I read and presented this paper in my lab class this past fall -- and I thought it was just one of the coolest papers I'd ever read. Cool concept, rational design, elegant solution to a Hard Problem, multiple benefits, modularity/composability, real-world results... this paper has it all.
You can look at cells as little factories, taking in raw materials and churning out interesting molecules. A cell's naturally occurring assembly lines are optimized by evolution in various ways, to increase efficiency and decrease interference ("cross-talk") with other processes in the cell. In particular, natural metabolic pathways are regulated so that they don't go wildly out of control and start overproducing whatever chemical, because that would be wasteful and expensive (not to mention potentially harmful).
But when you're putting an artificial assembly line into a cell, you have to undo some of these constraints and not others. You have to maximize efficiency, minimize cross-talk, and avoid making toxic products in the middle of the pathway as much as possible. These goals all line up with the goals of the cell. However, your main goal is different from the cell's goal of "produce just enough": you want to produce as much product as possible. More medicine. More biofuel. More super-protein-material-thing. More whatever. So this should be easy, right? The metabolic pathway is made up of enzymes that convert Chemical A to Chemical B to Chemical C, and you're inserting the genes for those enzymes into a bacterium. Why can't you just put very strong promoters in front of those genes, so you get massive quantities of each enzyme, and massive output?
This simple maxing-out approach causes several problems. First of all, it does nothing about the intermediate chemicals along the pathway -- they could still be toxic, or even just float away and go to waste. Second of all, this approach doesn't bother to optimize the ratio of the two enzymes. (If Enzyme 1 is half as efficient as Enzyme 2, then you ought to have twice as much of Enzyme 1.) Third of all, this doesn't do anything to stop the pathway cross-talking with other pathways. Fourth, and possibly most important, there's no guarantee that forcing each individual cell to make as much product as it possibly, possibly can is the most efficient way to convert cell food into useful chemicals. It's probably more efficient to let the cell divert plenty of energy into maintaining its own health and into spawning more cells, so you end up with more product overall.
To solve this dilemma, Dueber et al borrowed a trick that cells often use to regulate their own pathways. A scaffold is a structural protein that grabs on to all the enzymes in a given pathway, and holds them together into something like an assembly line or a bucket brigade.
This neatly solves several of our difficulties. Intermediate chemicals proceed right down the pathway instead of floating off to get lost or wreak havoc. If the system needs more copies of Enzyme 1 and fewer of Enzyme 2, the scaffold can simply include an extra binding site for Enzyme 1. Neither the enzymes nor the intermediate chemicals can run off and cross-talk with other processes -- isolating things on a scaffold is easy, although rather leaky. And because of all these gains in efficiency, it's no longer necessary to max out production of every enzyme in order to make the cells produce a lot of product. So the cells are happier and probably end up producing more total product anyway.
(Side note: scaffolds are a particular instance of a a more general strategy called substrate channeling. There are a lot of synthesis pathways that literally take place in a tunnel, forcing the molecules to stick around and undergo reactions. Tryptophan synthesis is one such pathway.)
But how do you make a scaffold? A scaffold has to hold on to its enzymes as tightly as a lock holds a key. We simply don't know how to design proteins that well. (Predicting protein folding is more or less The Famous Unsolved Problem of biology today.)
Dueber et al's really elegant, clever idea was this: borrow naturally occurring lock-and-key pairs (thoroughly massaged by evolution), and hang the keys as tags on the enzymes we want. It's brilliant, and so simple that I'm frankly astounded it worked, never mind worked as well as it did.
The 'keys' are short protein chains, less than 20 amino acids long. The DNA that encodes the 'key' can be easily added to the end of the gene for each metabolic enzyme, so you end up with an enzyme with a little key-tag hanging off one end. The 'locks' are chunks of other proteins, that recognize and bind the keys for their own inscrutable (read: irrelevant) purposes. You can slap the 'lock' genes together into a giant fusion protein, and you have an artificial scaffold!
This is just fabulously elegant. It provides all the advantages of scaffolds, and what's more, it is entirely modular. You can swap the locks around, put them in a different order, add more of them, delete a few, etc, without messing up the rest of the scaffold. This makes it really easy to optimize things like the ratio of Enzyme 1 to Enzyme 2: if you don't know exactly what it should be, you can just try a bunch of different ratios and see what works. The reason it's so easy to change things around is that the scaffold is nothing more than a chain of protein chunks strung together. In particular, the folding of any given protein chunk doesn't depend on the adjacent chunks, as it would if the scaffold was built as a single large unit instead of a composition of smaller units.
So now we get to the literal million-dollar question: does it work? The authors chose to test their scaffold by synthesizing mevanolate, a precursor to artemisinin (the antimalarial). Using this scaffold, they got 77 times as much product as they did from cells with no scaffold.
Figure 3: Each bar is a slightly different scaffold configuration. The best one rocked all the others into the ground.
What a cool piece of work. A really simple, elegant idea that solves so many complicated and subtle problems in one neat package, and produces real-world-usable results.