Never underestimate a cell

DODI: One thing we know in our industry is not to underestimate cells.

CONOR: Yes, our knowledge and use of cells forms the basis of our industry. Cells are the building blocks of all life as we know it.

DODI: It doesn't get more basic than that. Today, we're going to focus on two ways in which cells are proven to be anything but boring or average.

CONOR: And that’s what matters on today’s episode of Discovery Matters.

BRETT KAGAN: So, my name is Dr. Brett Kagan, I am the chief scientific officer over at Cortical Labs. We are a small but three-years-old startup based in Melbourne, Australia. We are interested in exploring biological intelligence and how it might be harnessed, and what one could do if one could harness biological intelligence in a dish.

DODI: Now, Brett is part of a team harvesting brain cells and teaching them to play the classic video game 'Pong'. Do you remember Pong, Conor? Did you ever play it?

CONOR: I played Pong. It was fantastic. You know those kind of like twisty hand things, I used to get very frustrated when they got stuck. So yes, Pong was a great one.

DODI: And I think I can mentally hear the 'burr...plonk'.

CONOR: Anyway, I think most of the audience are like, 'What are they talking about?' I think.

DODI: But you can find the research papers in our show notes. So do have a look at what Brett and his team were doing. Fun fact, it's been 50 years since Pong was released.

BRETT KAGAN: Really, all it is, is a toy example to answer this question that has sort of been proposed before, can we actually get biological neurons to show us that they can act intelligently? One thing we know about biology is that it can lead to a generally intelligent creature, so flies, cats, rats, humans, whatever are all able to show a degree of general intelligence that we're not yet able to recreate with machine learning. The question becomes, how is it possible?

CONOR: Explain the relation between brain cells, understanding intelligence and machine learning. How does Brett bridge the brain cells in a dish, which are biological and are a living form, and machine learning, which is all about the inanimate?

BRETT KAGAN: A lot of people look at neuroscience and biology as an inspiration for machine learning. And in a lot of ways, that's how we started out. There were a lot of calls from a number of very prestigious researchers and developers in the machine learning space saying, let's advance machine learning by reengaging back in the neurosciences, and our group took that incredibly literally, probably more literally than anyone else it might be fair to say.

DODI: The team set out to figure out the algorithms of intelligence as it were, which if implemented to machine learning, you can just imagine how powerful that would be.

BRETT KAGAN: You can look at artificial neural networks as an example of a huge reach in that direction. However, our thinking around that has changed. Now we're less about can we implement this into machine learning? Because I think our conclusions become, you probably can't. And that's simply because biology at its basic physical level, is very plastic, it's changeable, it's adaptable, the number of connections between any neuron and its fellow neurons is huge. Now, we don't have this in neuromorphic, we don't have this in machine learning with any reasonable scale. So, we've kind of concluded that the right question isn't necessarily, you know, we really want to recreate this in machine learning, but what can biology do that machine learning can't do? That's a different sort of question because it's like saying, why mimic or recreate what you can harness? Is it even possible to recreate them or mimic them?

DODI: This involves using two different types of cells, taken from either an embryonic mouse or are human induced pluripotent stem cells. So, this is a type of pluripotent stem cell, which in theory means it could turn into anything else in the body. Brett and his team differentiated it into a neuron using small molecules, they used a multi electrode array, essentially a computer chip that can sense the smallest of changes in electrical activity.

CONOR: So, I presume that can tell them when the brain cells are active and when the neurons generate small pulses of electrical activity, you can see that they're alive and they're doing their thing precisely.

DODI: That's it and that allowed the team to read the cells to see how they're behaving and how they are active. So, with this technology, you can deliver small electrical pulses at certain locations at certain times, and that is what the team computed into the chip.

BRETT KAGAN: We have our cells, we have them on a chip, and we can read and write information through electricity, because electricity is a shared language between the silicon and the biology to communicate with them. Now, of course, the real challenge as it were, because people have been doing that for a long time, is to figure out how do you actually communicate. What language are they speaking to actually convince a group of brain cells, which are otherwise just happily do their own thing, to do something very specific. That's where this whole idea that we talked about in the paper comes in. This is a very broad principle-based theory about how intelligence might arise. But in the simplest possible terms suggests that biological systems such as us, such as cells, and everything in between, what they want to do is try to create a predictable world. So, we set it up so that to create a predictable world for themselves, these neurons must behave in such a way as to hit the ball, there's no reward or punishment. They're very simple systems. And so, we're trying to do this at a very, very fundamental level. I think that was the really exciting thing to show in this work.

CONOR: So why Pong? Why pick a 50-year-old albeit awesome computer game that is simply about making sure that you hit the ball into the right place?

DODI: So, Pong has incredibly clear win or lose conditions, and Pong happens in real time. So, you do not have to wait for a response to then make the subsequent move.

CONOR: Okay. So that's why it's not chess, or backgammon, right?

DODI: That's right, because those games are not continuous, you make a move, then you wait for your next turn, kind of like children and dogs, it's hard to teach cells to wait for their turn.

BRETT KAGAN: It's like playing with your cat. Cats don't understand taking turns. There's no reason a group of cells in a dish would either but every single system does learn to operate in real time. Yeah, flies or worms, they move around and in real world continuously, and so we needed a game where there was a close to continuous control and gameplay and Pong fit the bill, easily recognizable one of the first games used for machine learning one of the first games at all. So that was why we picked it, you know, they'll start to definitely play a role, but it also met a bunch of other criteria.

CONOR: This is hilariously creative. It's a fun study. But surely the team aren't driven entirely by fun.

DODI: Now I think Conor, you can probably tell who was asking the questions with Brett on the phone, because I was like, 'Whew, that just sounds like never ending fun'. In reality, Brett was keen to point out that everyone on the team is passionate about what they do. They are driven by their impact that they can bring to the world and people in need. So, in fact, a very serious mission for this team.

BRETT KAGAN: I sat down with Hon, who's the CEO, not too long after I started, and I said, 'Why are you doing this? You know, you could make more money elsewhere? You could do this; you could do that. Why? Why is this like what you want to be part of?' For me, I knew I wanted to be part of it. Why do you want that sort of legacy? I want to leave a legacy of having done something in the world. That's exactly how I felt and, and I think that really does guide the team, we're here because we see something in the world that could maybe be brought out to make it better, in maybe a very small way, maybe not. But that excites us and drives us forward. You know, we think you can do that with some good humor and interesting approaches. Let’s see what the world takes from it.

CONOR: So, this then brings us to what Brett sees as the future for the team's research. What is it?

BRETT KAGAN: I say is it is so hard to know what 10 years down the line will bring. I'm a scientist at my heart and I've always been fascinated by intelligence, so if we could figure out some unified theory or understanding of how intelligence is arising at its most fundamental level, and how we can use it 10 years from now would be a very big goal. But that would be an amazing dream, to come true to understand how this could be done. The implications for that, as well as for drug discovery disease modelling, hopefully we can start to tackle some of the things that really afflict people in a more personalized way. Currently drug discovery disease modelling is done by averages. But we know that people in a clinic, they don't respond based on what the average is. They respond based on how they respond in a personalized way. We need to now move, and I'm not the only one saying this of course, but I am a big believer that we need to move the testing in the clinical trial and the actual clinic and leveraging the idea of the average responses to your genetic response is most likely going to be that we can test it, we can culture cells from you in a dish, and so like these aspects, I think are the most exciting things for 10 years from now.

DODI: And that would be personalized medicine at a crazy scale.

BRETT KAGAN: You know, induced pluripotent stem cells are an amazing technology. I think when you couple it with this synthetic biological intelligence approach, it opens so many options for people.

DODI: Brett's comment on averages is a perfect segue into our next conversation. Now we started off this episode talking about how important cells are to our understanding of life. When it comes to the relationship between cell growth, DNA replication, and division in a bacterial system, there has been a tendency to base our understanding on the "average" life of a cell.

CONOR: So, it's kind of important then for us to know what's really happening inside each individual living cell, right?

DODI: Yes. In a new paper, a team of biologists and physicists from Washington University in St. Louis and Purdue University used real single cell data to create an updated framework for better understanding this relationship in bacteria cell growth.

CONOR: And again, as ever, you can find the paper in our show notes.

DODI: Do go and read our show notes and read these papers in full. There were four authors in volved in this study: Petra Levin, Srividya Iyer-Biswas, Kunaal Joshi, and Sara Sanders. We spoke with Petra and Kunaal.

PETRA: I'm at Washington University in St. Louis. I work on bacteria, some pathogens, some non-pathogens. E.coli can do both pathogenesis and non-pathogenesis. I really got interested in biology and being a biologist, although I'm the daughter of a biologist. I got really interested in it , in college because I had some great teachers. I got to design experiments myself. Just having the ability to think through a problem and think how to solve that problem and being able to execute it in a laboratory situation is awesome.

KUNAAL: I'm a physics PhD student at the Iyer-Biswas lab at Purdue and well, my childhood dream was to be a physicist but not necessarily a biophysicist. I actually joined this biophysics field after I came to Purdue. Before coming to Purdue, I wasn't that exposed to biophysics, I was exposed to more of the popular science fields. So, I guess I didn't have a chance to appreciate the stuff that's going on in biophysics before then. But then after coming to Purdue. I mean, I guess I got exposed to it, more exposure to stuff that happens here in this field.

DODI: They wanted to figure out how these stochastic cells — meaning having a random probability distribution pattern - manage to coordinate DNA replication with growth and division, so that overall events happen in the right sequence despite the noisiness of each process.

CONOR: What are they getting at? Why do they need to understand this?

DODI: Here is one way of looking at it. If you are trying to design antibiotics, you cannot just target the ‘average’ cell because then it will not work on all cells. It's kind of like aiming for a common denominator that just doesn't exist. One analogy that came from our producer, Beth Armitt-Brewster who said that while she was studying, they looked at a story from the Second World War when the UK Royal Air Force tried to design cockpit seats that all pilots would fit without having to adjust the seats. So, they took an average of all the pilots' measurements, and it did not fit a single pilot. It fit no one because they aimed for the average.

CONOR: So, everyone was either too big, or too small, or too wide. Okay, so I guess you can't do the same thing and hope that it works for all bacteria.

KUNAAL: If you design an antibiotic that essentially kills only the average cell, but then actual cells are different from average cells, it might not work on all the cells, right?

PETRA: What Kunaal just said is true, because you have slow growing cells and fast-growing cells in the population when you hit them with an antibiotic. To the ones that are growing slowly, those guys are much less sensitive to the antibiotic. So, your analogy is perfect.

CONOR: Explain some of the observations behind their study.

DODI: Well, you might know E. coli as that wicked bacteria that results in recalls of lettuce or fresh produce. But for scientists, it's delightful because E.coli reproduces very quickly at a large scale and in gene cloning, due to the high efficiency of introduction of DNA molecules into cells. So, the team looked at single cell growth data from E. coli collected by the June Laboratory at the University of California, San Diego. They then constructed a mathematical model that captured the complex, stochastic behaviors of individual cells. So, the observations are much more about how to remove that average and really try to identify what's happening at an individual level?

KUNAAL: We were looking at results from population-based experiments initially, that other groups have already found. And so, this replication speed, which is the speed at which the replication progresses, in the DNA replication process, and what was known at that time was that there are two types of replication single food and multiferroic replication, in the bacteria we are studying. Essentially that means that in single fork replication, there is only one. The DNA only doubles once in a cell cycle, and then added in multiferroic replication, what that means is that the DNA can double and multiply multiple times. It can do this actively for future divisions as well.

PETRA: Yeah, I think there are different modes of replication where the cells can start one round and complete around between birth and division, when cells are, but they can also sometimes start a new round of replication before they finished the previous one. They're doing two at once and there used to be that only fast-growing cells do we start new rounds before they finish old grounds, and slow growing cells always finish around before they started a new round. If you go into the single cell data, for example, that's not true.

DODI: But if you look at population average data, it's true. That's a great example of where there was a disconnect because single cells which start new rounds before, they finish old rounds.

PETRA: You didn't have to be fast growing to be starting new rounds before you finish old rounds. So that was already a disconnect in the literature. There were other disconnects, that in the literature on population average, it looked like cells initiated new rounds of replication, and then wait a certain amount of time before dividing as if those two things, we call them, timers are connected, like you are born, you wait a certain amount of time, and or you know, grow a certain amount, then you initiate around a replication and then you have a certain amount of time. Then you finish and then you have a certain amount of time, and you divide.

KUNAAL: So, that was extremely surprising. Essentially it means two things. One, the plateauing out of the replication speed at higher population base growth rate isn't caused due to this multifocal application directly at least, and then the second thing was why is it that we observed this shift between single and multifocal application at this point, so what's the reason behind the multifocal replication? Initially the general assumption in the community was that, if the replication speed increases beyond that threshold, which would give rise to multifocal application period, but now we notice that in the same experiment we have both single and multifocal replication within the same replication speech, so, replication speed is not this direct factor that we thought it was.

DODI: Essentially, they found that, unlike others who believed DNA replication and cell division are dependent on each other, there is a correlation and independent. It reminded me of when we talk about manufacturing, when we talk about starting processes before another process has completed, so that you really can make the overall process a lot more efficient.

CONOR: Kunaal could predict the sequence of DNA replication initiation, the end of DNA replication and division based on when the three timers independently go off and reset? That kind of shifts how we understand the basic process of a cell’s life cycle.

DODI: And that's the whole point of their paper. That's right.

CONOR: What have you learned this week Dodi?

DODI: You know, I was looking at science.org, which is a great place to go down rabbit holes and look at some interesting studies that scientists are doing. There was a study done in the University of Würzburg about the total biomass on the earth and comparing wild mammals to humans. What is the total biomass of humans compared to the total biomass of wild mammals? What do you suppose is more?

CONOR: There can't be more of us, although maybe with current eating habits, maybe there is more biomass of humans.

DODI: That's the thing. So, as I understand it, the human biomass comes in at 390 million tons. Then when you add on urban rats and livestock you come up to 630 million tons.

CONOR: No!

DODI: The wild mammals are not that much. Wild land mammals have a total biomass of just 22 million tons.

CONOR: Is that it?

DODI: That's it and marine mammals just 40 million tons. So, it's just amazing.

CONOR: Think of the resources that we're consuming, it's insane. It's extraordinary. Wow, that's kind of terrifying. You know, so you add humans and then what they eat, and you know what the earth is supporting in terms of wild mammals. Given that learning is what we're really focused on, and every day is a school day who knew that I would learn from my dearest darling daughter that wearing an eye mask during overnight sleep improves your episodic learning and alertness the following day. Wearing an eye mask, you know, those kind of funky little eye masks? My daughter has one with eyes painted on it.

DODI: And what you get on an airplane if you want to sleep.

CONOR: So, this is based on the fact that ambient light influences sleep. Even if your eyes are closed, if there's ambient light in your bedroom from an alarm clock, or from under a blackout blind or even a flashing light on a fire alarm or a router, or whatever it is that you may have in your room that can affect your sleep, because that light can shine through your eyelids. So, this was published last year in December, but this team looked at how ambient light influences sleep structure and timing. They found that if you wear an eye mask, you learn better and you are more alert and more focused the following day when you sleep. So, I immediately tried this last night, I put a mask on because I have a few of those lurking around in travel bags and stuff. And you know what I woke up this morning feeling refreshed, having deeply slept and not at all bothered by the flashing light of the alarm clock always telling me how little sleep I have left. So, there you go. I confirm the study. Wear an eye mask when you sleep kids.

DODI: Well, we will do that. And you know what if you want to wear an eye mask while you listen to Discovery Matters, we're open to that focus with us. Our producer is Beth Armitt-Brewster. Editing, mixing and supervision by Ulrika Svensson and Tom Henley from Banda Productions, music from Epidemic Sound. My name is Dodi Axelson

CONOR: And I'm Conor McKechnie, make sure you rate us on Spotify or whichever platform you use. We'll see you when we come back with another episode of Discovery Matters. Bye for now.

Listen to more podcast episodes.