AFM Data in ChimeraX

A structural biology compatible file format for atomic force microscopy. Jiang Y, Wang Z, Scheuring S. Nat Commun. 2025 Feb 15;16(1):1671.

  • Simon Scheuring is a Professor of Physiology and Biophysics in Anesthesiology, Weill Cornell Medicine Graduate School of Medical Sciences

  • cryo-EM, crystallography, and NMR structural data have established file formats, crucial for sharing and methods development

  • atomic force microscopy (AFM) has lacked a standard format for interfacing with structural biology methods and is generally published as 2D image panels

  • recent developments allow extracting high-resolution structural data

  • the authors propose a 3D density format (.afm) based on MRC, and have developed a pipeline for generating it from the raw AFM data and a ChimeraX bundle for reading it [github]

  • the 3D map format facilitates using structural biology methods developed for cryo-EM such as Chimera(X) visualization and MDFF flexible fitting, as well as inclusion in integrative modeling


[back to paper list]

AFM for Structural Biology: Pros & Cons

  • AFM is a surface technique:
    • molecules have to be surface-adsorbed
    • gives a single height per X,Y position
    • in conventional AFM, topographic data are convoluted/smeared out by tip shape
    • not enough information alone to solve a 3D atomic structure, but may contribute significantly to an integrative, cross-methodology approach
    • allows following surface shape changes at the single-molecule level

  • AFM allows close-to-physiological, dynamic conditions:
    • ambient temperature and pressure (not cryogenic, vacuum, etc.)
    • physiological buffer
    • molecules not constrained by crystal lattice
    • proteins can be reconstituted in lipid membranes of controlled composition
    • within a lipid membrane on freshly cleaved mica, most membrane proteins have been observed to diffuse and change conformation freely

  • recent advances in AFM:
    • high-speed AFM (HS-AFM) allows up to 100 acquisitions per second with shorter cantilevers for much greater sensitivity than conventional AFM
    • localization AFM (LAFM) overcomes convolution by tip shape; topographic peaks are extracted from individual particle observations and merged into a super-resolution probability map approaching atomic resolution
  • Example dataset: annexin V, a membrane protein that forms lattices, detected at 2.5 Å/pixel

  • a: LAFM detections aligned into a x,y,n stack (n=number of detections) in which the values are heights
  • b: detections converted into an x,y,h stack in which the values are counts of how many detections measured that height at that x,y position). Yellow dots are local maxima (counts) for high, low, medium heights.
  • c: top and side views of the resulting 3D stack, darker for higher values
  • d,e: Fourier Shell Correlation (FSC) analysis of detection half-stacks (for electron density, FSC curves estimate isotropic resolution; for LAFM, the interpretation is not so simple, but conceptually similar)

  • ...however, this 3D stack is still a bunch of dots... (movie: afm-detection.avi)

  • a→b: the 3D detection stack (movie: afm-detection.avi) is converted to a 3D density map (movie: afm-density.avi) by applying a 3D Gaussian with σ determined from the FSC analysis. Selected slices shown from highest topography to lowest going from left to right.

  • c: the annexin V monomer has 4 domains, repeats I, II, III, and IV

  • d: molecular surface of trimer colored by height, with dashed line around one monomer

  • e,f: 3D-LAFM density map displayed in Chimera in image style and surface style (height-colorized), respectively

  • g: comparison of monomers from d-f (rotated)

  • h: FSC analysis of masked density half-maps; half-bit wavelength (somewhat like resolution) ~1.4 Å
(movie: afm-annexin-md.mp4)
    a: the AFM map can be used in flexible fitting, with force proportional to the gradient in the ~20-Å slab containing the height range of the map; below that, the "density" is set to a constant background value
    • 60ns simulation with symmetry restraint (the LAFM data is symmetric), with minimization at 20 and 40ns and domain restraints in the first 40ns
  • b: the protein is drawn to the AFM surface within 1ns, and c-e: equilibrium is mostly reached within 10ns
  • f: initial vs. final structure
  • g, histogram on the far right: MDFF reproduced the relative heights of domains from AFM, with a flatter arrangement of the repeats than in the starting, crystal structure (PDB)
  • Example: a membrane transporter with large rearrangements between inward-facing state (IFS) open and IFS closed; AFM allows tracking the state of a single molecule over time

  • a: the transporter has a three transport domains and a central trimerization domain

  • b: molecular surfaces (left) and AFM snapshots (right) colored by height

  • c: main difference is change in tilt angle of the transport domains; the AFM snapshots could be sorted into the two states

  • d-i: in a nutshell, MDFF using a map from the closed-state AFM drives an open input structure to closed, and using a map from the open-state AFM drives a closed input structure to open
    (movie: afm-closed-to-open.mp4)

    (Supplementary data shows that input structures are maintained when flexibly fit to the map based on the same state, suggesting it is not necessary to pre-judge the state of the input structure.)