Anthrobots
Anthro-whats?
This newsletter isn't about a new Transformers movie. It's not about an AI-powered device. It's about the early stages of a new medical treatment you may be grateful for someday.
The need
In the 1966 sci-fi classic Fantastic Voyage, a submarine and its crew are miniaturized and injected into the body of a scientist. Their plan is to travel into the scientist's brain and destroy a blood clot.
Fantastic indeed. But also grounded in a sensible premise: Some health problems call for solutions that are intelligent and operate on a very small scale. It's hard to have both at the same time.
For instance, surgery is intelligent – it's performed by an expert – but from a molecular perspective, the knife blade is still quite large and crudely directed.
Medicine works at a molecular level, but it's not intelligent. The molecules just do what they do. They can't monitor their own actions, change the treatment plan, exercise restraint, etc.
In short, Fantastic Voyage had the right idea. We need intelligent, micro-level treatments – and not just for blood clots. These treatments could repair damaged neurons, destroy cancer cells embedded in healthy tissue, deliver chemicals that promote cellular growth, etc. The possibilties are endless.
In this newsletter I'll be discussing a new study that develops the rudiments of such a treatment.
But no, it's not a study about minaturizing submarines. Rather, it starts with something that's already tiny: Living cells. The researchers show how ordinary human cells can be transformed into "bots" that move in potentially helpful ways. The researchers call the transformed cells Anthrobots.
The study has a surprise in it. The Anthrobots do more than just move around; they also promote healing.
This study also provides a nice illustration of a statistical method that has gone viral across the sciences since its invention in the early 20th century. If you've ever taken a standardized test (and of course you have), the statistical procedures used to develop that test included at least one that was the same as, or at least related to, one used to understand Anthrobot characteristics.
The new study
The new study, conducted in the lab of Dr. Michael Levin (Tufts, Harvard), was led by Gizem Gumuskaya (Tufts) and published last week in Advanced Science.
To create what they call Anthrobots, the researchers used living cells taken from the inner epithelium (surface tissue) of adult human lungs.
Why lung cells?
A key reason is that these cells have tiny hair-like structures called cilia. Cilia move in coordinated, wave-like fashion to carry mucus and other crap out of your airway.
Imagine removing one of these cells without damaging it. What you'd see is a tiny, roundish blob with hairs on it. (I think I just described my granddaughter.) But the "hairs" are moving. You couldn't help noticing that if they were just tweaked a bit, they might work well as legs.
In essence, this is what Gumuskaya and colleagues did. They grew lung cells under conditions that changed the positioning of the cilia, so that the cells could move under their own power.
This isn't genetic engineering. The researchers simply placed the cells in a sticky medium that twisted things around and re-positioned the cilia so that they could propel the cell body. After about three weeks, voilà, the researchers had lung cells that could move.
If you're one of my more squeamish readers, I can sympathize. It's slightly creepy to think of lung cells, umm, crawling out of a petri dish. (They actually swim rather than crawl, but I get that this might not reduce your squeamishness, so I'll move on.)
Adult human lung cells were also used because of what they're not. They're not embryonic stem cells. They're not animal cells. They're not synthetic. From a medical perspective, the raw materials sidestep a number of ethical and immunological challenges. In theory a person's own cells could be used someday to develop Anthrobots that are reintroduced to their body later for micro-level medical treatments. But I'm getting ahead of the story...
Findings
The researchers didn't know in advance how the Anthrobots would turn out. This was intentional. Here's how Dr. Levin, the senior researcher on this study, put it in a blog post last week:
"Next time you think about your body cells, quietly sitting in their tissues, you might wonder: what else are they capable of, if we were to let go of our tight focus on the genomic defaults of the hardware and gave them a chance to express new branchpoints of the software of life to see what journeys...they could undertake?"
Freedom to the cells!
The researchers did more than just ponder Levin's question. When they induced the cilia to protrude in leg-like fashion, they didn't micro-manage the process.
What Gumuskaya and colleagues had before them now were literally thousands of cells, many of which were moving. How to make sense of what these cells were doing? This is one place where probability theory and statistics helped. I'll describe the findings first, then touch on the stats.
1. To characterize movement, the researchers sampled 200 randomly chosen Anthrobots and videorecorded them for a period of 5 hours. Each Anthrobot engaged in one of four types of movement: circular, linear, curvilinear, or ecletic. Actual movement patterns are shown in the figure below.
Evidently, "eclectic" is a euphemism for "moves so unpredictably we may never find a practical use for it."
2. To characterize form, the researchers looked at a separate sample of 350 Anthrobots and identified three types. The main difference was in size and shape: Type 1 are small and regularly-shaped, Type 2 is the largest and least regular, and Type 3 falls between these extremes.
3. To determine the relationship between form and movement, the researchers examined several hundred additional Anthrobots. Here's what they found:
Type 1 Anthrobots either don't move, or they wiggle in place (which means that, medically speaking, they may not be useful).
Type 2 Anthrobots almost always move in linear fashion, while Type 3 almost always move circularly.
(Anthrobots that moved curvilinearly weren't analyzed, because their paths weren't distinct enough from the curved or linear variety. Eclectics weren't analyzed because, to put it bluntly, they don't seem useful either. Sorry guys.)
Why is this important?
So far, the findings may sound underwhelming. We've got modified lung cells swimming around in mostly straight or curved paths. Why refer to them as a type of bot, a term that implies intelligent, programmable behavior? More importantly, how will they ever unclog someone's arteries or excise their tumor?
Scientific American has already published a report on the new study that, although favorable on the whole, cites a few experts who sound unimpressed.
Here are two reasons why those critics are probably too critical:
1. "A journey of a thousand miles begins with one step."
The new study only marks the beginning of Anthrobot development. The researchers did not attempt to influence how the cells would end up moving, or to otherwise "program" intelligent behavior.
If you doubt that these critters will play at least a small role in the health care of the future, here's an anecdote for you:
In 1987, when I was a doctoral student in psychology at Cornell University, a prominent British scholar named Philip Johnson-Laird came to visit with a computer program that could improvise original Christmas carols.
Johnson-Laird played a couple of the carols for us. They were awful. I mean, they were clearly songs, but (a) they didn't sound Christmasy, and (b) they weren't anything you'd want to hear.
Now, less than 40 years later, AI is creating songs that are "good" enough to fool people into believing they were performed by famous musicians. AI-generated music is being purchased and (presumably) enjoyed.
The point of this anecdote is not just that digital technology progresses, but that it progresses in somewhat predictable, data-driven ways.
Johnson-Laird had to teach his program musical principles that allowed it to improvise, because the computers of the 1980s weren't powerful enough to store and extract necessary information from large musical databases.
In other words, the program just did its thing, based on general principles, without ongoing guidance from its creator or analyses of actual Christmas carols. Since then, we've created more malleable programs informed by millions of songs. No wonder the music is (slightly) better.
Right now, in a sense, the Anthrobots are like Johnson-Laird's program. They just do their thing, without ongoing guidance from their creator or anything else. But microbiologists are very good at getting cells to do things. They know a lot about what would make a swimming cell useful, and much is understood about how cells learn such things. We should be prepared for a future in which Anthrobots can be taught to move through the bloodstream (like the submarine in Fantastic Voyage) and take out a clot, repair damaged tissue, and so on.
2. An amazing finding.
When Anthrobots were placed together in a small dish, they spontaneously clustered together to form larger "superbots." The researchers then introduced the superbots to a layer of damaged neurons.
This neuronal layer, maintained in the lab, had been "scratched" by the researchers to a depth of 400-1000 microns. (This would be a needle-sized scratch, and thus larger than any individual nerve cell.)
The researchers placed some superbots at various points along the scratch. The superbots spontaneously moved partway along its length. That's remarkable. But here's the truly amazing finding: Within 72 hours, the scratch had begun to heal. Not everywhere, but only in the places the superbots had touched.
In the micrograph below, the green stuff is the neuronal layer, the black area is the scratch, and the red arrow points to a "bridge" of neuronal regrowth where a superbot had been moving.
Healing did not occur simply because time had passed. Nor was it caused by the mere weight of the superbots, as neutral substances placed on the scratch did not promote any healing.
Gumuskaya and colleagues assumed that the healing resulted from some sort of biochemical signaling, but the details remain mysterious.
Meanwhile....wow. Not to diminish the complexity of their lab procedures, but all the researchers did was to change the shape of some lung cells. The modified cells themselves, for whatever reason, move around, cluster, and induce healing. Can I say "wow" again?
Final thought: Should reliance on statistics worry us?
The researchers used a number of statistical procedures to analyze Anthrobot behavior. For instance, through Principal Component Analysis (PCA) they determined that Anthrobots tend to take on one of three forms, and that each form tends to move in a certain way.
Your life has been changed by statistical procedures like PCA, whether you realize it or not. These are the kinds of statistics that discern patterns in complex information by identifying what tends to cluster. To take just one example, PCA and related statistics are used in the development of standardized tests. Looking at the vast complexity of human beings and discerning certain personality types, or certain types of intelligence, is, statistically speaking, the same as looking at a complex array of Anthrobots and discerning certain types of bodies and movements.
You should be slightly worried now. The statistics I'm talking about here are among those used to manage uncertainty. The mere presence of these statistics tells you that some degree of uncertainty still exists. Do we really want Anthrobots to be used medically if they only "tend to" move in a certain way?
In the new study, almost all the findings are "tend to" findings. They're all statistically based; they all come with p-values. But if I had a problem in one of my kidneys, I wouldn't want to be treated with something that would "tend to" move toward that kidney.
I reached out to Dr. Levin, the senior author, for his thoughts on this. In my email I raised the broader question of how well Anthrobots need to be understood in order to be medically useful. Would it be enough, I asked, to rely on statistically-based descriptions of their behavioral tendencies? Here's part of his answer:
"Broadly, the relationship between input signals, and resulting cell morphology and behavior is the key to a lot of the regenerative medicine of the future. We need to understand it well, to have complete control of growth and form...But for specific applications, the history of medicine (e.g., aspirin) shows that deep understanding is not always required for useful applications. So, I think we will have biomedical treatments long before we know everything..."
In short, we'll use Anthrobots when we feel they work well enough, even if we're not sure why they work, or if their characteristics are statistically determined rather than known with certainty.
I can relate. For roughly two decades I've been taking anti-seizure medications. They work just fine, even though scientists still don't fully understand why they work (and even though, in my case, they don't know why I'd begun having mild seizures in the first place).
As for when Anthrobots might actually come to market, and how we might first see them applied, here's what Dr. Levin told me:
"I am not able to give reliable timelines because there are too many factors here out of our control. I think the earliest applications will be in vitro (bioengineered organs and such) and then in vivo uses for things like neural repair, sensing of disease states, etc."
I can relate to this business of scratched neurons and neural repair too. When I started having mild seizures, one of the neurologists speculated that a few of my neurons might have "scratches" too small to be detected by an MRI.
Although MRI imaging has improved since then, no damage has been spotted yet. But suppose they find something someday? Will I allow doctors to swab a few cells from my trachea, then inject Anthrobots into my carotid artery three weeks later, under the assumption that they would "probably" travel to the right place in my brain and heal the scratches?
Heck yes. Anthrobots swimming around in my brain sounds a little creepy, but so do a lot of other medical procedures when you ponder the details.
Besides, most of the medical treatments we currently rely on are only statistically likely to be effective; they don't work 100% of the time, and nobody from the FDA to your local health care provider expects them to. Crudely speaking, medical treatments consist of lots of "tend tos". I come away from this study persuaded that in the future, Anthrobots will take their place among our "tend to" health care resources.
Thanks for reading!