Shamini Bundell
Welcome back to the Nature Podcast, this week: supermassive black holes in the early Universe...
Nick Petrić Howe
...and putting an AI for heart health to the test. I'm Nick Petrić Howe.
Shamini Bundell
And I'm Shamini Bundell.
<Music>
Shamini Bundell
First up on the show this week, reporter Benjamin Thompson has been learning about a pair of supermassive black holes slowly approaching one another in the distant Universe.
Benjamin Thompson
Although they might differ in shape or size, galaxies across the Universe all seem to share something in common. At the centre of each of them lurks something huge, a supermassive black hole. Their name rather gives a hint as to what they're all about, as Yue Shen from the University of Illinois at Urbana-Champaign, explains.
Yue Shen
In terms of their mass, supermassive black holes are typically defined as between a million times the mass of the Sun to several billion times the mass of the Sun, and some of them might be even more massive. But also they are very, very compact compared to galaxies. So when you're talking about a 10 billion solar mass black hole, the size — if we talk about the event horizon of that black hole — it is not much larger than, say the size of the Solar System. So you're confining all that mass within a tiny region in space, that's what makes them so extreme.
Benjamin Thompson
There's still a lot that isn't known about supermassive black holes. One mystery the researchers have puzzled over is why there aren't more galaxies seen with more than one. But why would a galaxy have two? Well, galaxies have come together, have merged, throughout the Universe's lifetime, which should lead to situations where one galaxy has two of these black holes that are in the act of orbiting closer before ultimately merging themselves.
Yue Shen
It is kind of strange, right? We know that those galaxies have undergone a lot of mergers. And if we believe that there is a supermassive black hole at the centre of each and every one of those galaxies, then it should be quite common for those massive black holes to pair within merged galaxies, or merging galaxies, but we don't have a tonne of confirmed dual supermassive black holes.
Benjamin Thompson
Those that have been confirmed have been seen relatively close to Earth, meaning the galaxy merger happened in the relatively recent past, in astronomical terms. That's because due to the speed of light, things we observe close to Earth are more recent, and looking further away shows us more ancient events. And until now, there's been no concrete evidence of dual supermassive black holes being seen in the process of merging further away. In other words, we haven't seen any from a very long time ago, closer to the Big Bang. But that might be about to change. This week, Yue and his colleagues report evidence in Nature of a supermassive black hole pair in a galaxy merger that happened early in the Universe's evolution. It's a time period that researchers are keen to explore because it will help in understanding how, and why, the universe is the way it is now.
Yue Shen
So, the particular system we discovered was at a time when the Universe was only 3 billion years old. Compared to the 14 billion years age of its current age.
Benjamin Thompson
This system of two black holes has got the snappy name SDSS J0749 + 2255. And it's found within two merging galaxies tens of billions of light years away from Earth. However, despite their heft, supermassive black holes are actually pretty tough to detect directly. In this case, the team saw them thanks to the massive jets of electromagnetic radiation known as quasars, that active supermassive black hole spew out as they feed on gas within galaxies.
Yue Shen
And, for the quasars we're interested in, this radiation can be more luminous than all the stars combined in the galaxy. And it's not just in radio wavelengths, it's all wavelengths from gamma rays to X rays to ultraviolet to visible light to infrared and to radio. So, it is the entire electromagnetic spectrum you will see this bright quasar emission.
Benjamin Thompson
So, all in all, pretty hard to miss. And yet finding pairs of quasars, and thus, black holes far out in the universe, and far into the past has been hard. A few years ago, Yue was part of a team that identified SDSS J0749 + 2255 using data from the Gaia space observatory, and the Hubble Telescope. It appeared to be two separate quasars, but looking at the system led to a lot of questions.
Yue Shen
Are they a pair of quasars at the same distance? Or, maybe, there are one quasar in the background and a star in the foreground? Or maybe the are a pair of quasar images that is created by a gravitational lens. So if there is a foreground galaxy serve as a lens, it can create a multiple images of the same background source. So in order to answer this question, and to pin down the nature of this system, that is when we need this whole slew of observations to really convince us that, hey, this is really a pair of physical quasars at the same distance, and they are within a merging pair of galaxies.
Benjamin Thompson
And a slew of observations is what they did for their new paper. By using multiple telescopes and observatories, some on Earth and some in space, the team probed the quasars at different wavelengths to learn more about the system, showing that it is two distinct supermassive black holes that are early in the process of coming together.
Yue Shen
We know that they are actually not that different, they are actually comparable in their mass — hundreds of millions times the mass of the sun — and the host of galaxies are also comparable in mass. So they don't differ much, they are like twins, living in Twin Galaxies. And right now they are separated by about 10,000 light years, it still has a long journey before them, we have some estimates that it's going to take probably more than a billion years for them to eventually enter into that final stage of coalescence. And right now, they're just kind of orbiting inside of this merging galaxy, and slowly decay in their orbital separation. So, still a long way to go, but now we know it's going to happen.
Benjamin Thompson
Scientists have been looking for distant supermassive black hole pairs for a long time. So, the results of this paper are likely to be of great interest to the astronomy community. Other potential pairs have been put forward previously but haven't been confirmed experimentally. Yue is confident though, that SDSS J0749 + 2255 is indeed a pair of black holes.
Yue Shen
So to be honest, I think there's nothing we can say is 100% correct. So, I think we have a good confidence and welcome independent analysis of this data set. But if it turns out to be not a pair of supermassive black holes, then it will be rather strange, because the only other competing scenario is that this is a gravitational lens. But then it creates a very strange system where we don't see the lens galaxy. So either way, I think this dataset provides valuable information. Even if this turns out to be a lensed quasar. Then I think the ball is on the court of theorists; how you can make your model work to explain this very peculiar planning system? But I'm hoping that we are correct.
Benjamin Thompson
Cristiana Spingola also researches supermassive black holes and has written a News and Views article about the new work. She was impressed by the research and thinks that the team's combining of techniques shows how other such distant pairs could be seen in the future.
Cristiana Spingola
So, I was really amazed to read about this discovery because it has been really decades that scientists are searching to confirm such objects. And it's such an extremely difficult work, that it really changes the game here. It sets a method to find and confirm such systems. But also, the outstanding part is that from the theoretical point of view, this system is exactly what we hoped to observe for many, many years.
Benjamin Thompson
If the discovery stands up to scrutiny, this will be the furthest from Earth a pair of supermassive black holes has yet been detected. It's assumed that there are many more out there waiting to be found, but only time will tell if there are and how similar they are to the one that Yue and his colleagues have identified. Regardless, for Cristiana, this discovery, could help reveal a lot about the Universe in the past, and today.
Cristiana Spingola
Finding a pair of supermassive black holes in such a young age of the Universe is really important to test our current understanding of how galaxies form. And so we expect to observe such systems and we did observe one and that's really already a proof that our understanding of galaxy formation and evolution works. And it also opens different ways, let's say to test also general relativity, or how the merger process happens, or how galaxies really look like in such distant Universe because they are not similar to what we observe in the local Universe. And so that's important for us to understand that if our knowledge, or our theory so far is actually correct.
Shamini Bundell
That was Cristiana Spingola from the Institute for Radio Astronomy in Bologna, Italy. You also heard from Yue Shen from the University of Illinois at Urbana-Champaign in the US. You'll find links to Yue's paper, and Cristiana's News and Views article, in the show notes.
Nick Petrić Howe
Coming up, how researchers are testing whether an AI is ready for use in the real world. Right now though, it's time for the Research Highlights, with Noah Baker.
<Music>
Noah Baker
Can't keep off lost weight? Well, blame your hunger cells. A team based in Germany have identified a new neuronal mechanism which stimulates appetite right after a diet. A brain area called the hypothalamus is known to contain cells called AgRP neurons that play an essential part in promoting hunger after fasting. But it hasn't been clear whether these hunger neurons drive prolonged weight gain. Now, a team of researchers have examined the brains of mice that had fasted for 16 hours. They found an increased number of connections between another set of hypothalamus neurons and the AgRP neurons, which heightened their activity. What's more, the connectivity boost persisted for several days after the animals fast ended, and during this period, the mice consumed more than they had before the diet and gained extra weight as a result. Silencing the neurons in this circuit, however, prevented the post-diet binging. According to the researchers, this mechanism might serve as a potential therapeutic target to help maintain lost weight. You can find that paper in Cell Metabolism.
<Music>
Noah Baker
Ever wondered how gliding mammals developed their flaps for flight? The ability to glide or fly has independently evolved seven times in different groups of mammals. In each case, a flap of skin called the patagium develops between the forelimbs and the hindlimbs acting as an aerofoil. Now, a team based in the US has investigated how this skin flap develops in two mammals: the marsupial sugar glider and the Seba's short tail bat. These two species aren't closely related, their last known common ancestor lived 160 million years ago, but the author's identified a network of genes in both animals that drives the skin thickening required to kick off the formation of flight membranes in developing young. Within this network is one key gene, called Wnt5a, which also causes skin thickening in the ears of developing mice. The results suggest that the genetic toolkit for making flight membranes predates mammalian flight and was redeployed from other skin formation processes. You can find that study in Science Advances.
<Music>
Nick Petrić Howe
Next up on the show, an AI that measures heart function and how researchers are testing whether it's actually ready for use in hospitals. Whilst ChatGPT has recently launched artificial intelligences into headlines, AIs are actually already all around us. For example, a couple of years ago on the podcast, we talked about an AI that can measure something called the left ventricular ejection fraction, an important metric of heart health. Normally, this measurement is done initially by a sonographer who looks at an ultrasound of the heart as it's beating, and takes tracings by drawing around the ventricle to show the difference between its size when full and when nearly empty. These tracings are then checked by cardiologists, and from that you can find that ejection fraction, which gives clues as for the health of the heart. This whole process is all quite labour-intensive, though. So, the idea behind the AI was to automate things. It seemed to work pretty well, but the researchers weren't satisfied with just developing it.
David Ouyang
I think there's increasingly more interested in AI and healthcare. And part of our motivation is to increase the rigour of how clinicians and scientists evaluate AI in this space.
Nick Petrić Howe
This is David Ouyang, one of the developers of the AI. To rigorously evaluate their tool, he and his team have been pitting the AI against human experts in a large clinical trial in the hopes of determining if the AI is really ready to be used by clinicians. To find out more, I called him up and started by asking how they've been putting the AI through its paces.
David Ouyang
The standard American workflow is that a technician or sonographer scans the patient and comes up with a preliminary assessment that's ultimately reviewed and signed off by a cardiologist. We wanted to really test it head-to-head between the sonographer and the AI to really assess, you know, how well does it integrate into the clinical workflow and what are the opportunities as well as the risks.
David Ouyang
And the way that you sort of put it through its paces is you did something called a blinded, randomized, non-inferiority clinical trial. Now, there is a lot of adjectives there. So I was wondering if you could just sort of unpack what that is, for me.
David Ouyang
These are the pillars of clinical trial design. Blinding refers to the fact that the cardiologist was not told whether it was AI or sonographer that gave the interpretation and they actually had to assess which they guess it would be. This is important to minimise the bias that the cardiologists have, potentially if he or she either likes AI or dislikes AI might be harsher or more lenient on the assessment. Randomization speaks to essentially applying one -to-one to make sure that the set that the AI and the sonographer evaluates is similar. And non-inferiority means that we initially sought out to just say that it is almost as good or just as good as a sonographer, because it's something that can save time, but wanted to also assess whether it can be even better than sonographers.
Nick Petrić Howe
And so, when you conducted this clinical trial, how did the AI perform?
David Ouyang
There's a couple of key outcomes we're really happy to present in this study. First, we were able to show that cardiologists can't tell the difference between AI and sonographer. This is a sign that AI is already quite good because it's already to the point where cardiologist can't distinguish between the two. But our primary result was how often the cardiologist had to adjust the tracings done by either AI or sonographer, and we found that the AI tracings were changed 16% of the time, while the sonographer tracings were changed 27% of time.
Nick Petrić Howe
So, the AI was able to save them time?
David Ouyang
Yes, in addition to saving time, the AI was actually able to allow for more precise measurements.
Nick Petrić Howe
And were you surprised at all at how the AI performed or how well it performed?
David Ouyang
We were very pleasantly surprised. The trial was designed as a non-inferiority trial, because we really wanted to hit the bar that this was good enough, or this is similar to sonographers, and we were surprised to show that it actually performed better than sonographers in many aspects. And I think that this really speaks to the promise of the technology.
Nick Petrić Howe
And so when we last spoke about this on the podcast, one future challenge was it was not quite clear whether this would work in any hospital with any sort of equipment. Where are we now with this sort of challenge?
David Ouyang
Because this was such a big effort with so many sonographers and cardiologists; this is a single centre study, but it was an external validation study, meaning that the AI was trained with images at Stanford, and then the trial was run entirely with images, and cardiologists and sonographers at Cedars Sinai, but in other studies, we have shown that this AI seems to be working quite well, in many different datasets in many different populations. We're increasingly both seeing prospective evidence such as today, as well as retrospective evidence across many different centres, that this AI feels ready for primetime.
Nick Petrić Howe
So, one thing that comes up often when talking about AI, especially in the context of healthcare, are there are discussions of bias and things like that. Do you think there is a risk that this could perpetuate existing biases in the healthcare system?
David Ouyang
I think that having more technology allows for greater access to care. Because this is something that is already being done frequently, but maybe there's not enough trained cardiologists and sonographers to interpret these images, this allows for more patients to get the type of care that is optimal and faster. And second, that we were very careful in our analysis to show that the AI results were consistent across racial, demographic and age groups. And I think that this is something that is really important in the validation of AI. And we were fortunate to see that generally, this AI works well across all the subgroups that we tested.
Nick Petrić Howe
So, what would you say then the implications of this paper?
David Ouyang
I think there's a couple of implications I want to point out. First, I think that this shows that AI is definitely safe. It's something that we shouldn't worry about deploying in the echocardiography space. Second, when we use AI, we show that it actually saves both sonographers and cardiologists time, so potentially can optimize the workflow, prevent burnout and improve the quality of care. And third, I think that in our supplementary analysis, we now have a better sense of how many training examples it takes to get really good AI or AI to the level of clinicians. There are still some AIs are trained out there on the order of thousands of studies, our AI was trained on the order of 144,000 — so, close to 100x larger — but I would say that we're pretty confident that AIs trained with more than 100,000 studies will probably be near human level.
Nick Petrić Howe
That was David Ouyang from the Smidt Heart Institute at Cedars-Sinai Medical Center in the US. For more on that story, check out the show notes for a link to the paper.
Shamini Bundell
Finally on the show, it's time for the Briefing Chat, where we discuss a couple of articles that have been highlighted in the Nature Briefing. Nick, what have you been reading this week?
Nick Petrić Howe
Well, it's not so much reading this week, I've been hearing something so I'm actually going to send you a sound that I want you to listen to. And I'll play this for the podcast listeners as well.
<Popping sounds>
Shamini Bundell
Some sort of drumming happening, possibly.
Nick Petrić Howe
So, first question I have for you is what do you reckon this sound is?
Shamini Bundell
Um, like a really slow tap dancer?
Nick Petrić Howe
A really slow tap dancer, it's not a bad guess. My first thought was it sounds a bit like popcorn popping. But it's actually the sound of a stressed or very dry plant.
Shamini Bundell
How is the... how is the plant tap dancing? Sadly, and stressedly? I'm struggling to... unless it's like sadly tap dancing, I'm struggling to understand how a plant is making these weird noises at me.
Nick Petrić Howe
So, we don't actually know how the plants are making this sound. So, this is from a paper in Cell by Lilach Hadany and colleagues. And I was reading about it in a news article in Nature. And so plants get dry, they get stressed and, apparently — and we've never observed this before — they make sounds when they do so. And I must say as well. These have been edited so we can actually hear them, you can't normally hear them.
Shamini Bundell
I was gonna say is this one of those ones where it's like a really high frequency? Or is it just really quiet?
Nick Petrić Howe
It's at a very high frequency. So, it's around 20 to 100 kilohertz, which is just beyond the edge of human hearing. So, humans may be able to hear some of the sounds, but most of them are beyond our sort of range of hearing. But animals could some animals may actually be able to hear the sounds. But to return to your question of how exactly it works. They don't know, but they have a theory. So you might remember that plants have like phloem and xylem, these are tubes for transporting various things that they need, like water and nutrients. And so what they reckon is going on is that in the xylem, that transports water and nutrients from the roots, they reckon that when it's very dry, or the plant has been cut or something like that, then there are air bubbles forming. There's not enough water sort of air bubbles forming. And they're making these sort of pop sounds. And that's the sort of origin of the sounds, but we don't actually know.
Shamini Bundell
So the researchers are sort of, you know, doing a mental health questionnaire with the plants, are you feeling very stressed today, and then just sort of sticking a tape recorder in front of them recording their responses?
Nick Petrić Howe
No, not quite. Unfortunately, plants do not yet have the ability to fill in surveys, as far as we know. They do make these sounds apparently when they are cut or when they are water stressed. So, the way that they did this study is they positioned microphones around the plants in order to capture sounds from them. And they compared healthy plants that are well watered against ones that are in drought stress, so they don't have enough water, or ones that have recently been cut by something as well. And so when they were looking at this, the plants that were in this drought stress or have recently been cut, they're making around 35 of these sounds every hour, whereas the ones that were well watered — and you know sort of happy plants for lack of a better way of putting it — make about one sound per hour. So that makes it a lot more sounds when they're in some sort of stress.
Shamini Bundell
So, it's pretty interesting to think about plants making sounds at all. But you know, if a tree falls in the forest, and I'm standing there, I would definitely hear that. I suppose the key question is whether this sound is just a sort of byproduct or somehow deliberate? Like is anyone claiming that it's communication?
Nick Petrić Howe
Again, it's another thing we don't know. We do know — and this lab have actually published about this before — that plants will react to sounds. So in the past, they looked at a particular kind of primrose, and showed that when they played the sound of a flying bee to the primrose, it started releasing nectar. So, potentially there could be some sort of interplay going on and as I said—
Shamini Bundell
—Wow—
Nick Petrić Howe
—some animals like bats and things could very well hear this sound, but as to whether there's some actual communication going on, we don't really know. So, one person who was interviewed for this story who's a biologist, they said that they think it's unlikely that the animals are really going to be able to hear these sounds like, you know, you'd have to be very close to plant the plants and making these sounds very quietly. So, it's unlikely that they'll be able to hear them in their opinion, but we just don't know.
Shamini Bundell
And I suppose you would need to have a sort of beneficial outcome, if we're saying that it's sort of an evolved trait to make these sounds, they'd need to benefit in some way from other plants or creatures hearing them, which we don't have a mechanism for.
Nick Petrić Howe
No exactly, there would have to be some sort of benefit to it. But we just really don't know at this point, like plants have got all sorts of ways of communicating with each other and with other animals. This could be a way, but we really don't know. But what the team have done is they've trained a machine learning model to predict whether plants are dry or stressed based on these sounds. So, if you were to position microphones, say in a greenhouse or something, and were like, "oh, are my plant stressed?" they've made a machine learning model that could tell you this with about 70% accuracy. So it could be a way to sort of monitor how plants are doing in the future.
Shamini Bundell
That's ingenious. So it could actually be of use to us, this would be helpful to me, I've been repotting my house plants this weekend, and worrying about whether I know that they're happy or not. And I'm like trying to get a sense of like, are you enjoying life plant? Am I killing you? And I'm not very good at it. So I could certainly do with them. With borrowing this the plant microphones and the machine learning model.
Nick Petrić Howe
Well, maybe you can ask them shoved in perhaps they'll lead them to you. But I think that's more or less all I have this week Shamini, what story do you have for us?
Shamini Bundell
So I've been reading this news article in Nature about a Science Advances paper, where they've made what they're calling 3D printable glass, that's made from proteins. So it's kind of like biodegradable glass.
Nick Petrić Howe
Okay, that seems like it could be really useful for you know, waste management, stopping things building up, recycling, that sort of thing. Sounds very cool.
Shamini Bundell
I think ultimately, that's kind of where they're hoping the use will be. Glass is actually an easily recyclable material, but a lot of it still ends up in landfill, and it does take then thousands of years to actually break down. Whereas this new kind of material is very easy to break down, very easy to decompose, because it's entirely, you know, organic, it's made from chains of amino acids, which make up proteins, just like most of the materials in our body are made of proteins. And you know, we are something that microorganisms can very easily break down.
Nick Petrić Howe
I mean, that's this sort of a macabre way of putting it, but I guess you're right, so how have they managed to do this to make this sort of more easy to break down material?
Shamini Bundell
Well, the hard bit hasn't been making proteins that degrade easily — that that they'll do — the hard bit has been making it like glass, and in particular to make it transparent. And in this case, they can also 3D print it and cast this material in moulds. And the final material they've got is actually, in some ways similar to glass. So, I don't know whether you've ever heard people sometimes describe glass as just a really slow-moving liquid. Have you ever heard that?
Nick Petrić Howe
Yeah, I've heard that before.
Shamini Bundell
So, people say that the bottom of church windows are thicker than the top because actually the glass is flowing slowly downwards, which is not in fact true. But it is true that glass is a... it's not a liquid, it's called an a amorphous solid. And it's got this irregular pattern of atoms, which is part of the reason why light can travel straight through it. So what they've done here is they have changed the ends of these amino acid chains, change these molecules, so that they don't start to split up before they melt. They've then melted them into a liquid and then rapidly supercooled them. So it's below freezing, and actually solidifies while retaining this liquid like arrangement of the molecules. And they managed to get this to retain and stay solid, even when it returns to room temperature. And because you've stopped the amino acids from crystallizing and forming a crystal structure. That's why they're transparent, a structure would make the glass kind of cloudy.
Nick Petrić Howe
Right, okay, so they've got this sort of disordered structure going on and that lets the light through like it does with glass.
Shamini Bundell
Roughly, yes, there's nice picture in the article of this sort of very nice see through shells that they've sort of moulded out of this material, which in this case, looks totally clear, although sometimes the glass comes out coloured, and they can also to some extent, tweak the peptides, the amino acids, and change the properties in different ways.
Nick Petrić Howe
So, it seems like they've nailed down the transparent part of what glass does, but glass also has lots of other quite useful properties, so does it do everything that glass does?
Shamini Bundell
Well, they say that it could be beneficial in things like lenses. And actually, the fact that you can give it different properties, you could make less rigid glass. So you could use it in sort of miniature flexible devices. And of course, the key thing here is this sort of biodegradable, you can make it break down. And you again, you can sort of tweak that slightly, but you can create a material that will break down in a compost heap, or they also exposed it to to digestive fluids, and again, showed that you can break it down, although that is also to some extent, a sort of downside, you can't use this for everything; for example, they noted that you probably wouldn't want to use it for drinks bottles, because it would slowly just decompose in the drink.
Nick Petrić Howe
So maybe it's not ready to quite replace glass yet. So what are the sort of next steps for this work where they sort of taking it?
Shamini Bundell
Well, this is just very basic... the researchers are quoted as saying this is a very fundamental study. The article sort of describes it as just a lab curiosity at this stage, but this definitely opens new paths for researchers to explore. And I just think it's very cool the idea that something that appears to be glass could be made of sort of peptides the same way as you and I are.
Nick Petrić Howe
Well I think that's all we've got time for thanks so much for chatting to me today Shamini. Listeners for more on the stories, and where you can sign up to the Nature Briefing to get more like them, check out the show notes for some links.
Shamini Bundell
Well, that brings us to the end of this week's show. As always, though, you can keep in touch with us on Twitter, we're @naturepodcast, or you can send us an email to podcast@nature.com. I'm Shamini Bundell.
Nick Petrić Howe
And I'm Nick Petrić Howe. Thanks for listening.