An alanine scan replaces one residue at a time with alanine and measures what each replacement does to binding. The output is a list of positions and numbers. Those numbers are differences between two molecules measured under one set of conditions, not properties that belong to residues, and most misreadings of a scan begin by forgetting that.
This guide covers what a single alanine substitution physically removes, how an affinity ratio becomes a free-energy difference and what that decomposition assumes, the hotspot and O-ring picture and the evidence actually underneath it, the failure modes that produce confident but wrong residue assignments, and why removing a side chain and removing a residue are different experiments.
What one alanine substitution removes
Alanine truncates a side chain at the beta carbon. Everything past C-beta goes; the backbone, the chain length, the register and the L-configuration stay. That is the design logic — the smallest substitution that deletes side-chain chemistry without deleting the residue. Alanine is not nothing, though: the methyl group remains, so a scan reports the cost of exchanging a side chain for a methyl, not the cost of exchanging it for vacuum. This framing is standard bench convention rather than a published finding, and it carries three positions the method cannot address. Glycine has nothing beyond C-alpha to remove. Proline's side chain is part of its own backbone constraint, so replacing it changes conformational freedom as well as contacts. Cysteines in disulfides cannot be truncated without also breaking a crosslink. Competent scans skip these positions and say which ones they skipped.
The method was introduced on human growth hormone, where 62 single alanine mutations were made at residues within three discontinuous segments implicated in receptor recognition, and the authors found a cluster of a dozen large side chains that each showed more than four times lower binding affinity to the receptor when mutated [1]. Two features of that first report are still the template. First, the scan was targeted at three discontinuous segments already implicated in receptor recognition, not at the whole protein. Second, it carried a folding control: the mutants cross-reacted strongly with seven different conformationally sensitive monoclonal antibodies, so the authors could report the overall folding as indistinguishable from wild type [1]. A scan without a fold control is not interpretable, because a substitution that unfolds a domain and a substitution that removes a contact both reduce binding.
The same study also found at least one position, Glu174, where replacement by alanine increased receptor affinity by more than a factor of four [1]. Scans produce these; the sign of the effect is an experimental result, not an assumption.
From an affinity ratio to a free energy
Binding data are usually converted to a free-energy difference, conventionally written so that a positive value means the mutant binds more weakly. The conversion is standard thermodynamics and is stated here without citation: at 298 K, RT is about 0.59 kcal/mol, so a tenfold change in dissociation constant corresponds to about 1.36 kcal/mol. Checking the direction in both senses: a fourfold loss of affinity is about 0.82 kcal/mol, a 1.0 kcal/mol penalty is about a fivefold loss, and a 2.0 kcal/mol penalty is about a thirtyfold loss.
That arithmetic matters because different studies draw the hotspot line in different places, and the lines are far apart. The original hGH scan used a fourfold affinity criterion [1], which is about 0.8 kcal/mol — considerably more permissive than the 2.0 kcal/mol figure used elsewhere. One computational benchmark, for example, groups residues whose alanine mutation raises the binding free energy by more than 2.0 kcal/mol together as warm and hot spots [2]. Neither cutoff is a physical constant. Both are reporting conventions, and a scan that calls residues hotspots without naming its threshold has not reported a result.
Aggregated measurements exist. SKEMPI 2.0 contains manually curated binding data for 7,085 mutations in structurally resolved protein-protein interactions, including kinetic changes for 1,844 mutations and enthalpy and entropy changes for 443 [3]. It also records 440 mutations that abolish detectable binding [3] — about 6% of the curated 7,085, taking the full curated set as the denominator. Those 440 are censored values: the assay reports that the affinity fell below its detection floor, which bounds the free-energy penalty from below but does not measure it. Reporting such a position as a specific large number, rather than as a lower bound tied to the assay's limit, overstates what was measured.
Hotspots: the energy is not spread evenly
The central empirical result of alanine scanning is that contact area and binding energy are largely decoupled. In the growth hormone system, the crystal structure shows about 30 side chains from each protein in contact, yet individually replacing contact residues in the receptor's extracellular domain with alanine showed that a central hydrophobic region dominated by two tryptophans accounts for more than three-quarters of the binding free energy [4]. The authors named that small subset the functional epitope and described it as surrounded by less important contact residues that are generally hydrophilic and partially hydrated [4]. The structural epitope and the functional epitope are not the same object.
A survey compiled from 2,325 measured alanine mutants reached the same conclusion in aggregate: at the level of side chains there is little correlation between buried surface area and free energy of binding, and the free energy is not evenly distributed across interfaces but concentrated in a small subset of residues, enriched in tryptophan, tyrosine and arginine [5]. Note the scope — this is a composition trend across a heterogeneous compiled database, not a rule that predicts which tryptophan in a given interface is energetic.
Individual interfaces bear this out with specific numbers. A comprehensive experimental and computational scan of the interface between NEMO and IKK-beta reported that the NEMO-binding domain region of IKK-beta holds the highest concentration of hotspot residues, the strongest of them W739, W741 and L742, with penalties of 4.3, 3.5 and 3.2 kcal/mol, and identified two further hotspot regions centered on L708/V709 and L719/I723 that had not previously been recognized [6]. The largest of those, 4.3 kcal/mol, corresponds to roughly a 1,400-fold loss of affinity by the conversion above.
The O-ring picture and what it rests on
The O-ring hypothesis is the structural gloss on the hotspot finding: energetic residues sit in a densely packed center, and the ring of less important residues around them most likely serves to occlude bulk solvent, with occlusion of solvent reported as a necessary condition for highly energetic interactions [5]. It is a useful picture, and it is worth knowing what supports it. One follow-up study characterizes the hypothesis as resting mainly on accessible-surface-area analysis of 23 static crystallographic structures of protein complexes, and points out that protein flexibility allows temporary exposure of buried interfacial groups that a static surface calculation cannot see [7]. Those authors then ran molecular dynamics on the lysozyme HEL / antibody FvD1.3 complex and reported that within a flexible, dynamic framework the warm and hot spot residues were kept sheltered from bulk solvent throughout the simulation [7]. That is one complex, in silico, and it is consistent with the hypothesis rather than a general proof of it.
Additivity: the assumption underneath decomposition
Calling a per-residue number a contribution assumes the parts add. Double mutant cycles test that directly: mutate two residues singly and together, and compare the double effect against the sum of the singles. In the barnase-barstar interface, cycles built between a subset of five barnase and seven barstar residues showed that coupling energy decreases with distance — residues separated by less than 7 angstroms generally interact cooperatively, while at greater separations the effects of mutation are additive and the interactions are energetically independent [8]. The highest coupling energies in that system were between pairs of charged residues, from 1.6 to 7 kcal/mol, and three of the six strongest interactions detected had not been noted previously from the crystal structure [8].
Across a larger set, additivity holds on average and fails often enough to matter. An analysis of double mutants found a correlation of 0.90 between the double-mutant binding free-energy change and the sum of the corresponding singles, but a mean absolute deviation of 0.86 kcal/mol, with 25.6% of the double mutants deviating by more than 1 kcal/mol and therefore classified as non-additive; the non-additive pairs tended to be closer together and to have more contacts [9]. So roughly one double mutation in four, in that dataset, is not the sum of its parts — and 1 kcal/mol is about a fivefold error in affinity. Additivity is a working approximation whose failures cluster exactly where scans are most often used, at short residue-residue separations inside a hotspot.
Where alanine scans mislead
The most durable lesson from antibody-antigen work is that contact is not coupling. In the D1.3 antibody / HEL system, 12 of the 13 non-glycine contact residues on HEL were individually truncated to alanine, and only four positions showed a penalty greater than 1.0 kcal/mol, with Gln121 the most critical at 2.9 kcal/mol [10]. Double mutant cycles on 14 residue pairs, 10 of them in direct contact in the crystal structure, found that most cycles between directly contacting residues showed no significant coupling, including several hydrogen-bonded pairs that made no net contribution to complex stabilization [10]. The authors' conclusion is worth stating in their terms: the simple fact that two residues are in direct contact in a protein-protein interface cannot be taken as evidence that there necessarily exists a productive interaction between them [10].
Second, cavities get filled. In that same system, the HEL Asp18 to Ala mutation cost only 0.3 kcal/mol — about a 1.7-fold change — despite losing a hydrogen bond and seven van der Waals contacts, and a 1.5 angstrom structure showed three additional water molecules stably incorporated at the site of the mutation, filling the cavity and forming part of a rearranged solvent network [10]. A small measured penalty can mean the side chain did little, or it can mean water substituted for it.
Third, a residue can be energetically important without touching anything. In the NEMO/IKK-beta scan, residues D738 and S740 were important for binding but make no direct contact with NEMO, and the authors interpret them as likely acting to stabilize the active conformation of the surrounding residues [6]. In the same study, the computational fragment-mapping approach accurately quantified the energetic importance of all hotspot residues that make direct contact [6] — a boundary worth reading carefully, since indirect contributors are precisely what contact-based reasoning misses.
Fourth, structure can survive what stability does not. Scanning polyalanine mutations made within the strands and turns of the beta-sheet region of T4 lysozyme substantially reduced the stability of every construct, yet crystal structures showed each mutant retained the beta-sheet conformation [11]. Those were multi-residue polyalanine constructs rather than single substitutions, so the number does not transfer to a one-position scan; the transferable point is that a circular dichroism spectrum or a retained fold is not evidence that a substitution was energetically neutral.
Alanine substitution is not deletion
Deleting a residue is a different perturbation with different consequences. Substitution preserves chain length, so every downstream residue keeps its position, spacing and register; deletion shortens the chain, shifting everything after the deleted position and, in a helix, rotating the downstream face. That geometric distinction is definitional rather than a finding, and it means an alanine scan and a deletion scan answer different questions: the first asks what a side chain is worth, the second asks whether a position's presence and spacing are required.
Empirically, where both length-changing mutation classes have been scanned comprehensively, deletions are the less tolerated of the two, and tolerance varies from region to region. A near-comprehensive analysis of single amino acid insertions and deletions in TEM-1 beta-lactamase found insertions and deletions to be largely deleterious, with partially overlapping deletion-tolerant and insertion-tolerant regions throughout the protein, especially in unstructured regions and at the ends of helices; most regions tolerated insertions more than deletions [12]. That is a functional-selection readout in one bacterial enzyme, not a binding measurement, so it speaks to structural tolerance rather than to interface energetics.
Deletion scans are also harder to characterize. One study deleted every residue in turn from a small alpha-helical sterile alpha motif domain and used two-dimensional NMR spectroscopy and differential scanning fluorimetry to follow the structural and thermodynamic consequences, then benchmarked computational protocols for classifying the results [13]. The reason for that instrumentation is the point: with deletion you cannot assume the fold survived, so the structural readout is part of the experiment rather than a control on it.
Computed scans and their error bars
Computational alanine scanning estimates the same quantity from structures and simulation. One published protocol reported a mean unsigned error of 0.80 kcal/mol over a set of 46 mutations in three complexes, an overall success rate of 80%, and 82% for residues whose alanine mutation raises the binding free energy by more than 2.0 kcal/mol [2]. That protocol requires three different internal dielectric constants chosen according to the type of amino acid being mutated [2] — a tuned parameter, and one reason results from different implementations are not interchangeable. An error near 0.8 kcal/mol is roughly a fourfold uncertainty in affinity, which is adequate for ranking a strong hotspot against a neutral position and inadequate for resolving two residues a few tenths of a kcal/mol apart.
What is not settled
The hotspot concept is well supported as a description of measured interfaces and much weaker as a predictor. Solvent occlusion has been reported as a necessary condition for high-energy interactions [5], but necessary is not sufficient, and the underlying surface analysis was drawn from a small set of static structures [7]. Additivity fails for a substantial minority of residue pairs and fails preferentially at short separations [9]. Contact geometry does not identify energetic pairs [10], and non-contacting residues can carry energy [6]. Where a scan is used to define a functional epitope, the honest report names the assay and its detection floor, the threshold used to call a hotspot, the fold control, the positions skipped, and which residues were tested singly versus in combination — and treats every unmeasurable position as a bound rather than a value.