Waiting for the Electrician, or Someone Like Him

circuit.jpgIf our genes are wired like circuits, does that mean nature is an electrician?

One of the most important sorts of jobs that genes do is to switch other genes on or off. The classic example comes from Escherichia coli, and how it eats milk. (I’m afraid Escherichia coli will be progressively infecting this blog in the months to come, as I finish my new book on this remarkable bug.) Escherichia coli can make the enzymes necessary for digesting lactose, the sugar in milk. But it normally doesn’t, because it makes proteins that clamp onto its DNA next to these genes and prevent gene-reading enzymes from reaching them. If lactose should seep into the microbe, however, the molecule binds to the repressor proteins and causes them to change shape and fall off the DNA. If another protein, which signals a drop in other kinds of sugar, also grabs the DNA near the genes, Escherichia coli begins to crank out the lactose-digesting enzymes, and the milk feast begins.

It turns out that in any cell, many hundreds of genes may switch other genes on and off this way. A single gene may only make proteins if it is grabbed by many kinds of proteins, and its own protein may switch on many genes in turn. Those proteins may switch on still other genes, creating a cascade of switches.

Scientists have mapped the links between these genes in Escherichia coli, yeast, and other species. (Here’s a striking image of the networks in several species of bacteria.) They’re far from random. They have a hierarchy, with a few genes exerting major influences on the others. Even at their smallest scales, they show patterns. Consider just three genes. You can link them in many way (just try connecting three dots with arrows). It turns out that in cells, a few arrangements of arrows (known as motifs) turn up far more often in genetic networks than would be expected if the genes were linked randomly.

One of those common arrangements is particularly intriguing–shown here on the left of this figure. One gene, X, controls two other genes, Y and Z. Y, in turn, also controls Z. Electrical and computer engineers may have a sense of deja vu looking at this diagram. It’s a common element in circuit design, called a feed-forward loop. Engineers wire circuit components in feed-forward loops for many purposes. Uri Alon of Tel Aviv University and his colleagues were among the first to discover the abundance of feed-forward loops in nature. They proposed that these loops must serve some useful function that had been favored by natural selection. They zeroed in on a few key examples to test their hypothesis. And they gotten some pretty impressive results.

The genes that control when Escherichia coli builds its spinning tails, for example, are arranged in a feed-forward loop. After studying this loop closely, he and his colleagues concluded that it works as a noise filter. The gene X in this particular loop acts as a switch to turn on genes for constructing the tails (Z in this diagram). Gene X usually switches on in response to some kind of stress that Escherichia coli would do well to swim away from. X also causes gene Y to build up in the microbe, which also turns on the same tail building genes. If the stress should briefly drop, Escherichia coli stops expressing gene X. But that doesn’t shut down the tail construction right away, since there’s enough proteins from gene Y floating around to continue the project. If the stress disappears for good, however, Y disappears too, and the microbe abandons the project. Only strong signals of change get through this noise filter, which can ignore the minor meaningless static.

Alon’s work has become hugely influential, but it is now drawing some deep skepticism. Just because something looks like a circuit doesn’t mean it actually works like one.

Last year, French scientists considered the important of motifs–in evolution and in the functioning of a cell–by comparing several closely related species of yeast. If motifs played a vital role in survival, they might be conserved over evolutionary time. But the French scientists found that feed-forward loops and other motifs have been subject to as much evolutionary change as other interactions between genes.

The scientists then looked at what several motifs in yeast actually do (or don’t do). They chose motifs that are part of networks involved in functions such as pumping out toxins and starting the process of cell division. If you only looked at the shape of these motifs–the way the genes are joined–you might assume that they must be acting like a circuit, filtering and otherwise processing signals in important ways. But in each case, the scientists showed that none of the major steps of information processing take place in the motifs. The genes in the motifs are only important as parts of much larger networks. The scientists proposed that the abundance of feed-forward loops and other motifs was just a mirage produced by an incomplete understanding of how genes influence one another.

And a few days ago Dutch scientists launched another attack on the feed-forward loop, called simply “Feed Forward Loop Circuits as a Side Effect of Genome Evolution”. (It’s in press at Molecular Biology and Evolution, and you can get the full pdf for free here.)

The Dutch scientists wondered whether an abundance of feed-forward loops could emerge spontaneously, even if natural selection was not favoring them. They built a model for network evolution, based on an earlier one published in 2004. They built a network of genes, each of which had sites at which proteins could bind to them and cause them to be expressed. They then allowed the network to evolve according to rules based on what scientists know about how actual genes evolve. The genes could lose their binding sites or they could acquire an extra copy of a binding site. The entire genome could be accidentally duplicated. Genes sometimes disappeared, and sometimes proteins that switched on one gene began to switch on another.

yeast500.jpgInitially the model produced a random network, which came as no surprise. (The graph marked here as A.) But over time, something odd occurred. Some genes began to get more and more connected to other genes. And after about 100 generations, a large number of feed-forward loops appeared in the network, in what the scientists liken to an avalanche. In this picture, the feed-forward loops are marked in C as white circles. The scientists did not have to build in any advantage to feed-forward loops that could make them the object of natural selection. They emerged spontaneously from mutating networks. “Selection on individual circuits,” the scientists conclude, “is not needed to explain their abundance.”

I am very curious to see where this debate goes. One possibility that comes to mind is that the flagella-building feed-forward loop is a spandrel. In 1979 Stephen Jay Gould and Richard Lewontin spandrel. argued that many structures in living things got their start through no help from natural selection. They used the metaphor of the spandrel, the triangular space formed between an arch and its supporting wall. The spandrel was merely a side-effect of the arch, which painters made use of by covering them with images. Perhaps feed-forward loops emerge spontaneously through evolution, without any adaptive value of their own. In many cases they never become adaptations in their own right. But in some cases, they do turn out to process information in a useful way. I won’t predict where this debate will go–except perhaps to bet that scientists will not simply sit back and declare, “Design! My work is done.” I guess I like to stick to safe bets.

[Title thanks to Firesign Theater, soundtrack to my twisted youth.]

0 thoughts on “Waiting for the Electrician, or Someone Like Him

  1. Some of the more interesting work being done in the investigation of gene regulatory networks is by Michael Levine and Eric Davidson. What they have found, for example, is that there exist gene regulatory network kernels which tend to be conserved once they have been developed even among fairly distant species over hundreds of millions of years. Much of the rest of the network will change, but typically through the addition of later GRN kernels.

    Here is a link to one of their papers. I can look up more later.

    Gene regulatory networks for development
    Michael Levine, and Eric H. Davidson
    PNAS | April 5, 2005 | vol. 102 | no. 14 | 4936-4942

    Now it has been argued that modularity tends to be the result whenever gene duplication takes place, and this may very well be the case in the long run, and no doubt this tends to explain how much of the modularity we see comes into existence. However, in the case of the bat wing (as studied by comparing gene regulation in the fruit bat, a species of mouse and humans) there are times when two proteins will play essentially the same role in the network. This too, I believe is to be expected as the result of a form of network redundancy which results in developmental and environmental robustness.

    Taking this somewhat further, what we have seen so far in gene regulatory networks is a kind of scale-free, small world – similar to those found in many areas – and which tend to arise systemically simply as the result of the natural evolution of networks. (For comparison, bacteria evidently also employ the same sort of network for lateral gene transfer involving drug resistance and pathogenicity. Likewise, phages play an important role in such networks and appear to be fairly modular in terms of how they are constructed and trade in genetic components.) However,in some cases, there appear to be at least local violations to scale-free structures in gene regulatory networks. For example, I remember reading not too long ago that the network for yeast (which has been largely mapped) involves an exponential core – perhaps a preserved relic from the origin of the network itself.

    Two points which are a bit more tangential:

    1. It has been recognized that a great deal of genetic variation is cryptic and thus is not expressed at the level of phenotype. This, it has been suggested, is largely the result of of the canalization of such networks as is required for robustness. However, when perhaps a one or another key gene is knocked out or is subject to mutation, a fair amount of this variation is unmasked, consequently being made visible to selection and results in the release of evolutionary potential. At some later point, evolutionary developments are stablized by further mutations, resulting in the recovery of robustness.

    2. With respect to the bat wing, it appears that what occurs is a shift in the expression of HOXd13 towards the posterior late in the embryonic development of the bat, increasing the expression of bone morphogenic protein 2 by 30%, lengthening the “forearm” by a factor of 2 and at least two digits by a factor of 6 to 8. Such shifting HOX zones seems to be the cause of much of the modification of tetrapod bodyplans – and goes a long way towards explaining the allometric relationships between fairly divergent species which were discovered quite some time ago.

  2. Three engineers are discussing the nature of God. The electrical engineer says, “God is a EE! It’s all circuits and neural nets.”

    The chemical engineer replies, “Nah, God’s a ChemE. Those circuits all boil down to diffusion and chemical kinetics.”

    The civil engineer pauses, then says, “I’m pretty sure God is a civil engineer. Only a civil engineer would put a wastewater line through a recreational facility.”

  3. Also related would be John R. Koza’s work in Genetic Programming that has been able to reproduce a sizable number of electronic circuits using Darwinian principles. A partial online list is available at http://www.genetic-programming.com/humancompetitive.html.

    His book “Genetic Programming III” has some wonderful sequences of circuits gradually becoming more accurate and complex over a number of generations. For example, Chapter 31, “Synthesis of an Asymmetric Bandpass Filter”.

  4. you might have rushed to wikipedia too quickly. there are two meanings for feedforward. In the biological sciences (especially neural networks) feedforward usually just means “without feedback”, i.e. an output doesn’t loop back to provide input to an earlier stage of the chain. In engineering, and I believe this is the older and certainly less namby-pamby version, feedforward generally means a feedback control system where the thing that is being monitored is not the desired outcome variable, but a predictor of the desired outcome. For example, instead of a heater monitoring (and responding to) the temperature of a room, you can monitor the difference between outside and inside temperature, use the wall insulation value to predict the temperature in five minutes, and pump in the amount of energy which will counteract that predicted effect. The example you show in figure 1 is the first sort, but where there is any sort of loop you would usually be talking about the second sort, although I haven’t looked through the material enough to work out which sort any of the examples are talking about.

  5. I always thought of spandrels as more of a structural concept rather than an organisational one, which is what these feed-forward loops seem to be. I think this is an important conceptual difference since the way things are organised as opposed to the way they are built operate along a different premise.
    Spandrels are just thrown around liberally with no real policy as to what they actually constitute. If you can’t explain something by natural selection then does it have to be a spandrel? Perhaps an adaptation, but a gene network?
    The gene regulatory networks look remarkably like the diagrams that ecologists were doing for keystone species back in the 70s and 80s. This is where one species has a seemingly disproportionate effect on the ecosystem. Surely keystone species cannot be spandrels, but perhaps my analogy is wrong.
    I would still say an organisation principle is just an organisation principle, and a spandrel is just a fluke, a chance hapening that turns out to have some benefit, not something that you would expect again and again.

  6. Well, that settles it. God is a multi-disciplinary engineering design firm, with an EE department, chem eng department, and a civil (enviro) department. So does this prove that the correct form of ID is multiple designers?

  7. This looks like there’s a typo:

    In 1979 Stephen Jay Gould and Richard Lewontin spandrel. argued that many structures in living things got their start through no help from natural selection.

    It would make more sense if it read:

    In 1979 Stephen Jay Gould and Richard Lewontin argued that many structures in living things got their start through no help from natural selection.

  8. I’ve always though of engineering firms as somewhat monotheist, where the plebs do the designing and the general manager takes all the credit.
    Perhaps this is feudalism?

    I don’t really care, if God has to have some plebs do the dirty work for him so be it. He got the job done in the end


  9. “If our genes are wired like circuits, does that mean nature is an electrician?”

    Nah. Nature is a civil engineer. Who else would put a sewage canal right in the middle of a recreational area?

Leave a Reply

Your email address will not be published. Required fields are marked *