Computing Comes to Life
All of the molecular computing methods mentioned above envision that the computation will be done in vitro. Although the molecules are of biological origin, they are extracted from the cell, and the reaction takes place in laboratory glassware. But why not turn the living cell itself into a computer, powered by its own metabolism? Several research collaborations have done work pointing toward this possibility. Here I shall focus mainly on the ideas of a group at MIT, who have examined the computational aspects of the problem in great detail. The MIT group consists of Thomas F. Knight, Jr., Harold Abelson and Gerald Jay Sussman, and several of their present and former students, including Don Allen, Daniel Coore, Chris Hanson, George E. Homsy, Radhika Nagpal, Erik Rauch and Ron Weiss.
The first major goal of the MIT group is to develop design rules and a parts catalogue for biological computers, like the comparable tools that facilitate design of electronic integrated circuits. An engineer planning the layout of a silicon chip does not have to define the geometry of each transistor individually; those details are specified in a library of functional units, so that the designer can think in terms of higher-level abstractions such as logic gates and registers. A similar design discipline will be needed before biocomputing can become practical.
The elements of the MIT biocomputing design library will be repressor proteins. The logic "family" might be named RRL, for repressor-repressor logic, in analogy with the long-established TTL, which stands for transistor-transistor logic. The basic NOT gate in RRL will be a gene encoding some repressor protein (call it Y), with transcription of the Y gene regulated in turn by a different repressor (call it X). Thus whenever X is present in the cell, it binds near the promoter site for Y and blocks the progress of RNA polymerase. When X is absent, transcription of Y proceeds normally. Because the Y protein is itself a repressor, it can serve as the input to some other logic gate, controlling the production of yet another repressor protein, say Z. In this way gates can be linked together in a chain or cascade.
Going beyond the NOT gate to other logical operations calls for just a little more complexity. Inserting binding sites for two repressor proteins (A and B) upstream of a gene for protein C creates a NOR gate, which computes the negation of the logical OR function. With the dual repressor sites in place, the C gene is transcribed only if both A and B are absent from the cell; if either one of them should rise above a threshold level, production of C stops. In other words, C is transcribed only if neither A nor B is present. The NOR gate is said to be a universal logical element, because any Boolean function can be generated by linking together a series of NOR gates. The NAND gate (NOT AND) is also universal. Thus all that’s really needed to build the information-processing circuitry of a computer is the ability to make and connect NOR gates or NAND gates.
Pairs of NAND gates can also be coupled together to form the computer memory element known as a flip-flop, or latch. Implementing this concept in RRL calls for two copies of the genes coding for two repressor proteins, M and N. One copy of the M gene is controlled by a different repressor, R, and likewise one copy of the N gene is regulated by repressor S. The tricky part comes in the control arrangements for the second pair of genes: Here the repressor of M is protein N, and symmetrically the repressor of N is M. In other words, each of these proteins inhibits the other's synthesis. Here's how the flip-flop works. Suppose initially that both R and S are present in the cell, shutting down both of the genes in the first pair; but protein M is being made at high levels by the M gene in the second pair. Through the cross-coupling of the second pair, M suppresses the output of N, with the collateral result that M's own repressor site remains vacant, so that production of M can continue. But now imagine that the S protein momentarily falls below threshold. This event briefly lifts the repression of the N gene in the first pair. The resulting pulse of N protein represses the M gene in the second pair, lowering the concentration of protein M, which allows a little more N to be manufactured by the second N gene, which further inhibits the second M gene, and so on. Thus a momentary change in S switches the system from steady production of M to steady production of N. Likewise a brief blip in R would switch it back again. (S and R stand for "set" and "reset.")
One conclusion to be drawn from this synopsis of a few RRL devices is that a computer based on genetic circuits will need a sizable repertory of different repressor proteins. (I've used up a third of the alphabet already.) Each logic gate inside a cell must have a distinct repressor assigned to it, or else the gates would interfere with one another. In this respect a biomolecular computer is very different from an electronic one, where all signals are carried by the same medium—an electric current. The reason for the difference is that electronic signals are steered by the pattern of conductors on the surface of the chip, so that they reach only their intended target. The biological computer is a wireless device, where signals are broadcast throughout the cell. The need to find a separate repressor for every signal complicates the designer's task, but there is also a compensating benefit. On electronic chips, communication pathways claim a major share of the real estate. In a biochemical computer, communication comes for free.
Are there enough repressor proteins available to create useful computational machinery? Note that interference between logic gates is not the only potential problem; the repressor molecules taking part in the computation must also be distinct from those involved in the normal metabolism of the cell. Otherwise, a physiological upset could lead to a wrong answer; or, conversely, a computation might well poison the cell in which it is running. A toxic instruction might actually be useful—any multitasking computer must occasionally "kill" a process—but unintended events of this kind would be a debugging nightmare. You can't just reboot a dead bacterium.
Nature faces the same problem: A multitude of metabolic pathways have to be kept under control without unwanted crosstalk. As a result, cells have evolved thousands of distinct regulatory proteins. Moreover, the biocomputing engineer will be able to mix and match among molecules and binding sites that may never occur together in the natural world. The aim of the RRL design rules is to identify a set of genes and proteins that can be encapsulated as black-box components, to be plugged in as needed without any thought about conflicts.
Another important design tool is a simulator, which allows a device to be tested without the substantial effort of building a prototype. The world of electronics has long relied on a simulator called Spice, which models the physics of transistors and other electronic components. The MIT group is building a BioSpice simulator, which will model the dynamics of genetic circuits in a similar way.
So far, the MIT group has based their design work primarily on such simulations, but other groups have begun a few "wet" experiments. Michael B. Elowitz and Stanislas Leibler of Princeton University have created a free-running genetic oscillator in E. coli. Arranging three repressor genes so that they act on one another in turn, they observed periodic fluctuations in gene expression, with a frequency independent of the cell's reproductive cycle. In another E. coli experiment, James J. Collins, Timothy S. Gardner and Charles R. Cantor of Boston University built a genetic toggle switch much like the flip-flop described above, with two cross-coupled promoters and repressors. They report "robust bistability." Their eventual aim is the construction of "genetic applets"—self-contained program modules that could be "downloaded" into organisms.