How to superimpose and align protein structures

One worked example, open and closed adenylate kinase, in PyMOL, ChimeraX and the browser

12 min

A schematic protein with a rigid core and two lime-filled domains closed over it, each paired by lines with a dashed open copy swung upward about its hinge, the way open and closed adenylate kinase superimpose.

To superimpose two protein structures, fit one onto the other by pairing residues that correspond and rotating the first until the paired atoms sit as close as they can. In PyMOL that is align mobile, target, in ChimeraX matchmaker mobile to target, and in the browser it is the Superpose tool of the PDB viewer. Each reports an RMSD, the average distance between the paired atoms after the fit, in ångströms.

The number depends on what you pair and what you throw away, so the same two structures can give very different values. Adenylate kinase from E. coli shows this well. The ligand-free structure 4AKE is open, and 1AKE, crystallised with the bisubstrate inhibitor Ap5A, is closed.[5][6] Both are E. coli adenylate kinase, 214 residues per chain, yet fitting the whole chain gives one RMSD, and fitting only the rigid core gives a much smaller one.

PDB ViewerOpen and inspect protein structure files in a 3D viewer.Open tool

Superposition methods compared

Every method has two stages: decide which residue pairs correspond, then find the rotation and shift that minimise the distance between them. The methods differ in the first stage, and in whether they drop pairs that fit badly.

CommandPairs residues byOutlier rejectionChoose it when
PyMOL alignSequence alignmentYes, up to 5 cycles at a 2.0 Å cutoffThe sequences are similar (PyMOL suggests above 30% identity)
PyMOL superStructure, ignoring sequenceYes, refinement cyclesSequence similarity is low but the folds are alike
PyMOL cealignStructure (CE algorithm)No refinement cyclesLittle or no sequence similarity; slower
ChimeraX matchmakerSequence alignment scored with secondary structureYes, iterated at 2.0 ÅThe same cases as align, plus chain selection options
Viewer SequenceSequence alignmentYes, as PyMOL alignClose homologs, mutants and predicted models of the same protein
Viewer TM-alignStructureBuilt into the TM-align scoreLow sequence identity, or when you want a TM-score
Table 1. Superposition commands and what each does. Sources: PyMOL Wiki (align, super, cealign), UCSF ChimeraX User Guide (matchmaker), ProteinIQ PDB viewer documentation.[1][2][3][4]

Sequence-based methods are fast and give the pairing you expect when the two chains are the same protein, because residue 57 matches residue 57. Structure-based methods find a pairing from the shapes alone, which is the only option for distant relatives, and they can pair residues that the sequences would not.

The PyMOL pages say that align does well above about 30% identity, that super is more robust below that, and that cealign is the most robust on twilight-zone pairs but much slower.[1][2][3] The browser viewer's super and cealign commands keep PyMOL's names and argument order but both run TM-align, because it does not implement the CE algorithm itself.

Open and closed adenylate kinase

Adenylate kinase moves a phosphate between nucleotides, and it does so by closing two lid-like domains over its substrates. The chain splits into a CORE (residues 1 to 29, 60 to 121 and 160 to 214), an NMP domain (30 to 59) and a LID (122 to 159).[7] The core barely changes shape between states, while the lid and NMP domains move relative to it.

That makes 4AKE chain A against 1AKE chain A a good test. A fit over the whole chain tries to satisfy the moving domains and the rigid core together, so none of them lines up. A fit over the core alone should match tightly and leave the lid and NMP domains visibly apart.

How to superimpose structures in the PDB viewer

The PDB viewer opens both entries in the browser and shows the numbers in a side panel. It does not need an account.

  1. Open the PDB viewer and click Command line.
  2. Type fetch 4ake, press Enter, then type fetch 1ake and press Enter. Both structures now appear in the list of structures.
  3. Open the Superpose tool in the analysis group.
  4. Under Move pick chain A of 1AKE, and under Onto pick chain A of 4AKE.
  5. Choose Sequence as the method and click Superpose.

The side panel then lists the RMSD, the TM-score, how many residues were Aligned and the Method. Fitting 1AKE chain A onto 4AKE chain A with Sequence gave an RMSD of ‹N1› Å over ‹N2› aligned residues and a TM-score of ‹N3›, with ‹N4›% sequence identity.

Switching the method to TM-align and clicking Superpose again gives ‹N5› Å over ‹N6› residues and a TM-score of ‹N7›. The two methods pick different pairs and different atoms to drop, so the RMSD differs even though the structures are the same. Move back returns the structure to where it started, so you can try a method, read the numbers and undo it.

Fit only the rigid core

The viewer's command bar accepts PyMOL syntax, so the core fit is one command. Using the residue ranges from the domain definition above:

Text
align 1ake and chain A and resi 1-29+60-121+160-214 and name CA, 4ake and chain A and resi 1-29+60-121+160-214 and name CA

PyMOL prints an RMSD before and after outlier rejection. The same command in the browser fitted ‹N9› Cα atoms to an RMSD of ‹N8› Å. The lid and the NMP domain now sit visibly apart, which is the real difference between the states and what the whole-chain number hid.

How to superimpose structures in PyMOL

Load both entries and run the three commands. Each one moves the first object onto the second, except cealign, which takes its arguments the other way round.

Text
fetch 4ake, async=0
fetch 1ake, async=0
align 1ake and chain A, 4ake and chain A
super 1ake and chain A, 4ake and chain A
cealign 4ake and chain A, 1ake and chain A

align does a sequence alignment, superimposes, then runs up to five cycles that drop pairs further apart than 2.0 Å. The RMSD it prints is for the pairs left after that rejection, so it is lower than the all-atom value. Set cycles=0 to get the RMSD over every aligned atom with no rejection.[1] Add object=aln to draw lines between the paired atoms.

super aligns by structure instead of sequence and then refines in the same way, which is why PyMOL recommends it when sequence identity is low.[2] cealign uses the CE algorithm, which is robust for proteins with little or no sequence similarity.[3][9] The two objects must be different objects. To fit two chains of the same file, extract one chain into its own object first.

How to superimpose structures in ChimeraX

ChimeraX has one command for sequence-based fitting. Open both entries, then match the second model to the first.

Text
open 4ake
open 1ake
matchmaker #2/A to #1/A

matchmaker aligns the two chains' sequences, using secondary structure as well as residue type, then fits one Cα per residue and iterates with a 2.0 Å cutoff.[4] The Log shows the RMSD over the pruned pairs and, because the fit is iterated, the RMSD over all pairs as well. mm #2 to #1 without chains uses the best-scoring chain pair, and bring moves other models along with the one being fitted.

ChimeraX recommends against restricting the residues unless you want to, because full-length alignments tend to be better and the iteration already excludes badly fitting regions.[4] For this pair, a restricted core fit is exactly the case where you do want it.

Reading the numbers

The RMSD is the root mean square distance between paired atoms after the fit. It is in ångströms, lower is closer, and it is reported only over the pairs that were kept, so a lower value from a method that discarded more pairs is not a better match. Always read it with the number of aligned residues next to it. The RMSD guide covers the formula and what values mean.

The TM-score runs from 0 to 1 and does not depend on chain length the way RMSD does, which is why it is the better score when the structures differ in size or in a few flexible regions. Values above about 0.5 mean the same fold. For when each one misleads, see TM-score vs RMSD. TM-score comes from the TM-align algorithm.[8] The viewer reports it after either method, so you can compare the two on the same pair.

Sequence identity is the share of aligned positions with the same residue. 4AKE and 1AKE are the same protein, so it is high by construction. For a pair of homologs, identity tells you whether align is a safe choice or whether to switch to a structure-based method.

Several structures at once

When more than two structures are open, the Superpose tool can fit all of them onto one reference. This is a set of pairwise fits to a single reference, not a true multiple structure alignment that finds one consensus core for the whole set, and the result depends on which structure you pick as reference.

Add an AlphaFold prediction of the same protein, UniProt P69441, by typing fetch P69441 in the command line.[10] Choose All other structures under Move, pick 4AKE as the reference under Onto, and click the Superpose button.

The panel lists each structure's RMSD, TM-score and identity to the reference, and a Cα spread, the distance of each matched Cα from the average position of all the structures after the fit. For this set the mean spread was ‹N10› Å and the maximum ‹N11› Å. Colour by Cα spread paints the structures from blue where they agree to red where they differ, and the red lands on the lid and the NMP domain. The AlphaFold model fitted 4AKE with an RMSD of ‹N12› Å and a TM-score of ‹N13›.

From the command line, alignto 4ake does the same with TM-align and alignto 4ake, align uses the sequence fit. The RMSD matrix CSV button downloads the distance between every pair, which is what you need for a table in a paper. ChimeraX does it with mm #2-3 to #1.

Common mistakes

ProblemSymptomFix
Comparing an RMSD over different numbers of pairsA method looks better only because it dropped more residuesReport the aligned count and the TM-score beside the RMSD
Fitting the whole chain when domains moveHigh RMSD, and no region lines up wellFit the rigid core only, then look at the moving domains
Aligning two selections of one object in PyMOLExecutiveAlign: invalid selections for alignmentextract one chain into its own object first
Swapped argument order in cealignThe wrong structure movesGive the target first, then the mobile structure
Using align on distant homologsA poor pairing and a large RMSDUse super or cealign, or TM-align in the browser
Multi-state files in PyMOLAn NMR ensemble fitted across all statesSet mobile_state and target_state
Table 2. Superposition mistakes that give a misleading number.

After the superposition

A fitted pair is the starting point for a figure. Colour the two chains differently, ray-trace and export, as in how to make a publication-quality protein figure. The commands above are part of the PyMOL commands cheat sheet, which runs in the browser. To check whether a predicted model is close to an experimental one, the RMSD and TM-score here are the numbers to report, with pLDDT beside them to show which regions the model itself trusts.

PDB ViewerOpen and inspect protein structure files in a 3D viewer.Open tool

FAQ

How do I calculate RMSD between two protein structures?

Superimpose them with one of the commands above and read the RMSD from the output. PyMOL prints it after align, ChimeraX in the Log after matchmaker, and the PDB viewer in the Superpose panel. Make sure you know which residues were paired and whether any were rejected.

What is the difference between align, super and cealign in PyMOL?

align pairs residues from a sequence alignment, super pairs them by structure with no use of sequence, and cealign uses the CE algorithm and takes its arguments in the opposite order. align is fastest and best for similar sequences, cealign is slowest and best for the most distant pairs.[1][2][3]

Can I superimpose structures with different sequences?

Yes, with a structure-based method: super or cealign in PyMOL, or TM-align in the PDB viewer. Sequence-based methods need enough identity to align the sequences first.

What RMSD counts as a good match?

It depends on length and on how many residues were paired, so there is no single cutoff. Compare it with the TM-score, which treats a value above about 0.5 as the same fold regardless of size.

How do I align more than two structures?

Pick one as the reference and fit the others onto it. In the PDB viewer that is All other structures under Move, in PyMOL alignto, and in ChimeraX matchmaker all to followed by the reference.