和最大速度，而袋鼬表现出最大的转向能力。袋狸是三组中最慢、最不敏捷且最可预测的。袋鼬在机动性上胜过追捕捕食者的能力可能有助于解释其与结构复杂生境的关联，在这种生境中敏捷性应提供更大优势。相比之下，北褐袋狸的持续存活和丰度可能由其他生态特征所支持，因为其躲避追捕捕食者的运动能力似乎有限。这些发现表明，运动性能 alone 不能解释猎物种群结果，但可能与生境复杂性相互作用以有利于不同的逃避策略。
关键词：敏捷性，犬科，袋鼬科，逃避概率，运动，袋狸科
引言
捕食通过其对群落的直接影响驱动自然选择，极大地影响生态系统结构。捕食事件已被归类为伏击、弹道或追逐互动（Moore and Biewener, 2015）。伏击攻击被描述为隐蔽捕食者的一次动作，由通常未察觉的猎物的一个关键刺激引发。弹道拦截要求捕食者预测猎物的路径并致力于单一轨迹进行拦截，具有离散的成功或失败。在追逐捕食中，捕食者与猎物之间的互动延长，涉及多次刺激交换（Clemente and Wilson, 2015b; Moore and Biewener, 2015; Wilson et al., 2018）。
这些捕食类型的结果取决于捕食者与猎物之间的相对性能。对于伏击和弹道捕食者，初始的加速度和速度爆发是成功的重要组成部分，因此已记录了许多物种的最大速度（Garland, 1983）。追逐捕食事件的特征是高速转向，间以加速和减速时刻（Howland, 1974; Garland, 1983; Wilson et al., 2013; Clemente and Wilson, 2015a）。猎物决定了追逐中与转向相关的许多方面（例如转向时机、转弯速度）；为了成功，捕食者必须对猎物的刺激做出反应，最大化机会但最小化能量浪费（Lima and Dill, 1990; Clemente and Wilson, 2015b; Wilson et al., 2015, 2018）。因此，追逐捕食的结果将取决于参与者的相对速度和转向能力（Clemente and Wilson, 2015b; Wilson et al., 2015）。
速度与转向能力之间的生物力学权衡使预测追逐结果进一步复杂化。几项研究已在多种陆生物种中阐明了这些权衡，包括袋鼬（Dasyurus hallucatus）、宽足袋鼩（Antechinus mysticus）、猎豹（Acinonyx jubatus）、犬（Canis familiaris）和马（Equus caballus）（Tan and Wilson, 2011; Wilson et al., 2013; Wynn et al., 2015; Wheatley et al., 2018; Haagensen et al., 2022）。高速转弯的四个关键挑战是肢体力量限制、身体重新定向、转弯时倾斜和摩擦系数（Haagensen et al., 2022）。转弯需要在切向和向心轴上产生力，并受肢体可产生力量大小的限制；在最大速度下，向心力的增加导致切向力的减少（Greene, 1985; Weyand et al., 2000）。因此，随着肢体在旁矢状轴上功能更好（该轴平行于切向力），将身体重新定向到新航向角变得重要（Walter, 2003）。向转弯内侧倾斜是实现这种对齐的另一种方式；然而，需要适当的进入转弯的力量水平以避免侧向倾倒（Biewener and Patek, 2018）。可传递到基质的力水平由摩擦系数决定，而摩擦系数受足/垫形态和基质本身的影响（Russell and Bryant, 2001; Tan and Wilson, 2011; Wynn et al., 2015; Wheatley et al., 2018）。为提高转弯能力，个体必须克服这些挑战；然而，个体的质量和步态也促成成功。
体型大小在决定转向能力方面也很重要。一个比较不同体型捕食者（30 kg、250 kg）与猎物（3–200 kg）捕获概率的数学模型显示，当仅模拟直线追逐时，最大的捕食者能够捕获所有猎物；然而，当模拟转向追逐时，最大的捕食者无法捕获较小的猎物（Wilson et al., 2015）。在这些转向追逐中，较小的捕食者能够捕获除最小猎物（3 kg）以外的所有猎物。一项关于家犬转向的研究表明，中型犬（10–25 kg）具有更强的切向（直线）加速度和向心加速度能力（Haagensen et al., 2022）。较小的犬（<10 kg）受限于其力量产生能力，而大型犬（>25 kg）在产生向心加速度的能力方面受到限制。其他方面也有助于野外有效转向，包括失误以及对失误的恐惧（Amir Abdul Nasir et al., 2017; Wheatley et al., 2021）、以次优速度奔跑（Wilson et al., 2013）、转向策略（如时机）（Wilson et al., 2015）以及形态适应（如足部或爪的形态）；然而，关于不同形态和步态如何影响转向能力，人们所知甚少（Jindrich and Qiao, 2009）。
在澳大利亚，入侵性捕食者被认为与本土物种的灭绝和衰退有关（Doherty, 2015; Woinarski, Burbidge and Harrison, 2015; Doherty et al., 2019; Fleming et al., 2021）。对于处于关键体重范围（CWR）内的物种而言尤其如此，该范围提出了澳大利亚的一种趋势，即许多体重在35至5500 g之间的陆生物种相较于该范围之外的物种已经灭绝（Chisholm and Taylor, 2007）。运动表现的差异被认为可能是这些入侵性捕食者在捕猎澳大利亚动物区系时取得成功的一个潜在解释（Banks and Dickman, 2007; Gaschk and Clemente, 2022）。捕食的影响因土地开垦、火灾、气候变化和栖息地退化等其他促成因素而加剧（Allen, 2011; Griffiths and Brook, 2014; Johnstone, Lill and Reina, 2014; Murphy et al., 2019）。这些栖息地变化可能对陆地运动及捕食者—猎物相互作用产生影响（Winchell, Battles and Moore, 2020）。
物种生态学的复杂性意味着种群衰退在不同物种间并不一致，许多本土动物区系尽管处于CWR之内，却相对未受影响（Radford et al., 2018; Gaschk and Clemente, 2022）。北部长鼻袋狸（Isoodon macrourus——以下称袋狸；500–3100 g）严格陆生，且处于CWR之内。尽管如此，它们表现出极小的分布区消失，属于无危物种（Lunney et al., 2016）。相比之下，北部袋鼬（D. hallucatus——以下称袋鼬；300–900 g）属于濒危物种，其分布已缩减为澳大利亚北部破碎化的种群（Braithwaite and Begg, 1995; Oakwood, Woinarski and Burnett, 2016）。
澳洲野犬和野犬是常见的捕食者，其分布与袋狸和袋鼬的分布重叠，但它们有效追捕这两种猎物的程度尚不清楚。这两个物种都出现在热带和亚热带生物区的澳洲野犬和野犬的食谱中（Gaschk and Clemente, 2022）。如果运动表现（速度和机动性）是逃脱犬科捕食压力的重要决定因素，那么与不依赖运动表现的物种相比，这将增加多个季节存活的个体数量。我们假设袋狸相较于袋鼬会表现出更高的逃脱概率，反映在更高的最高速度或加速度上。雌性袋鼬的转向表现已与随后季节的存活相关联；然而，这尚未被量化并与犬科动物的转向表现进行直接比较，也未与同一岛屿上的袋狸建立联系（Rew-Duffy et al., 2020）。因此，我们预测袋鼬由于其较小的体型及所提示的重要性，将具有相对于犬类更强的转向能力。如果我们未观察到猎物物种有效的逃脱概率，我们可以假定其生态学的其他方面对个体和种群存活很重要。
材料与方法
视频整理
袋狸和袋鼬的视频来源于昆士兰大学Wilson Performance Lab。袋鼬视频取自Wynn et al. (2015)；袋鼬（Dasyurus hallucatus Gould 1842）在Groote Eylandt（澳大利亚北领地）捕获，并在由两条轨道（L: 3.4 m，H: 1.23 m，W: 0.4 m）构成的跑道上奔跑，两条轨道通过可拆卸的角度结构连接，形成一条限制在约0.15 m宽的转弯通道。跑道拐角调整为135度、90度和45度，地面覆盖橡胶垫。袋狸视频取自Del Simone (2024)；袋狸（Isoodon macrourus (Gould 1842)）在Groote Eylandt和Hidden Vale野生动物中心拍摄。袋狸沿着由星形桩构建的跑道奔跑，桩间及桩周拉紧遮阳布形成两侧边界（宽度0.75 m）。该跑道通向一个更大的区域，该区域使用与跑道相同的边界构建，使袋狸能够自行决定转弯的方向和角度。袋狸在壤土或黏土/草地基质上奔跑。袋鼬和袋狸均通过安装在转弯顶点处的摄像机以240帧/秒拍摄（Casio EXILIM HS EX-ZR200，512×384）。袋狸和袋鼬均通过人的存在鼓励其奔跑，利用其逃避捕食的本能。犬类视频来源于阳光海岸大学生物力学与生物机器人实验室（3.4–44 kg；品种列表见Haagensen et al., 2022）。犬（Canis familiaris Linnaeus 1758）在草地上围绕设置在转弯弧上的不同数量锥桶奔跑；锥桶数量决定转弯的急缓程度。犬由训练员鼓励奔跑——它们要么追逐物品（如球），要么跟随训练员的方向。使用受过训练的家犬来了解最大速度和转弯表现，作为野犬或澳洲野犬表现的近似参考。促使犬追逐物体或服从指令模拟了捕猎行为，因此，使用该方法进行的奔跑所显示的速度和转弯能力与野生猎犬密切相关。有关动物捕获方法、伦理和许可的更多细节可在各自论文中找到，犬品种列表见表S2。
视频分析
视频在MATLAB中使用DLTdv8（v.8.0）（Hedrick, 2008）进行二维标注。在整个转弯序列中，在动物身上标注三个点（鼻、尾、后足足迹；图1A），并在每个序列开始时额外标注两个点用于校准。校准点按已知间距标注。对于敏捷犬，这是地面上的测量距离[犬（D）：1.206 m，0.057像素 mm−1]，对于袋鼬和袋狸，这是跑道顶点宽度[袋鼬（Q）：0.15 m，0.311像素 mm−1；袋狸（B）：0.75 m，0.147像素 mm−1]。当外侧（转弯方向的对面）后足接触地面（足迹）时进行标注。一个步幅定义为从一个后足足迹到下一个后足足迹，每个性能指标均在单个步幅上计算。在整个转弯过程中逐帧连续标注鼻点和尾点：鼻尖和尾基部。对于犬，由于视野较宽，从第一次足迹开始直到加速离开转弯时，依次标注尾部和鼻部。所有个体均在转弯顶点前后至少一个步幅处进行标注。
[图：视频分析和计算指标的视觉展示。(A) 标注点包括鼻（栗色）、尾（粉色）和]
转弯变量的计算
步幅长度（m）通过将一个步幅内（足迹至足迹）连续鼻点之间的像素距离求和并转换为米来计算。步幅持续时间（s）通过将一个步幅的视频帧数总和除以每秒帧数来计算。频率（Hz）计算为1/持续时间。一个步幅的速度（v，m s−1）通过步幅长度除以每个步幅的持续时间来计算。步幅间速度（vBS，m s−1）通过计算两个连续步幅之间的速度差来获得，这也揭示了步幅间的加速和减速。支撑中期，即一个步幅内的中点（图1A），通过将每个步幅持续时间（以帧计）减半并加到原始足迹的帧数上来确定。支撑中期用于计算支撑中期持续时间（s）、支撑中期长度（m）和支撑中期速度（m s−1）。这些变量未进行正式分析，但用于生成其他变量。切向加速度（at，m s−2），即与转弯相切的加速度（前后方向），通过vBS除以支撑中期持续时间来计算。
体角（θB，度）通过计算单帧内连接尾点和鼻点的直线与每个视频y轴的夹角来确定，其中0度为负y方向（图1B）。体角在支撑中期计算（支撑中期θB，度），身体旋转（RoB，度）通过计算连续步幅支撑中期θB之间的差值来确定。序列中每个步幅的航向角（θH，度）通过使用连接连续鼻点的直线相对于负y方向来计算（图1A）。
航向变化（CoH，度）通过计算连续步幅间航向角的差值获得，并根据转向方向进行调整，使差值始终为正（顺时针、逆时针）。将CoH从度转换为弧度，再除以连续支撑中期之间的时间，计算角速度（ω，rad s−1）。转弯半径（r，m）按r=v/ω计算，向心加速度（ac，m s−2）按ac=vω计算，其中v为步幅速度。
摩擦系数可通过假设特定速度和半径的成功转弯需要最小摩擦力来估算（Alexander, 1982）。遵循这些方法，采用从0.01到4.96、增量为0.05的序列作为动物预期的摩擦极限。将每个摩擦极限下的预测角速度（ωexp，rad s−1）与记录的角速度（ωactual，rad s−1）进行比较，取ωactual超过ωexp的转弯的95%上限。
转弯的到达与离开
为比较转弯内相似的步幅，对步幅赋予字母数字标识符。步幅内最剧烈的CoH被视为转弯的顶点。进入顶点前的最后一个步幅标记为到达（A）1，离开顶点后的第一个步幅标记为离开（D）1。转弯内的步幅向远端递增（例如A2、A1、顶点、D1、D2）。
性能密度图
我们利用捕食者与猎物性能景观之间的相对重叠来推导估计逃逸概率。该指标代表基于比较运动性能预测的逃逸可能性，而非基于实际追捕结果的实证观察概率。为简便起见，全文将该指标称为估计逃逸概率。创建包含所有犬的性能密度图，x轴为速度（v，m s−1），y轴为角速度（ω，rad s−1）。使用ks包的‘kde’和base R的‘with’对数据计算核密度等值线图（https://CRAN.R-project.org/package=ks）。以5为增量对10至95的密度序列重复此步骤。使用sp包的‘Polygon’、‘Polygons’和‘SpatialPolygons’函数提取等值线数据并转换为shape文件（https://CRAN.R-project.org/package=sp）。对于10%的犬等值线图，为每个袋鼬受试者创建相同变量（x=速度，y=角速度）的性能密度图，并对数据计算10%的核密度等值线。使用raster包的‘intersect’计算每个袋鼬受试者10%等值线与整体犬10%等值线的重叠面积，并对所有个体的重叠取平均值（https://CRAN.R-project.org/package=raster）。使用10–95%的犬等值线图对10%至95%的袋鼬等值线图重复此步骤。对袋狸使用速度和角速度重复此过程，然后对两种猎物物种使用切向加速度和向心加速度再次重复，以创建热图。
统计分析
所有数据均在R版本4.0.5（2021-03-31）中分析和处理（https://www.r-project.org/）。分位数回归用于分析具有限制边界的数据；此类分析使用quantreg包（v.5.94）的‘rq’函数（https://CRAN.R-project.org/package=quantreg）。分位数回归使用代表边界限制的上限或下限5%分位数计算。分位数回归分析用于角速度和转弯半径。使用nlme包的‘lme’函数时，线性混合效应模型用于分析所有其他数据（https://CRAN.R-project.org/package=nlme）。
对于性能景观热图，对沿>55%估计逃逸概率线分布的每个网格计算努力差异（捕食者努力−猎物努力）。>55%的估计逃逸概率表明猎物物种逃逸的机会大于被捕获的机会。努力差异和密度图使用Welch's t检验进行比较。
计算每个物种的速度、角速度、转弯半径、CoH、RoB、θB、at和ac的密度曲线，并使用范围和高度比较步幅的变异性，作为可预测性的近似。将每个范围转换为均值的比例，以在不同单位间标准化。密度曲线的范围和高度使用Welch's t检验进行比较。
线性混合效应模型用于比较转弯序列内等效的步幅，以向心加速度、切向加速度、速度、角速度、步幅长度和步幅频率作为单独的因变量。物种和步幅位置作为固定效应，受试者作为随机效应。使用emmeans包进行估计边际均值的成对比较（表S1）（https://CRAN.R-project.org/package=emmeans）。
结果
比较速度与转弯指标
对三种动物的转弯动作进行了分析和比较，以帮助理解运动性能对逃脱概率的贡献。转弯性能反映了产生切向加速度和向心加速度的综合能力；因此，速度、转弯半径、角速度和身体重新定向均对整体性能有贡献。总体而言，我们分析了29只袋狸的387个步幅、17只袋鼬的467个步幅以及20只犬的869个步幅。犬的速度远快于两种猎物物种，平均（±s.d.）速度为3.55±1.27 m s−1，观测最大值为8.08 m s−1。袋鼬的平均速度为2.56±0.74 m s−1，最大速度为4.33 m s−1，袋狸的平均速度为2.38±0.76 m s−1，最大速度为4.84 m s−1（图2）。犬通过增加步幅长度（斜率=−0.33，R2=0.81；图2H）来提高速度，但未调整步幅频率（斜率=−0.02，R2=0；图2I）。袋鼬倾向于通过调节步幅频率来提高速度（斜率=0.63，R2=0.10；图2I），对步幅长度的调整最少（斜率=0.12，R2=0.54；图2H），而袋狸以相似的速率同时调节频率和长度来提高速度（频率：斜率=0.23，R2=0.04；长度：斜率=0.21，R2=0.64；图2H,I）。物种与步幅长度和步幅频率之间的交互作用在犬与猎物物种之间均显著（袋狸频率：t1717=3.94，P<0.001；袋鼬频率：t1717=10.94，P<0.001；袋狸长度：t1717=−10.4，P<0.001；袋鼬长度：t1717=−18.8，P<0.001）。
[图：北方棕袋狸和北方袋鼬与犬相比的单个步幅转弯变量。（A,B]
袋鼬的平均角速度最高（袋鼬：2.83±2.99 rad s−1；犬：1.43±1.07 rad s−1；袋狸：1.18±0.95 rad s−1；图2A,B）。由于转弯的限制（随着速度增加，转弯半径增加；图2C,D），比较了各物种转弯半径下限（5%）的斜率。袋鼬在三个物种中斜率最浅（斜率=0.29；下限0.20，上限0.36），与袋狸（斜率=0.48；下限0.41，上限0.64）和犬（斜率最高，斜率=0.81；下限0.62，上限0.91）均显著不同。袋鼬和犬在单个步幅内分别能够将转弯方向改变最多109度和103度（图2F），而袋狸的数值要低得多（75度；图2E）。尽管如此，袋狸（斜率=13.35；图2J）调整其CoH的速率快于犬（斜率=17.88），而袋鼬调整其CoH的速率快于两者（斜率=8.23；图2J）。袋鼬产生向心加速度的能力显著大于犬（图2G）。袋鼬的向心加速度（斜率=−0.53；R2=0.47；图2G）与切向加速度呈负相关，且斜率显著陡于犬和袋狸（犬斜率=−0.30；袋狸斜率=−0.13；t1717=−6.20）。
摩擦系数
袋鼬的摩擦系数（1.96±0.57）显著高于犬（0.88±0.22；t20.24=−7.29，P<0.001）和袋狸（0.69±0.22；t18.95=−8.75，P<0.001）（图3）。犬和袋狸之间也显著不同（t39.15=2.90，P=0.007）。袋鼬摩擦系数的增加可以解释其增强的转弯能力，而袋狸较差的摩擦系数可以解释其在速度和机动性方面的表现。
[图：预测北方棕袋狸、犬和北方袋鼬单个步幅的摩擦系数。Ve]
等高线重叠景观分析
通过比较所有犬只与平均个体猎物对象的密度等高线图之间的重叠情况，进一步探究了犬、袋鼬和袋狸的运动能力（图4）。该分析展示了猎物与捕食者之间的相对努力程度和估计逃脱概率。在比较切向加速度和向心加速度的性能景观时，袋鼬显示出较高的估计逃脱概率；然而，在比较角速度和速度时，这一概率有所降低（图4B,C）。袋狸在速度景观和加速度景观中均显示出极低的估计逃脱概率（图4D,E）。对于袋鼬而言，在两个性能景观中，55%的估计逃脱概率均超过了等效努力线（黑色），而袋狸则低于该线。计算了努力差异（捕食者努力−猎物努力），以了解达到55%估计逃脱概率所需的努力程度。在加速度景观中，袋鼬达到55%估计逃脱概率所需的努力比犬少14±9.5%，而袋狸达到相同估计逃脱概率所需的努力多36±11.2%（t26.13=−14.45，P<0.001；图4B,D）。在速度景观中，袋鼬达到55%逃脱概率所需的努力少3.2±8.0%，袋狸所需的努力多25.8±10.8%（t34.21=−9.99，P<0.001；图4C,E）。袋鼬在加速度景观中显示出较小的努力差异，而袋狸在速度性能景观中显示出较小的努力差异（t45=−4.35，P<0.001；图4B,C；t29.6=2.80，P=0.008；图4D,E）。基于这些估算，袋鼬似乎以远少于犬所需的努力即可逃脱捕食，而袋狸则需要接近最大努力才能逃脱未展现全速奔跑能力的犬。
[图：热图展示了等高线重叠景观分析的结果，该分析将犬的性能景观与以下物种的性能景观进行比较]
可预测性的密度
对密度图进行分析以展示转弯过程中选择的可变性（图5）。袋狸显示出比袋鼬（范围1.18±0.46，最大值0.14±0.15）和犬（范围1.12±0.51，最大值0.16±0.15）更小的比例范围（0.70±0.28）和更高的最大值（0.22±0.20）。然而，仅袋鼬与袋狸之间的比例范围显示出显著差异（t11.5=−2.467，P=0.03）。速度和转弯半径是袋狸显示出比袋鼬更低峰值的仅有两个性能指标；然而，袋鼬在两个图中均显示出相似或更宽的范围（图5A,D）。犬在速度和转弯半径方面显示出比其他物种更宽的范围。袋狸密度图范围有限且最大值增加，表明其逃脱轨迹比袋鼬更可预测。
[图：用于犬、北棕袋狸和北袋鼬转弯分析的性能指标密度图。（]
等效步幅分析
比较了转弯内的等效步幅（图6）。物种对所有变量均有显著影响，其中对步频的影响最大（at：f2,61=86.21，P<0.001；ac：f2,61=46.85，P<0.001；v：f2,61=22.18，P<0.001；ω：f2,61=95.52，P<0.001；步幅长度：f2,88=80.80，P<0.001；步频：f2,111=154.05，P<0.001）。步幅位置对所有六个变量均有显著影响，其中角速度观察到最大的F统计量（at：f6,1637=117.52，P<0.001；ac：f6,1637=171.15，P<0.001；v：f6,1637=112.39，P<0.001；ω：f6,1637=191.09，P<0.001；步幅长度：f6,1634=31.68，P<0.001；步频：f6,1634=16.48，P<0.001）。袋鼬在转弯顶点显示出最高的向心加速度，而袋狸在顶点显示出显著较低的向心加速度（图6A）。袋鼬在进入顶点时显示出比犬更大的负切向加速度，但在离开时以相似的速率加速。袋狸在转弯顶点期间显示出更大的切向加速度，与犬和袋鼬相比均显著（图6B）。袋鼬在接近顶点时速度与犬相似，但在离开顶点时显示出显著较慢的速度（图6C）。袋狸在步幅A1、顶点和D1中显示出显著低于犬的角速度，而袋鼬在顶点显示出显著更高的角速度（图6D）。在转弯序列中，袋狸在接近顶点时（A3–A1）较慢，且未使用那么多的负切向加速度。虽然其在顶点和D1步幅期间的速度与犬和袋鼬相当，但它们在离开顶点时（D1–D3）似乎并未增加切向加速度或速度。与两种猎物物种相比，犬在整个转弯过程中显示出增加的步幅长度和降低的步频（图6E,F；另见图2H,I）。
[图：犬、北棕袋狸和北袋鼬转弯内的比较步幅。（A）向心加速度（]
讨论
自然实验
我们最初的假设是，袋狸作为猎物物种应具有更强的整体运动能力；然而，袋鼬表现出更强的运动能力。基于运动性能，袋狸与袋鼬相比速度更慢。轮廓重叠分析显示，袋狸在被犬类追捕时的估计逃脱概率较低，而袋鼬的估计逃脱概率较高，主要通过其比犬类更快调整转向变化率（CoH）的能力实现。袋鼬相对于犬类的运动性能支持了Groote Eylandt种群的发现，在该地袋鼬显示出较高的朴素占有估计值，且转向能力与连续繁殖季节的补充量成正比（Rew-Duffy等，2020；Heiniger、Davies和Gillespie，2020a）。Groote Eylandt上的袋狸尽管相对于岛上存在的犬科动物其估计逃脱概率较低，但也具有较高的朴素占有估计值（Heiniger、Davies和Gillespie，2020a）。同一研究表明，两个物种在澳大利亚大陆的观测均减少，但袋鼬的估计值减少了约98.47%，而袋狸减少了52.9%，这并未反映这两个物种的运动性能。这很大程度上揭示了Groote Eylandt的情况——一个偏远的岛屿，拥有未受狐狸（Vulpes vulpes）、甘蔗蟾蜍（Rhinella marina）、入侵食草动物导致的栖息地退化以及猫存在受抑制的原始栖息地，所有这些因素都会导致这些动物群数量下降（Oakwood，1997；Heiniger等，2020b）。那么，在种群水平上，运动能力和逃避捕食对这些物种的贡献有多大？
运动能力是否有助于北袋鼬的存活率？
虽然袋鼬具有更强的转向能力，但与犬类相比其较低的最高速度会使它们在无障碍直线奔跑的情况下处于脆弱状态。Wilson等（2015）表明，猎物转向时机不当会导致捕食者“切角”，从而减小体型限制对转向的影响。如果不切角，捕食者必须匹配猎物的角速度，否则有转向过冲的风险，从而增加猎物与捕食者之间的距离（Wilson等，2015）。在大陆上，袋鼬在高度复杂的环境（岩石露头）中密度更高，而在开阔林地或栖息地复杂性降低的地方较少发现（Oakwood，1997；Oakwood等，2016）。一些捕食者，如狼蛛和猎豹，在复杂栖息地中捕猎时狩猎成功率降低（Mills等，2004；Morice等，2013）。栖息地复杂性的益处尚未完全被理解，提出的理论包括视觉或物理障碍、避难所丰富度或资源丰富度（Bonenfant和Kramer，1996；Laundré、Hernández和Ripple，2010；Morice等，2013）。尽管如此，短尾矮袋鼠（Setonix brachyurus）等猎物物种更常见于复杂栖息地，并被建议利用这些栖息地类型来限制狐狸的捕食（Hayward等，2005）。对于袋鼬而言，更强的转向性能在捕食者无法切角且直线奔跑受限的复杂栖息地中会产生更大的效果（Wilson等，2015；Rew-Duffy等，2020；Wheatley等，2020）。这可以解释在复杂岩石露头中报道的袋鼬密度更高，这将有利于其增强的机动性。
与袋鼬偏好岩石栖息地相比，袋狸更喜欢林地遮荫下茂密的地面植被，用于觅食和躲避捕食者；然而，它们适应性很强，可以栖息于广泛的环境范围（Kemper等，1989；FitzGibbon、Wilson和Goldizen，2011）。在Groote Eylandt上，袋鼬会占据具有茂密地面植被的类似栖息地（Heiniger等，2020b）。在澳大利亚大陆，火灾和入侵食草动物对茂密植被的破坏导致了袋鼬数量的下降，而袋狸则表现出更强的适应性（Oakwood，1997；Kemper等，1989；FitzGibbon、Wilson和Goldizen，2011）。
运动能力是否有助于北方棕色袋狸的存活率？
在转向能力方面，袋狸的表现不如袋鼬和犬类，且这一结果无法归因于其选择了次最大速度。猎物通常会选择以次最大速度奔跑，以减少失误并降低可预测性（Wynn et al., 2015; Amir Abdul Nasir et al., 2017; Wilson et al., 2013）。本研究中记录的袋狸平均速度与其他关于袋狸运动的研究中记录的平均速度相似，因此它们不太可能是在限制速度（Tay et al., 2022）。增加逃逸轨迹的变异性也与追捕交互中猎物的存活率密切相关（Moore et al., 2017）。然而，在比较各物种间转向参数的变异性时，袋狸是最具可预测性的，且不太可能在逃逸轨迹中增加变异性（图5）。袋狸无法超越犬类，加之其可预测的轨迹，使其成为澳洲野犬/野狗的常见目标。这反映在犬类食谱中袋狸类物种的出现频率高于袋鼬类（包括袋鼬）；然而，袋狸在澳大利亚大陆仍维持着庞大的种群数量（Gaschk and Clemente, 2022）。
尽管面临犬科捕食压力和较差的运动表现，袋狸仍能维持广泛的分布，这可能与其生态学的多个方面有关。其繁殖策略——每胎4-5只幼崽、相对较短的妊娠/断奶期（约72天）以及每年4.7胎的年繁殖力——将使种群得以保持稳定（Gemmell, Veitch and Nelson, 1999; FitzGibbon, 2015; Tay et al., 2022）。
袋狸的存活也可能归因于其适应性觅食行为，以限制被捕食者发现或缩短与避难所的距离（Lima and Dill, 1990）。恐惧景观是一个概念，认为动物会主动避开距离已知避难所过远或缺乏足够复杂性的觅食区域（Laundré, Hernández and Ripple, 2010）。袋狸在捕食者气味存在时会改变其觅食行为，在靠近气味地点觅食时增加警觉性，或完全避开该地点（Bytheway and Banks, 2019）。研究还表明，袋狸会根据降雨变化将其活动高峰期从双峰型转变为单峰型，这与饮食变化相关（Diete et al., 2017）。因此，多功能的觅食行为和可靠的繁殖很可能支撑了种群在犬科捕食压力下的持续存在，而速度和机动性对个体存活的贡献较小。
澳洲野犬和家犬
澳大利亚动物区系在澳大利亚面临三种犬科动物的捕食：家犬/野狗、澳洲野犬和狐狸。犬类是澳洲野犬捕食压力的良好近似代表，但同时也直接构成这种压力。澳洲野犬与自由放养野狗之间骨骼形态的许多差异已在颅骨中被识别；然而，这些差异处于家犬品种的形态范围内（Smith et al., 2019; Jackson et al., 2021; Koungoulos et al., 2024）。澳洲野犬与家犬的基因组比较表明，两者在遗传上可以区分（Smith et al., 2019; Cairns et al., 2023; Souilmi et al., 2024）。然而，尽管澳洲野犬与自由放养野狗之间存在这种可区分的差异，它们在相似的生态系统和条件组合中共享相似的捕猎生态；它们是机会主义猎手，在某些情况下具有共栖行为（Jackson et al., 2021; Gaschk, 2023）。由于澳洲野犬表现出相似的捕猎生态、条件性社会性、不可区分的形态以及与澳大利亚犬种的历史杂交，本研究中使用的家犬范围很可能涵盖了澳洲野犬的运动能力（Smith et al., 2019）。
摩擦力、质量和基底
北部长鼻袋鼬在三个物种中表现出最大的转向能力，这似乎与摩擦系数的增加相关。摩擦力允许在不打滑的情况下向基底施加更大的力，在转向过程中至关重要，因为施加的力可以垂直于初始方向（Alexander, 1982; Haagensen et al., 2022）。袋鼬的机动性得益于其较小的体型、蹲伏姿势和足垫形态，使其能够将质心从原始方向（动量）偏转到新方向（Alexander, 1982; Wynn et al., 2015; Wheatley et al., 2018）。此外，它们调节步频的能力增加了转向序列中质心偏转的潜在次数，从而产生更高的角速度和更剧烈的质心高度变化。
犬仅在改变速度时调节步幅长度；加速时，这限制了转弯过程中质心偏移的机会，并增大了这些偏移之间的距离。由于体型更大且姿态更直立，犬在转弯时还必须应对高得多的质心和更大的动量（Haagensen et al., 2022）。相比之下，袋狸姿态低伏且体型小得多，降低了质心，因此应在转弯能力上具有优势。尽管袋狸姿态低伏且体型较小，其摩擦系数却最低（图3）。然而，犬和袋狸的摩擦系数都远低于袋鼬。这可以用这些实验中使用的基质类型来解释。
袋鼬的试验在薄橡胶垫上进行，而袋狸和犬的试验在草地-土壤基质上完成，部分袋狸在壤土上奔跑。一项研究拉雪橇跨越不同基质任务的研究发现，草地的摩擦系数为0.45±0.1，而在橡胶田径跑道上为0.58±0.1（Linthorne and Cooper, 2013）。比较两种基质上的表现时，橡胶表面产生的摩擦力比草地表面高29%。将这一校正应用于袋鼬在橡胶上记录的摩擦系数（1.96/1.29），可估算出在草地上转弯的值为1.52。这仍显著高于犬和袋狸的估算值。因此，尽管橡胶垫提供了一定的机械优势，但并未实质性改变总体结果。
结论
我们探究了运动表现对澳大利亚本土物种逃避犬科捕食概率的影响。犬表现出更高的最高速度，袋鼬表现出更好的机动性和不可预测性，而袋狸则全面处于劣势。对于袋狸，运动表现似乎并不有助于种群存活，这很可能与其生态学的其他方面有关。袋鼬的运动表现，尤其是其机动性，支持其利用复杂栖息地（如岩石露头）来逃避追捕者。未来需要研究以确定在同样退缩至避难栖息地的物种（如短尾矮袋鼠）中，运动表现是否是种群存活的可靠预测因子。本研究强调，尽管运动表现是捕食者-猎物相互作用结果的关键决定因素，但仅凭运动表现可能无法准确预测种群的丰度或分布。
补充材料
[内容请查看原文]
致谢
我们谨此感谢Groote Eylandt的传统土地所有者Anindilyakwa人民以及Hidden Vale的传统土地所有者Yagara人民，慷慨允许我们在其土地上开展本研究。我们感谢Anindilyakwa土地与海洋护林员研究站、Anindilyakwa土地委员会提供设施和支持；Hidden Vale野生动物中心和Turner家族基金会允许使用其设施（https://hiddenvalewildlife.uq.edu.au/）；以及昆士兰敏捷犬俱乐部和Ruth Raymond允许我们拍摄其训练课程。
脚注
[内容请查看原文]
参考文献
[内容请查看原文]
关联数据
补充材料
[内容请查看原文]
查看英文原文▾
ABSTRACT
Predation can strongly influence population outcomes, and locomotor performance may be especially important during pursuit–evasion interactions. We hypothesised that differences in performance-based escape capacity could contribute to the abundance and distribution of prey populations. We compared the locomotor performance of northern brown bandicoots (Isoodon macrourus), which persist across mainland Australia, and northern quolls (Dasyurus hallucatus), which have declined on the mainland, with that of domestic dogs (Canis familiaris; multiple breeds, 3.4–44 kg) as an approximation of canid performance. Velocity, acceleration and changes in heading were extracted from top-down video footage and used to estimate relative escape performance. Dogs achieved the highest mean and maximum speeds, whereas quolls showed the greatest turning capacity. Bandicoots were the slowest, least agile and most predictable of the three groups. The ability of quolls to outmanoeuvre a pursuing predator may help explain their association with structurally complex habitats, where agility should provide a greater advantage. By contrast, the persistence and abundance of northern brown bandicoots are likely supported by other ecological traits because their locomotor capacity for avoiding pursuit predators appears limited. These findings show that locomotor performance alone cannot explain prey population outcomes, but may interact with habitat complexity to favour different escape strategies.
Keywords: Agility, Canids, Dasyurids, Escape probability, Locomotion, Peramelids
INTRODUCTION
Predation greatly influences the structure of an ecosystem through its direct effect on a community, driving natural selection. Predation events have been categorised as ambush, ballistic or pursuit interactions (Moore and Biewener, 2015). Ambush attacks are described as one motion from a concealed predator initiated by one key stimulus from the often-unaware prey. Ballistic interception requires the predator to predict the path of the prey and commit to a single trajectory to intercept, with discrete success or failure. In pursuit predation, the interaction between predator and prey is extended, involving multiple exchanges of stimuli (Clemente and Wilson, 2015b; Moore and Biewener, 2015; Wilson et al., 2018).
The outcome of these predation types depends on the relative performance between the predator and prey. For both ambush and ballistic predators, initial bursts of acceleration and speed are an important component for success, and so maximum speed has been recorded for numerous species (Garland, 1983). Pursuit predation events are characterised by high-speed turning, interspersed with moments of acceleration and deceleration (Howland, 1974; Garland, 1983; Wilson et al., 2013; Clemente and Wilson, 2015a). The prey determines many aspects within a pursuit that relate to turning (e.g. timing of turns, speed through the turns); to be successful, the predator must respond to their prey's stimuli, maximise opportunity but minimise wasted energy (Lima and Dill, 1990; Clemente and Wilson, 2015b; Wilson et al., 2015, 2018). Thus, pursuit predation outcomes will depend on both the relative speed and turning ability of participants (Clemente and Wilson, 2015b; Wilson et al., 2015).
Predicting the outcome of pursuits is complicated further by the biomechanical trade-offs between speed and turning ability. Several studies have illustrated these trade-offs in a variety of terrestrial species including quolls (Dasyurus hallucatus), antechinus (Antechinus mysticus), cheetahs (Acinonyx jubatus), dogs (Canis familiaris) and horses (Equus caballus) (Tan and Wilson, 2011; Wilson et al., 2013; Wynn et al., 2015; Wheatley et al., 2018; Haagensen et al., 2022). Four key challenges of turning at high speed are limb force limitations, reorientation of the body, leaning during a turn and coefficient of friction (Haagensen et al., 2022). Turning requires force production in both tangential and centripetal axes and is limited by the amount of force that can be produced by the limbs; at maximum speed, an increase in centripetal force results in a decrease of tangential force (Greene, 1985; Weyand et al., 2000). Therefore, reorienting the body to the new heading angle becomes important as limbs function better in the parasagittal axis, which is parallel to tangential forces (Walter, 2003). Leaning into a turn is an alternative way to achieve this alignment; however, the appropriate level of force leading into the turn is required to avoid lateral toppling (Biewener and Patek, 2018). The level of force that is transferable to the substrate is determined by the coefficient of friction, which is affected by foot/pad morphology and the substrate itself (Russell and Bryant, 2001; Tan and Wilson, 2011; Wynn et al., 2015; Wheatley et al., 2018). To increase turning ability, individuals must overcome these challenges; however, mass and gait of an individual also contribute to success.
Body size is also important in determining turning ability. A mathematical model comparing the probability of capture with different sizes of predator (30 kg, 250 kg) and prey (3–200 kg) showed that the largest predator was able to catch all prey when only straight-line pursuits were simulated; however, when turning pursuits were simulated, the largest predator was unable to catch smaller prey (Wilson et al., 2015). During these same turning pursuits, the smaller predator was able to capture all except the smallest prey (3 kg). A study of turning in domestic dogs revealed that medium-sized dogs (10–25 kg) had a greater capacity for tangential (straight line) and centripetal acceleration (Haagensen et al., 2022). Smaller dogs (<10 kg) were limited by their force production, while large dogs (>25 kg) were limited in their ability to produce centripetal acceleration. Other aspects contribute to effective turning in the wild, including mistakes and the fear of making them (Amir Abdul Nasir et al., 2017; Wheatley et al., 2021), running at suboptimal speeds (Wilson et al., 2013), turning strategies (e.g. timing) (Wilson et al., 2015) and morphological adaptations (e.g. foot or claw morphology); however, much less is known about how the different morphologies and gait affect turning ability (Jindrich and Qiao, 2009).
In Australia, invasive predators have been implicated in the extinction and decline of native species (Doherty, 2015; Woinarski, Burbidge and Harrison, 2015; Doherty et al., 2019; Fleming et al., 2021). This is particularly true for species within the critical weight range (CWR), which proposes a trend in Australia in which many terrestrial species between 35 and 5500 g have gone extinct in comparison to those outside that range (Chisholm and Taylor, 2007). The difference in locomotor performance has been suggested as a potential explanation for the success these invasive predators possess when hunting Australian fauna (Banks and Dickman, 2007; Gaschk and Clemente, 2022). The effect of predation is compounded by land clearing, fire, climate change and habitat degradation, among other contributing factors (Allen, 2011; Griffiths and Brook, 2014; Johnstone, Lill and Reina, 2014; Murphy et al., 2019). These habitat changes likely have an effect on terrestrial locomotion and predator–prey interactions (Winchell, Battles and Moore, 2020).
The complicated nature of a species ecology means that decline of populations has not been consistent across species, with many native fauna remaining relatively unaffected, despite falling within the CWR (Radford et al., 2018; Gaschk and Clemente, 2022). Northern brown bandicoots (Isoodon macrourus – from here on bandicoots; 500–3100 g) are strictly terrestrial, and within the CWR. Despite this, they show minimal range extirpation and are of least concern (Lunney et al., 2016). In contrast, northern quolls (D. hallucatus – from here on quolls; 300–900 g) are endangered and their distribution has been reduced to fragmented populations across northern Australia (Braithwaite and Begg, 1995; Oakwood, Woinarski and Burnett, 2016).
Dingoes and wild dogs are common predators whose distribution overlaps with that of the bandicoots and quolls but the extent to which they effectively pursue these two prey species is unknown. Both species are present in the diets of dingoes and wild dogs in the tropical and subtropical bioregions (Gaschk and Clemente, 2022). If locomotor performance (speed and manoeuvrability) is an important determinant of escape probability from canid predation pressure, this would increase the number of individuals that survive for multiple seasons compared with a species that does not rely on locomotor performance. We hypothesised that bandicoots would display a greater escape probability in comparison with quolls, reflected in higher top speeds or accelerations. The turning performance of female quolls has been linked to survival in subsequent seasons; however, this has not been quantified and compared directly with that of canids, and no link has been made with bandicoots from the same island (Rew-Duffy et al., 2020). As such, we predicted that quolls would have a greater turning ability in relation to dogs because of their smaller size and suggested importance. If we do not see an effective escape probability for prey species, we can assume some other aspect of their ecology is important for individual and population survival.
MATERIALS AND METHODS
Video collation
Videos of bandicoots and quolls were sourced from the Wilson Performance Lab, University of Queensland. Quoll videos were obtained from Wynn et al. (2015); quolls, Dasyurus hallucatus Gould 1842, were captured on Groote Eylandt (Northern Territory, Australia) and run on a trackway constructed using two tracks (L: 3.4 m, H: 1.23 m, W: 0.4 m), joined via a removable angle structure, creating a turning lane that was confined to approximately 0.15 m. The corner of the track was adjusted to 135 deg, 90 deg and 45 deg, and the floor was covered in a rubber matting. Bandicoot videos were sourced from Del Simone (2024); bandicoots, Isoodon macrourus (Gould 1842), were filmed on Groote Eylandt and at Hidden Vale Wildlife Centre. Bandicoots ran down a trackway constructed of star pickets with shade cloth pulled tight between/around each picket to create the sides (width 0.75 m). This trackway opened into a larger area, which was constructed with the same boundary as the trackway, allowing the bandicoots to determine the direction and angle of their turns. Bandicoots were run on a loamy soil or a clay/grassy substrate. Both quolls and bandicoots were filmed via a mounted camera at the apex of the turn at 240 frames s−1 (Casio EXILIM HS EX-ZR200, 512×384). Both bandicoots and quolls were encouraged to run by the presence of a human, taking advantage of their instinct to flee from predation. Dog videos were sourced from The Biomechanics and Biorobotics Lab, University of the Sunshine Coast (3.4–44 kg; for list of breeds, see Haagensen et al., 2022). Dogs, Canis familiaris Linnaeus 1758, were run on a grass field around a varying number of cones situated in a turning arc; the number of cones determined the severity of the turn. Dogs were encouraged to run by their trainers – they were either chasing an item (e.g. ball) or following the direction of the trainer. Trained domestic dogs were used to understand maximum speeds and turning performance as an approximation for performance in wild dogs or dingoes. Prompting dogs to chase objects or take orders simulates the hunting of prey and, as such, runs undertaken using this methodology would show speeds and turning ability closely associated with that of wild hunting dogs. More details on methods of animal capture, ethics and permits can be found in the respective papers and a list of dog breeds can be found in Table S2.
Video analysis
Videos were annotated in 2-dimensions, in MATLAB using DLTdv8 (v.8.0) (Hedrick, 2008). Three points (nose, tail, hind footfall; Fig. 1A) were annotated on the animal throughout the turning sequence, with an additional two points at the start of each sequence for calibration. Calibration points were annotated at a known distance apart. For the agility dogs, this was a measurement on the ground [Dog (D):1.206 m, 0.057 pixels mm−1], for the quolls and bandicoots this was the track apex width [Quoll (Q): 0.15 m, 0.311 pixels mm−1; Bandicoot (B): 0.75 m, 0.147 pixels mm−1]. The outside (opposite side to turning direction) hindfoot was annotated when contacting the ground (footfall). A stride was considered from one hindfoot footfall to the subsequent hindfoot footfall and each performance metric was calculated on individual strides. Nose and tail points were annotated continuously for each frame throughout a turn: on the tip of the nose and the base of the tail. In dogs, because of the wider field of view, the tail and the nose were tagged sequentially from the first footfall until acceleration occurred out of the turn. All individuals were tagged at least one stride before and after the apex of the turn.
[图：Visual display of video analysis and calculated metrics. (A) Annotated points included the nose (maroon), tail (pink) an]
Calculations for turning variables
Stride length (m) was calculated by summing the pixel distance between successive nose points throughout a stride (footfall–footfall) and converting to metres. Stride duration (s) was calculated by totalling the number of video frames of a stride and dividing by the number of frames per second. Frequency (Hz) was calculated as 1/duration. Velocity (v, m s−1) of a stride was calculated by dividing stride length by the duration of each stride. The between stride velocity (vBS, m s−1) was calculated by finding the difference in velocity between two consecutive strides, which also revealed acceleration and deceleration between strides. Midstance, the halfway point within a stride (Fig. 1A), was found by halving each stride duration (in frames) and adding it to the frame number from the original footfall. Midstance was used to calculate midstance duration (s), midstance length (m) and midstance velocity (m s−1). These variables were not formally analysed but were used to create other variables. Tangential acceleration (at, m s−2), the acceleration tangential to the turn (fore–aft), was calculated by dividing the vBS by the midstance duration.
Body angle (θB, deg) was calculated by finding the angle of the line that joins the tail and nose point within a single frame with the y-axis of each video, where 0 deg was negative y (Fig. 1B). Body angle was calculated at midstance (midstance θB, deg) and rotation of the body (RoB, deg) was calculated by finding the difference between midstance θB of consecutive strides. The heading angle (θH, deg) of each stride within a sequence was calculated by using the line that connects consecutive nose points relative to negative y (Fig. 1A).
Change of heading (CoH, deg) was calculated as the difference between the heading angles of consecutive strides, adjusting for the direction of the turn so that the difference was always positive (clockwise, counterclockwise). CoH was converted from degrees to radians and divided by the time between consecutive midstances to calculate angular velocity (ω, rad s−1). Turn radius (r, m) was calculated as r=v/ω, and centripetal acceleration (ac, m s−2) was calculated as ac=vω, where v is stride velocity.
The coefficient of friction can be estimated by assuming that a successful turn of specific speed and radius requires a minimum friction (Alexander, 1982). Following these methods, a sequence in 0.05 increments from 0.01 to 4.96 was used as expected friction limits on the animals. Predicted angular velocity at each friction limit (ωexp, rad s−1) was compared with the recorded angular velocity (ωactual, rad s−1) taking the 95% upper limit of turns where ωactual exceeds ωexp.
Arrival and departure of turns
To compare similar strides within a turn, strides were given alphanumeric identifiers. The most severe CoH within a stride was considered the Apex of the turn. The last stride before entering the Apex was given Arrival (A)1 and the first stride exiting the Apex was given Departing (D)1. Strides within a turn increased distally (e.g. A2, A1, Apex, D1, D2).
Performance density maps
We used the relative overlap between predator and prey performance landscapes to derive an estimated escape probability. This metric represents the likelihood of escape predicted from comparative locomotor performance rather than an empirically observed probability based on actual pursuit outcomes. For simplicity, we refer to this metric throughout as estimated escape probability. A performance density map containing all dogs was created with velocity (v, m s−1) on the x-axis and angular velocity (ω, rad s−1) on the y-axis. A kernel density contour plot was computed over the data using ‘kde’ from the ks package and ‘with’ from base R (https://CRAN.R-project.org/package=ks). This step was repeated for densities in a sequence of 10 to 95 in increments of 5. The contour data was extracted and converted to a shape file using the functions ‘Polygon’, ‘Polygons’ and ‘SpatialPolygons’ from the package sp (https://CRAN.R-project.org/package=sp). For the dog contour plot at 10%, a performance density map of the same variables (x=velocity, y=angular velocity) was created for each quoll subject and a kernel density contour of 10% was computed over the data. The area of overlap of each quoll subject's contour at 10% with the overall dog contour at 10% was calculated using ‘intersect’ from the package raster and the overlap of all individuals was averaged (https://CRAN.R-project.org/package=raster). This step was repeated for quoll contour plots from 10% to 95% using the 10–95% dog contour plots. This process was repeated for bandicoots using velocity and angular velocity and then again for both prey species using tangential acceleration and centripetal acceleration to create heatmaps.
Statistical analyses
All data were analysed and processed in R version 4.0.5 (2021-03-31) (https://www.r-project.org/). Quantile regressions were used to analyse data that possessed a limitation boundary; such analyses used the function ‘rq’ from the package quantreg (v.5.94) (https://CRAN.R-project.org/package=quantreg). Quantile regressions were calculated using the upper or lower 5% quantile that represented the boundary limitations. The quantile regression analysis was used for angular velocity and turn radius. Linear mixed-effects models were used to analyse all other data when using the function ‘lme’ from the package nlme (https://CRAN.R-project.org/package=nlme).
For the performance landscape heatmaps, the effort differential (predator effort−prey effort) was calculated for each box that fell along the >55% estimated escape probability line. The estimated escape probability at >55% indicates the prey species has a greater chance of escaping than of being caught. Effort differentials and density plots were compared using Welch's t-tests.
Density curves of velocity, angular velocity, turn radius, CoH, RoB, θB, at and ac were calculated for each species and the range and height were used to compare the variability of strides as an approximation for predictability. Each range was converted to a proportion of the mean to standardise between different units. The ranges and heights of density curves were compared using Welch's t-tests.
Linear mixed-effects models were used to compare equivalent strides within a turning sequence, with centripetal acceleration, tangential acceleration, velocity, angular velocity, stride length and stride frequency treated as separate dependent variables. Species and stride position were included as fixed effects, and subject was included as a random effect. Pairwise comparisons of estimated marginal means were performed using the package emmeans (Table S1) (https://CRAN.R-project.org/package=emmeans).
RESULTS
Comparative speed and turning metrics
The cornering manoeuvres of three species were analysed and compared to help understand the contribution of locomotor performance to escape probability. Turning performance reflects the combined capacity to generate tangential and centripetal acceleration; consequently, speed, turn radius, angular velocity and body reorientation all contribute to overall performance. Overall, we analysed 387 strides from 29 bandicoots, 467 strides from 17 quolls, and 869 strides from 20 dogs. Dogs were much faster than both prey species, with a mean (±s.d.) speed of 3.55±1.27 m s−1 and observed maximum of 8.08 m s−1. Quolls showed a mean speed of 2.56±0.74 m s−1 with a maximum speed of 4.33 m s−1, and bandicoots’ mean speed was 2.38±0.76 m s−1 with a maximum speed of 4.84 m s−1 (Fig. 2). Dogs increased stride length (slope=−0.33, R2=0.81; Fig. 2H) to increase speed but did not adjust stride frequency (slope=−0.02, R2=0; Fig. 2I). Quolls favoured modulation of stride frequency (slope=0.63, R2=0.10; Fig. 2I) to increase speed and showed the least amount of adjustment to stride length (slope=0.12, R2=0.54; Fig. 2H), while bandicoots modulated both frequency and length at similar rates to increase speed (frequency: slope=0.23, R2=0.04; length: slope=0.21, R2=0.64; Fig. 2H,I). The interaction between species and both stride length and stride frequency was significant between dogs and prey species (bandicoot frequency: t1717=3.94, P<0.001; quoll frequency: t1717=10.94, P<0.001; bandicoot length: t1717=−10.4, P<0.001; quoll length: t1717=−18.8, P<0.001).
[图：The turning variables of individual strides in northern brown bandicoots and northern quolls in comparison to dogs. (A,B]
Quolls had the highest average angular velocity (quolls: 2.83±2.99 rad s−1; dogs: 1.43±1.07 rad s−1; bandicoots: 1.18±0.95 rad s−1; Fig. 2A,B). The slopes of the lower limit (5%) of turning radii were compared among species because of the limitations of turning (as speed increases, turning radius increases; Fig. 2C,D). Quolls showed the shallowest slope of all three species (slope=0.29; lower 0.20, upper 0.36) and were significantly different from bandicoots (slope=0.48; lower 0.41, upper 0.64) and dogs, which had the highest slope (slope=0.81; lower 0.62, upper, 0.91). Quolls and dogs were able to change the heading of their turns up to a maximum of 109 deg and 103 deg within a single stride, respectively (Fig. 2F), while values for bandicoots were much lower (75 deg; Fig. 2E). Despite this, bandicoots (slope=13.35; Fig. 2J) adjusted their CoH at a faster rate than dogs (slope=17.88), and quolls adjusted their CoH at a faster rate than both (slope=8.23; Fig. 2J). Quolls showed significantly greater ability to produce centripetal acceleration than dogs (Fig. 2G). Quolls’ centripetal acceleration (slope=−0.53; R2=0.47; Fig. 2G) was negatively correlated to tangential acceleration and was significantly steeper than that of both dogs and bandicoots (dogs slope=−0.30; bandicoots slope=−0.13; t1717=−6.20).
Coefficient of friction
Quolls showed a significantly higher coefficient of friction (1.96±0.57) than dogs (0.88±0.22; t20.24=−7.29, P<0.001) and bandicoots (0.69±0.22; t18.95=−8.75, P<0.001) (Fig. 3). Dogs and bandicoots were also significantly different (t39.15=2.90, P=0.007). The increased coefficient of friction of quolls could explain their increased turning ability, while the poor coefficient of friction for bandicoots could explain their performance regarding speed and manoeuvrability.
[图：The coefficient of friction for individual strides predicted for northern brown bandicoots, dogs and northern quolls. Ve]
Contour overlap landscape analysis
The performance capabilities of dogs, quolls and bandicoots were further explored by comparing the overlap between density contour plots of all dogs and averaged individual prey subjects (Fig. 4). This analysis showed the comparative effort and estimated escape probability between prey and predator. Quolls showed a high estimated escape probability when comparing the performance landscape of tangential and centripetal acceleration; however, this was reduced somewhat when comparing angular velocity and velocity (Fig. 4B,C). Bandicoots showed minimal estimated escape probability in both velocity and acceleration performance landscapes (Fig. 4D,E). For quolls, in both performance landscapes, the 55% estimated escape probability exceeded the line of equivalent effort (black), while in bandicoots, this was reduced. The effort differential (predator effort−prey effort) was calculated to understand how much effort is required to achieve a 55% estimated escape probability. For acceleration landscapes, quolls were able to achieve a 55% estimated escape probability by using 14±9.5% less effort than dogs, while bandicoots used 36±11.2% more effort to achieve the same estimated escape probability (t26.13=−14.45, P<0.001; Fig. 4B,D). For velocity landscapes, quolls required 3.2±8.0% less effort to achieve a 55% escape probability and bandicoots required 25.8±10.8% more effort (t34.21=−9.99, P<0.001; Fig. 4C,E). Quolls showed decreased effort differentials in acceleration landscapes, while bandicoots showed decreased effort differentials in velocity performance landscapes (t45=−4.35, P<0.001; Fig. 4B,C; t29.6=2.80, P=0.008; Fig. 4D,E). Using these estimations, quolls seem to escape predation using much less effort than would be required by dogs, while bandicoots need to use close to maximum effort to escape dogs that do not exhibit full running capacity.
[图：Heatmaps displaying the results of a contour overlap landscape analysis comparing dog performance landscapes with those ]
Densities for predictability
Density plots were analysed to show the variability of choices during turning (Fig. 5). Bandicoots showed a smaller proportional range (0.70±0.28) and a higher maximum (0.22±0.20) than both quolls (range 1.18±0.46, maximum 0.14±0.15) and dogs (range 1.12±0.51, maximum 0.16±0.15). However, only the proportional range between quolls and bandicoots showed a significant difference (t11.5=−2.467, P=0.03). Velocity and CoH were the only two performance metrics where bandicoots showed lower peaks than quolls; however, quolls either showed a similar or broader range in both plots (Fig. 5A,D). Dogs showed a broader range in velocity and turn radius than the other species. The limited range and increased maximum of bandicoot density plots suggests that their escape trajectories would be more predictable than those of quolls.
[图：Density plots of performance metrics used for turning analysis of dogs, northern brown bandicoots and northern quolls. (]
Equivalent stride analysis
Equivalent strides within a turn were compared (Fig. 6). There was a significant effect of species on all variables, with the greatest effect on stride frequency (at: f2,61=86.21, P<0.001; ac: f2,61=46.85, P<0.001; v: f2,61=22.18, P<0.001; ω: f2,61=95.52, P<0.001; stride length: f2,88=80.80, P<0.001; stride frequency: f2,111=154.05, P<0.001). Stride position had a significant effect on all six variables, with the largest F-statistic observed for angular velocity (at: f6,1637=117.52, P<0.001; ac: f6,1637=171.15, P<0.001; v: f6,1637=112.39, P<0.001; ω: f6,1637=191.09, P<0.001; stride length: f6,1634=31.68, P<0.001; stride frequency: f6,1634=16.48, P<0.001). Quolls displayed the highest centripetal acceleration at the apex of the turn, while bandicoots showed a significantly lower centripetal acceleration at the apex (Fig. 6A). Quolls showed more negative tangential acceleration than dogs when entering the apex but accelerated at similar rates when departing. Bandicoots showed greater tangential acceleration through the apex of the turn, which was significant compared with both dogs and quolls (Fig. 6B). Quolls had similar speeds to dogs when approaching the apex but showed significantly slower speeds departing the apex (Fig. 6C). Bandicoots showed significantly lower angular velocity in strides A1, Apex and D1 in comparison to dogs, and quolls showed a significantly higher angular velocity at the apex (Fig. 6D). During a turning sequence, bandicoots were slower when approaching the apex (A3–A1) and did not use as much negative tangential acceleration. While their velocity was comparable to that of both dogs and quolls during strides at the apex and D1, they did not seem to increase tangential acceleration or speed when departing the apex (D1–D3). Dogs showed increased stride length and decreased stride frequency compared with both prey species throughout the turns (Fig. 6E,F; also see Fig. 2H,I).
[图：Comparative strides within a turn of dogs, northern brown bandicoots and northern quolls. (A) Centripetal acceleration (]
DISCUSSION
The natural experiment
Our original hypothesis was that the bandicoots should possess a greater overall locomotor performance of the prey species; however, quolls showed greater capacity for locomotion. Based on locomotor performance, bandicoots were slower in comparison to quolls. The contour overlap analysis showed that bandicoots had a low estimated escape probability when pursued by dogs, while quolls had a higher estimated escape probability, mainly through their ability to adjust their change of heading (CoH) at faster rates than dogs. The locomotor performance of quolls, in comparison to that of dogs, supports findings for Groote Eylandt populations, where quolls show high naive occupancy estimates, and turning ability is proportional to recruitment into successive breeding seasons (Rew-Duffy et al., 2020; Heiniger, Davies and Gillespie, 2020a). Bandicoots on Groote Eylandt also have high naive occupancy estimates despite their poorer estimated escape probabilities relative to canids present on the island (Heiniger, Davies and Gillespie, 2020a). From the same study, both species showed reduced observations on mainland Australia, but quoll estimates were reduced by approximately 98.47% as opposed to 52.9% in bandicoots, which does not reflect the locomotor performance of these two species. This reveals much about Groote Eylandt, a remote island which holds pristine habitat free from foxes (Vulpes vulpes), cane toads (Rhinella marina), habitat degradation due to invasive herbivores, and suppressed presence of cats, all things that cause declines in these fauna (Oakwood, 1997; Heiniger et al., 2020b). How much then does locomotion and escaping predation contribute to these species at a population level?
Does locomotion contribute to northern quoll survival rate?
While quolls have a greater turning ability, their low top speed in comparison to dogs would leave them vulnerable in situations during unobstructed straight running. Wilson et al. (2015) showed that mistimed turns by prey would result in the predator ‘cutting corners’, reducing the effect of size limitations for turning. Without cutting corners, the predator must match the angular velocity of the prey or risk overshooting the turn, increasing the distance between prey and predator (Wilson et al., 2015). On the mainland, quolls are in greater densities in highly complex environments (rocky outcrops) and are less commonly found in open woodlands or where habitat complexity has been reduced (Oakwood, 1997; Oakwood et al., 2016). Some predators, such as wolf spiders and cheetahs, experience reduced hunting success when hunting in complex habitats (Mills et al., 2004; Morice et al., 2013). The benefit of habitat complexity is not completely understood, with suggested theories of visual or physical obstruction, refuge abundance or resource abundance (Bonenfant and Kramer, 1996; Laundré, Hernández and Ripple, 2010; Morice et al., 2013). Despite this, prey species such as quokkas (Setonix brachyurus) are found more commonly in complex habitats and are suggested to use these habitat types to limit predation from foxes (Hayward et al., 2005). For quolls, greater turning performance would have an increased effect in complex habitats where predators cannot cut corners and straight running is limited (Wilson et al., 2015; Rew-Duffy et al., 2020; Wheatley et al., 2020). This could explain greater densities of quolls reported in complex rocky outcrops, which would favour their increased manoeuvrability.
In comparison to quolls' preference for rocky habitat, bandicoots prefer dense ground vegetation in the shade of woodlands for both foraging and protection from predators; however, they are very adaptable and can inhabit a broad range of environments (Kemper et al., 1989; FitzGibbon, Wilson and Goldizen, 2011). On Groote Eylandt, quolls will occupy similar habitats with dense ground vegetation (Heiniger et al., 2020b). On mainland Australia, the destruction of dense vegetation through fires and invasive herbivores has caused declines in quolls, while bandicoots show more versatility (Oakwood, 1997; Kemper et al., 1989; FitzGibbon, Wilson and Goldizen, 2011).
Does locomotion contribute to northern brown bandicoot survival rate?
Bandicoots performed poorly in comparison to quolls and dogs in terms of their turning ability, which could not be related to choosing submaximal speeds. Prey will often choose to run at submaximal speeds for the purpose of reducing mistakes and decreasing predictability (Wynn et al., 2015; Amir Abdul Nasir et al., 2017; Wilson et al., 2013). The average speeds recorded here for bandicoots are similar to average speeds recorded in other studies on bandicoot locomotion, and thus it is unlikely that they are limiting speed (Tay et al., 2022). Increasing the variability of the escape trajectory has also been strongly associated with survival rate of prey within pursuit interactions (Moore et al., 2017). However, when comparing the variability of turning parameters among species, bandicoots were the most predictable and are unlikely to be increasing variability in escape trajectories (Fig. 5). Bandicoots’ inability to outperform dogs, in combination with their predictable trajectory, makes them common targets for dingoes/wild dogs. This is reflected in the increased appearance of bandicoot-like species in dog diets when compared with dasyurids (including quolls); however, bandicoots still persist in large populations on mainland Australia (Gaschk and Clemente, 2022).
Bandicoots' ability to persist in broad distributions, despite the pressure of canid predation and their poor locomotor performance, could be related to many aspects of their ecology. Their reproductive strategy, which comprises 4–5 joeys per litter, a relatively short gestation/weaning period (∼72 days) and an annual fecundity of 4.7 litters per year, would allow populations to remain stable (Gemmell, Veitch and Nelson, 1999; FitzGibbon, 2015; Tay et al., 2022).
Bandicoots' survival could also be due to adaptable foraging behaviour to limit detection from predators or reduce distance to refuges (Lima and Dill, 1990). The landscape of fear is a concept that suggests animals will actively avoid foraging areas that are too far from known refuges or lack adequate complexity (Laundré, Hernández and Ripple, 2010). Bandicoots change their foraging behaviour in the presence of predator scents, increasing vigilance when foraging close to, or outright avoiding, the scented location (Bytheway and Banks, 2019). Bandicoots have also been shown to change their peak temporal activity period from bimodal to unimodal in response to rainfall variation, correlating with a change in diet (Diete et al., 2017). Thus, versatile foraging and reliable reproduction are likely to support population persistence through canid predation, while speed and manoeuvrability contribute less to individual survival.
Dingoes and domestic dogs
Australian fauna experience predation from three forms of canids in Australia: domestic/wild dogs, dingoes and foxes. Dogs are a good approximation for the predation pressure from dingoes but also contribute to this pressure directly. Many of the differences between the skeletal morphology of dingoes and free-roaming feral dogs have been identified in the cranium; however, these differences are within the morphological range found in domestic dog breeds (Smith et al., 2019; Jackson et al., 2021; Koungoulos et al., 2024). Genomic comparisons of dingoes and domestic dogs show that the two are distinguishable genetically (Smith et al., 2019; Cairns et al., 2023; Souilmi et al., 2024). However, despite this distinguishable difference between dingoes and free-roaming feral dogs, they share a similar hunting ecology within comparable combinations of ecosystems and conditions; they are opportunistic hunters with commensal behaviours under certain circumstances (Jackson et al., 2021; Gaschk, 2023). With dingoes showing a similar hunting ecology, conditional sociality, indistinguishable morphology and historical interbreeding with Australian dog breeds, the range of domestic dogs used in this study likely encompasses the performance capability of dingoes (Smith et al., 2019).
Friction, mass and substrate
Northern quolls showed the greatest turning capacity of the three species, which seemed to correlate with an increased coefficient of friction. Friction allows for a greater capacity to apply force to the substrate without slipping and is vital during turning because the force applied can be perpendicular to initial direction (Alexander, 1982; Haagensen et al., 2022). Quoll manoeuvrability is facilitated by their small size, crouched posture and foot pad morphology, allowing them to deflect the centre of mass from its original heading (momentum) to the new direction (Alexander, 1982; Wynn et al., 2015; Wheatley et al., 2018). Further, their ability to modulate stride frequency increases the potential number of centre of mass deflections within a turning sequence, resulting in higher angular velocities and more severe CoHs.
Dogs only modulate stride length when changing speed; when gaining speed, this limits the opportunity for deflection of the centre of mass during a turn and increases the distance between these deflections. Dogs also have to contend with a much higher centre of mass and greater momentum through the turns because of their larger size and more upright posture (Haagensen et al., 2022). Bandicoots, in contrast, have a crouched posture and are much smaller, lowering their centre of mass, and as such should have an advantage in turning ability. Despite their crouched posture and smaller body size, bandicoots had the lowest coefficient of friction (Fig. 3). However, the coefficient of friction of both dogs and bandicoots was far lower than that of quolls. This could be explained by the type of substrate used in these experiments.
Quolls performed their trials on a thin rubber matting, while bandicoots and dogs completed their trials on a grassy-soil substrate, with some bandicoots running on a loamy soil. A study looking at the task of pulling sleds across substrates found that the coefficient of friction for a grass pitch was 0.45±0.1 while on a rubber athletics track it was 0.58±0.1 (Linthorne and Cooper, 2013). When comparing performance across the two substrates, the rubber surface generated 29% more friction than the grass surface. Applying this correction to the coefficient of friction recorded for quolls on rubber (1.96/1.29) yields an estimated value of 1.52 for turns performed on grass. This remains substantially higher than the values estimated for both dogs and bandicoots. Therefore, although the rubber matting provides some mechanical advantage, it does not materially alter the overall result.
Conclusion
We explored the effect of locomotor performance on the escape probability of native Australian species from canid predation. Dogs showed higher top speeds, quolls showed better manoeuvrability and unpredictability, while bandicoots were wholly outperformed. For bandicoots, locomotor performance does not seem to contribute to population survival, which is likely related to other aspects of their ecology. Locomotor performance for quolls, particularly their manoeuvrability, supports their use of complex habitats (e.g. rocky outcrops) to escape pursuers. Future research is needed to determine whether performance is a reliable predictor of population survival in species that have similarly receded to refuge habitats (e.g. quokkas). This study highlights that locomotor performance alone may not be able to accurately predict the abundance or distribution of populations, despite being a key determinant in the outcomes of predator–prey interactions.
Supplementary Material
[内容请查看原文]
Acknowledgements
We would like to acknowledge the Anindilyakwa people, the traditional landowners of Groote Eylandt, and the Yagara people, the traditional landowners of Hidden Vale, for graciously allowing us to conduct this research on their land. We thank the Anindilyakwa Land and Sea Ranger Research Station, Anindilyakwa Land Council for providing facilities and support; Hidden Vale Wildlife Centre and the Turner Family Foundation for use of their facilities (https://hiddenvalewildlife.uq.edu.au/); and Agility Dog Club of Queensland and Ruth Raymond for allowing us to film their training sessions.
Footnotes
[内容请查看原文]
References
[内容请查看原文]
Associated Data
Supplementary Materials
[内容请查看原文]