How We Know: Detecting a Signal

Fasting blood holds insulin at a few tens of picomolar: about one insulin molecule for every trillion molecules of water, with the blood protein albumin outnumbering it roughly ten million to one. Every pathway in this course rests on someone measuring a thing that scarce, or picking one modified protein out of the thousands in a cell. This lesson is about how that is done and how much each kind of result is allowed to prove.

Measuring almost nothing: let the hormone compete with a labelled copy

The method grew out of an unexpected by-product of a different experiment. In the mid-1950s Rosalyn Yalow, a physicist, and Solomon Berson, a physician, worked at the Veterans Administration hospital in the Bronx. They were testing the proposal that people with adult-onset diabetes destroy insulin too fast. They tagged insulin with radioactive iodine, injected it, and followed the radioactivity as it left the blood.

It left more slowly, and in one group: people who had already been treated with insulin. When their serum was sorted by paper electrophoresis, the radioactivity travelled with a globulin, the class of serum protein that includes antibodies. Something in the blood of treated patients was holding the injected insulin. Immunologists of the day doubted that a protein as small as insulin, 51 amino acids, could raise antibodies. Science rejected the paper. The Journal of Clinical Investigation rejected it at first, then printed it in 1956 with the word “antibody” removed from the title.

Yalow and Berson then used the unwelcome antibody as a scale. Put a fixed, small amount of antibody in a tube with a fixed, small amount of radioactive insulin, and add the blood sample. The sample’s own insulin carries no label, and it competes with the labelled insulin for the same few binding sites. The more insulin the patient had, the less radioactivity ends up on the antibody. Separate bound from free, count each, and read the concentration from a curve built with known amounts.

This is the radioimmunoassay. It works from nanomolar down to a fraction of a picomolar, and the binding obeys the rules in binding and dose-response. Its selectivity is the antibody’s: Yalow pointed out in her Nobel lecture that an antibody can tell cortisol from corticosterone, two steroids that differ by one hydroxyl group. The bioassays in finding a signal measured what a hormone does. This one counts how much is there.

Seeing one protein among thousands: sort by size, then ask an antibody

A cell extract holds thousands of different proteins. Prof. Deppmann asks his students to describe three methods that follow one of them.

SDS-PAGE sorts. Heat the extract in SDS, a detergent that unfolds each protein and coats it with negative charge roughly in proportion to its length. Pull the mixture through a polyacrylamide gel with an electric field, and small proteins travel farther than large ones.

The western blot finds. Move the sorted proteins from the gel onto a membrane, then soak the membrane in an antibody against your protein. A second antibody, which carries an enzyme that makes light or colour, marks where the first one stuck. You see a band at one size.

Immunoprecipitation fishes. Here the extract is not heated. An antibody fixed to beads pulls its protein out of the mixture, and partners bound to that protein come out with it. Blot the catch for a suspected partner and you learn that the two were in one complex.

Among what he calls “other ways to detect phosphorylation events” is the tool behind much of this course’s evidence: the phosphospecific antibody. It is raised against a short piece of a protein with a phosphate on one residue, and chosen so that it binds only when the phosphate is present. A blot with it reads one site, from cells, minutes after a signal arrives. Blot the same samples with an antibody to the whole protein as well, or you cannot tell more phosphate from more protein.

An antibody has limits. It answers the question you built it for, so it cannot find a site nobody suspected. It reports that a band got darker, not what fraction of the molecules carry the phosphate. It can bind the wrong protein, which is why a careful blot includes a sample in which the protein or the phosphate is missing. And a blot averages millions of cells; watching one living cell takes the reporters of the next lesson.

”How do we determine where the protein is modified?” Weigh the pieces

Prediction tools such as Scansite search a sequence for the patterns that known kinases prefer, and databases such as UniProt list sites that have been reported. They are a start. A predicted site is a guess until someone finds the phosphate on it.

Mass spectrometry finds it by weighing. Cut the protein into peptides with trypsin, an enzyme that cuts after lysine and arginine, usually not when proline follows. The instrument weighs each peptide (strictly, it measures mass divided by charge) closely enough that one phosphate shows.

mass of a phosphopeptide = mass of the peptide + about 80 daltons for each phosphate

In tandem mass spectrometry (MS/MS) the instrument then selects one peptide, breaks it along its backbone and weighs the fragments. The fragments spell out the sequence, and the ones carrying the extra 80 daltons show which residue holds the phosphate. No antibody is needed and no site has to be known in advance. Once the rare phosphopeptides have been enriched, one experiment on a whole extract can report thousands of sites. A site that is not reported may still exist, because some peptides are never detected.

A worked case: BATF

Prof. Deppmann’s 2003 paper in the Biochemical Journal, with Tina Thornton, Fransiscus Utama and Elizabeth Taparowsky, asks these questions of the transcription factor BATF. BATF from cells ran on a gel as shifted bands that calf intestinal alkaline phosphatase removed, which showed phosphorylation in vivo. A gel run with reversed polarity showed that serine 43 and threonine 48, both in the DNA-binding domain, carry phosphate. The paper reports phosphate on several serines and threonines and on at least one tyrosine; phosphoamino acid analysis on a thin-layer plate, which shows what kind of amino acid carries the phosphate, and kinase assays were among its methods. Then serine 43 was changed to aspartate, whose negative charge imitates a phosphate, to ask what the modification does. Each method is explained in the phosphorylation switch.

Which kinase? Give one kinase an ATP of its own

“What if you have a kinase but don’t know the substrates?” The approaches he lists each have a weakness. A purified kinase in a test tube shows what it can phosphorylate, which is more than it does in a cell. Prediction from sequence gives candidates. Finding the kinase bound to a protein shows contact, which is not catalysis. Removing the kinase with an inhibitor, RNA interference or a knockout changes many proteins, and the direct substrates look the same as those three steps downstream.

Kevan Shokat’s laboratory found a way through, which Prof. Deppmann introduces as “organic chemistry meets protein manipulation”. Kinases share an ATP pocket, and one bulky residue in it, the gatekeeper, limits what fits. Replace the gatekeeper with glycine or alanine and the pocket gains a hole. An ATP with a bulky group added to its adenine fits the enlarged pocket and fits normal kinases poorly. Such a kinase is called analog-sensitive. The laboratory introduced the approach in 1997 with the tyrosine kinase v-Src.

Design problem

A cell has about five hundred kinases, and you want the direct substrates of one of them. Say why adding radioactive ATP to a cell extract cannot give the answer. Then design a version of your kinase that can use a form of ATP that the other four hundred and ninety-nine cannot, and say what control proves that the labelled proteins are its direct substrates.

What an inhibitor proves, and what “necessary” and “sufficient” mean

“How do we determine what effect a phosphorylation event is having on a protein?” Prof. Deppmann’s answer has two parts: mutate the site to imitate the gain or the loss of the phosphate, and inhibit or activate the kinase. The second part has misled many laboratories.

Most kinase inhibitors work by plugging the ATP pocket, and about five hundred kinases share that pocket. Staurosporine, a natural product, binds most of the kinases it has been tested against. In 2000 Philip Cohen’s group in Dundee took compounds sold as selective inhibitors of named kinases and tested each against a panel of purified kinases. Apart from rapamycin and PD 184352, even the most selective inhibited at least one other kinase. Jenny Bain led further rounds in 2003 and 2007. The 2007 paper judged many widely used compounds too non-specific for any conclusion except one: ruling a kinase out. In 2008 Mazen Karaman, Patrick Zarrinkar and colleagues at the company Ambit Biosciences measured thirty-eight inhibitors against 317 kinases and concluded that “small assay panels do not provide a robust measure of selectivity.”

So an inhibitor result has a direction. If a dose that shuts off kinase X leaves the response intact, the activity of X is not needed. If the response stops, X or any other target of the compound may be the cause. To pin it on X, use two inhibitors of unlike structure, which are unlikely to share their other targets. Better, express a mutant of X that the drug cannot bind: if the response returns while every other target stays blocked, the drug was acting through X. Then add genetics, a knockout or a knockdown, which goes wrong in different ways.

Two words organise all of this. A step is necessary if removing it removes the response: the knockout, the controlled inhibitor, the site changed to alanine. A step is sufficient if turning it on by itself produces the response without the signal: a kinase that is always active, or a site changed to aspartate. A step can be one without being the other. Every method has its own way of misleading you, so a claim earns belief when two methods that fail differently agree.