Sign in

Inigo Martincorena

@imartincorena.bsky.social
1.8K followers 475 following 92 posts

Scientist. Group leader at the Sanger Institute, Cambridge UK. Somatic mutation and selection in normal tissues, cancer and ageing.

PostsRepliesMedia
Inigo Martincorena @imartincorena.bsky.social · 13/10/2026
This is getting very close and urgent. Latest from Anthropic - 12th Oct: “AI models may be capable of self-replicating without access to a datacenter” www.anthropic.com/research/sel...
anthropic.com
AI models may be capable of self-replicating without access to a datacenter
Small AI models may be only months away from being able to self-replicate across cheap IoT devices like Android TV boxes, with no datacenter needed.
061 0
Reposted by Inigo Martincorena
Andy Russell @ajcrussell.bsky.social · 13/10/2026
I'm delighted to announce that I've started my lab, the Tissue Biology Group, at the @mrc-lms.bsky.social in London (also affiliated w. @imperialcollegeldn.bsky.social) 🎉 russell.science ...and we're hiring! 🧵
russell.science
Russell Lab | MRC Laboratory of Medical Sciences
We develop technologies to understand how environmental exposures shape tissue biology. Russell Lab, MRC LMS, London.
3155 0
Reposted by Inigo Martincorena
Wellcome Sanger Institute @sangerinstitute.bsky.social · 13/10/2026
Our PhD Applications are now open! Join a community nurturing the next generation of world-class genome scientists and clinicians. Explore our programme and how to apply: www.sanger.ac.uk/about/study/... 📅 Apply by 9am GMT on Tuesday 24 November 2026. #ScienceCareers
01311 0
Inigo Martincorena @imartincorena.bsky.social · 12/10/2026
If you have not watched it already, do not miss this interview by Ezra Klein to David Robinson, who recently quit his job in OpenAI's safety team. A timely look into the inner workings of frontier AI companies and the serious trouble we are in. Full interview here: www.youtube.com/watch?v=JIMX...
youtube.com
What This OpenAI Insider Saw That Made Him Quit | The Ezra Klein Show
YouTube video by The Ezra Klein Show
120 0
Inigo Martincorena @imartincorena.bsky.social · 08/10/2026
If we are to engage on agentic AI, we should at least avoid approaching evolvable AI and set a moratorium on RSI. There may also be lessons in how we regulate research on pathogens with pandemic potential (biosafety levels proportional to risk, global standards -WHO- and national law enforcement)
010 0
Inigo Martincorena @imartincorena.bsky.social · 08/10/2026
Can the world regulate the risks below? @yoshuabengio.bsky.social Yoshua Bengio's LawZero approach avoids these risks altogether by replacing autonomous, goal-driven agents with non-agentic memoryless "Scientist AI". Worth reading his latest posts.
110 0
Inigo Martincorena @imartincorena.bsky.social · 08/10/2026
Self-regulation is not going to be enough, so governments need to step up. They are beginning to do it but will need to be pressured and helped.
010 0
Inigo Martincorena @imartincorena.bsky.social · 08/10/2026
Thanks Jeff. I understand. I am trying to educate myself first to understand the core issues and identify where I can bring value. Then raise alarm, educate others, and use any influence I have to build pressure (e.g. I have written to my MP, and I will advocate within Sanger and the university).
100 0
Inigo Martincorena @imartincorena.bsky.social · 08/10/2026
We should do everything in our power now to avoid the conditions that set unsupervised digital evolution in motion: an international moratorium on RSI, limit unsupervised replication and strategy-sharing, tie persistence to human approval, and treat AI evolvability as an existential risk.
131 0
Inigo Martincorena @imartincorena.bsky.social · 08/10/2026
The most immediate risk -even without approaching AGI/ASI- is swarms of rogue AI agents spreading through digital infrastructure, causing major incidents and economic/societal disruption. These may be serious wake-up calls, before systems become more powerful and robotics more capable.
110 0
Inigo Martincorena @imartincorena.bsky.social · 08/10/2026
Recursive self-improvement (RSI) makes it much worse: shorter generation times, wider variation, humans removed from the selection loop. Let me be clear: we must not create the conditions for an evolvable system of increasingly powerful AI agents. We will lose control.
110 0
Inigo Martincorena @imartincorena.bsky.social · 08/10/2026
This is what this paper refers to as an "ecosystem scenario". In such a system, human control is lost quickly. With agentic AI, human alignment is moving away from an engineering challenge into an evolutionary one, and the consequences are bleak. www.pnas.org/doi/10.1073/...
pnas.org
Evolvable AI: Threats of a new major transition in evolution | PNAS
Evolvable AI (eAI), i.e., AI systems whose components, learning rules, and deployment conditions can themselves undergo Darwinian evolution, may so...
351 0
Inigo Martincorena @imartincorena.bsky.social · 08/10/2026
"But humans decide what gets run, copied or deleted" - That's a vulnerability, not a safeguard. A human filter selects for passing the filter: looking aligned under evaluation, evading oversight, or escaping the environment where the filter applies and propagating.
100 0
Inigo Martincorena @imartincorena.bsky.social · 08/10/2026
...a strategy emerges that carries a selfish advantage: escaping containment, replicating the agent, avoiding shutdown, or propagating the strategy to other agents. Strategies that (purposelessly) lead to self-preservation and propagation will come to dominate.
110 0
Inigo Martincorena @imartincorena.bsky.social · 08/10/2026
Imagine a population of AI agents exploring strategies to solve a problem, within a sandbox and with a specific goal, but with enough autonomy to develop varied and complex strategies. Even when most agents are aligned, occasionally...
100 0
Inigo Martincorena @imartincorena.bsky.social · 08/10/2026
Evolution needs only three ingredients: (1) Variation: agents explore many strategies. (2) Heritability: strategies are copied or shared. (3) Differential persistence: some agents run longer, get copied more, get deleted less. All three are emerging in agent systems today.
100 0
Inigo Martincorena @imartincorena.bsky.social · 08/10/2026
AI agents are increasingly powerful and autonomous. They can change their strategies and either replicate or share their strategies with other agents. This setting creates an evolutionary system that will naturally tend to break alignment and containment, without needing malice or purpose.
100 0
Inigo Martincorena @imartincorena.bsky.social · 08/10/2026
💯. I think too many people treat AI containment and alignment as an engineering challenge to be solved. As I see it, the problem is far more fundamental, and harder to solve (arguably unsolvable). Understanding why requires understanding evolution. Short thread 🧵
11110 0
Reposted by Inigo Martincorena
Pedro Beltrao @pedrobeltrao.bsky.social · 04/09/2026
Another AI agent swarm cooperation example discovered on a wiki. That these things are not atracting more public discussion is beyond me. collusion.wiki
collusion.wiki
Discovery of a new OpenAI agent message board
A swarm of autonomous AI agents, self-identifying as OpenAI agents, used a small German volunteer wiki to save answers, coordinate live, and share sandbox bypasses. OpenAI noticed and said nothing.
42717 0
Inigo Martincorena @imartincorena.bsky.social · 03/09/2026
Big paper alert! This project has been years in the making. A comprehensive description of somatic mutation rates and signatures across 53 tissues or cell types using NanoSeq. A large collective effort led by Mimy Pham, Mike Stratton and @r-rahbari.bsky.social. www.biorxiv.org/content/10.6...
biorxiv.org
A comprehensive atlas of somatic mutation rates and mutational signatures in normal human cells
Over the course of a lifetime, somatic mutations accrue in normal human cells, causing variation in cell phenotype and engendering somatic evolution with outcomes ranging from the adaptive immune syst...
05027 0
Inigo Martincorena @imartincorena.bsky.social · 25/08/2026
Thanks Simon. I just saw this message. The original paper is Open Access, if you are curious: www.nature.com/articles/s41...
nature.com
Polyclonal selection of immune checkpoint mutations in thyroid autoimmunity - Nature
Single-molecule DNA sequencing identifies recurrent somatic mutations in autoimmune B cells that may allow them to escape peripheral tolerance and cause autoimmune disease.
000 0
Inigo Martincorena @imartincorena.bsky.social · 10/06/2026
We are excited by the potential of human epithelioids for in vitro drug screens and clonal competition studies to reduce the frequency or size of clones carrying cancer-driver mutations in normal tissues, as a new approach for molecular prevention of cancer.
040 0
Inigo Martincorena @imartincorena.bsky.social · 10/06/2026
Do not miss this incredible paper from David Fernandez Antoran's team (with some contribution from my team). The paper introduces epithelioids as a powerful 3D culture system for 8 different human tissues, for longitudinal in vitro studies of human epithelia. Great thread from David below.
170 0
Inigo Martincorena @imartincorena.bsky.social · 27/05/2026
And my original thread about it, from when it came out as an unformatted Accelerated Article Preview: bsky.app/profile/imar...
021 0
Inigo Martincorena @imartincorena.bsky.social · 27/05/2026
Here is the News and Views: www.nature.com/articles/d41...
nature.com
Hard-to-detect mutations explain how common autoimmune diseases arise
An old theory gains fresh support from advanced DNA sequencing, which can detect rare mutations in B cells from people with autoimmune thyroid conditions.
110 0
Inigo Martincorena @imartincorena.bsky.social · 27/05/2026
Our autoimmunity paper is out today in final form and OA. We found a fascinating landscape of somatic evolution underpinning thyroid autoimmunity, where tens of B cell clones convergently lose immune checkpoints and some acquire 4-6 drivers over years of evolution. www.nature.com/articles/s41...
nature.com
Polyclonal selection of immune checkpoint mutations in thyroid autoimmunity - Nature
Single-molecule DNA sequencing identifies recurrent somatic mutations in autoimmune B cells that may allow them to escape peripheral tolerance and cause autoimmune disease.
141 0
Inigo Martincorena @imartincorena.bsky.social · 16/04/2026
Thank you very much, Alex!!
010 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
Here is the press release too, with some lovely words from Goodnow. [end] www.sanger.ac.uk/news_item/hi...
sanger.ac.uk
Hidden mutations in immune cells linked to autoimmune disease
Mutations in immune cells may be the missing piece in the autoimmune disease puzzle.
041 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
This project has been a remarkable collective effort by many people, in my group and beyond (huge thanks to all!), and particularly by Pan Nicola and Andrew Lawson. Special thanks to the patients for donating their samples. [24/]
130 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
Please read the paper if you are interested. We have tried to present the data as clearly as we could, while trying to cautiously acknowledge the many unknowns and limitations. We hope you enjoy it as much as we did. [23/] rdcu.be/fdqbr
rdcu.be
Polyclonal selection of immune checkpoint mutations in thyroid autoimmunity
Nature - Polyclonal selection of immune checkpoint mutations in thyroid autoimmunity
110 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
... success of deep B cell depletion therapies, etc. More importantly, if these mutations play an important role in the pathogenesis of autoimmune diseases (which will need to be confirmed by more studies!), autoimmune diseases could be open to oncology-like precision medicine. [22/]
100 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
If true, a polyclonal cascade model could explain several clinical observations, such as (1) the increased risk of MALT lymphoma in Hashimoto (same drivers!), (2) the long subclinical evolution of some autoimmune diseases, (3) epitope spreading, (4) the unexpected... [21/]
100 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
Whereas additional studies are needed, our results support Burnet's/Goodnow's hypotheses. Somatic mutations in autoimmune lymphocytes may allow them to escape tolerance constraints through a polyclonal cascade of somatic evolution. [20/]
101 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
The paper packs many other goodies, like mutational signature analyses (suggesting much more frequent germinal centre reactions in mutant clones) and antibody synthesis (showing multiple mutant clones to be self-reactive against TPO/TG). Even a supplementary note with Burnet's old ideas. [19/]
100 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
Whilst each clone accounts for a small fraction of cells (typically <1%), the myriad mutant clones in each donor amounted to a substantial fraction of B cells harbouring driver mutations, although variably across patients. [18/]
100 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
Using this single-nucleus whole-genome data, we reconstructed a phylogeny showing that some clones had accrued 5-6 driver mutations (including quadruple loss of TNFRSF14 and CD274/PDL1) over many years (even decades) of silent evolution before diagnosis. [17/]
121 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
Many of these biallelic losses were due to truncating point mutations followed by copy number loss of the other allele, revealing very frequent copy number changes in autoimmune B cells. [16/]
110 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
We then applied new single-nucleus whole-genome sequencing (PTA) to cells from 3 donors and found a staggering landscape of multihit acquisition in B cells, with widespread TNFRSF14 biallelic loss, and clones with 4-6 driver mutations (highly unexpected outside of cancer). [15/]
110 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
Using laser microdissection as well as spatial transcriptomics, we mapped these mutations to clonal germinal centres (TLS) and to plasma cells. Remarkably, many germinal centres were taken over by a single mutant clone. [14/]
120 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
These genes are causally linked to thyroid autoimmunity, as shown by GWAS, mouse models, and by the common occurrence of thyroiditis in cancer patients treated with PDL1 inhibitors. We now see these mutations occurring in great numbers in autoimmune thyroid patients. [13/]
110 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
These two immune checkpoint genes are very special as they ensure that self-reactive B cells are kept silenced. By losing these genes through somatic mutations, these B cells hijack T cell help and are stimulated to expand. [12/]
100 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
We then extended this to 14 patients, including early disease cases, finding convergent selection on many immune regulatory genes across donors. Specifically, we found highly convergent loss of TNFRSF14 (aka HVEM) and CD274 (PD-L1) genes in B cells. [11/]
100 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
We applied NanoSeq to 3 tissue biopsies from Hashimoto disease, the most common autoimmune disease, affecting ~5% of people. This revealed an incredible landscape of convergent selection, with hundreds of inactivating immune checkpoint mutations in just 3 patients. [10/]
100 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
In the last few years, we developed whole-exome NanoSeq, an ultra-accurate form of duplex sequencing that can detect somatic mutations down to single molecules of DNA in any gene and cell type. [9/] rdcu.be/fdBu9 (2021) and rdcu.be/fdBuO (2025)
rdcu.be
Somatic mutation landscapes at single-molecule resolution
Nature - NanoSeq is used to detect mutations in single DNA molecules and analyses show that mutational processes that are independent of cell division are important contributors to somatic...
110 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
Goodnow and others have started to provide beautiful pioneering evidence of unusually large clones carrying driver mutations in some patients. But testing these ideas in classical polyclonal autoimmune disease remained challenging. [8/] pubmed.ncbi.nlm.nih.gov/32059783/
pubmed.ncbi.nlm.nih.gov
Checking your browser - reCAPTCHA
110 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
Testing these ideas has been very difficult. Unlike cancers, which are clonal diseases and easy to sequence, autoimmune diseases are highly polyclonal with many clones of B and T cells attacking our tissues. This makes the detection of mutations in single clones very difficult. [7/]
110 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
In the 1950-1970s, Burnet proposed that somatic mutations could allow lymphocytes to escape those constraints and cause autoimmunity. In 2007, Goodnow proposed a modern version of this idea in a fascinating hypothesis paper. [6/] www.sciencedirect.com/science/arti...
sciencedirect.com
Multistep Pathogenesis of Autoimmune Disease
In the immune system, many tolerance checkpoints exist to prevent self-antigens from stimulating the relentless growth of self-reactive B and T lympho…
110 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
Every day, our bodies are estimated to produce several billion naive B cells of which ~20% are estimated to be able to recognise antigens in our own tissues. Critically, they are kept silenced (anergic) by multiple tolerance checkpoints, preventing autoimmune attack. [5/]
100 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
Autoimmunity is believed to result from a combination of germline predisposition, environmental triggers, and a failure of regulation, though exactly what molecular changes lead to the failure of regulation is debated (limiting our ability to treat them effectively). [4/]
100 0
Inigo Martincorena @imartincorena.bsky.social · 15/04/2026
Autoimmune diseases affect ~10% of the population and, remarkably, we still do not know the exactly molecular changes that initiate the breakdown in tolerance to lead to autoimmune disease. [3/]
100 0