Transplanted human brain-tissue takes root in mice without a cortex

0:00Rubrik is the security and AI operations company built for what happens after an attack hits not just the moments before. AI has turned the threat landscape into quicksand, moving too fast for any human to fully predict. That's why an agentic cyber resilience platform matters. Automated recovery, clean data, a business that keeps moving no matter what hits. One platform, not a patchwork of stitched together tools and gaps. Don't wait for the next attack. Secure and accelerate your business at Rubric.com. Again, Rubric.com. While every other channel is fighting for your customers' attention, podcasts are where they've already given it.

0:34No one accidentally listens to a podcast for 45 minutes. They choose to be here. They trust the voice in their ears. And when that voice talks about your brand, it doesn't sound like advertising. It sounds like a recommendation from a friend. Acast gives you that trust at scale. Digital precision, post-read authenticity, and performance data that proves it worked. Don't fight for attention. Buy it with Acast. Learn more by visiting acast.com slash advertise.

1:29Welcome back to the Nature Podcast. This week, transplanting human brain organoids into mice and how to turn a scientific paper into an AI you can talk to. I'm Nick Patrick Howe and I'm Benjamin Thompson.

1:53First up this week, we hear how researchers have transplanted human brain-like tissue into mice and watched it grow and develop, a technique they hope will aid in the future understanding of brain development or disease. Now, one of the ways that scientists have been studying the complexities of the human brain is by using organoids. Derived from human stem cells, these small 3D clumps of brain-like tissue are grown in a dish, either individually or combined into more complex structures called assembloids. But while these systems have transformed the way that researchers study developmental processes or disease progression, their in vitro nature means they may not be able to mimic some of the specifics of brain function, meaning they may not be applicable for use in every situation.

2:47Even in these preparations that can be kept for like years in a dish, we realize that there are a number of aspects that kind of prevent their full potential. This is Sergio Paschka from Stanford University in the US. One of them is quite interestingly, the fact that cells can like sit in a dish, but there are like no physiological inputs and there are no outputs. And we know that those are really important for the brain to truly develop. Back in 2022, Sergio and his colleagues looked to get around this by implanting a human brain organoid into the brain of a newborn rat. He spoke to us on the podcast about it at the time.

3:22This method is something he and others have been using in their research, but it too has some limitations. Namely, that human brain cells grow a lot slower than rodent brain cells, and essentially the implanted organoid gets out-competed, losing the chance to make connections and the space it needs to grow. Now, Sergio and his colleagues have looked to overcome this issue with a different approach, and they have a paper out about it in Nature this week. And so the idea was, like, could we actually create a strategy that would allow us to create some of, like, this space? And could we actually do it in a clean, precise way?

3:58So we thought that the best way would be not to do it surgically, but to actually do it genetically. And this is what they've done, working in mice this time, using a genetic approach to prevent part of the animal brains from growing. So we removed parts of the nervous system that is called a pallium. So this is a structure that is present early in development and will give rise to the cerebral cortex, the outer layer of the brain, the hippocampus, which is very important for memory. And so what we essentially did here is use the strategy that allows us to delete a gene that is very important for cells to divide, but only in the cells that are going to form these brain regions in development.

4:42So in other words, the moment the cells are actually born and they're ready to make a cortex, they can divide. So they're just stuck there in very few early cells. Despite losing almost half of their brain volume through the loss of their palliums, these apallial mice survive. And although they show reduced body weight compared with control mice, they were broadly similar in terms of basic mobility and sensory responses. The team then transplanted human stem cell-derived cortical organoids into newborn apaleal pups that were between 5 and 17 days old and watched the cells develop. We essentially place about four cortical organoids in that vacant space.

5:25Each of them has probably, you know, 100 ,000 cells or so, so they're not a lot of cells. They already have been differentiated so that they're going to make cortical cells, So they're on that path to form cortex. And essentially, we just place them in, and that's it, through one single injection. So it turns out that the human cells, they will divide, and they will start differentiating. They're going to become vascularized by the host. Immune cells from the mouse are actually going to move inside the human graft. Over the course of a few months, the organoid cells divided to take up empty space.

5:59So like at the end, you know, we estimated that probably one of these animals, which we called xenocortical mice, because they have a cortex of a foreign origin, the mouse will have 70 million neurons total. And about 14 million are cortical and hippocampal cells. So those are largely removed. And they're replaced by, you know, about 4 million human neurons. Now, of course, human cells are larger, as you can imagine. And of course, they're not densely packed. So the replacement is not with the same number of cells, but it's still a large number of human neurons. As well as multiplying, the team showed that these neurons were active and formed connections with the mouse brain, or sometimes the spinal cord.

6:43But what did this mean for the mice? To find out, the team conducted a series of behavioral experiments comparing control mice, a paleal mice that lacked almost the entirety of their cortex, and the xenocortical mice that had had organoid implantation. In one set of experiments, the team filmed the different mice, then let machine learning hunt for subtle signals of differences in behaviour. These differences were seen in the apaleal mice compared to the controls, but what about the xenocortical mice? Would they be like the wild-type controls or the apaleal mice? You know, they're neither. They're still like in a third behavioural state.

7:22And that got corroborated also by experiments done afterwards, where let's say if the apaleal animals had deficits in a working memory task, the xenocortical animals did not. It seems as if functions was preserved in the animals that received human cortical cells, unlike the apaleal cells. So it's probably contributing to some of it. Of course, it's not a perfect restoration, but it tells us that you can actually integrate them functionally and have behavioral consequences for that. These results suggest then that the implanted organoids were ultimately doing something in the mice, which was further illustrated when the team used their system to model a type of brain injury.

8:03Hypoxia, when the body is starved of oxygen, is serious and can lead to lifelong medical conditions in humans. Mice, on the other hand, aren't affected by low oxygen levels nearly as much. When the three types of mice here were exposed to low oxygen conditions for a few hours, the team saw a difference between them. It turned out it was enough to induce a hypoxia response in human cells, but the mouse brain was relatively resilient. And then if you come just two days later, if you put them in a motor test, it turns out that wild-type mice and apalial mice have nothing. But xenocortical mice that carry human cells now start having deficits in the way they walk and the way they coordinate gait.

8:46So now this tells us that it's likely the human cells that are causing this. And I think this is going to be a very powerful system for asking questions, not just about how this takes place, but actually for testing therapeutics, which is an equally difficult challenge. So this provides a platform for actually testing whether any of this will have any benefits before you would actually go into a clinical trial. Aparna Baduri from the University of California in the US studies human brain development and cancer. She says that one of the positives she sees in this work is that one of the cell types that the human organoids differentiated into looked similar to a very rare type of neuron that researchers have struggled to grow in vitro The loss of these neurons is a key factor in a particular form of dementia so better understanding them could help in the development of future treatments But while this work may represent a step towards where the field would like these organoids to be in terms of making them useful in things like understanding brain biology, Aparna says there's a way to go yet.

9:49You know, I think that this is a really expansive effort in terms of the series of things that they have put together. But I think that there are also things about the cortex that are interesting and important, and this is actually noted in the paper, such as some of the layer structures, such as some of the different areas of the cortex that are involved in processing. They're not well preserved in this system yet in a way that would resemble more of the features of the human brain. And so I think that those would be some of the things that are not yet fully developed in terms of where we would want the system to go.

10:27And then the other thing I would say is that even though there is the emergence of some of these important and interesting cell types, the authors clarify that the developmental stage at which the organoid gets to is still during some time in the middle of development this is not making a fully mature cortex and so that has been the persistent challenge with these organoids and it seems that there's a little bit of advancement here but you're not going all the way to making a developing brain but there will be questions as to what these animals are. Are they a human-mouse hybrid? Because of the state of the organoid and its development, Aparna doesn't think so.

11:08I want to say very clearly this is not a human brain in a mouse. I think the reasons for that are a couple. One is that it's not mature. It doesn't have the complexity of cell types. It's missing some cell types. It doesn't have the maturation of those cell types. And then also, while there are some connections with the rodent, it's not as connected or oriented in the same way that a human brain would be. So both because of the complexity, the maturation and the connectivity, I would say that this is not a human brain in a mouse, but rather is a organoid that is being pushed to progress in its development.

11:49Other researchers that Nature has spoken to echo this sentiment, saying that these are not fully formed functional human cortexes. Aparna says she commends the researchers behind the work for their engagement with ethicists throughout this work, something echoed by Emily Jackson from the London School of Economics, part of whose work focuses on the regulation of stem cell related research, which includes brain organoids. So one thing that's really, really striking about this paper, but also about the work of Sergio Pascua's lab in general and other work that he's done is how focused this group is on the importance of ethics and the importance of regulation.

12:29So I think sometimes people wrongly think that regulation is a hindrance to scientists and I think that's a real mistake. In my experience scientists working at the cutting edge of biotechnology as this clearly is are very keen on the involvement of ethicists and very keen on in a sense knowing where the red lines are so that they get on with their work knowing that society is happy with what they're doing. So this group in particular, I think, is very focused on the need to take ethics very seriously. The work published today is an early step in this strand of organoid research. Ultimately, it could be used to learn more about how the brain functions in different situations.

13:08This will require further questions to be asked. The key thing here, if you're replacing part of a rodent's brain with human brain cells, is what sort of harms might that cause to the animal. And in particular, might that cause a new form of harm that you're not so readily able to detect? And we need to take that very, very seriously because although this research potentially could have tremendous benefits, it's really important to take seriously the well-being of animals that are involved in this research. Emily also stresses the need to undertake quality engagement with the public on future research in this area to communicate the potential benefits but also understand any concerns, which may be on topics different from what researchers assume them to be.

13:53These are all things that need to be considered moving forward. Both Sergio and Aparna are of a mind that this type of organoid implementation is unlikely to replace other in vitro or in vivo brain organoid research, but rather they will each have specific uses. Sergio, for one, is excited to see what his new platform will help researchers uncover about brain developments and how it might help in the treatment of neuropsychiatric conditions. I think it would be amazing to see it really put to work towards understanding how these devastating disorders are really arising. I think that's my goal.

14:29That has been my goal from the very beginning and it will continue to be until we will find therapeutic solutions for these conditions. That was Sergio Pasca there. You also heard from Aparna Badouri and Emily Jackson. To read Sergio's paper, look out for a link in the show notes. Coming up, a static research article may become a thing of the past, thanks to a new AI workflow. Right now though, it's time for the Research Highlights with Dan Fox.

15:00Building on wetlands is hard work. But 5 ,000 years ago, people in what is now Eastern China were finding sophisticated techniques to build on marshes. Researchers studied layers of earth from the ancient Liangzhu city settlement. Their analysis suggests that builders created raised mounds by piling up bundles of wet clay wrapped in grasses. Hundreds of these bundles could be assembled into blocks, making them easier to move and helping them hold together on soggy ground. The surfaces of these mounds were then finished with mixtures of clay and sand to make them more durable. The authors say these findings suggest that the building of Liangzhu City was an engineering feat that relied on know-how preserved over generations as well as large-scale coordination of labour.

15:51If that sounds like the foundation of some interesting research, you can read the paper in the Journal of Archaeological Science.

16:01A powerful volcanic eruption in the South Pacific Ocean has revealed a way to detect a tsunami unleashed by such an event. Hidden in the depths of the ocean, underwater volcanoes are hard to monitor. And when one such volcano in Tonga erupted in January 2022, it triggered tsunamis that struck the coasts of nearby islands with little warning, producing waves more than 40 metres high. To understand how the 11-hour eruption played out, researchers studied data captured by seismic sensors hundreds to thousands of kilometres away. They spotted T-waves, high-frequency, slow-moving seismic signals created as the eruption's energy radiated out into the ocean.

16:47These T-waves were strongest about an hour after the eruption began when the top of the underwater volcano collapsed. This collapse also generated the largest and most destructive tsunami of the eruption. Because T-waves travel at least seven times faster than tsunamis do, they could serve as early warnings of tsunamis generated by underwater eruptions. You can read that research in full in Science Advances.

17:22The humble scientific paper, where research heads out to meet the world in a formulaic document that scientists can pore over, analyse and assess for rigour. A paper is primarily the way in which science is communicated to the rest of the world. For some, though, papers are relics of the past. One can trace their origins to periodicals collating various insights from people of means who had an interest in this newfangled science thing. An author wrote something down, it was distributed, and readers consumed it. Research articles have undergone some evolution in the past few hundred years, but they still basically exist as a static document.

18:07And in a world where we're increasingly able to have conversations with artificial entities, do scientific papers have to be so passive? Well, not everybody thinks so. Our idea is to try to convert and transform all of these static artifacts of knowledge from passive papers into some sort of dynamic and living knowledge in the form of knowledge agents. As you can think of, for example, each knowledge agent of a paper is almost like the virtual corresponding author of that paper. This is James So, a researcher who develops scientific AI agents. These are AIs that are a bit more self-sufficient than your run-of-the-mill chatbot.

18:53They can do tasks on their own and even bring other tools into the mix. these are a relatively new invention leapfrogging off other advances like large language models and for james they're an opportunity a way to turn that static old paper into something more dynamic something you can talk to and this week he and a team of researchers are publishing in nature in a regular old paper exactly how you would do that they created a workflow called paper to agent. This essentially agentifies the original research article, turning it into an AI agent in its own right. How this is actually done is fairly complicated, but basically it involves using teams of AI agents that will pour over the paper, work out how it was put together, and recreate the original article's results.

19:49That knowledge is then captured in a server, which essentially acts as a translator between the human written paper and any AIs that you want to use. That allows things like large language models that power chatbots like ChatGPT and Claude to then step in. Armed with the knowledge captured on that server, they can act as a virtual co-author, allowing you to ask them anything you like about the paper, or even run your own analyses or apply what the original authors did to your own data. And you may be thinking that that doesn't sound terribly new. After all, you can just put a paper into a chatbot and ask it questions about it.

20:29But James says that this new workflow goes a step beyond. So if I put just like a paper into a chat GP or a dual-native language model, then that language model is really just reading and interpreting that paper and be summarizing the results of that paper. But it doesn't really know how to generate the results or generate the figures or do the experiments or use the data and the code because it doesn't have that hands-on experience. The paper agent that we create here through our framework then actually learns and have that hands-on experience. So then it's actually then able to provide much more in-depth insights to the readers.

21:06And for example, the reader might say to this agent, can you help me to actually apply the code from this publication to my own data sets? the agent, our paper agent, would be able to do that automatically and generate reproducible workflows and results. James also thinks that paper to agent, with its focus on reproducing the original results, could help identify where things have gone wrong in producing a paper. If the agent is unable to reproduce the results, say because there's an error in the interpretation of the data, it could flag to the original authors that there's a mismatch between what the data says and what the results show.

21:43This process also helps reduce hallucinations, according to James. These are phenomena where the AI essentially just makes something up. But by going through a virtual version of what the original authors did, James says that the agent is much more grounded, its conclusions being validated against the paper's results. James and the team also tried out paper to agent in the real world, and they found that these agentified papers could work together and uncover new insights. We have one case study, it's pretty interesting, where we actually created two paper agents, right? One agent's actually looking at alpha genome, which is recently published by Google DeepMind as a tool for interpreting a place of different mutations.

22:30So with paper to agent, we automatically converted of a genome into an agent that knows how to use that tool. There's a separate research study that published a large genetic study data set about genetic risks for ADHD. So that was also automatically converted into a paper agent that knows how to use that data set. So in this case, those two agents actually started talking to each other, and they realized that they have some common interests, and they can actually collaborate. And in this case, they actually discovered a new splicing mutation associated with increases in risks for ADHD, which is not known before.

23:04So I think this is kind of an interesting example of kind of agent-to-agent collaboration that's possible now because knowledge is in some sense living rather than being passive artefacts. But if papers are no longer passive artefacts, what does that mean for conventional research articles? I'm Magdalena Skipper. I'm Editor-in-Chief of Nature. Magdalena is someone who has thought a lot about how research articles are presented. She's been an editor for 25 years and editor-in-chief at Nature for eight years. For transparency's sake, she's also my boss's boss. Anyway, in those 25 years, there's been a lot of discussion about what the future of scientific papers might be.

23:49Magdalena doesn't think that the conventional paper is going away anytime soon. Storytelling for her is central to papers and how humans share knowledge. But she is optimistic about what this new way of agentifying papers might mean and how it could change the future of sharing knowledge. There is a prospect that that future will look different, that there will be something genuinely influential, that will change the way knowledge is disseminated. but importantly, the way that knowledge is interacted with, engaged with. I think that's a real potential benefit of this paper. She also thinks that the process of reproducing the work of the original paper may be a good way to check the robustness of the results and could help with making the papers more accessible.

24:40A paper you can talk to may be easier for more scientists to interact with than a static document. and she does think that there may be a role that journals play here. There is an interesting open question to my mind about this paper. Central to the paper is the idea of creating this AI connector, if you like, that plugs in all the different elements of the paper, the data, the code, the information that is described in the paper, And that then creates an interface with an LLM so that a user can use language to query the paper and interact with it. What is not clear to me from the paper, whose responsibility it is to build this tool.

25:28So is it imagined that in the future it's the authors themselves, the researchers who have done that piece of research, they need to create that? That would be quite an onerous task, and I would imagine most researchers, depending on the area of expertise, would not be equipped to build it. If it's adopted widely, then maybe we will see journals evolve. Maybe the publishers will offer this kind of a service, overlaying on top of the papers that they publish. James's ambitions are to use this technology to make research more available and robust. Right now, these papers, it's often required to have a data availability statement or sometimes a code availability statement.

26:12But in the future, I think it would be great if the paper also has an agent availability statement that says here's the agent associated with this paper. The agent then is the virtual author of this paper who's then responsible for making sure that this paper is reproducible, but can also help to answer any questions about this paper with readers and collaborators. So that would actually greatly reduce the barrier to knowledge dissemination and also make knowledge much more robust. Now, paper tube agent doesn't work for every bit of research. James and the team tried out converting 100 papers in computational biology into agents with their workflow.

26:51But for 26 of them, it didn't work. James says this is down to missing information in the paper, or some inconsistency in the code which was used in the article. Because of that, he thinks that even when Paper2Agent fails, it's informative, as it could tell you that there's something in the paper that's stopping it from being reproduced. And Paper2Agent is now out there, ready for researchers to try out if they've got the necessary computing chops. James says it costs around a dollar or two to convert a paper into an agent. The price of the tokens you need to run the various AI processes required.

27:30So we'll likely find out soon if this changes how knowledge is disseminated. For James, this is just the beginning. I think this is the opportunity for us to reimagine what knowledge can look like in the future. It does not have to be static artifacts. It could actually become this dynamic knowledge agents, in some sense like a living knowledge. And this living knowledge, I think, has many benefits that we discussed. One benefit would be it can make sure that knowledge is actually reproducible, it's robust, can be updated continuously over time. A second big benefit that this enables a new form of collaboration where the agents of different knowledge can directly collaborate and generate new discoveries, new findings.

28:11If you think about it, in the past, if two human teams want to collaborate, they have to read each other's papers and somehow send emails back and forth to set up a meeting. and most of the time people may not have time to actually meet and may not have time to collaborate. But now with these paper agents, those two paper agents can directly talk to each other and generate some new insights without being bottlenecked by our own limited calendars and scheduling constraints. That was James So from Stanford University in the US. You also heard from Nature's Magdalena Skipper. For more on that story, check out the show notes for some links.

28:47And that's all for this time If you'd like to stay in touch with us, you can We're on email, podcast at nature.com Or you can find us on social media, at naturepodcast I'm Benjamin Thompson And I'm Nick Petra-Chow, thanks for listening

29:18Here's a tip for you. There's a podcast out there with fans waiting to be your next customer. They tune in every week. They trust the host. And that host wants to talk about brands like yours in their own words to their audience. The problem is you just haven't been introduced yet. We're Acast, where that introduction happens. As the world's largest podcast marketplace, we let you browse shows, see who's listening, and book host-read sponsorships or run your own ads all from one platform. Transparent pricing, real-time data, complete control. Start advertising on podcasts by visiting acast.com slash advertise.