内容来源：https://www.quantamagazine.org/how-does-life-unfold-a-landscape-metaphor-comes-into-its-own-20261009/

内容总结：

生命是如何展开的？一种景观隐喻终于有了用武之地。

到20世纪中叶，生物学似乎已近乎被完全破解。查尔斯·达尔文的自然选择进化论解释了生物如何变化和适应。现代遗传学则在分子层面揭示了其运作机制。控制生物性状的基因从亲代传递给子代，随机突变产生变异，自然选择从中挑选出“最适者”。自然选择与遗传学在生物学家朱利安·赫胥黎1942年的著作中所命名的“现代综合”中融为一体。1953年DNA被证明是基因的编码物质之后，生物学的其余部分看起来不过是在拼凑细节。

但并非所有人都感到满意。英国生物学家康拉德·哈尔·沃丁顿——朋友和同事称他为“沃德”——认为关注基因固然很好，但他和许多人一样并不清楚：基因究竟如何塑造生物体的形态和特征？在胚胎发育过程中，基因如何可靠地在正确的位置生成身体的各种组织？大约在赫胥黎提出“现代综合”的同时，沃丁顿提出了一种关于这一过程如何展开的新图景。

他将发育过程想象为一片带有山丘和山谷的景观，山谷像河网一样不时分叉。在最高山的顶端，他想象有一个球，代表处于最早胚胎形态的一群细胞——即我们现在所称的干细胞或多能细胞，它们有潜力变成任何细胞类型。细胞向各自命运推进的过程——比如变成皮肤细胞、肌肉细胞或神经细胞——就像球滚下山坡穿越这片景观。

当球到达分叉点时，模型的“重力”将其拉向一条轨道或另一条：用科学术语来说，它必须分化。沃丁顿认为，通过一系列这样的分叉决策，细胞特化为成熟类型，各自表达特定的基因组。细胞被限制在少数明确定义的状态中，因为它们在沃丁顿的术语中是“渠化”的：被山谷两侧困住并引导。这样，有限数量的细胞类型就能可靠地从大量基因的活动中产生。

伦敦弗朗西斯·克里克研究所的发育生物学家詹姆斯·布里斯科表示，沃丁顿的景观“作为概念框架极为有用”。“核心思想——发育是渐进的、细胞状态是离散的、分叉代表细胞命运决定、渠化意味着鲁棒性——所有这些都经受住了考验。”

西雅图干细胞与再生医学研究所的生物化学家苏珊娜·拉费尔斯基说：“沃丁顿试图传达的一些概念内涵极为丰富。”

但这一概念更像是一种视觉隐喻，而非对实际生物学过程的表述。至少看起来如此。过去几年中，研究人员已根据胚胎发育过程中从数千个细胞收集的实验数据，绘制出了细胞状态空间的真实地形——实际上就是沃丁顿的景观。这些分析揭示，发育不仅仅是遗传程序的读取，而是一个动态过程：细胞群体通过相互协商和互动，构建自身以及它们所穿越的景观，最终形成一个生物体。

这张发育地图正在帮助研究人员理解早期胚胎中相同的细胞如何发育成复杂生物体中不同的组织和细胞类型。布里斯科说，这对再生医学和干细胞工程意义重大：“如果你了解景观的拓扑结构并知道分叉点在哪里，原则上就可以设计方法，精确地将细胞群体引导到期望的状态，而不是靠反复试验。”

因此，沃丁顿的隐喻可能是理解和重塑细胞、组织和胚胎可能性的关键。

绘制景观图

20世纪40年代，遗传学和胚胎学是两门独立的科学；甚至并非所有人都相信基因在发育中起重要作用。然而，沃丁顿自己的胚胎学实验以及其他人的实验使他确信，胚胎中组织的形成和模式化确实受基因控制。他认为，不同组织的细胞在染色体中有不同组的基因被激活。分化过程逐渐使细胞特化为皮肤细胞、神经细胞等。

沃丁顿认为，发育路径——例如从一层称为外胚层的胚胎组织形成神经管，最终发育为中枢神经系统——一旦开始就是不可避免的，就像水流沿河谷下泄时受周围环境约束一样。路径可能略有蜿蜒，但山谷使河流保持方向。分化对应山谷的分叉，最终目的地是身体的成熟细胞类型。

沃丁顿在绘图中将这一过程视觉化为景观，但宾夕法尼亚州斯沃斯莫尔学院的发育生物学家斯科特·吉尔伯特在1991年发表于《生物学与哲学》的文章中写道，这些绘图“没有物理现实基础”。它们只是思考发育问题的有用示意图，仅此而已。

布里斯科说，“没有数学内容”，这个隐喻“无法区分不同机制、做出定量预测或被证伪”。因此，他说，沃丁顿的景观“有时变成了陈词滥调，而非有意义的解释”。

例如，是什么决定了山丘和山谷的地形？在1957年出版的《基因的策略》一书中，沃丁顿将他所设想的景观描绘为从下方看的样子：一张被绳索网络拉扯成形的薄片，绳索固定在桩子上。桩子代表单个基因，绳索代表它们对发育的影响。表达特定基因——将其转化为相应蛋白质——就像拉动其绳索改变薄片的形状。

沃丁顿意识到，实际情况要复杂得多，因为绳索连接成一个网络。任何一根绳索都可能在多个点影响景观，而景观上的任何一点都可能连接多根绳索。

这些隐藏的连接对应着研究人员现在所称的基因调控网络：基因之间的相互作用，其中一个基因表达的增加可能改变或调控其他基因的活性。发育的隐秘复杂性正在于此。特定过程（如神经管的形成）可能受许多相互作用基因的影响；基因型与决定形态的发育景观之间不存在简单关系。基因调控网络在其中起中介作用。

因此，在整个生物体中寻找专门负责制造特定组织、器官或结构的基因是毫无意义的，这也是沃丁顿的隐喻被证明有用的原因之一。布里斯科说，景观图像“在提供替代纯粹基因中心发育观方面很有价值”。通过强调景观的形状，它指向了“无法从任何单个基因读取的系统层面属性”。

沃丁顿最初提出景观和绳索网络时，基因调控尚未被理解。几年后，在20世纪60年代初，人们开始清楚一个基因可以开启或关闭另一个基因。复杂性理论家斯图尔特·考夫曼等研究人员开始构建基因相互作用网络的简单数学模型。但当时对真实基因网络的了解太少，无法将抽象理论与实验联系起来。吉尔伯特写道：“鉴于通道和球体没有物理现实，胚胎学家该怎么用它们呢？”

过去二十年，由于表征细胞状态的新实验工具的出现，答案开始浮现。生物学家终于能够将实验数据映射到真实的数学分化景观上，并证明沃丁顿多么有先见之明。

命运的吸引子

定义胚胎中特定细胞状态的一种方法是其基因的表达水平。现在可以使用一种称为单细胞RNA测序的技术同时测量许多细胞中的这些水平，该技术提供每个细胞所含所有不同RNA分子的快照。RNA分子（或转录本）是基因的一种拷贝，用于将其翻译成功能性蛋白质，通常RNA转录本越多意味着该基因表达越高。每种细胞类型都有特定的基因表达设置，并在细胞分裂时被继承。这确保肝细胞在生物体发育过程中始终保持为肝细胞。

原则上，RNA测序数据可以建模为高维数学空间，其中每个轴表示单个基因的RNA转录本数量。这样的图可能有数千个维度来表示数千个基因，无法可视化，但可以通过数学方法投影到较少维度的空间，通常只有两个维度，产生类似复杂3D物体阴影的东西。

在这种称为UMAP的图上，不同细胞类型表现为点簇——比如血细胞聚集在一起，肌肉或前脑细胞也是如此——因为它们共享基因表达谱。随着发育进行，细胞谱系沿着路径穿过这张图。当数据按时间映射时——即所谓的网络流模型——看起来非常像一群细胞滚过沃丁顿式景观，被引导沿着山谷流向对应不同细胞类型的盆地。

不过，我们不应假设这张基因表达图就是字面意义上的沃丁顿景观。华盛顿大学的生物物理学家朱莉·塞里奥特警告说，单细胞RNA测序并非表征细胞状态的唯一甚至最佳方法。她说，这项技术起初“令人惊叹”，但RNA转录本并不完美映射基因表达。此外，当研究人员将数据压缩为低维图时，他们往往以反映先前关于细胞如何聚类为不同类型的假设的方式进行。

尽管如此，可以明确的是，正如沃丁顿所设想的那样，细胞在成熟和发育过程中经历状态变化——由哪些基因活跃或哪些蛋白质被制造来定义——这些变化具有时间轨迹，将它们带向有限数量的最终状态。那么，这个可能性景观是如何形成的？是什么决定了它提供的路径？

2007年，现就职于西雅图系统生物学研究所的生物学家黄穗描述了沃丁顿所提出的那种细胞状态分叉。他发现称为髓系祖细胞的血细胞分化为两种特化类型，使它们走上变成红细胞或白细胞的路径，这一过程由两种蛋白质GATA1和PU.1主导。因此，细胞分化景观可以仅用两个值来表示：这两种蛋白质的浓度。更多GATA1促进红细胞形成，更多PU.1则将细胞推向白细胞命运。

黄穗及其同事利用RNA测序实验数据和理论计算绘制了景观图。起初只有一个盆地，对应髓系祖细胞池。但随着分化进行，这个盆地相对于两个新的相邻盆地变得不那么稳定——不那么深——这两个新盆地分别对应红细胞（高GATA1）和白细胞（高PU.1）状态。想象一张从四角水平悬挂的薄片，中间积了一池水。如果你从下方推水池，水会向两侧流向新的更低的池子。这些盆地就是数学家所称的吸引子：细胞不可避免地被吸引向的状态，就像水在景观中遵循重力流动一样。

研究人员现在已经记录了许多可以用这种方式描述的细胞命运决定：基因活性的变化改变了景观的形状。这些变化产生新的吸引子，对应低洼山谷中稳定的细胞类型；不稳定的基因活性配置则构成高地。

布里斯科是一个研究小组的成员，该小组用动力系统理论的语言推进景观思想，探索由相互作用变量集合描述的系统如何演化。与布里斯科合作的英国华威大学数学家戴维·兰德说：“我们的工作从沃丁顿的隐喻出发。”他说，这不仅仅是一个方便的意象：它“揭示了发育系统结构的某些根本性东西”。

结合实验和理论，他们也发现少数关键基因驱动细胞命运决定。网络的其余部分执行决定，其他基因在下游被激活或抑制，但并不主动参与决定。布里斯科说：“我们认为这反映了基因调控网络组织方式的某种真实情况。”

从进化角度看，这是合理的。对于任何多细胞生物，各种细胞类型必须可靠地生成并保持稳定。布里斯科说，具有深且分离良好的吸引子的基因调控网络将被自然选择保留。

这些深井代表了沃丁顿所设想的细胞命运决定的渠化。分子世界充满噪声，充满随机波动。同一类型的两个细胞在任何时刻都不可避免地具有略微不同的基因表达水平。在景观图中，它们不必占据完全相同的位置。但只要一组细胞在同一盆地中，它们就会遵循相同的基本轨迹。

布里斯科说：“如果命运决定对任何分子波动都高度敏感，发育就不会是鲁棒的。”

自然之力

沃丁顿引人入胜的直观观点留下了一些未解问题。景观的高度究竟代表什么？当他的球滚下斜坡时，发育中什么相当于时间流逝？

布里斯科和兰德尔的动力系统观点提供了一些答案。它表明景观的高度反映细胞状态的稳定性，最不稳定的状态在最高峰，最稳定的在最低谷。时间不是字面时间，而是一种“伪时间”，反映所有细胞的基因表达谱如何变化。理论上，所有细胞可以沿单一伪时间轴排序，细胞状态变化越大，沿轴越靠后。但这些变化不会同步，因此不存在适用于所有细胞的滴答时钟——故称伪时间。

问题是这种时间概念并不总是有效，兰德说。当细胞穿越景观中高而不稳定的点时，它适用。但当细胞处于低洼吸引子盆地时，它们的表达谱不会稳定为单一模式，而似乎更随机而非系统地演化，仿佛“时间”在蜿蜒。

那么在这幅图中什么扮演重力的角色？什么力量推动细胞沿发育路径前进？拉费尔斯基说：“在发育生物学中，重力相当于多细胞生物中分化的必然进程，从少数细胞类型开始。”这一过程似乎有不可阻挡的方向和动力。塞里奥特说：“大多数时候，细胞遵循可预测的分化顺序。”但这种顺序并非编程在基因组中；毕竟，生物体中的所有细胞都有相同的基因组。相反，细胞命运决定是集体做出的，细胞通过细胞表面读取的化学信号以及组织生长折叠时细胞膜的拉扯和挤压相互交流和互动。

塞里奥特说：“在发育中，总有反馈，细胞内的东西可以根据外部输入进行调整。”

因此，虽然沃丁顿的景观看起来像等待穿越的静态领土，但真实的景观随着细胞发育而变化。兰德说：“在我们的框架中，景观是每个细胞的特征，并非固定不变。吸引子通过分叉出现和消失，连接它们的路径也在变化。”就好像景观不断被底层构造力重塑。可以说，沃丁顿的绳索网络本身在不断运动，将薄片拉向各个方向。

兰德说，在这方面，沃丁顿的图景并不完整。从来不是某种普遍重力使球滚动；相反，细胞谱系在生命展开的过程中都穿越一个不断变化的景观。

通过这种方式，现代数据既证实了沃丁顿强大隐喻的洞见，也增加了其复杂性。理论与实验相结合的方法帮助阐明了复杂但至关重要的过程：像我们这样的多细胞生物的细胞如何从全能的多能干细胞导航到明确定义的特化组织。例如，布里斯科发现神经管模式化中有两种方式到达景观中的同一位置——这一发现调和了先前相互矛盾的观察。他说，没有景观框架，他的团队无法理解其结果。

黄穗现在将景观图应用于癌症发展。传统观点认为癌细胞是功能失调的，通常因为基因突变破坏了细胞分裂使其不受控制地进行。这种观点将癌症视为异常状态。但黄穗认为它只是景观中的另一个吸引子——尽管可能是灾难性的——可以通过许多不同途径轻易到达。在这种观点中，癌变只是我们细胞所做的可能事情之一：可以说是作为多组织生物体的必然结果。

然而，这并不意味着我们对此无能为力。一旦将癌症视为吸引子状态，新选择就成为可能：引导这些细胞离开癌症盆地，进入更良性的盆地——细胞什么都不做、经历自发死亡，甚至恢复健康状态。塞里奥特说：“与其说需要杀死所有癌细胞，不如想：我们能否驱动它们分化？”

这种对景观轨迹的控制也可用于组织再生，或将一种细胞类型转化为另一种——例如再生受损神经。塞里奥特说，在农业中也可能有用途，比如选择性增强植物的可食用部分。

布里斯科说，“景观框架为这种发育设计提供了理性基础”，开启了“生成性”生物学的前景，即可以按需塑造组织甚至生物体。若如此，沃丁顿的景观模型就远不止是隐喻——它可能是一片迄今未绘图区域的地图集。

中文翻译：

生命如何展开？一幅景观隐喻终于实至名归。

到20世纪中叶，生物学似乎已经几乎被解决了。查尔斯·达尔文的自然选择进化论解释了生物如何变化和适应。现代遗传学表明了这一过程在分子尺度上如何运作。控制生物性状的基因从亲代传递给子代，随机突变产生变异，自然选择从中挑选出“最适应者”。自然选择与遗传学相互契合，生物学家朱利安·赫胥黎在其1942年的著作中将其命名为“现代综合”。1953年DNA被证明是基因的编码物质之后，生物学的其余部分看起来不过是拼凑细节而已。

但并非所有人都感到满意。英国生物学家康拉德·哈尔·沃丁顿——朋友和同事都叫他“沃德”——认为这种对基因的关注固然很好。但对他来说，对任何人来说都一样，并不清楚基因究竟如何塑造生物体的形态和特征——它们如何在胚胎发育过程中，可靠地在所有正确的位置生成身体所有不同的组织。大约在赫胥黎揭幕现代综合的同时，沃丁顿提出了一幅新颖的图景，展示了这一过程可能如何展开。

他将发育过程想象为一片带有山丘和山谷的景观，山谷时不时地分叉，就像河网中的那样。在最高山丘的顶端，他想象了一个球，代表处于最早胚胎形态的一群细胞——即我们现在所称的干细胞或多能细胞，它们有潜力变成任何细胞类型。细胞向其命运推进——比如变成皮肤细胞、肌肉细胞或神经细胞——就像球从山坡上滚下，穿过这片景观。

当球到达一个分支点（分叉）时，模型的“重力”将它拉向这条或那条轨道：用科学术语来说，它必须分化。沃丁顿论证说，通过一系列特定的此类分支决策，细胞特化为其成熟类型，即命运，每种类型表达一组特定的基因。细胞被限制在少数几个明确的状态中，因为它们是“渠道化的”——用沃丁顿的术语来说：被山谷两侧困住并引导。通过这种方式，一支基因大军在活动中可靠地产生了有限数量的细胞类型。

沃丁顿的景观“作为概念脚手架极为有用”，伦敦弗朗西斯·克里克研究所的发育生物学家詹姆斯·布里斯科说。“核心思想——发育是渐进的、细胞状态是独特的、分叉代表（细胞）命运决定、渠道化意味着稳健性——所有这些都经受住了考验。”

“沃丁顿试图分享的一些概念内涵极为丰富，”西雅图干细胞与再生医学研究所的生物化学家苏珊娜·拉费尔斯基说。

但这一概念更像是一个视觉隐喻，而非对实际生物学过程的表征。至少看起来是这样。在过去几年中，研究人员已经从胚胎发育过程中从数千个细胞收集的实验数据中，绘制出了细胞状态空间的真实地形——实际上就是沃丁顿的景观。这些分析揭示出，发育不仅仅是遗传程序的读取，而是一个动态过程，细胞群体通过相互协商和互动，构建自身以及它们所穿越的景观，最终形成一个生物体。

这种发育的制图学如今正在帮助研究人员理解早期胚胎中相同的细胞如何发育为复杂生物体中不同的组织和细胞类型。布里斯科说，这对再生医学和干细胞工程意义重大：“如果你理解了景观拓扑结构，知道分叉在哪里，原则上你就可以设计方法，精确地将细胞群体引导到期望的状态，而不是靠反复试验。”

那么，沃丁顿的隐喻可能是理解和重塑细胞、组织及胚胎可能成为什么的关键。

绘制景观图

20世纪40年代，遗传学和胚胎学是分开的科学；甚至并非所有人都相信基因在发育中扮演重要角色。然而，沃丁顿自己在胚胎学中的实验以及他人的实验使他确信，胚胎中组织的形成和模式化确实受基因控制。他认为，不同组织的细胞在其染色体上有不同组的基因被激活。一个分化过程逐渐将细胞特化为皮肤细胞、神经细胞等等。

沃丁顿认为，发育路径——例如，从一层称为外胚层的胚胎组织形成神经管，最终发育为中枢神经系统——一旦过程开始就是不可避免的，就像水流沿着河谷下泄受周围环境约束一样。路径可能略有蜿蜒，但山谷使河流保持航向。分化对应于山谷的分叉，最终目的地是身体的成熟细胞类型。

沃丁顿将此在绘图中可视化为一片景观，但它们“没有物理现实的基础”，宾夕法尼亚州斯沃斯莫尔学院的发育生物学家斯科特·吉尔伯特在1991年发表于《生物学与哲学》的一篇文章中写道。它们只是思考发育问题的有用示意图，仅此而已。

“没有数学内容”，这一隐喻“无法区分替代机制、做出定量预测或被证伪”，布里斯科说。因此，他说，沃丁顿的景观“有时变成了陈词滥调，而非有意义的解释。”

例如，是什么决定了山丘和山谷的地形？在他1957年的著作《基因的策略》中，沃丁顿将他所设想的景观描绘为从下方看的样子：一张被钉子上系着的绳索网络拉扯成形的薄片。钉子代表单个基因，绳索代表它们对发育的影响。表达一个特定基因——将其转化为相应的蛋白质——就像拉动其绳索来改变薄片的形状。

沃丁顿意识到，实际情况比这复杂得多，因为绳索连接成一个网络。其中任何一根都可能在多个点影响景观，而景观上任何给定点都可能连接着多根绳索。

这些隐藏的连接对应于研究人员现在所称的基因调控网络：基因之间的相互作用，其中一个基因表达的增加可能改变或调控其他基因的活性。发育的隐秘复杂性就在这里。一个给定的过程，比如神经管的形成，可能受到许多相互作用基因的影响；基因（基因型）与决定形态的发育景观（表型）之间没有简单的关系。基因调控网络在两者之间起中介作用。

因此，在整个生物体中寻找专门负责创建特定组织、器官或结构的基因是没有意义的，这也是沃丁顿的隐喻被证明有用的一个原因。景观图像“在提供一种替代纯粹以基因为中心的发育观方面很有价值”，布里斯科说。通过强调景观的形状，它指向了“无法从任何单个基因中读出的系统层面属性”。

沃丁顿最初设计他的景观和绳索网络时，基因调控还不为人所理解。几年后，在20世纪60年代初，人们清楚地认识到一个基因可以开启或关闭另一个基因。研究人员，如复杂性理论家斯图尔特·考夫曼，开始构建基因相互作用网络的简单数学模型。但当时对真实基因网络的了解太少，无法将抽象理论与实验联系起来。“鉴于通道和球体没有物理现实，”吉尔伯特写道，“胚胎学家该怎么用它们呢？”

过去二十年中，由于表征细胞状态的新实验工具的出现，答案开始浮现。生物学家现在终于能够将实验数据映射到一个真实的数学分化景观上，并展示沃丁顿是多么有先见之明。

命运的吸引子

定义胚胎中给定细胞状态的一种方法是其基因的表达水平。现在可以使用一种称为单细胞RNA测序的技术在许多细胞中同时测量这些水平，该技术提供每个细胞所含所有不同RNA分子的快照。一个RNA分子（或转录本）是基因的一种拷贝，用于将其翻译为功能性蛋白质，一般来说，更多的RNA转录本意味着该基因更高的表达。每种细胞类型都以特定的基因表达设置为特征，这种设置在细胞分裂时被继承。这确保了肝细胞在生物体发育过程中始终保持为肝细胞。

原则上，RNA测序数据可以被建模为一个高维数学空间，其中每个轴表示单个基因的RNA转录本数量。这样的图，可能有数千个维度代表数千个基因，是不可能可视化的，但可以通过数学方法投影到维度更少的空间，通常只有两个维度，产生类似复杂3D物体阴影的东西。

在这样一张称为UMAP（均匀流形近似与投影）的图上，不同的细胞类型表现为点簇——比如血细胞聚集在一起，肌肉或前脑细胞也是如此——因为它们共享基因表达谱。细胞谱系随着发育的推进沿着这张图上的路径前进。当数据通过时间映射时——在所谓的网络流模型中——它看起来非常像一群细胞滚过一片类似沃丁顿的景观，被沿着山谷引导到对应于不同细胞类型的盆地。

然而，我们应该警惕不要以为这张基因表达图就是字面意义上的沃丁顿景观。华盛顿大学的生物物理学家朱莉·塞里奥警告说，单细胞RNA测序不是表征细胞状态的唯一甚至最佳方法。她说，这项技术起初“令人惊叹”，但RNA转录本并不能完美地映射到基因表达。此外，当研究人员将数据压缩为低维图时，他们通常采用反映先前关于细胞如何聚集成不同类型假设的方式。

尽管如此，正如沃丁顿所设想的那样，细胞在成熟和发育过程中确实经历状态变化——由哪些基因活跃或哪些蛋白质被制造来定义——而这些变化有一条穿越时间的轨迹，将它们带向有限数量的最终状态。那么，可能性景观是如何形成的？是什么支配着它提供的路径？

2007年，现就职于西雅图系统生物学研究所的生物学家黄穗描述了沃丁顿所提出的那种细胞状态分叉。他发现，称为髓系祖细胞的造血细胞分化为两种特化类型，使它们走上成为红细胞或白细胞的路径，这一过程由两种蛋白质GATA1和PU.1主导。因此，这个细胞分化景观可以仅用两个值来表示：这两种蛋白质的浓度。更多的GATA1促进红细胞的形成，而更多的PU.1将细胞推向白细胞命运。

黄穗和他的同事利用RNA测序实验数据和理论计算绘制了这片景观。起初只有一个盆地，对应于髓系祖细胞池。但随着分化的推进，这个盆地相对于两个新的相邻盆地——分别对应于红色（高GATA1）和白色（高PU.1）血细胞状态——变得不那么稳定——不那么深。想象一张从四角水平悬挂的薄片，中间积了一滩水。如果你从下方顶起水滩，水就会向两侧流向新的、更低的池子。这些盆地就是数学家所称的吸引子：细胞不可避免地被吸引向的状态，就像水在景观上遵循重力流动一样。

研究人员现在已经记录了许多可以用这种方式描述的细胞命运决定：作为改变景观形状的基因活性变化。这些变化产生新的吸引子，对应于低洼山谷中的稳定细胞类型；不稳定的基因活性构型构成了高地。

布里斯科是一个研究小组的成员，该小组用动力系统理论的语言来探索景观思想，该理论研究由相互作用变量集描述的系统如何演化。“我们的工作从沃丁顿的隐喻出发，”与布里斯科合作的英国华威大学数学家戴维·兰德说。他说，这不仅仅是一个方便的意象：它“揭示了发育系统结构的某些（根本性）东西”。

结合实验和理论，他们也发现少数关键基因驱动细胞命运决定。网络的其他部分执行决定，因为其他基因在下游被激活或抑制，但并不主动参与其中。“我们认为这反映了基因调控网络组织方式的某些真实情况，”布里斯科说。

从进化的角度来看，这是合理的。对于任何多细胞生物体，各种细胞类型必须被可靠地生成并保持稳定。布里斯科说，具有深而分离良好的吸引子的基因调控网络将被自然选择。

这些深井代表了沃丁顿所设想的那种细胞命运决定的渠道化。分子世界是嘈杂的，充满了随机波动。同一类型的两个细胞在任何时刻都不可避免地会有略微不同的基因表达水平。在景观图景中，它们不必占据完全相同的位置。但只要一组细胞在同一个盆地中，它们都将遵循相同的基本轨迹。

“如果命运决定对任何分子波动都高度敏感，”布里斯科说，“发育就不会是稳健的。”

自然之力

沃丁顿引人入胜的直觉性观点留下了一些未解的问题。景观的高度实际上代表什么？当他的球滚下斜坡时，发育中什么等同于时间的流逝？

布里斯科和兰德尔的动力系统观点提供了一些答案。它表明景观的高度反映了细胞状态的稳定性，最不稳定的状态在最高峰，最稳定的在最低谷。同时，时间不是字面意义上的时间，而是一种反映所有细胞基因表达谱如何变化的“伪时间”。理论上，所有细胞可以沿单一伪时间轴排序，使得给定细胞状态变化越大，它在轴上就越靠后。但这些变化不会全部同步，所以在任何瞬间没有一个适用于所有细胞的滴答作响的时钟——因此称为伪时间。

问题是，这种时间概念并不总是有效，兰德说。当细胞穿越景观中高的、不稳定的点时，它是适用的。但当它们处于低洼的吸引子盆地中时，它们的表达谱不会稳定为单一、稳定的模式，而似乎更随机而非系统地演化，仿佛“时间”在蜿蜒流淌。

那么在这幅图景中，什么扮演重力的角色？什么力量推动细胞沿其发育路径前进？“在发育生物学中，重力（相当于）多细胞生物中分化的不可避免的推进，从少数细胞类型开始，”拉费尔斯基说。这个过程似乎有一种不可阻挡的方向和动力。“大多数时候，细胞遵循一个可预测的分化顺序，”塞里奥说。但这个顺序并非编程在基因组中；毕竟，生物体中的所有细胞都有相同的基因组。相反，细胞命运决定是集体做出的，因为细胞通过在被细胞表面读取的化学信号以及组织生长和折叠时发生在细胞膜上的拉扯和挤压，相互互动和交流。

“在发育中，总是有反馈，”塞里奥说，“（细胞中）的东西可以根据外部输入进行调整。”

因此，虽然沃丁顿的景观看起来像一片等待被穿越的静态领地，但真实的景观随着细胞发育而变化。“在我们的框架中，景观是每个细胞的特征，不是固定的，”兰德说。“吸引子通过分叉出现和消失，连接它们的路径也在变化。”就好像景观被底层的构造力不断重塑。你可以说，沃丁顿的绳索网络本身就在不断运动，将薄片拉向这边那边。

在这方面，兰德说，沃丁顿的图景是不完整的。从来就不是某种普遍重力使球滚动；相反，细胞谱系在生命展开的过程中都穿越着一片不断变化的景观。

通过这种方式，现代数据既证实了又复杂化了沃丁顿强有力的隐喻。理论和实验方法的结合帮助阐明了复杂但至关重要的过程，通过这一过程，像我们这样的多细胞生物体的细胞从全能的多能干细胞导航到明确定义的特定组织。例如，布里斯科发现在神经管模式化中有两种方式可以到达景观中的同一位置——这一发现调和了先前矛盾的观察。他说，如果没有景观框架，他的团队无法理解他们的结果。

黄穗现在正在将景观图景应用于癌症的发展。传统观点认为癌细胞是功能失常的，通常是因为基因突变破坏了细胞分裂使其不受控制地进行。这种图景将癌症视为一种异常状态。但黄穗认为它只是景观中的另一个吸引子——虽然是一个可能灾难性的吸引子——可以通过许多不同途径轻易到达。在这种观点中，癌变只是我们细胞所做的可能事情之一：你可以说，这是作为一个多组织生物体的不可避免的后果。

然而，这并不意味着我们对此无能为力。一旦我们将癌症视为一种吸引子状态，一种新的选择就成为可能：引导那些细胞走出癌症盆地，进入另一个更良性的盆地——一个细胞什么都不做、经历自发细胞死亡、甚至恢复健康状态的盆地。“与其只是说我们需要杀死它们全部，”塞里奥说，“这个想法是：我们能否驱动它们分化？”

这种对景观轨迹的控制也可能用于组织再生，或将一种细胞类型转化为另一种——例如，再生受损的神经。塞里奥说，这在农业中也可能有用途，比如选择性地增强植物的可食用部分。

布里斯科说，“景观框架为这种发育设计提供了理性基础”，开启了“生成性”生物学的希望，在这种生物学中，组织和可能的生物体可以按订单被塑造。在这种情况下，沃丁顿的景观模型远不止是隐喻——它可能是一片迄今未知领域的地图集。

英文来源：

How Does Life Unfold? A Landscape Metaphor Comes Into Its Own.
By the middle of the 20th century, it had begun to seem that biology was almost solved. Charles Darwin’s theory of evolution by natural selection explained how organisms change and adapt. The modern science of genetics showed how it works at the molecular scale. Genes governing the traits of organisms get passed down from parent to offspring, and random mutations create the variations from which natural selection picks the “fittest.” Natural selection and genetics fitted together in what the biologist Julian Huxley christened “the modern synthesis” in his 1942 book. Once it was shown in 1953 that the genes are encoded by DNA, the rest of biology looked like little more than piecing together details.
But not everyone was satisfied. The British biologist Conrad Hal Waddington, known to his friends and colleagues as Wad, thought this focus on genes was all very well. But it wasn’t clear to him, or anyone else, exactly how genes shape the forms and features of organisms — how they reliably generate all the different tissues of the body, in all the right places, during embryonic development. Around the same time that Huxley unveiled the modern synthesis, Waddington presented a novel picture of how this might unfold.
He visualized the developmental process as a landscape with hills and valleys that split every so often like those in a river network. At the very top of the highest hill he imagined a ball, which represented a population of cells in their earliest embryonic form — what we now call stem or pluripotent cells, which have the potential to become any cell type. The progression of the cells towards their fates — to become, say, skin cells, muscle cells, or nerve cells — is like the ball rolling downhill through the landscape.
When the ball reaches a branching point (bifurcation), the model’s “gravity” pulls it to one track or the other: In scientific terms, it must differentiate. Through a particular sequence of such branching decisions, Waddington argued, cells specialize into their mature types, or fates, each expressing a particular set of genes. The cells are restricted to a small number of well-defined states because they are “canalized,” in Waddington’s lexicon: trapped and channeled by the sides of the valley. In this way, a limited number of cell types reliably arise from the activity of an army of genes.
Waddington’s landscape has been “remarkably useful as a conceptual scaffold,” said James Briscoe, a developmental biologist at the Francis Crick Institute in London. “The core ideas — that development is progressive, that cell states are distinct, that bifurcations represent [cell] fate decisions, that canalization implies robustness — all survive.”
“There’s so much richness to some of the concepts Waddington was trying to share,” said Susanne Rafelski, a biochemist at the Institute for Stem Cell and Regenerative Medicine in Seattle.
But the concept was more of a visual metaphor than a representation of an actual biological process. Or so it seemed. In the past several years, researchers have mapped out the real topography of cell-state spaces — in effect, Waddington’s landscapes — from experimental data collected from many thousands of cells over the course of embryonic development. Those analyses are revealing that development is not a mere readout of a genetic program, but rather a dynamic process in which communities of cells build themselves, and the very landscape they navigate, by mutual negotiation and interaction, into an organism.
This cartography of development is now helping researchers understand how identical cells in the early embryo develop into the distinct tissues and cell types of a complex organism. This matters enormously for regenerative medicine and stem-cell engineering, Briscoe said: “If you understand the landscape topology and know where the bifurcations lie, you can in principle design methods that steer cell populations to desired states with precision, rather than by trial and error.”
Waddington’s metaphor, then, could be the key to understanding and reshaping possibilities for what cells, tissues, and embryos can be.
Charting a Landscape
In the 1940s, genetics and embryology were separate sciences; not everyone even believed that genes played a significant role in development. Yet Waddington’s own experiments in embryology, and those of others, convinced him that the formation and patterning of tissues in an embryo were indeed controlled by genes. The cells of different tissues had different groups of genes activated in their chromosomes, he thought. A process of differentiation gradually specialized cells into skin cells, nerve cells, and so on.
Waddington figured that the developmental pathways — for example, the formation of the neural tube from a layer of embryonic tissue called ectoderm, which ultimately develops into the central nervous system — are inevitable once the process has begun, just as water flowing down a river valley is constrained by its surroundings. The path might meander a little, but the valley keeps the river on course. Differentiation corresponds to the branching of a valley, and the final destinations are the mature cell types of the body.
From The Strategy of the Genes by C.H. Waddington
Waddington visualized this as a landscape in drawings, but they had “no grounding in physical reality,” wrote Scott Gilbert, a developmental biologist at Swarthmore College in Pennsylvania, in a 1991 essay in Biology & Philosophy. They were useful schematics for thinking about the problem of development, and nothing more.
“Without mathematical content,” Briscoe said, the metaphor “could not distinguish between alternative mechanisms, make quantitative predictions, or be falsified.” As a result, he said, Waddington’s landscape “has sometimes become a cliché rather than a meaningful explanation.”
What, for example, determines the topography of the hills and valleys? In his 1957 book The Strategy of the Genes, Waddington depicted the landscape as he figured it might look from underneath: as a sheet that is tugged into shape by a network of ropes attached to pegs. The pegs represented individual genes, and the ropes their effects on development. Expressing a particular gene — turning it into its corresponding protein — is like pulling on its ropes to change the shape of the sheet.
From The Strategy of the Genes by C.H. Waddington
It’s rather more complex than that, Waddington realized, because the ropes are connected in a network. Any one of them might influence the landscape at several points, and any given point on the landscape might be attached to several ropes.
These hidden connections correspond to what researchers now call gene regulatory networks: interactions among genes whereby an increase in the expression of one might change, or regulate, the activity of others. Herein lies the cryptic complexity of development. A given process, such as formation of the neural tube, might be influenced by many interacting genes; there’s no simple relationship between genes (genotype) and the developmental landscape that determines form (phenotype). Gene regulatory networks mediate between them.
It makes no sense, then, to look for genes dedicated to creating specific tissues, organs, or structures in a whole organism, which is one reason Waddington’s metaphor proved useful. The landscape image was “valuable in providing an alternative to a purely gene-centric view of development,” Briscoe said. By emphasizing the landscape’s shape, it pointed “toward system-level properties that cannot be read off from any single gene.”
Gene regulation wasn’t understood when Waddington first devised his landscape and rope network. A few years later, in the early 1960s, it became clear that one gene could turn another on or off. Researchers, such as the complexity theorist Stuart Kauffman, began to construct simple mathematical models of the networks of gene interactions. But too little was then known about real gene networks to connect abstract theory to experiment. “Given that the channels and spheres had no physical reality,” Gilbert wrote, “what was an embryologist supposed to do with them?”
Answers have begun to emerge over the past two decades thanks to new experimental tools for characterizing cell states. Biologists are now finally becoming able to map experimental data to a real mathematical landscape for differentiation and show just how prescient Waddington was.
Fated Attractors
One way to define the state of a given cell in an embryo is by the expression levels of its genes. These levels can now be measured simultaneously in many cells using a technique called single-cell RNA sequencing, which supplies a snapshot of all the different RNA molecules each cell contains. An RNA molecule (or transcript) is a kind of copy of a gene used to translate it into a functional protein, and generally, more RNA transcripts means higher expression of that gene. Each cell type is characterized by particular gene-expression settings, which are inherited when a cell divides. This ensures that a liver cell will stay a liver cell as it progresses through an organism’s development process.
In principle, RNA-sequencing data can be modeled as a high-dimensional mathematical space in which each axis denotes the number of RNA transcripts of a single gene. Such a plot, with maybe thousands of dimensions to represent thousands of genes, is impossible to visualize, but it can be mathematically projected onto a space with fewer dimensions, usually just two, to produce something like the shadow of a complex 3D object.
On such a plot, called a UMAP (uniform manifold approximation and projection), different cell types appear as clusters of points — blood cells cluster together, say, as do cells of the muscle or forebrain — as they share a gene expression profile. Cell lineages follow paths through this map as development proceeds. When the data is mapped through time — in what’s known as a network flow model — it looks very much like a population of cells rolling through a Waddington-like landscape, being channeled along valleys that lead to basins that correspond to distinct cell types.
We should be wary of supposing, though, that this gene-expression map is literally Waddington’s landscape. Julie Theriot, a biophysicist at the University of Washington, cautioned that single-cell RNA sequencing is not the only or even the best way to characterize cell states. The technique was “mind-blowing” at first, she said, but RNA transcripts don’t map perfectly to gene expression. Plus, when researchers collapse the data into low-dimensional maps, they often do so in ways that reflect prior assumptions about how cells cluster into different types.
Still, it’s clear that, as Waddington supposed, cells undergo changes in state, defined by which genes are active or which proteins are made, as they mature and develop, and that these changes have a trajectory through time that carries them toward a limited number of final states. So how is that landscape of possibilities shaped? What governs the paths it offers?
In 2007, the biologist Sui Huang, now at the Institute for Systems Biology in Seattle, described a bifurcation in cell state of the sort proposed by Waddington. He found that blood-forming cells called myeloid progenitor cells differentiate into two specialized types, setting them on course to become red or white blood cells, in a process dominated by two proteins, GATA1 and PU.1. This landscape of cell differentiation can thus be represented with just two values: the concentrations of these two proteins. More GATA1 promotes the formation of red blood cells, while more PU.1 sends the cell toward a white-blood-cell fate.
Huang and his colleagues mapped out the landscape using data from RNA-sequencing experiments and theoretical calculations. At first there is just a single basin, corresponding to the pool of myeloid progenitor cells. But as differentiation proceeds, this basin becomes less stable — less deep — relative to two new adjacent basins, which correspond to the red (high GATA1) and white (high PU.1) blood-cell states. Think of it like a sheet suspended horizontally from its corners, with water pooled in the middle. If you push up on the pool from below, the water flows out sideways toward new, lower pools on either side. These basins are what mathematicians call attractors: states toward which the cells are inevitably drawn, like water following gravity across a landscape.
Allison Kudla
Researchers have now documented many cell-fate decisions that can be described in this way: as changes in gene activity that alter the shape of a landscape. The changes produce new attractors that correspond to stable cell types in the low-lying valleys; unstable gene-activity configurations constitute the high ground.
Briscoe is part of a research group that has pursued the landscape idea using the language of dynamical systems theory, which explores how systems described by sets of interacting variables evolve. “Our work starts from Waddington’s metaphor,” said David Rand, a mathematician at the University of Warwick in the United Kingdom, who collaborates with Briscoe. It’s more than a convenient image, he said: It “reveals something [fundamental] about the structure of developmental systems.”
Combining experiments and theory, they too have found that a small number of key genes drive cell-fate decisions. The rest of the network enacts the decision, as other genes are activated or suppressed farther downstream, but isn’t active in it. “We think this reflects something real about how gene regulatory networks are organized,” Briscoe said.
This makes sense from an evolutionary perspective. For any multicellular organism, the various cell types must be reliably generated and stable. Gene regulatory networks that have deep, well-separated attractors will be selected for, Briscoe said.
These deep wells represent canalization of cell-fate decisions in the way Waddington supposed. The molecular world is noisy, filled with random fluctuations. Two cells of the same type will inevitably have slightly different levels of gene expression at any moment. In the landscape picture, they don’t have to occupy the exact same spot. But so long as a group of cells is in the same basin, they will all follow the same basic trajectory.
“If fate decisions were highly sensitive to any molecular fluctuation,” Briscoe said, “development would not be robust.”
Forces of Nature
Waddington’s alluringly intuitive view left some open questions. What does the height of the landscape actually represent? And what’s the developmental equivalent of time passing as his balls roll down the slopes?
Briscoe and Rand’s dynamical-systems view provides some answers. It suggests that the landscape’s height reflects the stability of the cells’ states, with the least stable states on the highest peaks and the most stable in the lowest valleys. Time, meanwhile, is not literal time, but a kind of “pseudo-time” that reflects how the gene-expression profiles of all the cells change. In theory, all the cells can be ordered along a single pseudo-time axis so that the greater the change in a given cell’s state, the farther along the axis it is. But these changes won’t all be in sync, so there’s no ticking clock that applies to them all at any instant — hence, pseudo-time.
Dave Guttridge/The Photo Unit
The trouble is that this notion of time doesn’t always work, Rand said. It fits when cells are crossing over high, unstable points in the landscape. But when they are in a low-lying attractor basin, their expression profiles don’t settle into a single, stable pattern but rather seem to evolve more randomly than systematically, as if “time” meanders.
And what plays the role of gravity in this picture? What is the force that moves cells along their development pathway? “In developmental biology, gravity is [equivalent to] the inevitable progression of differentiation in multi-celled organisms, starting from a small number of cell types,” Rafelski said. This process seems to have an inexorable direction and impetus. “Most of the time, cells follow a predictable order in which they differentiate,” Theriot said. But that order is not programmed into the genome; after all, the cells in an organism all have the same genome. Rather, the cell-fate decisions are made collectively, as cells interact with and communicate with one another through chemical signals that are read at the cell surface, and through the tugging and squeezing on cell membranes that takes place as tissues grow and fold.
“In development, there’s always feedback,” Theriot said, “and things [in the cell] can adjust depending on inputs from the outside.”
So while Waddington’s landscape looked like a static territory waiting to be traversed, the real landscape changes as cells develop. “In our framework, the landscape is a feature of each cell and is not fixed,” Rand said. “Attractors appear and disappear through bifurcations, and the routes connecting them change.” It is as though the landscape is constantly being reshaped by underlying tectonic forces. Waddington’s rope network, you might say, is itself in constant motion, pulling the sheet this way and that.
In this respect, Rand said, Waddington’s picture was incomplete. It was never really a kind of universal gravity that made the ball roll; rather, cell lineages all move through a constantly shifting landscape in the process of life’s unfolding.
In such ways, modern data has both confirmed and complicated Waddington’s powerful metaphor. Combined theoretical and experimental approaches have helped to elucidate the complex but vital process through which the cells of multicellular organisms like us navigate a path from all-purpose, pluripotent stem cells to well-defined and specific tissues. Briscoe, for instance, has found that in neural-tube patterning there are two ways of reaching the same place in the landscape — a finding that reconciled previously contradictory observations. His team could not have made sense of their results without the landscape framework, he said.
Huang is now applying the landscape picture to the development of cancers. The traditional perspective is that cancer cells are malfunctioning, generally because gene mutations have damaged cell division so that it proceeds unchecked. It’s a picture that sees cancer as an aberrant state. But Huang regards it as just another attractor in the landscape — albeit a potentially catastrophic one — that can be all too easily accessed by many different routes. In this view, becoming cancerous is just one of the possible things that our cells do: an inevitable consequence, you might say, of being a multi-tissue organism.
This doesn’t mean, however, that there’s nothing we can do about it. Once we view cancer as an attractor state, a new option becomes possible: to guide those cells back out of the cancer basin and into another that is more benign — one where the cells do nothing, undergo spontaneous cell death, or even regain a healthy state. “Rather than just saying we need to kill them all,” Theriot said, “the idea is: Can we drive them to differentiate?”
Such control of landscape trajectories might also be used for tissue regeneration, or for transforming one cell type into another — for example, to regrow damaged nerves. There could be uses in agriculture as well, for selectively enhancing edible parts of plants, Theriot said.
Briscoe said that “the landscape framework provides a rational basis” for this kind of developmental design, opening up the promise of a “generative” biology in which tissues and possibly organisms can be sculpted to order. In that case, Waddington’s landscape model is far more than metaphor — it could be an atlas of a hitherto uncharted terrain.