FlyTitle: Neuroscience

The biggest, most detailed map yet made of brain cells and the connections between them has now been released

迄今为止最大、最详细的脑细胞及其连接体图谱发布

经济学人双语版-果蝇脑图谱 Fly atlas

AT THE BEGINNING of the 20th century Santiago Ramón y Cajal, a Spanish neuroscientist, became known for his exquisite drawings of the branching, treelike cells of the brain and spinal cord. In 1906 he was awarded a Nobel prize for this work, which gave the world its first glimpse into the structure of these neurons, and an inkling of how they are arranged in an animal’s central nervous system.

二十世纪初,西班牙神经学家圣地亚哥·拉蒙·卡哈尔(Santiago Ramón y Cajal)以绘制精细的大脑和脊髓的树状细胞分支图而闻名。1906年,他凭此获得了诺贝尔奖。世人通过他的画第一次领略了这些神经元的结构,并初步了解到它们在动物中枢神经系统中的排列方式。

A century later Cajal’s legacy—supercharged by modern microscopy, heavy-duty robotics and a dollop of machine learning—is thriving. The objective now is to create connectomes. These are three-dimensional maps of all the neurons in entire brains, and how those neurons link together. Earlier this year saw the publication of an important step on the road to a complete brain connectome: a map of about a quarter of a fruit fly’s cerebral capacity.

一个世纪以后,在最新的显微镜技术、高工作量机器人技术,以及些许机器学习的助力下,卡哈尔的遗产正在发扬光大。当前的目标是创建出“连接组”,也就是展现整个大脑中所有神经元及其连接方式的三维图谱。今年稍早时,研究人员在通向完整的脑连接组的道路上迈出了重要一步:发布了约四分之一个果蝇大脑的图谱。

That map, of what its cartographers refer to as the fly’s hemibrain—a set of around 25,000 neurons in the centre of the organ—has been more than a decade in the making. It is the brainchild of Gerry Rubin, a biologist who was also responsible for mapping the fruit fly’s genome as a proof of principle for the Human Genome Project. Dr Rubin is now boss of the Janelia Research Campus in Virginia, a part of the Howard Hughes Medical Institute that is dedicated to neuroscience. The hemibrain connectome is the first phase of the campus’s FlyEM project, to map the fruit fly’s entire brain, which contains around 100,000 neurons. That is a drop in the ocean compared with the 85bn in a human brain, or even the 70m in a mouse brain. But, like the fly’s role in the Human Genome Project, it will be a proof of principle.

绘制者把这个图谱称为果蝇半脑图,包含了果蝇大脑中心的一组约2.5万个神经元。这项工作耗时十多年,是生物学家杰里·鲁宾(Gerry Rubin)的心血之作。他此前还负责绘制果蝇的基因组,为人类基因组计划提供原理验证。他目前是美国弗吉尼亚州珍妮亚研究所(Janelia Research Campus)的所长,该研究所隶属于专门研究神经科学的霍华德·休斯医学研究院(Howard Hughes Medical Institute)。果蝇半脑连接组是该研究所FlyEM项目的第一阶段,这个项目将绘制出果蝇的整个大脑,包含约10万个神经元。这与人类大脑的850亿个神经元相比简直是九牛一毛——就连小鼠大脑也有7千万个神经元。但是,就像果蝇在人类基因组计划中的作用一样,它将提供原理验证。

Flying high

高飞

Each of the hemibrain’s neurons is connected to hundreds of others through junctions called synapses, for a total of more than 20m synapses. These neurons and synapses form circuits that are responsible for a fly’s ability to learn, navigate, sleep and tell the time of day. The only full connectome created so far is that of C. elegans, a nematode worm which has either 302 or 385 neurons in its nervous system, depending on whether it is hermaphrodite or male (there are no purely female C. elegans). The neurons in C. elegans have around 7,000 synapses between them. Mapping the fly hemibrain is thus a big step forward.

果蝇半脑里的每个神经元都通过被称作“突触”的连接体与成百上千个其他神经元相连接。总共有2000多万个突触。这些神经元和突触形成了回路,负责果蝇的学习、导航、睡眠和判断时间等能力。迄今为止,人类创建的唯一完整的连接组是秀丽隐杆线虫的图谱,这种线虫的神经系统中有302或385个神经元——数量取决于它是雌雄同体还是雄性(没有纯雌性的秀丽隐杆线虫)。连接秀丽隐杆线虫神经元的突触总共有7000个左右。因此,绘制果蝇半脑图是向前迈进了一大步。

Elucidating the connectome of C. elegans involved techniques Cajal himself would have recognised. The researchers who did it sliced their worms into thin sections using diamond knives, stained the slices to show the cells within them up more clearly, and then took electron-microscope pictures of the result. Identifying neurons and synapses within the thousands of images thus obtained was a task for expert human eyes.

描绘秀丽隐杆线虫连接组所采用的方法,即便卡哈尔本人也认得出。研究人员用金刚石刀具将线虫切成薄片,将切片染色,让其中的细胞更清楚地显现出来,然后用电子显微镜对处理好的切片拍照,由此获得成千上万张图像。而要从中识别神经元和突触则要靠专业人员的肉眼。

Dr Rubin and his crew have automated things. One of the teams on the campus has, for example, developed a way to speed up the slicing and imaging part of the operation. This technique, which works like an atomic-scale sandblaster, fires a beam of gallium ions at a sample of brain tissue. That etches off a layer of the tissue a few nanometres thick from the sample’s surface. A scanning electron microscope (SEM) then takes a picture of the newly exposed surface. That done, the gallium beam etches away another few nanometres and the process is repeated until the whole sample has been studied.

鲁宾和他的团队则把流程自动化了。例如,研究所的一个团队发明了一种方法来加快操作过程中切片和拍照的速度。这项技术的工作原理仿佛一台原子级别的喷砂机,向脑组织样本发射镓离子束。这会从样本表面削掉几纳米厚的一层组织。然后扫描电子显微镜(SEM)对新暴露出来的表面拍照。之后,镓离子束再削掉几个纳米厚的组织,如此不断重复,直到整个样本研究完毕。

The microscopes involved have been built especially for FlyEM. They sit on air-filled pads to minimise vibrations that might ruin the images, and the room containing them rests on its own concrete slab, to separate it from the remainder of the laboratory. Moreover, while run-of-the-mill SEMs usually operate for hours at a time at most, the FlyEM machines are designed to operate continuously for months.

他们使用的显微镜是为FlyEM项目特制的。它们被放置在充气垫上,从而可以最大限度地减少可能有损成像的震动。放置这些显微镜的房间建在单独的混凝土板上,与实验室的其他部分隔开。此外,普通的扫描电子显微镜通常一次最多只能运行几个小时,而FlyEM所用的可以连续运行几个月。

The result is millions of high-resolution images that have been stitched together to create 3D representations of the fruit fly hemibrain (see above for a picture of the olfactory pathway). The next step was to label the neurons and synapses within. Doing that manually, in the way used for C. elegans, would have taken centuries, according to Stephen Plaza, the project’s manager. Clearly this was a non-starter. So he turned to Google for help.

研究人员将最终得到的几百万张高分辨率的图像拼接在一起,就形成了果蝇半脑的三维图(上图为嗅觉通路部分)。接下来的工作是将其中的神经元和突触标记出来。项目主管斯蒂芬·普拉萨(Stephen Plaza)表示,如果采用像秀丽隐杆线虫项目那样的手工操作,这一步可能需要几个世纪才能完成。这显然行不通。于是他向谷歌求助。

Computer vision has improved enormously in recent years and is routinely used to scan through hundreds of hours of CCTV or satellite images to identify objects of interest to the authorities. Modern artificial-intelligence (AI) algorithms perform better than people at classifying images and, between 2015 and 2018, doubled their performance in object segmentation, a trickier task that involves picking multiple objects from a single image. At Janelia’s behest, Google trained one of its AI algorithms to recognise neurons and synapses within the FlyEM images. As this algorithm scrolled through the pictures, it also attempted to trace the fibrous protuberances called dendrites and axons that connect one neuron to another.

近年来,计算机视觉技术突飞猛进,它通常被用来浏览数百小时的闭路电视或卫星图像,以识别官方部门感兴趣的目标。现代人工智能(AI)算法的图像分类能力优于人类。2015到2018年间,它们在目标图像分割方面的性能提高了一倍。目标图像分割涉及从单张图像中提取多个目标,是一项更复杂的任务。应珍妮亚研究所的请求,谷歌训练了一个AI算法来识别FlyEM图像中的神经元和突触。算法在图片上全面扫描的同时,还尝试追踪连接神经元的被称作树突和轴突的纤维突起。

To start with, the researchers trained the AI on pictures that had already been marked up by human experts. As it churned through further images, human proofreaders checked its decisions and fed errors back to it, so that it could improve its understanding of what neurons look like in different contexts. As the AI got better, the manual workload lessened and the speed with which images were correctly annotated shot up. With the AI’s help, Dr Plaza and his team of 50 proofreaders cut the time required for the annotation down from centuries to a couple of years.

首先,研究人员用已由专业人员标记过的照片来训练AI。在AI处理更多图像的过程中,校对人员会检查它的判断结果,并将错误反馈给它,这样它就能更好地了解神经元在不同环境中的样子。随着AI的进步,人工工作量减少了,正确标记图像的速度也大幅提高。有了AI的帮助,普拉萨和他50人的校对团队将标记时间从几个世纪缩短到了几年。

For all mankind

为全人类

The FlyEM data released this week are available to all neuroscientists, professional or amateur, to use as they see fit. Anyone with an internet connection can look up lists of neurons that are connected to each other and see 3D diagrams of what each of those cells, with its myriad dendritic and axonal branches, looks like.

任何专业或非专业的神经科学家都可以根据自己的需要使用此次发布的FlyEM数据。任何人都能上网查阅相互连接的神经元列表,并查看每个细胞及其大量树突和轴突的3D示意图。

At Janelia, several groups are already mining these data to glean insights. Vivek Jayaraman’s team, for example, studies how a fruit fly’s brain helps the insect first to understand its orientation in space and then to employ that information to help it navigate. Until now Dr Jayaraman has worked with theoretical models of which parts of the brain might talk to each other. The hemibrain map has shown him the actual physical connections between the neurons involved. He, Dr Rubin and other researchers at Janelia will publish their insights over the coming months.

珍妮亚研究所的几个团队已经在挖掘这些数据以获取洞见。例如,维韦克·亚拉曼(Vivek Jayaraman)的小组正在研究果蝇的大脑如何帮助它先了解自己的方位,然后利用方位信息为自己导航。亚拉曼过去一直在研究果蝇大脑的某些部分互相交流的理论模型,如今半脑图向他展示了相关神经元之间的实际连接情况。亚拉曼、鲁宾以及珍妮亚研究所的其他研究人员将在未来几个月发表他们的见解。

With the hemibrain complete, FlyEM’s researchers expect to finish the rest of the fruit fly connectome within the next two years, and thus to gain further insights into fly neurology. But other consequences of the project are crucial, too. The advances in automation and machine learning that are being made through it will be as valuable as the biological insights. And, as the technology gets better, connectome reconstructions will happen faster, allowing the mapping of bigger brains in larger numbers.

半脑图完成后,FlyEM的研究人员期望在未来两年内完成果蝇连接组的其他部分,进而对果蝇的神经系统有更深入的了解。但该项目带来的其他成果也至关重要。在其中使用的自动化和机器学习所取得的进展将和生物学上的成果一样有价值。而且,随着技术的进步,重构连接组的速度会加快,也就可以绘制更多、更大的大脑。

In the future, the aim is to obtain connectomes for several strains of mouse and, eventually, several people too. Looking at the differences in wiring between typical and atypical brains might shed light on conditions such as schizophrenia and autism. Looking at the differences between human brains and those of other species may help explain just what it is that makes humans neurologically special.

未来的目标是获得几种不同品种的小鼠的脑连接组,最终还要获得若干人脑的连接组。研究典型和非典型大脑之间的神经线路的差异,可能让人进一步认识精神分裂症和自闭症等病症。而研究人类和其他物种大脑的差异可能有助于解释究竟是什么让人类的神经系统与众不同。

Dr Rubin estimates that assembling a mouse connectome would cost around $500m (more than ten times what FlyEM will have cost when finished). He is confident such a project could be started within ten years. A human-brain connectome would be orders of magnitude more difficult. But not, he reckons, impossible. In 1990, at the beginning of the Human Genome Project, he recalls that many scientists thought sequencing animal genomes would always be too expensive and difficult. Those detractors said that biologists should choose between mouse and human, since it would probably be impossible to do both. “And now”, he points out, “we have projects where we’re going to do 10,000 human-genome sequences.”■

鲁宾估计,构建小鼠的脑连接组大约需要耗费五亿美元(是FlyEM完成时预计花费的10倍以上)。他很有信心这样一个项目能在十年内启动。构建人类大脑连接组的难度则要高几个数量级,但他认为并非不可能。他回忆道,1990年人类基因组计划开始时,许多科学家认为给动物基因组测序始终都会太贵、太难。这些反对者认为,生物学家应该在老鼠和人类之间做选择,因为两者或许不可能同时进行。“而如今,”他指出,“我们已经有了要对一万个人类基因组测序的项目了。”