What question does this answer? How different is the QTY variant's 3D structure from the native receptor's crystal structure? Far-UV CD tells us the QTY variants are helical, but helicity is a global, low-resolution property — a bundle of seven helices wrongly assembled can still give a clean α-helical spectrum. All-atom molecular dynamics (MD) answers the harder question at atomic resolution: does the variant, stripped of its membrane, still adopt the same fold as the receptor it was derived from?
1. The comparison we are actually making
The QTY code swaps the bulky hydrophobic residues — Leu, Ile, Val, Phe — for Thr, Gln and Tyr. Those are exactly the side chains that anchor a transmembrane helix in a lipid bilayer, so removing them is what makes the receptor water-soluble without detergent. The worry is that the same substitution could loosen the helix–helix packing that defines a chemokine receptor fold. To test that, we need a structural model of the variant and something to compare it against. The reference points are experimental structures:
| Receptor | Native reference | Variant model source | Sequence identity |
|---|---|---|---|
| CCR5 | PDB 4MBS | GOMoDo homology model | 78.12% (CCR5QTY vs CCR5) |
| CXCR4 | PDB 3ODU | GOMoDo homology model | ~70.74% (CXCR4QTY vs CXCR4) |
The initial QTY models came from the GOMoDo GPCR modelling server, which runs a homology-modelling plus ligand-docking pipeline: it threads the variant sequence onto a template GPCR structure, rebuilds loops, and docks the native ligand into the binding pocket. Identities of 78.12% and ~70.74% are high enough that the template backbone places the variant helices reliably, but low enough that side chains and loops are genuine predictions rather than copies.
2. Simulation setup: from model to dynamics
A homology model is a static guess; MD turns it into an ensemble and lets us ask whether the guess is a stable state. Simulations were run in YASARA using the AMBER14 (ff14SB) force field with YASARA's self-parameterising machinery, which fills in bonded and non-bonded terms for non-standard groups automatically rather than requiring hand-built parameter files.
Each system was placed in an explicit water box at 24.85 °C (298 K), pH 7.4 and 0.9% NaCl to approximate physiological ionic strength. Four proteins — CCR5QTY, CXCR4QTY, CCR10QTY and CCR7QTY — were each simulated for 1 µs. Explicit solvent is not optional: a QTY variant designed to be water-soluble must be solvated in water, because the whole hypothesis being tested is that water replaces the membrane as the medium that keeps the fold together.
3. What RMSD means and how to read it
Root-mean-square deviation (RMSD) is the average distance between corresponding atoms of two optimally superimposed structures, the standard single number for "how different are these conformations". Here the comparison is a superposition of a simulated variant on a native crystal structure, restricted to the parts that correspond.
A critical detail makes the comparison fair. The 4MBS and 3ODU crystals contain fusion inserts in the third intracellular loop — rubredoxin in the CCR5 structure, T4 lysozyme in the CXCR4 structure — added by crystallographers because flexible intracellular loops prevent lattice formation. These inserts are artefacts of crystallisation, not parts of the receptor. The QTY variants contain no such insert, so the fusion proteins were computationally removed before superposition; otherwise RMSD would be dominated by a 100-residue difference unrelated to the QTY code.
The result: CCR5QTY (teal) superimposed on native CCR5 (magenta) deviates by ~1.9 Å, and CXCR4QTY (blue) against native CXCR4 (green) by ~2 Å, shown from two side views and a top view. The top view matters most: looking down the membrane normal, the seven-helix bundle topology is unchanged — helices retain their relative angular arrangement rather than splaying apart. Roughly 2 Å is on the order of a Cα displacement seen between independent structures of the same protein, far smaller than the tens of ångströms a fold change would produce.
4. Honest framing of the result
A ~2 Å all-atom deviation sits inside the normal fluctuation band for "homology model versus experimental structure", and it supports the claim that QTY substitution does not break the overall fold. But the conclusion has a hard boundary worth stating plainly: the authors only ran MD from a homology-modelling start. Predicting a structure and then simulating that prediction can confirm self-consistency without ever confirming against a new experimental reference. There was no cryo-EM or crystal structure of the QTY variants to compare against — a technical boundary of the time, not a flaw in the reasoning, and the motivation for later structural work.
5. What 1 µs does and does not prove
One microsecond is enough to show the modelled fold is stable in a physiological-like environment — that it does not unfold, collapse or unravel within that window. It is not enough to sample large conformational transitions. GPCRs have activation pathways, loop rearrangements and ligand-induced shifts on timescales that can exceed microseconds, and a single 1 µs trajectory samples one basin. The statement the MD licenses is therefore narrow: the QTY variant is a stable, seven-helical, native-like fold in water. Anything broader — about dynamics, activation or allostery — is not supported by this dataset.
Troubleshooting / pitfalls
- Force-field limits. ff14SB was parameterised on soluble proteins; a helical bundle designed to dissolve in water sits outside its training distribution. Small systematic biases in helix packing are expected, and a low RMSD should not be read as force-field validation.
- Under-sampling. 1 µs per protein, one trajectory each, is a single sample of a broad ensemble. Agreement in one run does not rule out an alternative, biologically relevant state. Replicates are the cheapest fix.
- RMSD versus RMSF. RMSD is a global number and can hide compensating motions — one loop moving out while another moves in. Per-residue RMSF (root-mean-square fluctuation) localises which regions are actually mobile, and loops and termini dominate it if not trimmed.
- Superposition bias. The measured deviation depends on which atoms are aligned. Fit on the transmembrane bundle, not flexible loops, or you report loop motion as global drift; state the atom selection explicitly.
- Over-interpreting a single trajectory. One 1 µs run is a movie, not a distribution. Never quote one RMSD as a thermodynamic property.
- Fusion-protein handling. Rubredoxin/T4 lysozyme are crystallisation aids; forgetting them inflates RMSD enormously.
- Starting-state dependence. MD relaxes toward the nearest basin. A wrong template gives a wrong basin, and the simulation reports it as stable.
Taken with the far-UV CD result, the MD picture is coherent: the QTY variants keep their helical bundle, in the atomistic detail of helix packing. One experiment asks whether the protein is helical; the other asks whether the helices are where they should be. Together they argue that detergent-free solubilisation preserved the fold, not merely the sequence.
References
- Zhang S, Tao F, Qing R, et al. QTY code enables design of detergent-free chemokine receptors that retain ligand-binding activities. Proc Natl Acad Sci USA. 2018;115(37):E8652–E8659. doi:10.1073/pnas.1811031115
- Maier JA, Martinez C, Kasavajhala K, et al. ff14SB: improving the accuracy of protein side chain and backbone parameters from ff99SB. J Chem Theory Comput. 2015;11(8):3696–3713. doi:10.1021/acs.jctc.5b00255
- Krieger E, Vriend G. YASARA View—molecular graphics for all devices—from smartphones to workstations. Bioinformatics. 2014;30(20):2981–2982. doi:10.1093/bioinformatics/btu426
- Waterhouse A, Bertoni M, Bienert S, et al. SWISS-MODEL: homology modelling of protein structures and complexes. Nucleic Acids Res. 2018;46(W1):W296–W303. doi:10.1093/nar/gky427
- Pándy-Szekeres G, Munk C, Tsonkov TM, et al. GPCRdb in 2018: adding GPCR structure models and ligands. Nucleic Acids Res. 2018;46(D1):D440–D446. doi:10.1093/nar/gkx1109
- Greenfield NJ. Using circular dichroism spectra to estimate protein secondary structure. Nat Protoc. 2006;1(6):2876–2890. doi:10.1038/nprot.2006.202
- Krogh A, Larsson B, von Heijne G, Sonnhammer ELL. Predicting transmembrane protein topology with a hidden Markov model: application to complete genomes. J Mol Biol. 2001;305(3):567–580. doi:10.1006/jmbi.2000.4315
- Kelly SM, Jess TJ, Price NC. How to study proteins by circular dichroism. Biochim Biophys Acta. 2005;1751(2):119–139. doi:10.1016/j.bbapap.2005.06.005
- Jerabek-Willemsen M, André T, Wanner R, et al. MicroScale Thermophoresis: interaction analysis and beyond. J Mol Struct. 2014;1077:101–113. doi:10.1016/j.molstruc.2014.03.009