# Alanine scanning and hotspot residues in protein interfaces

> Alanine scanning replaces interface residues one at a time and measures what binding loses. In the growth hormone receptor complex, two tryptophans carry most of the energy of about 30 contacts. Learn what a scan measures, why hot spots act as a group and how well computation reproduces the experiment.

## What alanine scanning measures

Alanine scanning replaces one residue at a time with alanine and measures how much binding is lost. The loss is the change in binding free energy, ΔΔG, in kcal/mol, and it works as a price tag for that side chain. In the first scan of this kind, Cunningham and Wells made 62 alanine mutants of human growth hormone (hGH) across three segments linked to receptor binding. About a dozen large side chains each cost more than a factor of four in affinity.

The structure and the scan then disagreed in a useful way. Thirty-one hGH side chains are buried at the interface with the receptor, yet only about a quarter of them account for most of the binding energy.

The receptor side looks the same: roughly 30 side chains touch the hormone, and a central hydrophobic region dominated by two tryptophans, Trp104 and Trp169, accounts for more than three quarters of the binding free energy. Those two residues bury 246 Å² of the receptor's 1,322 Å² in the complex, or 19% (our calculation, shown below). Most contacts add little.

A common cutoff calls a residue a hot spot when its alanine mutant costs at least 2 kcal/mol. At room temperature that is a 29-fold weaker Kd, and a 4.5 kcal/mol loss is roughly 2,000-fold (Figure 1). Cutoffs vary between studies, so compare ΔΔG values and not labels. A [docking score is not a binding free energy](/guides/docking-score-binding-energy-affinity), but a measured ΔΔG is, and it converts directly into a change in Kd.

![Bar chart on a log axis of the fold increase in Kd for ΔΔG values of 1.0, 2.0, 3.0 and 4.5 kcal/mol, at 5, 29, 158 and about 2,000 times](/images/charts/alanine-scanning-ddg-to-kd-fold.webp "**Figure 1. What a ΔΔG means for Kd.** We calculated fold = exp(ΔΔG / RT) at 298.15 K. The usual 2 kcal/mol hot spot cutoff is a 29-fold loss of affinity.")

Three questions decide how far to trust a scan: what the alanine substitution does besides removing a side chain, whether hot spots act alone, and how well a computer can reproduce the experiment.

## What the substitution does and does not show

Alanine keeps the backbone and the first side-chain carbon and removes everything beyond. The aim is to delete a side chain without adding a new interaction, and that is why alanine is the default replacement. It also means a scan cannot test positions that are already alanine, and glycine and proline can change the backbone in ways that alanine does not.

The catch is that a mutant that folds badly looks like a hot spot. Cunningham and Wells checked their mutants against seven conformation-sensitive antibodies to confirm that the fold was intact. Any scan without such a control mixes binding with stability.

Alanine is also not always neutral. In the original scan, Glu174 to alanine raised affinity more than fourfold, so that side chain hindered binding. A scan reports the net effect of a residue at equilibrium. When the kinetics are measured, they add a second layer: for hGH, most of the effect of the substitutions was a faster off-rate and not a slower on-rate, and mutations at arginines mattered most for association.

## What hot spots look like

Bogan and Thorn compiled 2,325 alanine mutants with measured ΔΔG values. Binding energy is not spread evenly: it concentrates in a small subset of interface residues, and at the level of side chains there is little correlation between buried surface area and binding energy.

Hot spots are enriched in tryptophan, tyrosine and arginine, and are surrounded by energetically less important residues that probably shield them from bulk solvent. They found that occluding solvent is a necessary condition for a highly energetic interaction, so a hot spot is a feature of a patch and not of one atom.

The patch has a practical consequence for [antibody and antigen interfaces](/guides/antibody-vs-antigen), where a contact list is a map of touching atoms and not a list of what pays. It also matters for [protein design](/guides/protein-design), where a binder often needs to reproduce a few target hot spots and not every contact.

## Growth hormone and its receptor, read four ways

The hGH receptor shows how the methods disagree. The same complex gives four different answers to "which residues matter", and only one of them is an energy.

Contacts give the widest answer. In the 1:1 complex (PDB 1A22, 2.6 Å), we count 26 receptor residues with an atom within 4 Å of the hormone, and 36 that lose at least 1 Å² of accessible surface on binding. The published description of about 30 contact side chains fits.

Burial narrows the list, and here it finds the tryptophans. Trp104 buries 154 Å², more than any other receptor residue, and Trp169 ranks third at 92 Å² (Figure 2). But burial also puts Asn218 second, and it cannot tell which of these twelve pay.

![Horizontal bar chart of buried surface area for the twelve largest receptor residues in the hGH complex, led by Trp104 at 154 Å squared, Asn218 at 130 and Trp169 at 92, with Pro106 last at 36](/images/charts/alanine-scanning-hgh-receptor-buried-area.webp "**Figure 2. Burial ranks the tryptophans but not the reason.** We calculated the surface area each receptor residue loses on binding hGH in PDB 1A22 (Shrake-Rupley, 1.4 Å probe, Biopython). The [script](/data/guides/alanine-scanning-hotspots/hgh_interface.py) reproduces every number.")

The alanine scan gives the energy: a hydrophobic centre, led by the two tryptophans, holds more than three quarters of the binding free energy.

A structure of the 1:1 complex then showed that the scan overstates the direct part. Several hydrophobic residues orient the tryptophans, and kinetic analysis suggests that Pro106 locks the main chain of Trp104 into the conformation binding needs. Pro106 buries only 36 Å², rank 12 in Figure 2. The authors conclude that the true functional epitope may comprise as few as six side chains, mostly in alkyl-aromatic stacking.

## Hot spots act as a group

That last result says the hot spot is assembled cooperatively, and many residues contribute indirectly. Single-site ΔΔG values therefore do not add. Outside the functional epitope of hGH, a hepta-alanine mutant bound up to tenfold tighter, and the authors suggest the periphery attenuates affinity, perhaps to confer specificity.

Two habits follow. Read a scan as a ranking of residues one at a time, not as a budget that sums to the measured Kd. And test the interesting residues together, with double mutants or a mutant of the whole patch, before you build a design or a drug on them.

## Predicting hot spots computationally

Computational alanine scanning takes a complex structure and predicts which side chains would cost energy. Kortemme and colleagues' Robetta method called 79% of hot spots and 68% of neutral residues correctly on 233 mutations in 19 complexes. A machine-learning method that combined energy terms reached a precision of 56% and a recall of 65% at the 2 kcal/mol cutoff. At 56% precision, nearly half of the residues it flags are not hot spots.

The broader benchmark is less flattering. SKEMPI 2.0 holds 7,085 mutations in structurally resolved complexes. When Tsishyn and colleagues tested eight predictors on it, most were reasonably accurate but generalized poorly to unseen mutations and were biased towards destabilizing predictions. Choose a predictor for ranking residues to test, and confirm a handful in the lab or against the literature.

Different tools answer different parts of the question. [StaB-ddG](/app/stab-ddg) predicts the binding ΔΔG of a substitution from a wild-type complex. [gmx_MMPBSA](/app/gmx-mmpbsa) decomposes binding energy per residue from a simulation.

[ScanNet](/app/scannet) predicts where a partner binds, which is the surface you would scan, and not what each residue costs. [PRODIGY](/app/prodigy) and [PPAP](/app/ppap) estimate the affinity of a whole complex from its structure, one number per complex, so they say how tight the interface is overall and not which residues carry it.

## Beyond alanine

Deep mutational scanning extends the idea from one replacement to every one. Starr and colleagues measured how all amino acid mutations to the SARS-CoV-2 receptor binding domain affect both expression of folded protein and ACE2 affinity. Measuring folding next to binding addresses the stability confound that an alanine scan has to control for separately. It is also a source of data for the predictors above, since the reference set of mutations is far larger than 2,325 alanine mutants.

## Which approach to start with

  <caption><strong>Table 1. Which question each approach answers.</strong> Start with the left column and add the second only when the first leaves a gap.</caption>
  <thead>
    <tr>
      <th scope="col">Your question</th>
      <th scope="col">Start with</th>
      <th scope="col">Add</th>
      <th scope="col">Check</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>Which interface residues carry the binding energy?</td>
      <td>An alanine scan with a folding control, or StaB-ddG as a first ranking</td>
      <td>Double mutants of the top residues</td>
      <td>That ΔΔG values are not simply additive</td>
    </tr>
    <tr>
      <td>Where does a partner bind on my protein?</td>
      <td>ScanNet</td>
      <td>An alanine scan across the predicted patch</td>
      <td>That the patch matches known mutants or a structure</td>
    </tr>
    <tr>
      <td>How tight is the whole complex?</td>
      <td>PRODIGY or PPAP</td>
      <td>A measured Kd for the wild type</td>
      <td>Fold error against measurement</td>
    </tr>
    <tr>
      <td>What does each residue contribute in a simulation?</td>
      <td>gmx_MMPBSA decomposition</td>
      <td>An alanine scan of the top residues</td>
      <td>That the ranking holds in a second run</td>
    </tr>
  </tbody>

A predicted hot spot is a hypothesis about a mutant. The cheapest test of one is the mutant itself.

## Alanine scanning tools on ProteinIQ

Start with the interface and its overall affinity, then rank residues, then test the top few.

- [ScanNet](/app/scannet): predict where a protein binds its partners
- [PRODIGY](/app/prodigy): estimate the binding affinity of a protein complex from its structure
- [PPAP](/app/ppap): predict protein-protein binding affinity with a deep learning model
- [StaB-ddG](/app/stab-ddg): predict binding ΔΔG for single or combined substitutions
- [gmx_MMPBSA](/app/gmx-mmpbsa): decompose binding free energy by residue from a trajectory

## Frequently asked questions

### What is alanine scanning mutagenesis?

It replaces residues one at a time with alanine and measures the change in binding or activity for each. Cunningham and Wells introduced it in 1989 to map the residues of human growth hormone that its receptor recognizes. The change in binding free energy, ΔΔG, ranks the residues.

### Why use alanine instead of another amino acid?

Alanine removes the side chain beyond the first carbon without adding a new group, so the change is mostly the loss of what was there. The limits are that positions already alanine cannot be scanned and that a mutant that misfolds can mimic a hot spot, which Cunningham and Wells controlled with conformation-sensitive antibodies.

### What is a hot spot residue?

A residue whose alanine substitution costs a large share of the binding energy, commonly at least 2 kcal/mol, or about 29-fold in Kd. In 2,325 alanine mutants, hot spots were enriched in tryptophan, tyrosine and arginine and were surrounded by residues that occlude solvent.

### Do hot spots bury the most surface area?

No. Bogan and Thorn found little correlation between buried surface area and binding free energy for individual side chains. In the hGH receptor, Pro106 buries little area yet appears to hold Trp104 in its binding conformation.

### Can hot spots be predicted without experiments?

Partly. Robetta called 79% of hot spots correctly on 233 mutations, but predictors tested on SKEMPI 2.0 generalized poorly to unseen mutations and favored destabilizing predictions. Use them to rank candidates and test the top few.

### Does the sum of single alanine ΔΔG values equal the effect of mutating them all?

Not reliably. In hGH, mutating several peripheral residues to alanine together raised affinity up to tenfold, which no sum of single results would predict.
