Josh P. Roberts has an M.A. in the history and philosophy of science, and he also went through the Ph.D. program in molecular, cellular, developmental biology, and genetics at the University of Minnesota, with dissertation research in ocular immunology.
You know the old joke. There are 10 types of people in the world: Those who understand binary and those who don’t. (For those in the latter category, “10” is binary—base 2, the language of computers—for the number 2). The same can be said for studying protein-protein interactions (PPIs): There are those who do binary and those who don’t.
Broadly speaking, there are two complementary ways of studying protein-protein interactions (which, in aggregate, are called an interactome), binary screens and affinity purification. The binary method uses a split reporter tag to indicate whether two proteins are interacting. This method is exemplified by yeast two-hybrid (Y2H) screens, in which an open reading frame (ORF) library is fused to the activation domain of a transcription factor while another ORF library is fused to the transcription factor’s DNA-recognition domain. Interacting proteins bring the reporter halves together, enabling the now complete reporter to do its reporting—in this case, transcription of a protein necessary for the cells to thrive.
The power of such systems is throughput, says Marc Vidal, director of the Center for Cancer Systems Biology at the Dana-Farber Cancer Institute. “If you work appropriately, you can easily test tens of millions of combinations, and going to 200 million is feasible.”
Affinity purification (AP) is exemplified by co-immunoprecipitation, in which a protein of interest is captured by an affinity reagent (typically an antibody), and its interacting partners are taken along for the ride, to be identified by a downstream process like mass spectrometry (MS). Although binary studies require prior knowledge of the identity of both partners, affinity purification is “essentially a forward screen” in which the cells tell us what the partners are, explains Anne-Claude Gingras, Principal Investigator at the Samuel Lunenfeld Research Institute in Toronto.
Avoiding falsies
Regardless of the approach used, questions of finding and verifying a genuine protein-protein interaction loom large.
Gingras points out that in “a sample in an MS coming from an affinity purification [AP-MS]— it doesn’t matter what you’re purifying—you’re going to identify 300 or 400 proteins. Of course, most of them are contaminants.”
To sort these out, it’s imperative to have the proper controls, making sure at a minimum that they use the same tags purified from the same cell line using the same resin. It’s also important to run biological replicates, separated in time, she says. “If you do it that way, and only score those interactions that are coming in both of your biological replicates for your bait of interest, but never in your negative controls, you actually have come a long, long way into [ensuring] the quality of your data and filtering out your false-positive interactions.”
Vidal’s lab has developed positive and random reference sets (PRS and RRS, respectively) “composed of a couple hundred pairs of known, solid, widely accepted protein-protein interactions as a positive control, and then a few hundred pairs of proteins that are chosen randomly in the proteome” as a negative control, he says.
Vidal's team has used these sets to compare the performance of five different binary screening strategies. [1] Each assay was optimized in terms of stringency, so that the maximum number of interactions of the PRS and the minimum of the RRS (ideally, none) were detected. They found a “detectability rate” of about 20% for each, with no completely overlapping sets and no statistically relevant correlation with the type of interactions detected (membrane receptor vs. transcription factor vs. cytosolic, for example). If conditions are optimal, Vidal notes, there is no reason to discount PPI detected by only a single assay as being a false positive.
The PRS and RRS have been expanded since the original benchmarking studies were run in 2009, and several additional species’ datasets were added, but “essentially the trend remains the same,” Vidal says.
Gingras, too, has been moving toward using collections of different proteins for controls. But she realizes that “it’s much easier for me to filter out contaminants from my AP-MS experiment (that’s what we do for a living) than it is for somebody that is just doing one experiment and has one control.”
To help facilitate smaller AP-MS efforts, Gingras’ lab collaborates with Alexey Nesvizhskii’s group at the University of Michigan to collect the protocols from major labs doing this kind of experiment. These are organized in terms of cell line used, expression level, fusion tag, negative controls and what proteins came down across those negative controls, in a quantitative manner.
The freely accessible Contaminant Repository for Affinity Purification—the CRAPome (www.crapome.org)—has “an easy-to-use interface so that the small lab can actually use that as a resource to score their own interactions,” Gingras explains. “You can just essentially upload your own data and just use the CRAPome resource to … filter your PPI. This makes things much easier for somebody that doesn’t have a huge knowledge of computational bioinformatics.”
Too much
One key to identifying biologically meaningful PPI from AP-MS is expressing tagged proteins at (nearly) physiological levels, says Gingras. Different expression systems are available, but “one that seems to be leading right now” is the Flp-In™ T-REx™ system from Life Technologies. As Gingras explains, Flp-In uses a recombination-based strategy, in which a sequence of interest is transferred to the vector via “a kind of slip-in transfer.”
“You can express different proteins and they tend to be expressed at fairly low levels,” she says. Flp-In-compatible vectors are available from several academic labs, including Gingras’.
Promega’s HaloTag® system allow users to control protein expression levels by selecting from among a series of vectors a with different promoters. In this system, the protein of interest (bait) is genetically tagged at either the N- or C-terminus with HaloTag, a protein domain that covalently binds to a surface such as a resin or glass slide containing a synthetic chloroalkane ligand within 15 minutes. That, in turn, “allows us to start washing very early to remove the lysate and prevent binding of nonspecific interactions,” says Danette Daniels, group leader in functional proteomics R&D at Promega; her team developed the system.
At this point users can either enzymatically cleave the bait from the resin, enabling functional assays to be performed on the complex, or elute with a denaturant to release the protein. Because the bait remains bound to the resin, it will not interfere with downstream analyses such as Western blotting and mass spectrometry.
Another way to reduce sample contaminants is tandem affinity purifications using a combination of tags—purify using one tag, release from that tag and repurify using another tag to the same protein or another component of the complex. Such a method offers a huge advantage for cleaning of samples with stable complexes, says Gingras, but it’s important to know what you’re going after: “All the interactions that I’m interested in are labile, and you cannot do that. You’d lose all the interactions.”
What’s new?
Despite the abundance of PPI detection methods available, researchers have not been content to rest on their laurels. One “completely new” method, introduced last year by Brian Burke’s group at the Institute of Medical Biology in Singapore, relies on tagging bait proteins with a variant of biotin ligase, says Gingras. [2] “If you add biotin, the biotin ligase attached to your protein of interest will biotinylate proteins in the vicinity. This is great, because now you have this stable mark on everything your protein of interest has been seeing in the cell. You can lyse the cells under really harsh conditions, because biotin is not going to go away.”
Alice Ting at the Massachusetts Institute of Technology has since come out with “another variant of the same idea,” Gingras says. [3] “Those techniques in the next couple of years will explode—they enable us to probe interactions for proteins that essentially have failed before.”
References
[1] Braun P, et al., “An experimentally derived confidence score for binary protein-protein interactions,” Nat Methods, 6: 91–7, 2009. [PubMed]
[2] Roux KJ, et al., “A promiscuous biotin ligase fusion protein identifies proximal and interacting proteins in mammalian cells,” J Cell Biol, 196:801–10, 2012 [PubMed]
[3] Rhee HW et al., “Proteomic mapping of mitochondria in living cells via spatially restricted enzymatic tagging,” Science, 339:1328–31, 2013. [PubMed]
Image: Overview of the yeast two-hybrid assay. Source: Wikipedia.