Forms of life, forms of mind | Dr. Michael Levin

Have you ever wondered what the cells of multicellular organisms are really capable of? We know what they normally do in vivo, building default, familiar tissues and organs under the influence of their neighbors during normal embryogenesis. But what would they do if allowed to reboot their multicellularity – if liberated from the control of other cells that shape their behavior, and allowed to express their baseline forms?

This is a story about a path that could lead to personalized medicine using your own normal adult body cells; it is also a story about a platform for addressing the genome:anatomy:behavior relationships, and for facilitating experiments in diverse intelligence research.

Take a look:

What would you guess this was? Looks like an organism we fished out of a pond somewhere. What if I told you we know what its genome is. It’s Homo sapiens. 100% human, with no genomic editing or exogenous synthetic biology circuits.

But we know what the Homo sapiens genome can build, don’t we? Well we know what it usually builds; but that’s far from the whole story. I’ve talked before about how things like plant galls teach us about what novel things cell collectives can be hacked to do. And, in the Institute that Josh Bongard and I lead, our team has asked the broader question by creating Xenobots – motile biobots made of frog embryo epithelial cells which do things like create copies of themselves using loose cells they find lying around in their vicinity.

But one thing some people said about Xenobots was that it was kind of a specialized result: these are amphibian, embryonic cells – known to be plastic. Maybe this is a unique thing frog embryo cells do, irrelevant to the broader issues of genomes and functional anatomical order. Especially since, “animal caps” – pieces of frog embryo epithelia that have functional cilia (little hairs that move) have been studied by developmental biologists for decades. Maybe this is just an animal cap; why call it a Xenobot?

And here we see why terminology is crucial. Terminology expresses the mindset, the mental lens through which we see something. It is observer-dependent and strongly constrains or enables the things that we can do with any system; by choosing a term, you are not picking out an objective unique truth, a single correct categorization for something. You are making a claim about the conceptual and practical toolkit that you will use to move forward. If you think of Xenobots as “just animal caps”, what you are likely to do is study their normal developmental and cell biology, and perhaps use them to study airway diseases and disorders of the mechanisms of cilia. What it does not facilitate is creating Anthrobots, characterizing their autonomous behaviors, and finding out what they are capable of in novel interactions with their environment (see below). That requires thinking about cells and tissues in terms of their plasticity in general and the use of their collectives as a biorobotics platform with top-down (not micromanaged) control. Are Xenobots animal caps? Yes. Is that the only, or the most interesting framework with which to understand them? No. Thus, to illustrate the broader opportunity, the next step was to go as far away from embryos, and from amphibians as possible, to find a truly surprising example that illustrates the point of plasticity and self-assembly. So we made (or rather, facilitated the self-organization of), and then studied the behavior of, Anthrobots – autonomous biorobots built from adult (often elderly) lung epithelial cells.

Now, other people have made airway organoids before (for example, Walter Finkbeiner’s and Charlie Ren’s work). In fact organoids is a huge field with many people building spherical constructs made of patient cells. But most of these organoids have a kind of “locked in syndrome” – they sit in place generating physiological data that can be used for drug screening and cell biology, but they are not able to move around and reveal their functional behavior. We focused on them as individual proto-organisms, made them of cells that have cilia (and thus could perhaps move, if they coordinated their beating activity), carefully characterized their behavior types and anatomical classes, and asked what they can do if given a chance to interact with other tissue types.

It’s all detailed in this paper, which was first preprinted over a year ago. The first author is Gizem Gumuskaya, a very talented PhD student in my lab (with interests spanning architecture and the new world of synthetic morphology) who developed the protocol for enabling these constructs and studied their behavior. Our collaborator is Simon Garnier who, among other things, came up with algorithms for the quantification of form and function. And, to my delight, the paper has a number of undergraduate authors – Tufts students (some of whom have now moved on) who participated in the work.

Here are the highlights. The Anthrobots self-assemble from a single cell:

Our protocol causes them to evert themselves – turning inside out so that the cilia can point outwards and row against the water in which they live. Rowing in specific patterns an in unison, they then exhibit a variety of behaviors, in sequences that can be visualized as a sort of ethogram, showing transition probabilities between discrete behaviors like that described for animals such as fish for example:

Here are some more videos:

We found that their various behaviors are also related to their shape and distribution of cilia (they come in several discrete morphotypes or classes). But perhaps the most amazing behavior we’ve seen so far is what happens when you let them interact with an induced gap in a neural tissue (a 2-dimensional layer of human neurons grown in a petri dish, scratched to mimic a very simple version of a wound):

As single bots, they traverse the scratch. But if allowed to assemble into a “superbot” cluster, and placed in this environment, they can induce the neural cells next to them to grow across the gap – they push the neurons to heal across the scratch “wound” (or do the neurons use the bot to complete their repair? We will identify which side is the driver). We don’t know how exactly they do it yet, but we know it’s not simply mechanical (passive material used in their place doesn’t cause the effect). My personal hypothesis is that the bots are enabling the cells from each side to know that there’s something on the other side for them to connect to. We don’t know this yet, but we will test the hypothesis that they can work as an active communications pathway (perhaps bioelectrical) by which cells and tissues can talk to each other. Note from the images below that the new neurons don’t just grow under (in contact with) the superbot – they also appear in spaces between the bots (the second image shows what it looks like after the bot has been removed).

We envision many future uses in the human body – laying down pro-regenerative molecules, clearing plaque from arteries, healing spinal cord or retinal damage, dealing with cancer cells or bacteria in the gut, or informing us of the status of the surrounding tissues. It’s crucial to note that the effect we saw – healing the neuronal scratch – was not test #78 out of hundreds of things we attempted. This was one of the first assays we tried. Thus, I suspect that we didn’t just luck onto the one thing they happen to be able to do – I think this is probably one of a myriad of unexpected things that Anthrobots (and other biological constructions) can do in the context of other tissues. Also important is the safety factor: unlike transgenic bacteria, viruses, and genetically-modified plants and animals, these biobots have a wild-type genome and do not exponentially reproduce and just biodegrade at the end of their lifespan (measured in weeks).

One interesting thing to note is that as a possible medical intervention, this technology is at a higher level of competency than for example drugs, which have a prescribed but fixed molecular interaction with some target(s). Biobots are made of cells which have a myriad different natural sensors on their surface and machinery for information amplification, decision-making, memory, and other types of context-sensitive behavioral controls. We do not know yet (but are studying) what their behavioral repertoire is but it’s quite likely that if we pay attention to what the cell collectives are telling us, we will be able to reach a much more sophisticated level of control of processes in the body by using technology that is itself not passive, hardwired devices but flexible agential materials that share our body’s priors about health and disease.

This conjecture is the flip side of my claim that physics only sees mechanisms, not minds, because it uses low-agency tools (voltmeters and rulers and such), and that higher-agency detectors (i.e., other minds) are needed to detect cognition at various levels. Perhaps the same is true for functional control; maybe complex system-level outcomes (e.g., health), that are hard to ensure with simple mechanical (bottom-up) stimuli like drugs, can be reached via symbiosis and communication with tools like Anthrobots with more of their own collective intelligence. Of course this is, at this point, speculation to be tested. And, we are not making any claims about their agency or problem-solving potential – the degree of their competency in physiological, anatomical, behavioral, and transcriptional spaces is an empirical question which we are in the process of addressing. We don’t know yet what they are capable of, but the idea is that as we learn to program them, we will use not only bottom-up familiar tools of synthetic biology but also top-down controls (training and re-writing of set-point memories) that take advantage of higher levels of organization that may exist in this system.

What is the Anthrobots’ status? They are certainly alive, but are these proto-organisms? machines? biological robots? products of bioengineering? natural forms? Yes. They are all of those things, because none of these words have (useful and deep) binary definitions, but rather express a variety of vantage-points across a continuum. By taking each of these perspectives, and the terminology that drives it, we see a different part of reality, focus our attention on a different aspect, and thus enable/prevent specific next discoveries.

I see the Anthrobots (and synthetic morphology in general) as an exploration device, with which to begin to understand plasticity of the software of life, to probe the competency of the agential material from which are constructed, and to start to map out the region of morphospace and behavioral space around the default outcomes facilitated by the genomic hardware under standard conditions. I think they are also a new platform for the fields of regenerative medicine, evolutionary developmental biology, and diverse intelligence (basal cognition). Importantly, we have made no claims yet about their degree of intelligence – we simply don’t know yet whether they can learn, and if so, in what capacity, but we will find out. Perturbative experiments are necessary to get a true idea of their capabilities in various problem spaces.

What’s next? Many things. First, unlike current Xenobots, Anthrobots can be made in bulk, in whatever quantities are necessary. Second, being made of the patients’ own cells, their use in the body won’t require immunosuppression – think personalized medicine. So, we are now investigating:

• how exactly do they knit together neural scratch wounds? what, if anything, are they saying to the neuronal cells?
• what other beneficial functions might they have in damaged tissues of different kinds?
• can they be used as avatars for screening drugs that alter behavior, not just morphogenesis?
• what are their proto-cognitive properties? can they learn from experience? do they have behavioral repertoires in complex environments? preferences?
• can we learn to program their form and behavior – how much control, and by what stimuli, can we exert? in other words, what does the option space look like, around the pinpoint of Homo sapiens morphology that evolution and normal development reveal by default?
• what other cells can these be made of? what if we make them chimeric with other body cells, or give them a microbiome, or instrumentize them with optical or electronic readouts of their internal states?
• what is the role of bioelectricity as an interface to modulate their morphogenesis?
• how do Anthrobots interact with bacterial, immune, and cancer cells?
• what are the transcriptomic and proteomic consequences of their new lifestyle?
• and much more – stay tuned, all of these experiments are under way.

So, 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 in morphospace, physiological space, and behavioral space they could undertake?

And one last thing. In the ancient biblical story, illustrated in the famous etching below, it was up to Adam to name the animals – not God, not the Angels, Adam had to do it (I guess that it’s in part because he was the one that would be living with them – an embedded observer, and in part because as one of them, he would have a unique ability to comprehend their bodies and minds in a kind of resonance that the Celestials would not have). I think this was pretty prescient, because “naming” a thing means to discover fundamental aspects of its true and deep nature. And in the coming decades, we humans are going to have to “name” and seek to truly understand a very wide variety of new forms that have never existed before.

The “scientific version of the Blind Men and the Elephant” image courtesy of Jeremy Guay of Peregrine Creative. The biological images were produced by the co-authors of the Anthrobot paper and taken from that paper and the supplemental material.