How to Read a Kyte–Doolittle Hydropathy Plot

The Kyte–Doolittle scale

In 1982 Jack Kyte and Russell Doolittle published a hydropathy index for the 20 amino acids and a simple way of using it: average the index over a short stretch of sequence, slide the stretch along the chain, and plot the result. The method became one of the most cited in biochemistry, mainly because it made it possible to spot membrane-spanning segments in a sequence long before structures of membrane proteins were available.

The scale runs from +4.5 for isoleucine, the most hydrophobic residue, to −4.5 for arginine, the most hydrophilic. Kyte and Doolittle built it from two kinds of data: the free energies of transferring amino acid side chains from water into the vapour phase, measured by Wolfenden and colleagues, and the observed tendency of each residue to be buried inside or exposed on the surface of proteins of known structure. Where the two sources disagreed, the authors adjusted the values by judgement, so the scale is empirical rather than a direct physical measurement.

Two values often surprise people. Tryptophan, a large aromatic residue that is certainly not water-loving, scores slightly negative (−0.9), because its indole NH can hydrogen-bond and it is often found at membrane and protein surfaces rather than deep in the core. Cysteine scores strongly positive (+2.5), reflecting how often it is buried, frequently in disulfide bonds.

How the sliding window works

A single residue's score is too noisy to be useful: a hydrophobic segment can contain a glycine or a threonine without changing its character. The plot therefore uses the average over a window of w consecutive residues:

The window is odd-sized so that each average can be assigned to the residue at its centre. Some programs instead assign it to the first residue of the window, which shifts the whole profile by (w−1)/2 positions. The shape is identical, but peak positions differ between tools by that offset, so check which convention a figure uses before comparing residue numbers.

Averaging has a cost at the ends. The first and last (w−1)/2 residues have no complete window around them, so a profile with window 9 loses four residues at each end, and one with window 19 loses nine. For a 26-residue peptide, a window of 19 leaves only 8 points to plot.

Worked example: melittin, window 9

Melittin is the main toxic component of honeybee venom, a 26-residue peptide with the sequence GIGAVLKVLTTGLPALISWIKRKRQQ. It is a useful teaching example because its two halves are so different.

From residue 5 to about 17 the average stays between +0.7 and +2.0: this is the hydrophobic stretch that inserts into lipid membranes. The dip around residues 10–11 comes from the polar and helix-breaking residues Thr10, Thr11, Gly12 and Pro14; the proline also puts a kink into melittin's helix. After residue 18 the profile falls steeply to −2.33, driven by the cationic C-terminal tail KRKRQQ. This combination, a hydrophobic segment followed by a cluster of positive charges, lets melittin bind to negatively charged membrane surfaces and then insert into the bilayer.

Choosing the window size

The window acts as a filter. Short windows follow the sequence closely and highlight local features; long windows smooth them away and show only broad trends. Kyte and Doolittle tested a range of sizes against proteins of known structure and made two recommendations that are still the standard:

The number 19 has a physical basis. An α-helix rises about 1.5 Å per residue, so roughly 20 residues span the 30 Å hydrophobic core of a lipid bilayer. Averaging over 19 residues therefore asks a direct question: is there a stretch long and hydrophobic enough to cross the membrane as a helix?

Finding transmembrane segments

Glycophorin A, a protein of the red blood cell membrane, has a single transmembrane helix and is one of the classic test cases for hydropathy analysis. A synthetic peptide covering its transmembrane region, EPEITLIIFGVMAGVIGTILLISYGIRRLIKK, gives a textbook profile with a window of 19:

The same analysis applied to melittin illustrates the method's main weakness with short peptides. With a window of 19, melittin produces only eight data points, and one of them reaches 1.65, just over the threshold. Taken at face value, that would predict a transmembrane helix. In reality, at low concentrations melittin lies parallel to the membrane surface, and only at higher concentrations does it assemble into pores. A single window crossing the threshold in a short peptide is not evidence of membrane spanning.

GRAVY: one number for the whole sequence

The grand average of hydropathy (GRAVY), also introduced by Kyte and Doolittle, is simply the mean index over every residue in the sequence. Positive values indicate an overall hydrophobic chain; negative values a hydrophilic one. Most soluble proteins fall between about −1 and 0.

GRAVY is a rough guide to handling behaviour. Peptides with strongly positive values are often hard to dissolve in water and tend to aggregate, and they elute later from reversed-phase HPLC columns. It is only a first approximation: retention and solubility also depend on charge, length and secondary structure.

GRAVY also shows why an average can hide what matters. Melittin, one of the most membrane-active peptides known, has a GRAVY of only +0.27, because its hydrophobic half and its cationic tail cancel out.

What hydropathy plots cannot show

Amphipathic helices

Many membrane-active peptides are amphipathic helices: hydrophobic residues line one face of the helix and charged residues the other. Because an α-helix turns about 100° per residue, hydrophobic residues in such a helix recur every three or four positions. A sliding window averages across both faces and cancels them out.

Magainin 2 from frog skin, GIGKFLHSAKKFGKAFVGEIMNS, is a clear example. It forms a strongly amphipathic helix on bacterial membranes, yet its Kyte–Doolittle profile with a window of 9 never rises above +0.9. Detecting this kind of structure requires the hydrophobic moment, introduced by Eisenberg and colleagues in 1982, which measures how unevenly hydrophobicity is distributed around the helix axis rather than how much of it there is. The guide to antimicrobial peptides shows a helical wheel of magainin 2 that makes this segregation visible.

Signal peptides

The N-terminal signal sequences that target proteins for secretion contain a hydrophobic core of 7–15 residues. On a hydropathy plot they look much like a transmembrane segment near the start of the chain. Distinguishing the two needs dedicated predictors such as Phobius or SignalP, which model the charged N-region and the cleavage site as well.

Structure and context

The plot treats each residue independently of its neighbours, its position in three dimensions and its environment. It cannot predict secondary structure, and it will score a buried hydrophobic core of a soluble protein in the same way as a membrane anchor. For transmembrane topology, modern tools such as TMHMM, DeepTMHMM and Phobius are considerably more accurate. The Kyte–Doolittle plot remains valuable as a fast, transparent first look.

Other hydrophobicity scales

Dozens of scales have been published, each derived from different data and suited to different questions. Four are especially common:

• Hopp–Woods (1981) is a hydrophilicity scale designed to find antigenic sites, which are usually exposed and hydrophilic. It is typically used with a window of 6.
• Eisenberg consensus (1984) is a normalised average of several earlier scales and is the usual choice for hydrophobic moment calculations.
• Wimley–White (1996) is based on measured free energies of transfer of peptides from water into a lipid bilayer interface and into octanol, which makes it directly relevant to membrane-active peptides.
• Hessa et al. (2005), sometimes called the biological hydrophobicity scale, measures how each residue affects the insertion of a helix into the ER membrane by the Sec61 translocon.

Different scales rank some residues differently, particularly Trp, Tyr, Gly and Pro, so profiles made with different scales should not be compared value by value.

Common mistakes

1. Applying the 1.6 threshold to a window other than 19.
2. Interpreting a profile of a short peptide as though it were a protein, when only a handful of windows fit.
3. Comparing residue positions between tools that use different window-assignment conventions.
4. Concluding that a peptide is not membrane-active because its profile is flat. Amphipathic helices need a hydrophobic moment analysis.
5. Reading a hydrophobic N-terminal peak as a transmembrane helix without considering a signal peptide.

Frequently asked questions

References

• Kyte J, Doolittle RF (1982) A simple method for displaying the hydropathic character of a protein. Journal of Molecular Biology 157:105–132.
• Wolfenden R, Andersson L, Cullis PM, Southgate CCB (1981) Affinities of amino acid side chains for solvent water. Biochemistry 20:849–855.
• Hopp TP, Woods KR (1981) Prediction of protein antigenic determinants from amino acid sequences. PNAS 78:3824–3828.
• Eisenberg D, Weiss RM, Terwilliger TC (1982) The helical hydrophobic moment: a measure of the amphiphilicity of a helix. Nature 299:371–374.
• Eisenberg D, Schwarz E, Komaromy M, Wall R (1984) Analysis of membrane and surface protein sequences with the hydrophobic moment plot. Journal of Molecular Biology 179:125–142.
• Wimley WC, White SH (1996) Experimentally determined hydrophobicity scale for proteins at membrane interfaces. Nature Structural Biology 3:842–848.
• Hessa T, Kim H, Bihlmaier K, et al. (2005) Recognition of transmembrane helices by the endoplasmic reticulum translocon. Nature 433:377–381.

Related guides

Browse all 27 guides →