There have been several reports over the years of people engineering receptor proteins to make them do defined tasks. They've generally been using the bacterial periplasmic binding proteins (PBPs) as a starting point, attaching some sort of fluorescent group onto one end, so that when a desired ligand binds, the protein folds in on itself in a way to set off a fluorescent resonance energy transfer (FRET). That's a commonly used technique to see if two proteins are in close proximity to each other; it's robust enough to be used in many high-throughput screening assays.

So the readout isn't the problem. But something else certainly is. In a new PNAS paper, a group at the Max Planck Institute in Tübingen has gone back and taken a look at these receptors, which are reported to bind a number of interesting ligands such as serotonin, lactate, and even TNT and a model for nerve gas agents. You can see the forensic applications for those latter two if the technique worked well, and the press releases were rather breathless, as they tend to be. But not only did these workers claim a very interesting sensor system, but they also went out of their way to emphasize that they arrived at these results computationally:

Computational design offers enormous generality for engineering protein structure and function. Here we present a structure-based computational method that can drastically redesign protein ligand-binding specificities. This method was used to construct soluble receptors that bind trinitrotoluene, l-lactate or serotonin with high selectivity and affinity. These engineered receptors can function as biosensors for their new ligands; we also incorporated them into synthetic bacterial signal transduction pathways, regulating gene expression in response to extracellular trinitrotoluene or l-lactate. The use of various ligands and proteins shows that a high degree of control over biomolecular recognition has been established computationally.

The Max Planck group would like to disagree with that. Their PNAS paper is entitled "Computational Design of Ligand Binding is Not a Solved Problem". They were able to get crystals of the serotonin-binding protein, but could not get any X-ray structures that showed any serotonin binding in the putative ligand pocket. They then turned to a well-known suite of techniques to characterize ligand binding. One of these is thermal stability: when a protein is binding a high-affinity ligand, it tends to show a higher melting point, since its structure is often more settled-down than the open form. None of the reported receptors showed any such behavior, and all of them were substantially less thermally stable than the wild-type proteins. Strike one.

They then tried ITC, a calorimetry measurement to look for heat of binding. A favorable binding event releases heat - it's a lower-energy state - but none of the engineered receptors showed any changes at all when their supposed ligands were introduced. Strike two. And finally, they turned to NMR experiments, which are widely used to determine protein structure and characterize binding of small molecules. WIld-type proteins of this sort showed exactly what they should have: big conformational changes when their ligands were present. But the engineered proteins showed almost no changes at all. Strike three, and as far as I'm concerned, these pieces of evidence absolutely close the case. These so-called receptors aren't binding anything.

So why do they show FRET signals? The authors suggest that this is some sort of artifact, not related to real receptor binding and note dryly that "Our analysis shows the importance of experimental and structural validation to improve computational design methodologies".

I should also note a very interesting sidelight: the same original research group also published a paper in Science on turning these computationally engineered PBPs into a functional enzyme. Unfortunately, this was retracted last year, when it turned out that the work could not be reproduced. Some wild-type enzyme was still present as an impurity, and when the engineered protein was rigorously purified, the activity went away. (Update: more on this retraction here, and there is indeed more to it). It appears that some other results from this work may be going away now, too. . .