From August 24 to 25, 2026, Changping Laboratory hosted the “25 Years of Human Genome” International Symposium. During the conference, Professor Sten Linnarsson, Fellow of the Royal Swedish Academy of Sciences and Professor of Molecular Systems Biology at Karolinska Institute, was interviewed. He has long studied cell types and lineages in the nervous system, contributed to the construction of single-cell atlases of the mouse and human nervous systems, and developed technologies using UMIs (unique molecular identifiers) to improve the accuracy of transcript quantification.
In the interview, he used brain cancer research as an example to discuss how seemingly promising findings generated in the laboratory can be rigorously tested in real patients.
Sten Linnarsson: From Single-Cell Atlases to Clinical Treatment
The Appeal of the Brain
Linnarsson entered neuroscience early in his career for a simple reason: the brain underlies perception, emotion, cognition, and virtually all aspects of human social behavior. As his research progressed, however, his focus gradually shifted towards a different question: How can scientific discoveries be translated into therapies that genuinely benefit patients? Brain cancer is not among the most common cancers, but it is exceptionally difficult to treat. Many findings that appear effective in animal models fail to produce the same results in humans. It is even more complicated when the immune system is involved, as immune responses, the tumor microenvironment, and effective dosage can differ substantially from mice to humans. A “cure” for animals can therefore serve only as a guide; it must never replace evidence from clinical trials.
Clinical Validation
In Linnarsson’s view, clinical trials are not merely the final step for determining whether a drug works. They are also experimental settings for verifying the mechanism of disease and treatment.
As long as ethical requirements are satisfied, researchers should make the most of every trial to extract new knowledge: analyze patient tissue samples, compare subtle molecular and cellular level changes before and after treatment, and merge these observations to refine subsequent hypotheses. Tissue samples alone can establish only that two phenomena are connected. Genomic evidences may help researchers identify underlying causes, but they remain subject to limitations and can never replace administering a treatment to patients and observing its de facto clinical effects. The most important considerations are therefore rigorous trial design, careful sample preservation, and an appropriate attitude toward failure. Even when a trial does not succeed, researchers should extract new questions from the results and continue the investigation.
Reliable Data
Single-cell omics and spatial technologies have made biological observation increasingly precise. Researchers can distinguish different cell types, states, and spatial loci within the same tissue and examine how tumor cells interact with immune cells.
However, sophisticated technology does not automatically guarantee reliable data. Early projects had to scale measurements from relatively small numbers of cells to hundreds or thousands while also dealing with noise that emerged during amplification. UMI technology addresses this issue by attaching molecular tags to original molecules, allowing researchers to distinguish genuine molecular signals from PCR duplicates. Even so, its performance remains reliant on experimental design, sequencing depth, and sample quality.
Frozen postmortem tissue presents another major challenge. Storage time, cellular degradation, and sampling location can all affect RNA quality, while different brain regions may require different processing protocols. Once the data have been generated, cell types must also be annotated using the existing literature and expert knowledge. No single research group can fully characterize the entire brain. Cell-type annotations therefore need to be continuously revised as new samples and new markers become available. Generating data is only an intermediate stage of mapping an atlas; ensuring that datasets generated by different studies remain comparable also requires long-term maintenance.
Choosing One’s Path
Sten Linnarsson did not follow the conventional academic route of immediately pursuing a postdoctoral position. Instead, he first founded a company and entered industry. This experience helped him build up knowledge in management, project execution, and resource allocation. At the same time, his company remained deeply engaged in scientific research and contributed to academic publications.
After returning to academia, this unconventional industry experience became an advantage when he applied for academic positions and research. grants.
In his view, industry and academia have never been inherently conflicting paths. What ultimately determines the value of research is whether the research direction is clear, whether experimental findings can be independently reproduced, and whether a team can refine a research technology into a practical tool that can be used by the broader scientific community.
Conclusion
There is no automatic pathway from a single-cell atlas to clinical treatment. Animal models provide mechanistic clues; single-cell and spatial technologies reveal the cellular details of patient tissues; and clinical trials determine whether these findings can ultimately translate into improved patient outcomes. Linnarsson believes that the next breakthrough will depend not only on generating larger datasets, but also on integrating data acquisition, experimental design, and causal inference into a single iterative cycle, allowing basic research to continuously refine itself under conditions that more closely reflect the complexity of the human body.
Joseph Ecker: From Plants to the Human Brain
In his interview, Professor Joseph Ecker noted that the Human Genome Project provided researchers with a “partial list”: we know which genes are present in the genome, but we still do not know which genes function in certain cells, and when.
He described the dramatic increase in sequencing capacity over the past 25 years as a transition “from hundreds to hundreds of millions.” Sequencing is no longer simply a matter of reading a stretch of DNA. Researchers can now examine, at single-cell resolution, whether genes turn on, how DNA methylation changes, and how these changes relate to life, aging, and disease.
Developing the Methods First in Plants
Joseph Ecker’s story began with Arabidopsis thaliana, the mouse-ear cress, widely used as a model organism. Early in his career, he participated in sequencing chromosome 1 of Arabidopsis in collaboration with Craig Venter’s team. Once the plant sequence had been obtained, the team began asking a new question: How does DNA methylation affect gene function? To answer this question, they developed DNA methylation sequencing approaches such as MethylC-seq in plants. In 2008, the team generated a genome-wide DNA methylation map of the plant genome. By 2009, the same technological framework had been applied to the study of the human methylome.
In Ecker’s view, this was not a sudden transition from plants to humans. Developing the technologies in plants first gave the team the technical capability to apply them to the much more complex cellular landscape of the human body. In this sense, a single plant became a gateway to human genome research.
Brain Maturation Is Also Written in the Methylation Map
As the technology evolved, Joseph Ecker’s team began studying a form of non-CG DNA methylation known as mCH in the human brain. mCH occurs predominantly in neurons. After birth, it gradually accumulates and continues to increase until approximately 25 years of age.
This period corresponds to a critical stage in the maturation of neural circuits. Ecker explained that the overall number of neurons in the human brain is largely established by the time of birth. After birth, many of the major changes involve the remodeling and refinement of synaptic connections between neurons. Neurodevelopmental and psychiatric disorders such as autism and schizophrenia may also emerge during this period. Similarly, researchers have observed changes in DNA methylation during aging and in Alzheimer’s disease. The relationship between dynamic changes in DNA methylation and brain maturation therefore represents an important scientific question. However, temporal coincidence does not establish causality. Ecker’s strategy begins with constructing reference maps of normal neurons across different age groups and then uses these maps as baselines for comparison with neurons in disease states. Determining which molecular changes represent normal developmental processes and which are directly associated with disease will require long-term longitudinal follow-up and functional validations.
From Sequence to Knowledge: What Can AI Do?
Today, a single cell can generate multiple layers of multi-omic data, including DNA sequence, genetic variation, DNA methylation, and transcriptional information. As the amount of data continues to grow, however, interpretation has become an increasingly difficult challenge. Ecker believes that AI may help researchers identify patterns across vast numbers of cells and developmental time points, transforming sequence information into hypotheses that can subsequently be tested. But AI cannot automatically answer questions about disease. The prerequisite remains the establishment of reliable reference maps of normal cells, followed by comparisons of disease samples within the same cell types and developmental stages. Ultimately, experimental validation is required to determine whether the conclusions are correct.
Conclusion
From studying DNA methylation in a single plant to investigating cellular states in the human brain, Joseph Ecker’s career has not followed a predetermined path. Each time he solved a problem, he gained the ability to ask the next question. The next stage of genomics may therefore involve more than simply acquiring data. The greater challenge may be to understand those sequences in the context of cells, time, and disease.