Scientists are building a catalogue of every type of cell in our bodies

An adult human body consists of some 37 trillion cells. Not so long ago, these were thought to come in 220 different types. That number, the product of painstaking decades spent peering through microscopes at slides bearing tissue sections coloured by chemical stains, gave a sense of the division of cellular labour needed to keep a body running.

A sense, but only a superficial one. Tools now exist that are capable of looking inside the cells, breaking them open one at a time to release their complements of messenger RNA (mRNA), the molecule which carries genetic information from a cell’s nucleus to its protein factories. Molecules of mRNA indicate which genes are active, thus revealing a cell’s inner nature. Cells that look alike under a microscope often turn out to be quite diverse. The cell-type count has thus risen above 5,000.

The leader of this histological revolution is the Human Cell Atlas (HCA) consortium, which was set up in 2016 and currently involves more than 3,600 collaborators in 190 laboratories in 102 countries. Other cell-atlas projects are limited to mapping particular organs or types of tissue. The HCA aspires to catalogue the whole caboodle: identifying and locating all the cell types, healthy and diseased, in every human tissue over the course of a lifetime. Its remit extends even to “organoids”, science’s fumbling first attempts to grow living simulacra of organs.

The hope, according to Sarah Teichmann of Cambridge University and Aviv Regev of Genentech, an American biopharma firm, who set the whole thing up, is to have a first draft of the atlas available next year. Their latest progress report has just been published as a set of papers in Nature and several of its sister journals.

As Dr Teichmann and Dr Regev point out, HCA maps are of two sorts. One, similar in concept to geographers’ maps, ties each cell type to a four-dimensional site in the human body (sampling at different stages of life adds the dimension of time to those of space). The other sort are less familiar. These, called manifolds, are normally used by mathematicians to represent multidimensional mathematical hyperspaces. In the case of the HCA, the numerous dimensions in question are not space and time but, rather, molecular features, such as mRNA profiles, characteristic of different cell types. By plotting different cell types on the same map, charts of manifolds thus enhance understanding of their similarities and differences.

No cell left behind

The geography of the real world also plays a part. From the start, Dr Teichmann and Dr Regev have been determined not to oversample parts of the world (Europe, North America and certain bits of Asia) where scientists are concentrated. Instead, they have sought participants from all six inhabited continents—a decision that has already been rewarded with insights into the cellular basis of geographical differences in immune responses and susceptibility to breast cancer.

The subjects of this week’s papers show the scope of the endeavour. Placentas, the embryonic development of the skeleton, gut inflammation and the formation of the thymus (the organ which generates the immune system’s T-lymphocytes, the cells ravaged by AIDS) are all discussed.

The findings of these studies break new ground. They confirm earlier suspicions that some cellular processes involved in the formation of cancerous tumours are involved in the placenta’s rapid growth. They identify genes expressed in developing bone and cartilage cells that may lead to arthritis in later life. They show, by comparing healthy and unhealthy guts, that one source of disease-causing inflammation seems to be intestinal cells accidentally developing into a type normally found in the stomach. And they give a detailed description of the thymus based on a standardised representation of that organ.

Perhaps the most intriguing paper of the lot, though, is on brain-mimicking organoids. Organoids composed of human brain cells, themselves derived from laboratory-created stem cells, are the sort of thing to give bio-ethicists jitters. At the moment, bereft of the blood supply needed to grow, they reach only three or four millimetres across, so are unlikely to develop any form of consciousness. But some worry that larger versions might.

They are, however, useful for research, as they permit the study of living human brain tissue without the need to remove any. But they would be even better if the particular types of neuron in particular versions of them could be reliably predicted—for neurons collectively make up a large fraction of known cell types, and each has a different job to do.

The HCA will make this easier. A paper co-ordinated by Barbara Treutlein of the Federal Institute of Technology in Zurich looked at mRNA data from 36 such organoids, created using 26 different protocols. The researchers involved were able both to identify the neuron types generated in each organoid and to determine how closely they resembled their natural equivalents. The results, stitched together, create a single manifold chart for such organoids that shows the strengths and weaknesses of the various protocols, and will help with planning future research.

Besides publicising the project members’ latest findings (though the raw data have been online since they were collected), the papers also allow Dr Teichmann and Dr Regev to set out their vision for using artificial intelligence (AI) to turn the atlas into something closer to a model of how a human being works.

Both are computational biologists by training, and it was this background which led them to conceive the HCA in the first place. Without the software underpinning the project, which turns data into maps and permits those maps to be interrogated, the project would not exist. But the pair have bigger visions. They were early adopters of foundation models, a class of AI (such as the large language models that have gained prominence in recent years) which feeds on vast quantities of training data in order to recognise patterns not discernible to humans.

The HCA’s foundation models are trained not on passages of text, but collections of cells. And their goal is not human-like composition but the creation of better and more useful maps. Some learn from mRNA data about cell types. Others rely on conventional histology slides and more modern iterations thereof—such as light-sheet imaging, which scans sections through three-dimensional samples. These models are now good enough to be used to annotate the cells in new specimens, to search for similar cells in different specimens and to discover the gene programs behind particular characteristics. In the future they should be able to predict how cell lineages will develop and even to envisage as-yet-unknown varieties of cell. Such models are not only faster than human researchers, but can also perform tasks beyond human capability.

The result is a system that can be (and has been) used not just for enhancing the atlas, but putting it to work. Drug companies are already, for example, using HCA data and models to screen potential drugs “virtually” before they are tested experimentally; to predict side-effects by discovering non-target tissues where the gene a drug candidate interacts with is expressed; and, conversely, to spot opportunities in such non-target tissues to extend a drug’s range of therapeutic targets.

One day all this effort may contribute to a human “digital twin”, which would also incorporate foundation models about how proteins work (such as AlphaFold, a protein-folding model developed by Google DeepMind) and how bodies develop. That day is still far distant. But it now seems more likely to arrive.

© 2024, The Economist Newspaper Limited. All rights reserved. From The Economist, published under licence. The original content can be found on www.economist.com