常见生物医学实验 · 第 3 篇
行为学：哪些测验真的测得出差别
同一批脑卒中大鼠，16 项测验，测了两个月：有的测验一直稳定地检出缺损，有的只在第 1 天检出、第 7 天就恢复，有的方向完全相反，还有的从头到尾什么都没测出来。本篇精译 Trueman 等 2017 年的系统比较，并补上脑出血模型常用的四项量表与测验。
行为学是实验室里最容易被“看起来很有道理”糊弄过去的一步。你给大鼠做一个评分，得 12 分，对照组得 9 分，p 值小于 0.05，写进论文就是“治疗显著改善了神经功能”。但这 3 分的差别，可能来自应激、来自测试的时间点、来自今天操作的人手更重一点、来自观察者知道哪只鼠打了药——这些因素没有一个会出现在结果表里。
本篇要做两件事。第一，精译一篇“几乎把 MCAO 大鼠所有能做的行为测验都做了一遍”的论文，看它用同一批动物、同一套盲法、四个时间点，到底哪些测验检出了稳定的缺损。第二，把脑出血（intracerebral hemorrhage, ICH）研究里最常用的几个量表与测验（mNSS、Garcia、前肢放置、旷场、corner test、转棒）各自的“测什么、怎么算分、什么时候失效”讲清楚。
本篇精译自 Journal of Cerebral Blood Flow & Metabolism 2017 年的论文《Systematic and detailed analysis of behavioural tests in the rat middle cerebral artery occlusion model of stroke: Tests for long-term assessment》（Trueman RC, Diaz C, Farr TD, Harrison DJ, Fuller A, Tokarczuk PF, Stewart AJ, Paisey SJ, Dunnett SB；37(4): 1349–1361，DOI 10.1177/0271678X16654921，PMID 27317655），精译“Behavioural testing”、四类结果小节与“Robust tests / Discussion”的核心结论；脑出血部分精译自 Thangameeran 等 2023 年发表于 Tzu Chi Medical Journal 的综述（35(1): 1–10，DOI 10.4103/tcmj.tcmj_119_22，PMID 36866349）的“BEHAVIORAL ASSESSMENT”一节。英文原文见文末附录。
一、行为学是整条证据链里最便宜、也最容易骗人的一环
一篇动物实验论文的证据链大致是：造模成功（影像或病理）→ 分子层面有变化（组化、Western、qPCR）→ 整体动物有功能变化（行为学） → 因此推断“这个干预有治疗潜力”。行为学是这条链上唯一直接回答“动物好不好”的环节，也是成本最低、门槛最低、最容易被做坏的一环。
它为什么容易骗人？因为它测量的是“行为”，而行为是所有变量的总和。至少六类因素会混进来：
应激。抓取、运输、新环境、气味、同伴的叫声，都会改变动物当天的活动量与反应时。一只被拎着尾巴从笼子里拽出来的鼠，和一个被双手托起来放在测试场里的鼠，测出来的“探索行为”不是一回事。
昼夜节律。大鼠是夜行性动物，白天（光照期）活动少、焦虑水平高。同一批动物，上午 9 点测和下午 6 点测，结果可以差一个档次。原文的做法很具体：所有测试安排在 8:00–19:00 之间、按笼序进行——这不是随便写的，是在控制“时间”这个变量。
手感差异。两个人做同一个评分，一个给 2 分一个给 3 分，差别可能比治疗效应还大。评分量表越主观（“肢体无力程度”“姿势异常”），这种差异越明显。
样本量。行为学的个体差异极大。原文用到的组样本量在 9–24 只之间，仍然有若干测验“差一点显著”（例如 corridor task 的 p = 0.054）。当效应量小、样本量小、又做了多重比较时，最容易出现“这次显著、下次不显著”的结果。
非盲观察。知道分组的人，会不自觉地多看治疗组一眼、多给半分。原文明确要求“所有行为学测试由盲法观察者完成”——这不是客套，是方法学上的硬要求。
学习效应。同一只鼠重复做同一个测验，会越做越好（熟悉环境、熟悉操作）。如果治疗组和对照组的测试顺序不同，学习效应就会伪装成“治疗效应”。原文的对策是：术前把所有动物训练到稳定基线，同一个测验在所有时间点由同一流程执行。
把这六条摆在一起，你会发现行为学的问题不在“测不准”，而在“不知道测的是什么”。一个测验的分数变化，可能来自运动功能、感觉功能、注意力、焦虑、动机、甚至“今天想不想动”。所以任何行为学结论都必须回答两个问题：这个测验测的是哪个功能？它对这个功能特异吗？
二、一篇“把 16 项测验全做一遍”的论文
这篇论文的设计值得完整抄一遍，因为它是“系统比较”的样板：两批（cohort）雄性 Wistar 大鼠，线栓法大脑中动脉闭塞（middle cerebral artery occlusion, MCAO）30 分钟，24 小时后用 9.4 T MRI 确认梗死；术前把所有动物训练到稳定基线；术后第 1 天、第 7 天、1 个月（4–5 周）、2 个月（8–9 周）四个时间点测试；所有测试由盲法观察者完成；为了减少每只动物承受的测试数量，把 16 项测验分给两批动物做。
剔除标准也很具体：线栓拔除后脑血流（cerebral blood flow, CBF）未恢复的（死后解剖证实为蛛网膜下腔出血）、MCAO 不完全的、术后体重下降超过 20% 的、sham 组 MRI 异常的、以及手术并发症。最终纳入 44 只（第一批 11 只 MCAO、8 只 sham、6 只 naïve；第二批 9、5、5；sham 与 naïve 合并为对照）。
表 1 该文的两批动物与测验分配（据原文整理；样本量随时间点略有变化）
| 批次 | 动物数（对照 / MCAO） | 主要测验 | 特殊要求 |
|---|---|---|---|
| 第一批 | 14 / 11 | 阶梯测验（staircase）、笼内够取、走廊测验、握力、转棒、侧向跨步、改良神经评分 | 需限食至自由进食体重的 85%–90%，术前 1 周开始 |
| 第二批 | 10 / 9 | 游泳分析、锥形平衡木、步态、黏纸去除、圆柱测验、脱离注意任务、阶梯（不规则横档） | 不限食；安非他命与阿扑吗啡旋转在实验末期做 |
这 16 项测验覆盖了运动、感觉、感觉运动整合、注意力、平衡、步态、以及药物诱导的旋转行为。换句话说，它不是“随便挑几个测验”，而是把当时文献里能用的测验都试了一遍，然后按结果分成四类：稳定检出长期缺损的、只有轻微缺损或会自行恢复的、结果反直觉的、完全没检出差异的。下面按这四类讲，并且每一步都标明“这是该文的结论”，而不是教科书上的通用说法。
三、评分量表：mNSS 与 Garcia
评分量表是“把一只鼠的状态压缩成一个数”的工具。它的好处是快，坏处是丢信息。理解量表，关键是看它把哪些维度加在了一起、以及分数往哪个方向走。
mNSS：最常用，也最容易被误读
改良神经功能缺损评分（modified neurological severity score, mNSS）是脑损伤研究里用得最多的综合量表。它把三类检查拼在一起：运动（提尾悬空时肢体屈曲、行走时的偏转、肌张力）、感觉（触须刺激后的反应、本体感觉）、平衡与反射（在平衡木上的表现、耳廓反射、角膜反射、惊跳反射等）。多数版本是 18 分制：每一项按“无、轻、中、重”计 0–3 分，最后加总；分数越高，缺损越重。
它的优点是“全面、能做长期”。ICH 综述的表 2 对它的评价是：comprehensive evaluation with long term ICH outcome assessment（能全面评估，也能做长期结局）。缺点是同一张表里的另一句话：它评的是基底节区的神经缺损，不代表全脑所有区域——因为 ICH 的病灶在基底节（纹状体），而 mNSS 的很多条目（比如平衡木）反映的是更广泛的功能。
更隐蔽的问题是多维度混在一起。一只鼠可能运动分很差但感觉分正常，另一只相反，两者总分相同。这时“总分下降”到底意味着哪一项改善，你无法回答。所以用 mNSS 的正确方式不是只报总分，而是同时报分项（运动、感觉、平衡、反射各多少分），并且说明评分者是否盲法。
Garcia：简单，但吃不了长期
Garcia 量表（Garcia scale）包含六项：自发活动、肢体运动的对称性、前肢伸展、攀爬与握力、体态对称性、触须感觉。每项 0–3 分，总分 18 分；分数越高，功能越好——注意方向与 mNSS 相反，这是初学者最容易搞反的地方：一个量表分数越高越差，另一个越高越好，写论文时必须在方法学里写清楚。
ICH 综述对它的评价两极：优点是 easy to perform（容易做）；缺点是 cannot be used for long-term neurological outcomes（不能用于长期神经功能结局）。原因不难理解：它的多数条目在急性期有区分度，但随着水肿消退、神经功能自发恢复，评分会回到接近正常，天花板效应明显。
还有一个常被忽略的偏差：Garcia 的六项里，前肢相关的占了两项（前肢伸展、攀爬与握力），所以它对“前肢功能”的权重偏高。如果你的模型主要损伤后肢或躯干，Garcia 可能测不出来。
两个量表怎么选
一个实用的判断：要看急性期（1–7 天）的整体严重程度，用 Garcia；要看长期（数周至数月）的功能缺损，用 mNSS，并且把分项一起报。如果既要急性期又要长期，就两个都做——代价是测试时间翻倍，而且同一只鼠一天内做两个量表，疲劳与应激会污染第二个。
怎么读这个矛盾：ICH 综述说 mNSS“全面”，又说它“只反映基底节”；说 Garcia“容易做”，又说它“不能做长期”。这不是自相矛盾，而是在说两件不同的事——前一句说的是“覆盖的功能维度多”，后一句说的是“覆盖的脑区有限”；前一句说的是“操作门槛低”，后一句说的是“时间窗窄”。读方法学综述时，把“优点”和“缺点”各自对应到“维度、时间、脑区、操作”四个坐标上，矛盾就消失了。
四、简单测验：前肢放置、旷场、corner test
这三个测验的共同点是：不需要训练、几分钟就能做完、对动物伤害小。它们的共同缺点也一样：信息量少，只能回答一个很窄的问题。
前肢放置测验
前肢放置测验（forelimb placement test, FPT）的做法是：把动物提起，让它的前肢自由悬空，然后把它移到桌沿，用触须或视觉刺激诱发“伸爪去够桌面”的动作。正常鼠会在同侧前肢伸出并放置到桌面上；单侧损伤的鼠，对侧前肢的放置反应减弱或消失。评分通常是每侧 0–3 分（或“放置/不放置”的二分法），比较两侧的差值。
它的优点是 easy to perform, no training required（容易做、不需要训练），特别适合急性期和术后早期。缺点是 ICH 综述写得很直白：confined to forelimbs, not considered as a comprehensive evaluation（只限于前肢，不算全面评估）。所以它适合做“造模是否成功”的快速验证，不适合做“治疗是否有效”的主要终点。
旷场测验
旷场测验（open field test）把动物放进一个方形或圆形的开阔场地（常见尺寸 40×40×40 cm 到 50×50×38 cm，具体尺寸因实验室而异），自由活动 5–10 分钟，用录像追踪系统记录：总移动距离、平均速度、在中央区域停留的时间与次数、直立（rearing）次数、以及梳理（grooming）时间。
它测的是“自发活动与探索”，同时被广泛用作焦虑的指标（中央区停留越少，越焦虑）。这个双重身份是它的麻烦：一只鼠活动量下降，可能是运动功能受损，也可能是焦虑增加、或者单纯不熟悉环境。ICH 综述对它的评价是 easy to perform，但 results will be affected due to external influences like environmental factors（结果会受环境等外部因素影响）。
要把旷场用好，必须控制三件事：光照均匀（中央与边缘亮度差会直接改变动物在中央区的停留）、气味（前一只动物的气味会改变后一只的探索）、时间（同一时段、同一顺序）。这三件事没控制，旷场数据基本不可比。
corner test
corner test（角落测验）用两块纸板或木板拼成一个约 30° 的夹角，把动物放进角落。动物会本能地转身钻出来；正常鼠左右转向各约一半，单侧损伤的鼠会持续向一侧转。记录 10 次测试中向损伤侧/健侧转的次数，算偏向指数。
它的优点是能测“感觉运动整合”，而且 able to assess long-term neurological outcomes（能做长期结局）。缺点是 cannot be used with severely injured animals as it might affect the results（重伤动物不适用，因为动物根本不愿意动或转不了身）。这个限制很实际：造模太重的时候，corner test 的数据会被“动不了”污染，而不是被“不对称”污染。
五、需要训练的测验：转棒与水迷宫
需要训练的测验，信息量大，但成本也大：训练要几天到几周，训练本身会改变动物（学习、习惯化、甚至神经可塑性），而且训练不充分会让“没学会”和“有缺损”混在一起。
转棒
转棒（rotarod）是一根直径约 3 cm 的转杆，动物在上面走，转速逐渐加快（常见 4–40 rpm，具体程序因实验室而异），记录从杆上掉下来的潜伏期。它测的是运动协调、平衡与运动学习。
转棒在脑卒中研究里用得极多，但该文的结果值得仔细看：MCAO 组与对照组的跌落潜伏期总体有差异（F1,22 = 6.59，p < 0.05），但事后比较只在 2 个月时显著；而且这个差异不是 MCAO 组变得更差，而是对照组进步了——原文的原话是：this effect was not due to a further decrease in performance in the MCAO group, but rather an improvement in the ability to remain on the Rotarod in the Control group.
这句话是整节里最有价值的一句。它说明：转棒测出来的“组间差异”，可能只是对照组的运动学习能力更强（完整的大脑能学会，损伤的大脑学不会），而不是“损伤组的功能在恶化”。如果不像该文这样把原始数据画出来、把“谁在变、怎么变”讲清楚，你会把“学习能力差异”误读成“功能缺损加重”。
转棒的两个经典问题是天花板效应与地板效应：训练好的鼠能一直不掉（天花板），重伤的鼠一放上去就掉（地板），两端都没有区分度。所以转棒适合“中等程度损伤 + 训练充分”的场景，不适合重伤模型或未训练的动物。
水迷宫
Morris 水迷宫（Morris water maze, MWM）是一个圆形水池（常见直径 1.2–2 m，因实验室而异），水面下 1–2 cm 处放一个隐藏平台，水池周围挂视觉线索。动物被放进水里，靠空间线索找到平台；训练若干天后撤掉平台做“探查试验”（probe trial），看它在原平台位置附近游的时间与穿越次数。它测的是空间学习与记忆，不是运动功能。
这一点非常重要：水迷宫经常被拿来当“神经功能结局”，但它测的其实是海马依赖的空间认知。如果你的模型损伤的是纹状体（MCAO 的典型病灶、ICH 的典型病灶），水迷宫可能根本不敏感；反过来，如果水迷宫测出了差异，你也不能说“运动功能改善了”，只能说“空间记忆改善了”。
该文没有用水迷宫，ICH 综述的行为学表里也没有列它——这不是遗漏，而是因为两篇文献关注的都是感觉运动功能。写论文时如果要用 MWM，必须单独论证“为什么这个模型会影响空间记忆”，并控制水温（过低会造成应激性低体温）、光照、以及动物是否会被“训练成游泳健将”（反复测试会让成绩变好，这是学习效应，不是记忆改善）。
读法提示：本节列的“常见参数”都只是范围，不是标准。真正的标准是你自己实验室 SOP 里写死的那个数值，并且在你所有的实验里保持一致。跨实验室比较行为学数据，本质上是在比较两套 SOP，而不是比较两个结论。
六、该文的真实结论：哪些测得出，哪些反直觉
下面四类结果，全部来自该文（Trueman 等，2017），不是教科书上的通用说法。每一项我都给出原文的方向与统计量，方便你判断“这个结论有多硬”。
第一类：稳定检出长期缺损的测验
阶梯测验（staircase task）。这是该文结论里最“硬”的一个：MCAO 大鼠用对侧前爪取到的食丸数，在术后 2 个月显著少于对照组（Block 2–4：U = 38–41，p < 0.05；Block 5：U = 45，p = 0.08），而同侧取食没有缺损（U = 59–74，n.s.）。同侧正常、对侧减少，这种“单侧性”正是单侧损伤模型最想看到的模式。
阿扑吗啡诱导旋转（apomorphine-induced rotation）。MCAO 大鼠向同侧旋转的次数显著增加（189.0 ± 169.3 次 vs 对照 3.4 ± 9.23 次，U = 14，p < 0.01）。阿扑吗啡是 dopamine 受体激动剂，单侧黑质-纹状体通路损伤后，两侧受体敏感性不对称，动物就会持续向一侧转。这个测验的变异系数小，所以它“稳定”；但也正因为效应量不算大，原文估算要在治疗实验中检出效果，需要 n ≈ 50 只/组。
与病理的相关性是这个结论的加分项：阶梯测验最后 5 天对侧取食数与存活纹状体神经元数正相关（ρ = 0.8，p < 0.01），与 MRI 高信号体积负相关（ρ = −0.63，p < 0.05）。也就是说，行为学分数不是“碰巧显著”，它确实追踪了病灶。相反，阿扑吗啡旋转、转棒、前向侧向跨步与组织学指标不相关——这不代表它们没用，只代表它们测的是“功能代偿”而非“病灶大小”。
第二类：只有轻微缺损，或者会自行恢复
转棒：只在 2 个月时组间显著，而且效应来自对照组进步（见上一节）。侧向跨步（lateralized stepping）：前向跨步在 1 天与 2 个月显著（U = 72 与 100.5，p < 0.001），后向跨步在 7 天与 1 个月显著（U = 105 与 114，p < 0.05 与 < 0.01）——同一个测验的不同方向、不同时间点，结果对不上，说明它“不稳”。
改良神经评分（modified neuroscore）：1 天与 7 天显著（U = 104.0 与 123.5，p < 0.01），到 1 个月与 2 个月不再显著（U = 212.5 与 162.0，n.s.）。这就是自发恢复：急性期缺损重，随后自行好转。原文对此有一段很关键的警告：if a benefit from a therapeutic intervention is seen acutely post-MCAO, the experimenter will not know whether the treatment has truly improved functional capacity or has simply accelerated the spontaneous recovery（如果在 MCAO 后急性期看到治疗获益，实验者无法判断是治疗真的改善了功能，还是仅仅加速了本来就会发生的自发恢复）。
这一类里还有笼内够取（只在 1 个月时“达标时间”显著，U = 28，p < 0.05）、走廊测验（p = 0.054，差一点，原文估算需 n = 38）、脱离注意任务（1 个月时双侧反应时都延长；加巧克力干扰后同侧在 1 个月与 2 个月都显著，需 n = 31）、以及阶梯测验的不规则横档版本（后爪滑步增多但前爪滑步减少，总滑步只有 0.1–0.5 次/趟，两组都以 10–12 秒通过——说明这个版本太容易，没有区分度）。
第三类：结果反直觉
这一类最值得本科生读，因为它训练的是“别把方向当意义”。
游泳分析：划水次数（前肢、后肢，两侧）都没有差异，但 MCAO 大鼠游得更快（1 个月：2.26 ± 0.37 s vs 2.82 ± 1.77 s，U = 10，p < 0.01；2 个月：2.41 ± 0.34 s vs 3.03 ± 0.81 s，U = 14，p < 0.01）。原文的解释是：可能招募了前肢（大鼠本来游泳主要用后肢），也可能是焦虑增加——但该文没有测焦虑，所以这个解释只能算推测。
锥形平衡木：MCAO 组在 2 个月时对侧前爪滑步更少（U = 20.5，p < 0.05），而通过时间没有差异。原文的解释与步态分析对上了：MCAO 组 1 个月时后肢基底更窄（U = 14，p < 0.05）、步幅增大（U = 13.5，p < 0.05），后肢离横梁边缘更远，反而更不容易滑下去。也就是说，“滑步少”不是功能更好，而是步态代偿的结果。
握力：MCAO 组需要更大的力才松开杆（F1,22 = 4.36，p < 0.05），两侧没有差异。原文的解释是：不是“握力变强”，而是“松开这个动作启动不了”——类似于肌张力障碍（dystonia），即无法主动打开前爪。这个解释如果成立，握力测验测的其实是“运动启动”，不是“力量”。
第四类：什么都没测出来
安非他命诱导旋转：两组都没有净旋转不对称（对照 7.10 ± 46.75 次 vs MCAO −8.22 ± 121.2 次，n.s.）。原文的解释有说服力：安非他命旋转依赖背侧与内侧纹状体的多巴胺去神经，而该文 30 分钟的 MCAO 病灶偏外侧，正好避开了这个区域；文献里报告阳性的研究多用 60–120 分钟闭塞，损伤更重。
黏纸去除（adhesive removal）：两个时间点、两侧的接触潜伏期与去除潜伏期都没有差异。原文记录的“最常见的坑”很生动：大鼠（对照与 MCAO 都一样）直接无视了贴纸——the most common caveat we experienced with this test was that the rats (both Control and MCAO) ignored the stimuli and took quite a long time to contact them。作者推测，这批鼠被“摸得太熟”（well handled），根本不觉得贴纸讨厌。
圆柱测验（cylinder test）：术后 1 天有轻微的对侧放置缺损，但不显著；其余时间点都没有前肢使用偏向。原文的解释是：依赖自发探索行为的测验，在慢性期会因动物“对环境太熟悉、失去兴趣”而失效。
作者推荐的“稳健测验”与它们的代价
该文的结论可以压缩成三句话：阶梯测验最有用，但需要数周预训练与限食；阿扑吗啡旋转最稳定，但需要 n ≈ 50，且要考虑药物与治疗的相互作用；转棒可以用于长期随访，但必须把“对照组在进步”这件事讲清楚（原文估算，基于 2 个月数据，检出 50% 疗效需 n = 27）。
这三句话背后有一个共同的取舍：越“灵敏”的测验，越贵；越“便宜”的测验，越容易饱和。16 项测验里没有一个在所有维度上胜出，所以原文的建议不是“用最好的那个”，而是“按你要回答的问题选测验，并且同时用两个以上、互相印证的测验”。
七、盲法与测试顺序
这一节是操作层面的，几乎每一条都能直接抄进 SOP。
盲法不是“尽量”，是“必须”。分组在造模后立即编码（例如 A/B 两组，编码表由不参与行为测试的人保管），行为测试者只知道编号、不知道处理；所有评分（尤其是 mNSS、Garcia 这种主观量表）最好先录像，再由第二个不知道分组的人盲评。数据分析前不要解码，解盲要留记录。
先做应激小的测验。同一天要做多个测验时，顺序应该是：旷场（需要新鲜环境）→ 简单反射/放置类 → 转棒/平衡木 → 需要抓取或注射的测验（如阿扑吗啡旋转）。水迷宫应激最大，通常单独安排，且不在同一天与其它测验混做。
同一时段、同一操作者、同一顺序。原文的做法是 8:00–19:00 之间、按笼序测试（Behavioural testing was carried out between the hours of 8 a.m. and 7 p.m. and animals were tested in cage order.）。按笼序的好处是：同一笼的动物处理方式一致，且时间带来的漂移被均匀分配到各组；坏处是如果笼子位置本身有差异（靠窗、靠门），会引入系统误差——所以要把不同组的笼子交替摆放。
预实验定基线。该文所有动物在术前都训练到稳定基线，术后再测。没有稳定基线的动物，术后的“变化”可能只是“还没学会”。基线不稳定的动物应当剔除，而不是用统计方法“校正”。
给动物适应时间。把动物从饲养间搬到测试间，至少适应 30 分钟再开始；测试间温度、湿度、光照与饲养间尽量一致（原文环境为 21 ± 2 °C、湿度 55 ± 10%、14/10 小时光暗周期）。
记录一切可能影响结果的变量。日期、时间、操作者、室温、湿度、光照强度、笼号、动物体重、当天是否有异常（腹泻、伤口、打架）。这些记录在投稿时看起来没用，但在你发现“某一批数据整体偏移”时，是唯一能救你的线索。
统计上别把重复测量当独立样本。同一只动物在 1 天、7 天、1 个月、2 个月各测一次，这是重复测量数据，应当用重复测量方差分析或混合效应模型，而不是在四个时间点各做一次 t 检验——后者会让“至少有一个时间点显著”的概率大幅膨胀。
八、给本科生的建议：先学会打分
这一节写给刚进组、还没有独立课题的人。行为学是本科生最容易“真的做出贡献”的地方，但也最容易变成“帮忙抓老鼠”的地方。区别在于你有没有把一件事做到“可被引用”。
先彻底学会一种测验
不要一上来就学五种。选一个与你课题最相关、流程最标准化的测验（mNSS、Garcia、前肢放置，或者旷场），跟班至少 20–30 次，从搬笼、适应、操作、记录到清理，每一步都自己走一遍。然后做一件关键的事：和师兄师姐独立评分同一批动物或同一批录像，比对分歧。
分歧不可怕，可怕的是不知道分歧在哪。把每一处不一致记下来：是“这一项到底算 2 分还是 3 分”的边界问题，还是“根本没看清动物做了什么”的操作问题。前者要修订评分细则（写进 SOP），后者要改进录像角度或操作手法。
做到评分者内部一致性可接受
评分者内部一致性（intraclass correlation coefficient, ICC）是衡量“同一批样本、不同评分者/不同时间点评分是否一致”的指标。做法很朴素：取 10–20 段录像（覆盖轻、中、重各种缺损），两个不知道分组的人各自独立评分，算 ICC。经验上，ICC > 0.75 算可接受，> 0.90 算优秀；低于 0.75 就不要急着测实验组，先改细则、再训一轮。
为什么要算 ICC 而不是“我们俩差不多”？因为“差不多”没有数值，审稿人不会信；而 ICC 是一个可以写进方法学、可以被别人重复的数字。更重要的是，算 ICC 的过程会逼你把评分细则写清楚——这本身就是对实验室的贡献。
再扩到第二种测验
第一种测验做到 ICC 可接受之后，再学第二种。新测验先做预实验（每组 3–5 只），确认它能区分“明显缺损”和“正常”，再正式纳入。不要因为“文献里都用”就直接上——该文已经证明，同一个模型、同一批动物，有的测验就是测不出差异。
为什么“学会打分”比“学会做手术”更容易成为你的贡献
三个理由。第一，门槛与风险：动物手术需要资质、伦理审批、长期训练，失败成本是动物的生命；评分需要的是时间、耐心与一致性，失败成本只是重看几段录像。第二，稀缺性：会做手术的人不少，愿意做几十次盲法评分、把 ICC 算出来、把 SOP 写清楚的人很少。第三，不可替代性：盲法评分者的价值恰恰在于“不知道分组”，这个角色不能由做实验的人兼任；你做得越规范，数据越可信。
具体能做的事，按“贡献度”排：建立评分 SOP（含边界案例的图例）→ 做 ICC 并记录 → 承担盲法评分 → 录像与编号管理 → 数据录入与双人核查 → 预实验验证新测验。这六件事做完，你在论文里的位置就不是“致谢”，而是“行为学方法”那一节。
不要做的事也很明确：不要单独决定剔除标准（哪只动物算“没造模成功”）、不要单独修改评分细则、不要在数据分析前解盲。这三件事一旦由一个人私下决定，整批数据的可信度就没了。
九、原文自陈的局限
原文的“Limitations”一节写得很坦诚，几乎每一条都值得抄下来提醒自己：
品系。只用了 Wistar 大鼠。不同品系在行为测验上的表现差异很大，结论不能直接搬到 SD 大鼠、Long-Evans 或小鼠。
模型偏轻。30 分钟闭塞造成的是以皮层下为主的梗死，只有少量皮层损伤。这解释了为什么很多测验“测不出差异”——不是测验不灵敏，而是病灶本身不够大。原文也承认，安非他命旋转失败很可能就是这个原因。
蛛网膜下腔出血。约 20% 的动物因线栓拔除后 CBF 未恢复而被剔除，死后解剖证实为蛛网膜下腔出血；作者说明现在改用 Doccol 线栓后发生率显著下降。
测试在光照期进行。大鼠是夜行性动物，光照期测试可能低估或扭曲某些行为（尤其是认知类测验）。原文明确点出这一点，但没有改。
测验本身可能是一种“环境丰富化”。16 项测验、两个月、反复抓取与操作，对动物来说是一种持续的刺激与训练，可能本身就有康复治疗的作用。这是一个很难排除的混杂因素，也是所有“多测验长期随访”研究的共同问题。
只测了雄性。雌性在动情周期不同阶段的行为差异、以及对损伤的反应都不同，该文没有涉及。
最后一条不在“局限”里，但同样重要：这套结论来自 MCAO 模型。脑出血（ICH）的病灶机制不同（血肿占位 + 凝血酶毒性 + 血红蛋白降解），行为学表现的时间进程也不同（ICH 的自发恢复通常比 MCAO 慢，胶原酶模型甚至 21 天仍无明显恢复）。所以该文的“哪些测验稳健”，在 ICH 里只能当参考，不能当结论——这也正是本系列把造模、行为学、影像分开讲的原因。
▲ 本页翻译难点
behavioural tasksblinded observercage orderRotarodspontaneous recoverystaircase taskswimming analysistapered balance beamgrip strengthapomorphine-induced rotationadhesive removalsub-arachnoid haemorrhage附录 · 英文原文
⚠ 本附录逐字收录既定英文片段（PMC5453456.md 的 Behavioural testing、四类结果小节、Robust tests，以及 PMC9972928.md 的 BEHAVIORAL ASSESSMENT 一节，约 21.6K 字符），供对照与检索。机器抽取自公开全文，保留原文的引用编号与符号。
Behavioural testing
Methods and full protocols are provided for each behavioural test in the online supplemental data (Sections A and C), including practical advice and troubleshooting tips (C). All behavioural testing was performed by a blinded observer. Two cohorts of rats were used to reduce the number of tests each rat was exposed to.
Tests that uncovered significant and stable deficits in the MCAO animals
Staircase task
MCAO rats collected fewer pellets with their contralateral paw, than the corresponding paw in the Control group, on the staircase test at two months post-surgery (: Blocks 2–4: U = 38–41, p < 0.05, Block 5: U = 45, p = 0.08), whereas they showed no deficits in the number of pellets retrieved on the ipsilateral side (U = 59–74, n.s.). Figure 3.Summary of tests in which MCAO animals exhibited deficits. (a) Pellets retrieved with each forelimb in the staircase task at two months (Cohort 1: n = 14 Control, n = 11 MCAO). (b) Latency to fall from the Rotarod at one and two months (Cohort 1: n = 14 Control, n = 11 MCAO). (c) Forehand and backhand steps during lateralized stepping (both cohorts at one day: n = 19 Control, n = 20 MCAO, at seven days: n = 19 Control, n = 19 MCAO, at one month: n = 23 Control, n = 18 MCAO, and at two months: n = 24 Control, n = 20 MCAO). (d) Modified neuroscore (both cohorts: n = 24 Control, n = 20 MCAO). (e) Time to criterion and number of failed reaches in the cage reaching task at one and two months (Cohort 1: at 1 mo: n = 13 Control, n = 11 MCAO, at two months: n = 13 Control, n = 11 MCAO). (f) Bias in the corridor task at one and two months (Cohort 1: n = 14 Control, n = 11 MCAO). (g) Latency to respond with and without the chocolate distractor in the disengage task (Cohort 2: n = 10 Control, n = 9 MCAO, except at one month with chocolate distractor for the contralateral paw: n = 10 Control, n = 8 MCAO). (h) Latency to cross for regular and irregular rungs in the ladder test at one and two months (Cohort 2: n = 10 Control, n = 9 MCAO). Data are expressed as median, minimum and maximum values in the box and whisker plots and as means ± standard deviation (SD) with individual values in the scatter plots. Note: the dotted line represents 50% bias (no preference for either side or paw). * = p < 0.05, = p < 0.01, * p < 0.001.
Apomorphine-induced rotation
The MCAO animals exhibited a significant increase in the number of rotations to the ipsilateral side in response to apomorphine; this was in contrast to no net asymmetry in rotations in the Control animals (MCAO, 189.0 ± 169.3 vs. Controls, 3.4 ± 9.23 (Mean ± SD); U = 14, p < 0.01).
Tests in which the MCAO animals exhibited subtle deficits and/or recovery
Rotarod
Latency to fall off the Rotarod was significantly lower in the MCAO group when compared to the Control (, F1,22 = 6.59, p < 0.05). This deficit appeared to increase over time (MCAO × Time: F1,23 = 8.95, p < 0.01) such that the post hoc comparison between groups was only significant on the two-month time point (p < 0.05). However, this effect was not due to a further decrease in performance in the MCAO group, but rather an improvement in the ability to remain on the Rotarod in the Control group.
Lateralized stepping
The MCAO rats exhibited a paw bias, meaning they performed fewer steps with the contralateral paw than with the ipsilateral paw, although this deficit was not stable across all time points. They had a paw bias in the forehand direction at one day and two months (, U = 72, p < 0.001, U = 100.5, p < 0.001) and in the backhand direction at seven days and one month (, U = 105, p < 0.05, U = 114, p < 0.01).
Modified neuroscore
The score differed between MCAO and Control groups at one and seven days post-MCAO (; U = 104.0 and 123.5, respectively, both p < 0.01); at both time points, the MCAO animals scored higher (indicating reduced neurological function overall) but this was no longer the case at one and two months (U = 212.5 and 162.0, respectively, both n.s.).
Cage reaching task
MCAO rats took longer to achieve 20 successful reaches than Control animals at one month (, U = 28, p < 0.05). However, no other measures achieved significance.
Corridor task
MCAO animals retrieved fewer pellets from the contralateral side in the corridor task at both one and two months after surgery, in comparison to the absence of a side preference in the Control group. However, this difference did not achieve significance (, Groups, F1,22 = 4.14, p = 0.054).
Disengage task
Overall a larger response time to both contralateral and ipsilateral perioral stimulation without the chocolate distraction was evident in the MCAO group in comparison to the Control group at one month (, Contra: U = 15.50, p < 0.05, Ipsi: U = 13, p < 0.01), but not at two months (U = 43, 29, n.s.). When the chocolate distraction was introduced, there was a significantly higher response time to perioral stimulation in the MCAO group on the ipsilateral side in comparison to the Control group at one month (U = 11, p < 0.01) and two months (U = 10, p < 0.01).
Ladder test
On the regular spaced rungs in the ladder task, there was a small but significant increase in the number of hindpaw slips exhibited by the MCAO group at one month (Control 0.0 ± 0.0, MCAO 0.08 ± 0.08 (Mean ± SD); U = 20, p < 0.05). However, there was also a reduction in forepaw slips in the MCAO group at one month (Control 0.50 ± 0.40, MCAO 0.20 ± 0.36 (Mean ± SD); U = 20.5, p < 0.05). It should be noted that the number of overall foot slips made on the ladder with regular rungs was very low (means of 0.1–0.5 slips per traverse) and both groups crossed relatively quickly (10–12 s;). In comparison, rats took almost twice as long to cross the ladder with irregularly spaced rungs (, 18–25 s). This slowing was most apparent in the MCAO group, but failed to achieve significance (U = 29 & 24, n.s.).
Tests in which the results were counter-intuitive.
Swimming analysis
There were no significant differences in the number of fore or hindlimb strokes performed by the MCAO animals on either side when compared to the corresponding paws from the Control group (, U = 24–44, n.s.). However, the MCAO animals were significantly faster at crossing the swim tank than the control animals (1 month: 2.26 ± 0.37 vs 2.82 ± 1.77 s, U = 10, p < 0.01. 2 month: 2.41 ± 0.34 vs 3.03 ± 0.81 s, U = 14, p < 0.01 (Mean ± SD)) Figure 4.Summary of tests in which the MCAO animals exhibited no deficits (b and f) or tests that were counter-intuitive (a, c, d, and e). (a) Number of fore and hindpaw strokes during swimming analysis at one and two months (Cohort 2: n = 10 Control, n = 9 MCAO). (b) Forelimb exploring bias in the cylinder task (Cohort 1: at one day: n = 14 Control, n = 11 MCAO, at seven days: n = 12 Control, n = 11 MCAO, at one month: n = 14 Control, n = 11 MCAO, and at two months n = 10 Control, n = 8 MCAO). (c) Hind and forepaw slips and time to cross the tapered balance beam at one and two months(Cohort 2: n = 10 Control, n = 9 MCAO). (d) Paw base width, paw placement overlap, and stride length during gait analysis at one and two months (Cohort 2: at one month: n = 9 control, n = 9 MCAO, and at two months: n = 10 Control, n = 9 MCAO). (e) Force required to release and side bias during grip strength at one and two months(Cohort 1: n = 14 Control, n = 11 MCAO). (f) Contact and removal bias during adhesive removal at one and two months(Cohort 2: n = 10 Control, n = 9 MCAO, except at one month for remove: n = 9 Control, n = 9 MCAO). Data are expressed as means ± SD with individual values in the scatter plots. Note: The dotted line represents 50% bias (no preference for either side or paw). * = p < 0.05, = p < 0.01, * p < 0.001.
Tapered balance beam
The MCAO group made significantly fewer contralateral forelimb slips on the beam in comparison with the control group at two months (; U = 20.5, p < 0.05). However, MCAO animals were no faster in their latencies to cross the beam (F1,15 = 3.57, n.s).
Gait analysis
A number of measures of gait were assessed. Distance between forepaws (base width) was unaltered in the MCAO group compared to the Control group (; n.s.). Hindlimb base width (distance between the two hind limbs) was narrower in the MCAO group than the Control rats at one month only (Hind base width,: U = 14, p < 0.05). Overlap of hind and forepaws during the walking cycle appeared to be reduced in the MCAO animals in comparison to the Control rats; however, this only achieved significance on the ipsilateral side at one month (U = 18, p < 0.05). Additionally, stride length was only significantly increased in the MCAO group in comparison to the Control group at the one-month time point (; U = 13.5, p < 0.05).
Grip strength
The MCAO group required significantly more force than control animals to release the bars at both one and two months (; Groups; F1,22 = 4.36, p < 0.05). Moreover, they exhibited no significant asymmetry between the two grip strength bars (Bias: n.s.) indicating no difference in grip strength between sides.
Tests in which the MCAO animals did not exhibit significant deficits.
Amphetamine-induced rotations
Neither group exhibited any rotational asymmetries in response to amphetamine (net rotations: Control: 7.10 ± 46.75 vs. MCAO: −8.22 ± 121.2, (Mean ± SD); n.s.).
Adhesive removal
The adhesive removal (‘sticky label’) test of somatosensory neglect showed no differences between groups at either time point (one or two months) in either the latency to contact or remove the labels affixed to either of their forepaws ().
Cylinder task
Although there appeared to be a small contralateral deficit in paw placing on the surface of the cylinder among the MCAO animals at one day, this was not significant and no significant bias in forelimb exploratory use was observed at any of the other post-surgical time points ()
Robust tests
Apomorphine-induced rotations robustly detected rotational bias in the MCAO animals at two months, which is consistent with other reports.^[6] The variability of this test was low, and as such achievable group sizes (n = 50,) would be required in order to test a therapeutic intervention. However, the interaction between apomorphine with any therapy should be carefully considered. The other test that proved to be useful at long-term time points was skilled reaching using the staircase chambers. This is not entirely surprising as skilled reaching requires a high degree of motor coordination and tests like this have been consistently used by several groups to reveal subtle impairments and compensatory behaviours in many different animal models of neurological disease and degeneration.^[7] Unfortunately, tests such as these are not routinely used in pre-clinical stroke research. This might be due to the fact that pre-training the animals to perform in the test apparatus can be quite labour intensive and may extend over several weeks in order to first achieve stable baseline performance. Furthermore, there is the additional requirement of food restriction in order to motivate the animals to perform. This may conflict with the ability of this test to be employed in the acute MCAO post-operative period and may render the tests undesirable for many groups. Furthermore, as food restriction has been shown to induce long-term alterations in rodents, which may be beneficial, such as improving behavioural Furthermore, enhancing neuroprotection and increasing longevity.^[8],[9] As is the case with the majority of behavioural assessments, there is variability between animals, so large group sizes are often required and this can be even more discouraging with tests that require large investment such as skilled reaching. Within our study, the staircase deficit was stable across blocks 2–4 of training, and approached significance on block 5. Therefore, it would be recommended to increase the sample size for this test, indeed large group sizes would be required for testing treatments (). A second type of skilled reaching, caged skilled reaching, did not reveal stable deficits. This may have been due to the variability of the data and the propensity of the rodents to compensate by using the non-affected paw (a factor that is not contributing in the staircase task where pellets on each side of the body can only be retrieved with the corresponding paw). As well as increasing the group size to account for this variability (n = 37,), some groups lesion the side of the brain that corresponds to the preferred paw of each individual rodent.
BEHAVIORAL ASSESSMENT
Standard behavioral assessment techniques evaluate rodents’ neurological, motor, and sensorimotor functions from the neurological scores assessed clinically to conclude the stroke outcomes in human patients. Motor and sensorimotor analyses are carried out in an experimental setup to determine rodent stroke outcomes.
summarizes the common techniques available for ICH evaluation in the experimental setup with the advantages and disadvantages of each techniques used.
[Table 2] Common techniques for intracerebral hemorrhage evaluation in experimental setup
ICH parameters | Tests/techniques | Advantages | Disadvantages | Reference number Behavioral assessment | | | | Neurological score | mNSS | Comprehensive evaluation with long term ICH outcome assessment | Can assess the neurological deficit in basal ganglia and not all parts of the brain region | [[53][54][55]] | Garcia Scale | Easy to perform | Cannot be used for long-term neurological outcomes | [[56]] Functional assay | Forelimb placement test | Easy to perform, no training required | Confined to forelimbs, not considered as a comprehensive evaluation | [[57][58]] | Open field test | Easy to perform | Results will be affected due to external influences like environmental factors | [[59][60]] | Corner test | Able to assess long-term neurological outcomes | Cannot be used with severely injured animals as it might affect the results | [[58][61]] | Rotarod test | Able to assess long-term neurological outcomes, objective and sensitive | Training required and an experimental setup is required to perform the test | [[62]] Hematoma/lesion volume | Morphological measurements | Most common in practice, easy to perform | Animals have to be sacrificed at certain time points to get the volume of the hematoma | [[55][63]] | MRI | Noninvasive and accurate | Expensive to perform | [[64]] | PET/CT | Noninvasive and cost effective when compared to MRI | Combination of PET/CT scan cannot match the accuracy of anatomical and physiological features provided by a single MRI scan | [[79][80]] | Cresyl violet staining | Traditionally used method to analyze the loss of neurons in a particular brain area | Not accurate | [[65]] | H and E staining | Most common method used | Has to be combined with MRI to confirm the volume of the hematoma as H and E staining provides just the histopathology of the tissue | [[66]] Brain water content (edema) | Dry/wet method | Easy to perform | Misinterpretation of the impact of minute changes in percentage brain water content | [[67][68]] | MRI | Noninvasive, assessment of hemispheric volumes on MRI which allows a direct quantification of the space-occupied effect caused by edema formation in experimental ICH | Expensive as it involves MRI machine in the experimental setup | [[69]] BBB disruption | Evans blue leakage method | Most commonly used method with a standard protocol | Potential toxicity as it is injected in vivo, free dye being present in the animal following injection | [[67][68]] | FITC-dextran permeability assay | Most commonly used for quantitative studies of BBB permeability | Toxic | [[64]] | Microscopic assessment | Realtime analysis of BBB disruption | Poor tissue penetration | [[73][74][75]] | Measurement of blood proteins | No exogenous tracers required | BBB disruption cannot be analyzed at a particular time point | [[75][76][77]] | Assessment of BBB disruption with MRI | Precision and accuracy | Needs experience to handle the machine and expensive | [[75][78]]
The modified neurological severity score
The modified neurological severity score (mNSS) is the most commonly used neurological test in rodents [[55]]. The mNSS can be performed both for ischemic stroke and ICH, including motor, sensory, and reflex functions. The higher the score, the more severe the neurological deficit. In most experimental brain injury models, mNSS is used to assess long-term outcomes. In ICH models, rats display a higher mNSS score as early as 1 day post-ICH, which indicates massive neurological impairment. The observational behavior of mNSS is based on the different parameters that are used in the test such as the hindlimb motor function, forelimb motor function, and body asymmetry test; mNSS is a useful tool to assess the long-term stroke outcome; and it can be used to evaluate the neuroprotective effects of novel treatments []. The disadvantage of this method is the complexity, as it involves multiple neurological aspects. Hence, individual scores have to be taken into account to analyze the neurological deficits of specific brain regions [[50][55][56]].
Garcia scale
The Garcia scale was developed to assess rats’ motor and sensorimotor functions [[59]]. In this model, the animals are tested for their motor functions with spontaneous activity (0–3), symmetry of limb movement (0–3), forelimb outstretching (0–3), climbing and grip strength (0–3), climbing and grip strength (0–3), body symmetry (0–3), and sensory function of vibrissae (0–3). The collective score is represented in the Garcia scale []. Unlike mNSS, in the Garcia scale, high scores correspond to less neurological deficits. The significant advantage of this model is its simplicity. The Garcia scale is easy to perform, providing motor and sensorimotor functions. The major disadvantage is Garcia scale focuses more on the forelimb than the hindlimb [] [[57]].
Forelimb placement test
The forelimb placement test (FPT) is based on the vibrissae-elicited forelimb placing test [[58]]. Animals respond to vibrissae stimulation, making it easier to check ICH effects. When the animals are placed on the edge of the table, as soon as the vibrissae touch the table, normal animals will place the ipsilateral forelimb on the table, with ICH animals showing less or no response. FPT is easy to perform; however, the major disadvantage is that FPT is confined to the forelimb. It has to be combined with other neurological functional assays to assess the animals’ complete motor or sensorimotor functions [] [[58][59]].
Open field test
An open field test was developed to assess the locomotor activity of rodents [[60]]. A wooden or plastic open field maze (50 cm [length] × 50 cm [width] × 38 cm [height]) is used. The animals are placed in the open field to explore for 10 min; locomotor activity is measured with the help of a video recorder. The major goal of this test is to analyze the movement of the animals placed in the chamber. With the help of video recording and automated systems, the distance, route, immobile time, and rear-up behavior can all be tracked and analyzed to evaluate the locomotor ability of the normal and ICH animals. Open field test is easy to perform and minimize human errors. The disadvantage of this method is the environmental factors that might affect the results [] [[60][61]].
Corner test
The corner test is used to analyze the sensorimotor and postural asymmetry [[59]]. The setup for the corner test is to attach two cardboards 30 cm × 20 cm at a 30° angle. Animals are analyzed based on the side they turn upon vibrissae stimulation after placing them in the middle of the two cardboards. Able to access long-term outcomes and a practical test to analyze the sensorimotor functions post-ICH [] [[59][66]].
Rotarod test
The rotarod test is used to assess motor coordination. The animals are placed in a 3 cm (diameter) × 40 cm (length) rod, which rotates with the help of an electric motor. The speed of the motor increases gradually, and the latency before the animal falls is recorded. The rotarod test is sensitive as it accurately measures the motor coordination of the animals, with a major limitation of long training trials required before the ICH surgery [] [[64]].