structural-biology/ai in a supporting role/BMC Biology 2023 · v2
Cryo-EM maps and AlphaFold models explain how two bacterial helper proteins activate an enzyme
Researchers purified the bacterial proteins NorQ and NorD, imaged them by electron microscopy at low resolution, and used AlphaFold to predict their structures. The predictions were fitted into the blurry density, and mutations tested the arrangement they suggested.
spectrum · one line per step, placed by what the step does · bright lines used AI
Insights into the structure-function relationship of the NorQ/NorD chaperones from Paracoccus denitrificans reveal shared principles of interacting MoxR AAA+/VWA domain proteins
BMC Biology, 2023
doi:10.1186/s12915-023-01546-w · record aix-00092 v2 · checked 2026-10-08
- AI was for
- Structure determination
- Model family
- Transformer
- Checked by
- Experimental
- Code
- not reported
AI processed or interpreted data, but the main finding does not rest on it.
What this research was about
Many enzymes cannot assemble themselves. They need helper proteins, called chaperones, to fold them correctly or to insert the metal atoms at their core. One such enzyme in the bacterium Paracoccus denitrificans is cNOR, which converts nitric oxide as part of how these bacteria breathe without oxygen. Two partner proteins, NorQ and NorD, are needed for cNOR to work. NorQ belongs to a family of motor proteins that form a ring of six copies and burn ATP, the cell's chemical fuel, to pull or push on other proteins. How NorQ and NorD grip each other, and how they reach cNOR, was not clear.
The authors set out to see the NorQ-NorD pair directly. They purified the two proteins from engineered bacteria and froze them in thin ice for electron microscopy, a technique that averages images of many individual particles into a three-dimensional map. The maps they obtained were blurry, roughly 8 ångström for the NorQ ring alone and roughly 10 ångström for the pair, which shows overall shape but not the individual chemical groups. They then tested the resulting picture by deliberately altering single building blocks in the proteins and measuring whether cNOR still worked.
Where AI came in
AlphaFold, a neural network that predicts a protein's three-dimensional shape from its sequence of amino acids, filled the gap the microscopy left. Run through the ColabFold web service with default settings, it predicted the shape of full-length NorD, including a floppy linker and a protruding feature the authors call a finger. A version built for assemblies, AlphaFold-multimer, predicted how NorQ and NorD fit together: single copies paired up, NorQ with separate pieces of NorD, and six NorQ chains with NorD. A second tool, FoldDock, produced further models for comparison; only one of its five placed NorD in the centre of the ring.
The predicted pairing was then fitted into the 10 ångström density with conventional modelling software, which does not learn from data. In effect the predictions supplied the atomic detail the experiment could not resolve, and the density said which prediction was consistent with the real particles. The reading that NorD's finger-bearing domain plugs the hole through the middle of the NorQ ring rests on that combination. The authors went on to mutate the residues the model put at the contact surfaces, and those changes stopped cNOR working.
Written by AIxSci from the checked record below, to give context for readers outside the field. It is not part of the record.
The work
Technical · from the record
The authors purified the MoxR AAA+ ATPase NorQ and its partner NorD from Paracoccus denitrificans and determined low-resolution cryo-EM maps, ~8 Å for the NorQ Walker B variant hexamer and ~10 Å for the NorQ-NorD complex. AlphaFold and AlphaFold-multimer were used to predict the structure of NorD and of NorQ-NorD complexes; the predictions placed the NorD VWA domain, carrying a protruding 'finger', in the centre of the NorQ hexameric ring, and were fitted into the cryo-EM density. Mutating the MIDAS residues T534 or D562 in NorD, the Walker B residue E109 in NorQ, or the NorB surface residues D220 or E222 abolished cNOR activity, and the D220A and E222A variants had reduced non-heme iron content. On this basis the authors propose a model in which NorD binds cNOR through its MIDAS site while ATP hydrolysis in NorQ drives conformational change via the VWA finger.
How AI was used
AlphaFold, run through the ColabFold web resource, was used to predict the structure of full-length NorD from its sequence, and AlphaFold-multimer with default settings was used to predict complexes of NorQ with NorD: a monomer-monomer pair, a NorQ chain with the separated N-terminal and VWA domains of NorD, six NorQ chains with the NorD VWA domain, and six NorQ chains with full-length NorD. Multimer models were scored with the predicted TM score. The FoldDock protocol was also run to build complexes and its five models were compared with the experimental data. The resulting NorQ-NorD prediction, with the N-terminal domain omitted, was combined with a NorQ hexamer built by superimposing the predicted NorQ monomer onto the protomers of the ClpX structure PDB 6SFW, and the assembled model was flexibly fitted into the ~10 Å cryo-EM map with iMODFIT. Separately, a homology model of NorQ was built with SWISS-MODEL from PDB 6L1Q and disordered regions of NorD were predicted with the PrDOS server. Cryo-EM particle picking, classification and refinement used conventional software (Xmipp, Relion, Scipion, cryoSPARC).
The shape of the work
Structural · the record, drawn
no AI
Express and purify NorQ, NorD and variants
Physical execution, by hand or by robot.
Co-expression of 6-His-tagged NorD and untagged NorQ led to purification of the NorQD via affinity chromatographywhere the paper describes this · verbatim
no AI
Collect single-particle cryo-EM data
Obtaining raw data, whether by measurement, download or retrieval.
Data acquisition of NorQD was performed on a Titan Krios G3i (Thermo Fisher Scientific) microscope using a K3 detectorwhere the paper describes this · verbatim
no AI
Pick, classify and refine particles into 3D maps
Cleaning, filtering, normalising or labelling data already obtained.
From 23018 micrographs, ~14 million particles were blob-picked on-the-fly, but only ~500,000 were selected using 2D classificationwhere the paper describes this · verbatim
AI
Predict the structure of full-length NorD
Running a trained model over new data to predict, classify or score. The AI stood in for unresolved measurement.
we first predicted its structure using the recently developed tool AlphaFoldwhere the paper describes this · verbatim
AI
Predict NorQ-NorD complex models
Running a trained model over new data to predict, classify or score. The AI stood in for unresolved measurement.
For predicting different versions of a complex between NorQ and NorD, we used AlphaFold-multimer.where the paper describes this · verbatim
no AI
Build and fit the 6NorQ-NorD model into the cryo-EM map
Extracting understanding from model behaviour.
The 6NorQ-NorD model was flexibly fitted into the cryo-EM map using iMODFIT.where the paper describes this · verbatim
no AI
Test proposed interaction residues by mutagenesis and activity assays
Testing outputs against ground truth.
Among four candidate residues, only mutation of D220 or E222 abolished cNOR activity completelywhere the paper describes this · verbatim
What the record says
Technical · every part carries its own basis
+ in the paper~ our reading− not reported
How to read the quotations. A quotation shows where the paper describes something. It does not quote every value beside it: one passage locates a part of the work, and values without their own quotation are our reading of that passage.
AlphaFold predictions were used to interpret low-resolution cryo-EM density and to build the NorQ-NorD model; the density, the biochemistry and the mutagenesis are experimental, but the assignment of the pore density and the 'finger' feature rests on the predictions
built using AlphaFold-multimer original or updated versions with default settingswhere the paper describes this · verbatim
however, for the resulting models only one out of five had the NorD in the center of the NorQ hexamerwhere the paper describes this · verbatim
Additional file 5. AlphaFold-multimer models related to Additional file 1: Table S2.where the paper describes this · verbatim
What this paper did not report
Technical · absence is published deliberately
- CodeWhether the code is available is not stated.
- Trained model weightsWhether the trained model is available is not stated.
- ComputeThe hardware or time used is not stated.
- How many were testedThe paper gives no count of what was tested.
- Version of AlphaFold (via ColabFold)Which version of the model was used is not stated.
- Version of AlphaFold-multimerWhich version of the model was used is not stated.
- Version of FoldDockWhich version of the model was used is not stated.
About this article
Record aix-00092, version 2, checked by a person on 2026-10-08. The record describes the paper; it does not assess whether the paper's findings are right. The paper is published under CC-BY; quotations are at most 25 words. How we work · Report an error