How MIT’s new biological ‘computer’ works, and what it could do in the future
by Graham Templeton
For years now, scientists have been working to make cells into computers. It’s a logical goal; cells store information in something roughly approximating memory, they behave due to the strict, rules-based expression of programming in response to stimuli, and they can carry out operations with astonishing speed. Each cell contains enough physical complexity to theoretically be quite a powerful computing unit all on its own, but each is also small enough to pack by the millions into tiny physical spaces. With a fully realized ability to program cell behavior as reliably as we do computer behavior, there’s no telling what biological computing could accomplish.
Now, researchers from MIT have taken a step toward this possible future, with cellular machines that can perform simple computational operations and store, then recall, memory. In principle, they provide the sort of control we’d need to design and build real cellular computers, but they could just revolutionize cell biology long before that future comes about.
MIT has been one of the most prolific sources of research on this topic. In 2013, this same team designed the computing strain that preceded this one: a biological “state machine.” A state machine is a straightforward (though not necessarily simple) form of computer or computer model in which the machine is only ever in one of a finite list of possible states, and can transition between these states according to input variables.
The classical example of a state machine is a vending machine. The counter doesn’t actually do math, but rather simply knows that if it already has five cents, and receives another five cents, it’s supposed to switch into “I have 10 cents” mode. This mode overwrites the “I have five cents” mode, and has its own set of associated behaviors accounting for all the possibilities for the next coin. This is distinct from having any sort of robust mathematical brain, any concept for five or 10, or the relationship between them.
Taken to an almost absurd level of sophistication, this basic system of combining input with a single changing internal state, over and over (and over), is behind modern speech recognition algorithms. To a great extent, machine learning is the automated process of building such chains of reactive states, which finally delivered the sci-fi ability for a computer to quickly and accurately narrow in on the identity of a spoken word. This is all to say: State machines may be old and straightforward, but they’re by no means useless if you can build them well enough, and elaborately enough.
This idea of a state machine can be applied in a living cell with the use of profoundly newfangled techniques in genetic manipulation; the DNA genome provides all the functionality necessary to make a useful state machine, we just have to use those functions correctly.
In the case of this new MIT paper, their highly customized strain of e.coli is designed with specifically arranged “target sequences” spaced very carefully throughout the genome; when scientists provide a specific combination of chemical signals, old and boring techniques in genetic engineering lead the cell to release a specific “recombinase,” a type of enzyme that can invert the orientation of a pre-programmed stretch of DNA, or remove it entirely. It’s the action of these recombinase enzymes, and their interaction with the short target sequences, that allow all-new abilities in cellular computing.
In response to each input variable, probably a chemical agent, a recombinase will either delete or invert its associated portion of the genome — and crucially, that portion of the genome itself contains targets that dictate later recombinase binding. So, the action of any recombinase will change the environment that the next recombinase will find upon activation, thus changing how that later recombinase interacts with the genome. If recombinase A inverted sequence A, then recombinase B can bind there; if it instead deleted sequence A, then recombinase B cannot bind there, and will instead go do something else, or perhaps nothing at all.
What this means is that the chain of responses to each new variable should be preserved in the sequence of the bacterial DNA itself, retrievable by sequencing the genome. More usefully, by coupling each state to the production of a specifically colored fluorescent protein, scientists can visibly reveal the sequence of the cell’s states, in real-time, with no ambiguity. For instance, providing input A followed by input B results in the production of red and green fluorescent proteins, while cells that receive the same two inputs in the reverse order produce red and blue fluorescence.
Most immediately, this provides a nice means to easily track gene expression — something that still largely confounds molecular biologists. In particular, the pattern of gene expression necessary to develop a stem cell into, for instance, a healthy neuron in the cerebral cortex, is very difficult to track. If we knew the pattern that dictates how cells traverse this path naturally, we could quite quickly go about trying to replicate that path synthetically. Implemented on a much more complex scale than in this experiment, MIT’s cellular state machine could offer a means to record even extremely rapid and complex patterns of gene expression, and take down a permanent record of these crucially important natural processes.
This experimental bio-state machine uses only three colors of fluorescence (red, green, and blue), and so by combing these colors it can only visually differentiate between a relatively small number of inputs — certainly not the full complement of hormones, transcription factors, and other signaling molecules that would need to be tracked to fully record a cell’s path through differentiation. But the researchers designed their system so it can be scaled up in complexity, and with a good-enough application, it could add a powerful new tool for studying cellular development and gene expression.
Cells are inherently programmable, so once you can reliably store information in the genome, doing simple in-out operations with that information requires the use of only long-standing techniques in biology. So the question becomes: What could be done with a sufficiently programmable cell or, ideally, a communicating group of cells? Put differently: We already have computers. Why is it worth reinventing the computational wheel inside a living cell? Gene expression is quick, but modern computer processors are quicker. And even with fluorescent reporting, reading the information output of a cell with never be as efficient as electrical pulses down a wire.
One major advantage of life over modern engineering is power efficiency. Running artificial intelligence algorithms takes many gigawatt-hours of electricity every year, and extremely long and complex problems could end up being vastly more affordable to solve, using biotech. Perhaps your vat full of computing e.coli is only a thousandth as fast as that Google data center down the street — but every one of their supercomputers costs millions of dollars in energy every year, while your bio-computer runs on just a few common, cheapo metabolites.
Life is also rugged; we find living cells at the bottom of the ocean and the top of the atmosphere, the mouths of active volcanoes and in ancient lakes under kilometers of arctic ice. Here’s an experiment: You want to know the response of a lake to acid rain. Release your investigative strain of e.coli; come back after a few weeks and a few rains; collect a sample; strain out your pet microbes; sequence their DNA; do a statistical analysis of the (hopefully) thousands of reporter genomes in your sample, each a detailed report about acidity since its host microbe’s release.
More important than environmental science, however, is medical science, since life can of course also exist inside other life. It could one day be possible to use programmable bacteria to read aspects of human biochemistry in living patients, from within their bloodstream — certainly, that seems like a path with less inherent resistance than building micro-robots to accomplish the same thing.
But in general, cells are simply different than computers. There’s really no telling what some visionary coder could do with algorithm designed from the ground up to use millions or even billions of simple, networked computers. Even if each computer is relatively slow, or limited, the technique could offer uniquely efficient ways past previously difficult or impassible barriers, from efficiently routing millions of packages around the United States to brute-force attacking strong encryption.
That’s all a long way out — but biotechnology researchers are taking the first crucial steps toward that goal. They’re building their awkward, jury rigged proofs of concept, living versions of the vacuum tube computers of old. There’s no telling whether these simple biological machines will go on to have the same sort of impact as computers, but the potential is undoubtedly there.
Top image credit: Liang Zong and Yan Liang