research letters\(\def\hfill{\hskip 5em}\def\hfil{\hskip 3em}\def\eqno#1{\hfil {#1}}\)

IUCrJ
Volume 9| Part 6| November 2022| Pages 728-734
ISSN: 2052-2525

Analytic modeling of inhomogeneous-resolution maps in cryo-electron microscopy and crystallography

crossmark logo

aCentre for Integrative Biology, Institute of Genetics and Molecular and Cellular Biology, Illkirch 67404, France, bDépartement de Physique, Université de Lorraine, Vandoeuvre-lès-Nancy 54506, France, and cInstitute of Mathematical Problems of Biology RAS, Keldysh Institute of Applied Mathematics of Russian Academy of Sciences, Pushchino 142290, Russian Federation
*Correspondence e-mail: sacha@igbmc.fr

Edited by A. Thorn, University of Hamburg, Germany (Received 21 April 2022; accepted 18 August 2022; online 28 September 2022)

Refinement of macromolecular atomic models versus experimental maps in crystallography and cryo-electron microscopy is a critical step in structure solution. For an appropriate comparison, model maps should mimic the imperfections in the experimental maps, mainly atomic disorder and limited resolution, which are often inhomogeneous over the molecular region. In the suggested method, these model maps are calculated as the sum of atomic contributions expressed through a specifically designed function describing a solitary spherical wave. Thanks to this function, atomic contributions are analytically expressed through their atomic displacement parameter and local resolution, a value now associated with each atom. Such a full analytic dependence of inhomogeneous-resolution map values on model parameters permits the refinement of all of these parameters together.

1. Introduction

Macromolecular atomic models are obtained using maps of electron or nuclear scattering density distributions in macromolecular crystallography, or those of an electrostatic scattering potential in cryo-electron microscopy (cryoEM) and microcrystal electron diffraction. Information extracted from these models depends on the accuracy of their parameters. Owing to the `resolution revolution' in cryoEM (Kühlbrandt, 2014[Kühlbrandt, W. (2014). Science, 343, 1443-1444.]) and to recent progress in structure prediction of the protein components of macromolecular complexes (Jumper et al., 2021[Jumper, J., Evans, R., Pritzel, A., Green, T., Figurnov, M., Ronneberger, O., Tunyasuvunakool, K., Bates, R., Žídek, A., Potapenko, A., Bridgland, A., Meyer, C., Kohl, S. A. A., Ballard, A. J., Cowie, A., Romera-Paredes, B., Nikolov, S., Jain, R., Adler, J., Back, T., Petersen, S., Reiman, D., Clancy, E., Zielinski, M., Steinegger, M., Pacholska, M., Berghammer, T., Bodenstein, S., Silver, D., Vinyals, O., Senior, A. W., Kavukcuoglu, K., Kohli, P. & Hassabis, D. (2021). Nature, 596, 583-589.]; Baek et al., 2021[Baek, M., DiMaio, F., Anishchenko, I., Dauparas, J., Ovchinnikov, S., Lee, G. R., Wang, J., Cong, Q., Kinch, L. N., Schaeffer, R. D., Millán, C., Park, H., Adams, C., Glassman, C. R., DeGiovanni, A., Pereira, J. H., Rodrigues, A. V., van Dijk, A. A., Ebrecht, A. C., Opperman, D. J., Sagmeister, T., Buhlheller, C., Pavkov-Keller, T., Rathinaswamy, M. K., Dalwadi, U., Yip, C. K., Burke, J. E., Garcia, K. C., Grishin, N. V., Adams, P. D., Read, R. J. & Baker, D. (2021). Science, 373, 871-876.]), these maps have become especially important to correct and refine initial atomic models (Diamond, 1971[Diamond, R. (1971). Acta Cryst. A27, 436-452.]; Chapman, 1995[Chapman, M. S. (1995). Acta Cryst. A51, 69-80.]; Murshudov, 2016[Murshudov, G. N. (2016). Methods Enzymol. 579, 277-305.]; Afonine, Poon et al., 2018[Afonine, P. V., Poon, B. K., Read, R. J., Sobolev, O. V., Terwilliger, T. C., Urzhumtsev, A. & Adams, P. D. (2018). Acta Cryst. D74, 531-544.]; Urzhumtsev & Lunin, 2019[Urzhumtsev, A. G. & Lunin, V. Y. (2019). Crystallogr. Rev. 25, 164-262.]; Yamashita et al., 2021[Yamashita, K., Palmer, C. M., Burnley, T. & Murshudov, G. N. (2021). Acta Cryst. D77, 1282-1291.]; Roversi & Tronrud, 2021[Roversi, P. & Tronrud, D. E. (2021). Acta Cryst. D77, 1497-1515.]; Palmer & Aylett, 2022[Palmer, C. M. & Aylett, C. H. S. (2022). Acta Cryst. D78, 136-143.]) and to validate the results (Helliwell, 2022[Helliwell, J. (2022). IUCr Newsl. 30(1).]). The experimental maps are subject to dynamic and static atomic positional disorder and are available at limited resolution, which often varies from one macromolecular region to another (Cardone et al., 2013[Cardone, G., Heymann, J. B. & Steven, A. C. (2013). J. Struct. Biol. 184, 226-236.]) (Fig. 1[link]). In order to refine an available model, we define a respective score function by comparing a map calculated from the model with the experimental one and minimize it by varying the model parameters. For an appropriate quantitative comparison, the model map should mimic imperfections in the experimental map. If the model map values are expressed analytically through the model parameters, this could drastically simplify the map calculation and model optimization. Since mathematically the problem and its solution are the same for all methods of structure determination, herein we use the term `density' for both the electrostatic potential and the electron-density distributions.

[Figure 1]
Figure 1
Maps of inhomogeneous resolution. (a) A fragment of a cryoEM map (von Loeffelholz et al., 2018[Loeffelholz, O. von, Papai, G., Danev, R., Myasnikov, A. G., Natchiar, S. K., Hazemann, I., Ménétret, J. & Klaholz, B. P. (2018). J. Struct. Biol. 202, 191-199.]; EMDB 4261) illustrates a decrease of the local resolution from the molecular center to the periphery. (b) Test model Ω-map (equation 15[link]) of an inhomogeneous resolution varying from 2 Å in the molecular center to 5 Å at the molecular periphery calculated in a single run by the shell decomposition (middle), and the maps calculated by the Fourier procedure with the resolution of 5 Å (left) and 2 Å (right). Red arrows mark the similarity of different parts of the Ω-map with the Fourier maps of different resolution. (c) Fragments of the 2 Å resolution maps contoured to show an equal volume (Urzhumtsev et al., 2014[Urzhumtsev, A. G., Afonine, P. V., Lunin, V. Y., Terwilliger, T. C. & Adams, P. D. (2014). Acta Cryst. D70, 2593-2606.]). The map in the middle was calculated by the standard Fourier procedure. The left-hand map calculated as the sum of the Gaussian approximation, M = 1, to the atomic images (no ripples included) reveals the density for some side chains poorly. The right-hand map was calculated by the shell decomposition (equation 15[link]) with M = 5 and reproduces the Fourier map correctly. Poor density for the Arg106 residue in the Ω- and Fourier maps can be attributed to large displacement parameters of its atoms. This figure was prepared with PyMol (DeLano, 2002[DeLano, W. L. (2002). PyMol http://www.pymol.org.]).

To mimic the limited resolution and general positional disorder of the experimental map, we start by calculating the exact, theoretical density ρ(r) from the model and obtaining its Fourier coefficients. In crystallography, these are known as structure factors. To model the general dynamic and static positional disorder of atoms (or uncertainties in atomic positions) these structure factors are multiplied by the Gaussian function, which is equivalent to blurring the distribution by convolution with the Gaussian function. The required limited-resolution map ρd(r) is then calculated by the inverse Fourier transform with the set of structure factors cut at the resolution D = dhigh. Overall, the procedure requires two Fourier transforms, does not provide simple analytic expressions for the derivatives of the score function and significantly complicates obtaining a map of an inhomogeneous resolution, such as those in cryoEM.

Instead, a map can be calculated as the sum of atomic contributions of the respective resolution D and the individual degree of disorder characterized by the atomic displacement factor B (we call these contributions `atomic images'). Both the resolution cutoff and the positional disorder blur atomic images, but the resolution cutoff also results in ripples, spherical shells of locally sign-alternative high density. Several known approximations to atomic images either model only the central peak in the image (e.g. Lunin & Urzhumtsev, 1984[Lunin, V. Yu. & Urzhumtsev, A. G. (1984). Acta Cryst. A40, 269-277.]; Mooij et al., 2006[Mooij, W. T. M., Hartshorn, M. J., Tickle, I. J., Sharff, A. J., Verdonk, M. L. & Jhoti, H. (2006). ChemMedChem, 1, 827-838.]; see also Sorzano et al., 2015[Sorzano, C. O. S., Vargas, J., Otón, J., Abrishami, V., de la Rosa-Trevín, J. M., del Riego, S., Fernández-Alderete, A., Martínez-Rey, C., Marabini, R. & Carazo, J. M. (2015). AIMS Biophys. 2, 8-20.], and references therein), interpolate the images precalculated with different B values at a given common resolution D (DiMaio et al., 2015[DiMaio, F., Song, Y., Li, X., Brunner, M. J., Xu, C., Conticello, V., Egelman, E., Marlovits, T., Cheng, Y. & Baker, D. (2015). Nat. Methods, 12, 361-365.]), or use a step-function approximation to the atomic scattering function and the integrals of this approximation at a chosen resolution (Chapman, 1995[Chapman, M. S. (1995). Acta Cryst. A51, 69-80.]; Chapman et al., 2013[Chapman, M. S., Trzynka, A. & Chapman, B. K. (2013). J. Struct. Biol. 182, 10-21.]; Sorzano et al., 2015[Sorzano, C. O. S., Vargas, J., Otón, J., Abrishami, V., de la Rosa-Trevín, J. M., del Riego, S., Fernández-Alderete, A., Martínez-Rey, C., Marabini, R. & Carazo, J. M. (2015). AIMS Biophys. 2, 8-20.]).

We suggest a method to calculate model maps at every point, extending the concept of a local resolution (Kucukelbir et al., 2014[Kucukelbir, A., Sigworth, F. J. & Tagare, H. D. (2014). Nat. Methods, 11, 63-65.]; Vilas et al., 2018[Vilas, J. L., Gómez-Blanco, J., Conesa, P., Melero, R., Miguel de la Rosa-Trevín, J., Otón, J., Cuenca, J., Marabini, R., Carazo, J. M., Vargas, J. & Sorzano, C. O. S. (2018). Structure, 26, 337-344.]; Ramírez-Aportela et al., 2019[Ramírez-Aportela, E., Mota, J., Conesa, P., Carazo, J. M. & Sorzano, C. O. S. (2019). IUCrJ, 6, 1054-1063.]) further and presenting the image of every atom n in the map with its own resolution Dn, in line with Chapman et al. (2013[Chapman, M. S., Trzynka, A. & Chapman, B. K. (2013). J. Struct. Biol. 182, 10-21.]). We found an explicit expression for the map values using a specially designed analytic function for the atomic position rn, its individual displacement parameter Bn and individual resolution Dn. This expression allows accurate calculation of the inhomogeneous resolution map in a single run, without Fourier transforms. Moreover, it allows simple analytic expressions for the gradient of the score function (Urzhumtseva et al., 2022[Urzhumtseva, L., Lunin, V. Y. & Urzhumtsev, A. (2022). arXiv:2206.08935.]) that rules the refinement of the model parameters. Finally, the Dn values can be refined and reported together with other atomic parameters such as coordinates and Bn, and then deposited in databases. The inhomogeneous-resolution maps in cryoEM mean that the suggested method is tailored to this experimental technique while it can also be applied to other types of structural studies in biology and physics.

This article presents the concept of this approach and its basic proofs addressing the direct problem: how to efficiently calculate a map given an atomic model with variable parameters.

2. Shell decomposition of oscillating functions

2.1. Maps and atomic contributions

The contributions of atoms to the density ρ(r) are usually described by spherically symmetric analytic functions [\rho _n^0({\bf r})], their atomic densities; here n is the consecutive atomic number. A limited-resolution map ρd(r) of the density ρ(r) can also be seen as the sum of atomic contributions [\rho _n^d({\bf r})], this time atomic images, which should reproduce respective map distortions. While the atomic positional disorder blurs atomic densities, the resolution cut-off, alongside this, generates Fourier ripples in the resulting image [Figs. 2[link](a) and 2[link](b)]. The ripples are observed as spherical waves of a slowly decreasing amplitude; they significantly contribute to the map quite a distance from the atomic center [Fig. 1[link](c)]. Thus, substitution of one type of map distortion by another (Jakobi et al., 2017[Jakobi, A. J., Wilmanns, M. & Sachse, C. (2017). eLife, 6, e27131.]) is not fully appropriate. However, both types of distortion can be described by the same mathematical operation of a convolution but with different functions.

[Figure 2]
Figure 2
Atomic images and Fourier ripples. (a) Electron-density distributions for an immobile carbon atom and its images affected by disorder and resolution cutoff. (b) Corresponding scattering functions [Fourier transform of the functions shown in (a)]. (c) Radial part of the interference function G(x) and (d) this function in the two-dimensional section z = 0. Approximation to (e) the first negative and (f) the first positive ripples of G(x) by the weighted functions Ω(x; μ, ν) according to Table 1[link]. Two-dimensional section z = 0 is shown.

2.2. Resolution and harmonic disorder

The term `resolution' in structural biology has different meanings (Urzhumtseva et al., 2013[Urzhumtseva, L., Klaholz, B. & Urzhumtsev, A. (2013). Acta Cryst. D69, 1921-1934.]; Afonine, Klaholz et al., 2018[Afonine, P. V., Klaholz, B. P., Moriarty, N. W., Poon, B. K., Sobolev, O. V., Terwilliger, T. C., Adams, P. D. & Urzhumtsev, A. (2018). Acta Cryst. D74, 814-840.]). Traditionally in this field, when an experimental map is represented by a Fourier series, the resolution cut-off for this series is defined as the shortest period of the Fourier harmonics included in the calculation, and is considered the map resolution (e.g. Rupp, 2010[Rupp, B. (2010). Biomolecular Crystallography. Principles, Practice, and Applications to Structural Biology. NY: Garland Science.]); the number of Fourier harmonics eventually missed is supposed to be negligible. The resolution effect on the image can be mathematically described by the convolution

[\rho _n^d({\bf r};D) = \rho _n^0({\bf r})*{\delta ^d}({\bf r};D) \eqno (1)]

of the atomic density [\rho _n^0({\bf r})] with the spherically symmetric function

[{\delta ^d}({\bf r};D) = {{4\pi } \over {3{D^3}}}G\left({{{\bf{r}} \over D}} \right).\eqno (2)]

Here * stands for the convolution operation, D is the resolution cut-off in the Fourier space, D = dhigh, and G(x) is the three-dimensional interference function,

[G({\bf x}) = 3{{\sin \left({2\pi | {\bf x}|} \right) - \left({2\pi | {\bf x}|} \right)\cos \left({2\pi |{\bf x}|} \right)} \over {{{\left({2\pi |{\bf x}|} \right)}^3}}}. \eqno (3)]

Function G(x) has a large peak in the origin surrounded by a number of spherically symmetric positive and negative ripples in space (Fig. 2[link]). Function δd(r; D) may be interpreted as the D-resolution image of a virtual immobile point atom, the density of which can be described by the Dirac delta function δ(r).

If an experimental map is obtained by an alternative method to that of a single Fourier series, its different regions may reveal patterns typical for maps of a different resolution [Figs. 1[link](a) and 1[link](b)]. Such variation of the local resolution is usually illustrated by colored maps and is less commonly available from structural databases.

For the bulk of macromolecular studies, a dynamic and static disorder of atomic positions, individual for each atom, is modeled by a convolution of the respective contributions with a three-dimensional Gaussian function centered in the origin and that is isotropic (isotropic atomic displacement),

[g({\bf r};B) = {\left({{4\pi} \over B} \right)^{3/2}}\exp \left({ - {{4{\pi ^2}{{|{\bf r}|}^2}} \over B}} \right),{\bf r} \in {{\bf{R}}^3}.\eqno (4)]

The Gaussian model describes not only the motion of a particular atom around its central position but also the uncertainty of this position, i.e. that between numerous copies of the same atom over the sample or different parts of a given sample. An important feature of the Gaussian distribution is that its convolution with other Gaussian functions does not change its form but simply modifies the value of its parameter

[g({\bf r};b)*g\left({\bf r};B_n \right) = g\left({\bf r};b + B_n \right).\eqno (5)]

In particular, the density of a Gaussian atom

[\rho_n ({\bf r}) = ag\left({{\bf{r}};b} \right),\eqno (6)]

with an isotropic positional disorder described by the atomic displacement parameter Bn can be expressed by

[\rho_n \left({\bf r};B_n \right) = ag\left({\bf r};b + B_n \right).\eqno (7)]

Since the convolutions are commutative, we can combine the two principal sources of the image distortion, (1[link]) and (5[link]), in the most convenient order. The obstacle is an absence of an analytic expression for a convolution of the interference and Gaussian functions.

2.3. Shell decomposition of the interference function

To overcome the latter obstacle, we represent the interference function G(x) (3[link]) by a linear combination

[G({\bf x}) \simeq \mathop \sum \limits_{m = 1}^M {\kappa _m}\Omega \left({{\bf{x}};{\mu _m},{\nu _m}} \right)\eqno (8)]

of the terms expressed by the function

[\eqalignno { \Omega ({\bf x};\mu, \nu) = & \ {1 \over {| {\bf x}|\mu }}\bigg({1 \over {4\pi \nu }} \bigg)^{1 / 2} \Bigg \{ \exp \bigg[ - {{4\pi^2{(| {\bf x}| - \mu )}^2} \over \nu } \bigg] \cr & - \exp \bigg[- {{4\pi^2{(| {\bf x}| + \mu )}^2} \over \nu } \bigg] \Bigg\} & (9)}]

(Fig. 3[link]). The function Ω(x; μ, ν) is the uniform distribution at the spherical surface of the radius μ, blurred by the convolution with the Gaussian function g(x, ν). It has a convolution property similar to (5[link]), a `disorder transferability':

[\Omega \left({{\bf x};\mu, \nu } \right)*g\left({{\bf x};{\nu _0}} \right) = \Omega \left({{\bf x};\mu, \nu + {\nu _0}} \right).\eqno (10)]

This results in

[G({\bf x})*g\left({{\bf{x}};{\nu _0}} \right) \simeq \mathop \sum \limits_{m = 1}^M {\kappa _m}\Omega \left({{\bf x};{\mu _m},{\nu _m} + {\nu _0}} \right).\eqno (11)]

To obtain the explicit form of function Ω(x; μ ν), we started from the uniform distribution in a thin spherical shell of radius μ and width Δ, and then applied the convolution theorem and took the limit [\Delta \to 0]. The Gaussian function g(x; ν) is a limit case of Ω(x; μ, ν) when [\mu \to 0]. The sum (8[link]), which we call a shell decomposition, goes beyond the approximation of its central peak by three-dimensional Gaussian functions including also the terms with μm > 0. The values of the parameters μm, νm and κm (Table 1[link]) can be obtained by minimizing the difference between the two sides in (8[link]) and can be modified for other desired accuracy or range of x values (Urzhumtseva et al., 2022[Urzhumtseva, L., Lunin, V. Y. & Urzhumtsev, A. (2022). arXiv:2206.08935.]).

Table 1
Coefficients of the shell decomposition (8[link]) of the function G(x) obtained for |x| ≤ 10, M = 21 [see Fig. 3[link](b)]

The central peak of G(x) is represented by the sum of the Gaussian function [{\kappa _1}\Omega \left({\bf x};0,\nu_1 \right) = \kappa_1 g\left({\bf x};\nu_1 \right)] and the correcting function [{\kappa_2}\Omega \left({\bf x};\mu_2,\nu_2 \right)]. Each Fourier ripple is represented by one [\kappa_m\Omega \left({\bf x};\mu_m,\nu_m \right)] term alternating the sign of κm. The maximal error of the approximation is about 2 × 10−4 of the maximum of G(x) equal to 1

M μ ν κ M μ ν κ M μ ν κ
1 0.000 10.131 0.693 8 3.492 2.795 0.485 15 6.989 1.670 −0.368
2 0.339 3.216 0.026 9 3.971 2.882 −0.476 16 7.490 1.509 0.356
3 0.873 4.819 −0.797 10 4.471 2.022 0.401 17 7.991 1.369 −0.334
4 1.439 3.622 0.595 11 4.995 1.620 −0.371 18 8.493 1.248 0.326
5 1.979 3.616 −0.599 12 5.504 2.317 0.416 19 8.995 1.146 −0.332
6 2.462 4.143 0.623 13 5.980 2.062 −0.407 20 9.494 1.060 0.333
7 2.953 3.047 −0.534 14 6.490 1.849 0.392 21 9.978 0.811 −0.290
[Figure 3]
Figure 3
Shell decomposition of the interference function. (a) Radial component of the function Ω(x; μ, ν) with different μ and ν values indicated on the plot. (b) Radial component of the difference between the left and right sides in the decomposition (8[link]) for different numbers M of the terms included.

An important feature of function (9[link]) is that, similar to the Gaussian function, it conserves its form under rescaling,

[\Omega \left({{{\bf r} \over D};\mu, \nu } \right) = {D^3}\Omega \left({{\bf r};\mu D,\nu {D^2}} \right). \eqno (12)]

2.4. Analytic form of an atomic image

Combining (2[link]), (8[link]) and (12[link]) we obtain the image of an immobile point atom at resolution D as

[\delta^d({\bf r};D) = {{4\pi } \over 3}\mathop \sum \limits_{m = 1}^M {\kappa _m}\Omega \left({{\bf r};{\mu_m}D,{\nu_m}{D^2}} \right).\eqno (13)]

Similarly, the image of a Gaussian atom (6[link]) in the position rn and possessing the atomic displacement parameter Bn is

[\rho_n^d\left({{\bf r};{B_n},{D_n}} \right) \simeq {{4\pi a} \over 3}\mathop \sum \limits_{m = 1}^M {\kappa _m}\Omega \left({{\bf r} - {{\bf r}_n};{\mu _m}{D_n},b + {B_n} + {\nu _m}D_n^2} \right).\eqno (14)]

Except for Bn, Dn and the coordinates of the atomic center rn, all to be refined, the other parameters in (14[link]) are external and known in advance.

3. Analytic calculation of model maps

In different experimental methods, the contribution [\rho _n^0({\bf r})] of an immobile atom to the exact density (or equivalently its scattering function) is often approximated by a weighted sum of a few Gaussian functions with the coefficients [a^{(k)},b^{(k)},k = 1, \ldots, {K^{\rm Gauss}}] tabulated for each type of atom or ion (e.g. Doyle & Turner, 1968[Doyle, P. A. & Turner, P. S. (1968). Acta Cryst. A24, 390-397.]; Agarwal, 1978[Agarwal, R. C. (1978). Acta Cryst. A34, 791-809.]; Waasmaier & Kirfel, 1995[Waasmaier, D. & Kirfel, A. (1995). Acta Cryst. A51, 416-431.]; Peng, 1999[Peng, L.-M. (1999). Micron, 30, 625-648.]; Grosse-Kunstleve et al., 2004[Grosse-Kunstleve, R. W., Sauter, N. K. & Adams, P. D. (2004). IUCr Comput. Comm. Newsl. 3, 22-31.]; Brown et al., 2006[Brown, P. J., Fox, A. G., Maslen, E. N., O'Keefe, M. A. & Willis, B. T. M. (2006). International Tables for Crystallography, Vol. C, pp. 554-595. Chester: International Union of Crystallography.]). This approximation may vary with the atom environment and may contain positive Gaussians as well as negative ones.

Expressions (9[link]) and (14[link]) allow us to present an atomic model density map, distorted by restricted resolution and positional disorder, in a closed analytical form as

[\rho^d({\bf r}) \! = \! {{4\pi } \over 3} \mathop \sum \limits_{n = 1}^{{N^{\rm atoms}}} \mathop \sum \limits_{k = 1}^{{K^{\rm Gauss}}}\! \! a^{(k)}\! \! \mathop \sum \limits_{m = 1}^M \! {\kappa _m}\Omega \left[{{\bf r} \! - \! {{\bf r}_n};{\mu _m}{D_n},b^{(k)} \! + \! B_n \! + \! {\nu _m}D_n^2} \right],\eqno (15)]

with the function Ω defined in (9[link]), and the values μm, νm and κm of the shell decomposition (8[link]) of the interference function universally calculated for the required accuracy (e.g. Table 1[link]). The parameter Bn manifests the uncertainty in the position of the nth atom while Dn describes the features of the experimental map in its local environment. These values can be refined together with the atomic position rn, depending on the amount of experimental data available. Besides the calculation of the distorted electron-density map, analytical expression (15[link]) allows us to obtain simple analytic formulae for the derivatives of the map values with respect to all variable parameters. As a consequence, this provides an analytic expression for the gradient that rules the minimization of the discrepancy between the experimental and model maps.

When the variation of the resolution may be neglected either over the whole molecule or locally over a region of interest, the shell decomposition can be used more efficiently. First, one calculates numerically the image [\rho _n^d({\bf r};0,D)] of an immobile atom of every required type at the given resolution D as the limited-resolution Fourier transform of the given atomic scattering function. Then the shell decomposition

[\rho _n^d({\bf r};0,D) \simeq \mathop \sum \limits_{m = 1}^M \kappa _m^{(n)}\Omega \left[{{\bf r};R_m^{(n)},B_m^{(n)}} \right]\eqno (16)]

is built directly for each of these few oscillating images. This operation is performed only once, for the chosen resolution D. For each atom of the model, the only further adjustment required is increasing the values of the respective parameters Bm(n) by the atomic Bn. Such an approach can be also used in situations when the multi-Gaussian approximation to these functions is poor or when the scattering functions are defined numerically (e.g. Fox et al., 1989[Fox, A. G., O'Keefe, M. A. & Tabbernor, M. A. (1989). Acta Cryst. A45, 786-793.]; Brown et al., 2006[Brown, P. J., Fox, A. G., Maslen, E. N., O'Keefe, M. A. & Willis, B. T. M. (2006). International Tables for Crystallography, Vol. C, pp. 554-595. Chester: International Union of Crystallography.]; Sorzano et al., 2015[Sorzano, C. O. S., Vargas, J., Otón, J., Abrishami, V., de la Rosa-Trevín, J. M., del Riego, S., Fernández-Alderete, A., Martínez-Rey, C., Marabini, R. & Carazo, J. M. (2015). AIMS Biophys. 2, 8-20.]; Murshudov, 2016[Murshudov, G. N. (2016). Methods Enzymol. 579, 277-305.]). The number M of terms to calculate ρd(r) is reduced roughly by KGauss times compared with the general scheme, since the summation over Gaussians is no longer required. This accelerates the calculations and may improve iterative refinement procedures due to a smaller number of parameters. Although such a simplified version of the shell decomposition seems to be useful at intermediate stages of model refinement, we expect that the full version with the refinement of Dn parameters would be important at the final stages.

To illustrate the efficiency of our approach we used a test protein model (PDB entry 1zud; Burley et al., 2021[Burley, S. K. (2021). Nucleic Acids Res. 49, D437-D451.]) artificially placed in a unit cell including a single molecule, space group P1, and for which we calculated synthetic data and an exact `diffraction map' of 2 Å resolution. Then we calculated a series of Ω-maps (15[link]) with all Dn equal to 2 Å and with M that varied from one (the Gaussian peak only) to six. The map calculated, without taking the ripples into account, M = 1, was inaccurate in a number of regions whereas inclusion of a few first ripples by an increase to M = 5 made the differences with the exact map negligible [Fig. 1[link](c)]. This proves a need for modeling the Fourier ripples which are ignored in a number of existing methods of map calculation.

To illustrate the possibility of reproducing a map of a prescribed inhomogeneous resolution, we assigned the resolution Dn varying from 2 Å in the model center to 5 Å at its periphery and calculated a respective model map. The central image in Fig. 1[link](b) shows a fragment of the Ω-map with respective resolutions and calculated in a single run. Indeed, this map is the same as the 2 Å resolution Fourier map in the model center [Fig. 1[link](b), right] and coincides with the exact 5 Å resolution Fourier map at the periphery [Fig. 1[link](b), left]. In between, where the local resolution is intermediate, the Ω-map is different from both control Fourier maps. More examples of calculating an Ω-map where its resolution varies from one molecular region to another are given by Urzhumtsev et al. (2022[Urzhumtseva, L., Lunin, V. Y. & Urzhumtsev, A. (2022). arXiv:2206.08935.]). These examples illustrate a solution of the so-called `direct problem': starting from an atomic model, calculate a map for comparison with an experimental map of an inhomogeneous resolution regardless of the way in which this experimental map has been obtained and the nature of the technique, under the assumption that the atomic scattering functions are known.

Finally, we validated the capacity of (16[link]) to approximate the atomic image for different Bn values. For this goal, we calculated the image of the immobile carbon atom at a resolution of 2 Å and found the coefficients of its shell decomposition. With M = 12 terms up to the distance r ≤ 8 Å, the relative accuracy with respect to its value in the atomic center [\rho_n^d(0;0,{D_n} = 2)] was 10−4. For the carbon images with Bn = 10, 20, 30 Å2, these coefficients approximated the carbon image (16[link]) with an accuracy close to 5 × 10−6 with respect to [\rho_n^d(0;B_n,D_n = 2)]. A similar accuracy was observed for r ≤ 4 Å with M = 7 terms.

4. Results and perspectives

The method developed for calculation of atomic model maps has a number of features crucial for an efficient real-space refinement of atomic models. Such refinement becomes the key to selecting hypothetical protein models suggested by structure prediction methods and to building other model components using various experimental techniques. Our method is aimed at structural studies using cryoEM however manipulating inhomogeneous resolution maps is applicable to other techniques such as X-ray, neutron or electron diffraction. It is based on the hypothesis that principal map distortions are caused by harmonic disorder of the structure and limited resolution which may vary over regions of the map.

First, the method does not require any Fourier transforms as it can reproduce atomic images very accurately (Figs. 1[link] and 3[link]). Second, this method gives an analytic expression for the map values and for their derivatives with respect to all atomic parameters (Urzhumtseva et al., 2022[Urzhumtsev, A. G., Urzhumtseva, L. M. & Lunin, V. Y. (2022). bioRxiv, 10.1101/2022.07.17.500345.]). Third, the method suggests how to model the heterogeneity of resolution of experimental maps and describes this effect quantitively by the values of the Fourier resolution attributed individually to each atom. Fourth, our method does not only make it trivial to calculate, in a single run, a map with the local resolution that varies from one region to another [Fig. 1[link](b)], but also to adjust this resolution on-the-fly, to refine it according to the experimental map in the environment of a given atom, and also to deposit it. This `inverse problem' of obtaining parameter values from the given data will be discussed separately.

Above, we considered the basic situation in cryoEM and crystallography. However, the method can be routinely extended to more complicated situations. When increasing the resolution, anisotropic scattering factors or atomic displacement parameters may be required (see, e.g. Merritt, 2012[Merritt, E. A. (2012). Acta Cryst. D68, 468-477.], for a discussion) that currently the suggested decomposition fails to address. However, moving to sub-atomic resolution, macromolecular studies may require taking the atomic environment into consideration and use, for example, multipolar models (Hansen & Coppens, 1978[Hansen, N. K. & Coppens, P. (1978). Acta Cryst. A34, 909-921.]) both for crystallography (Jelsch et al., 1998[Jelsch, C., Pichon-Pesme, V., Lecomte, C. & Aubry, A. (1998). Acta Cryst. D54, 1306-1318.]) and for cryoEM macromolecular projects (Yonekura et al., 2015[Yonekura, K., Kato, K., Ogasawara, M., Tomita, M. & Toyoshima, C. (2015). Proc. Natl Acad. Sci. USA, 112, 3368-3373.]). It has been shown that instead, one may describe a loss of spherical symmetry in [\rho _n^0({\bf r})] due to density deformation by interatomic scatterers (Afonine et al., 2007[Afonine, P. V., Grosse-Kunstleve, R. W., Adams, P. D., Lunin, V. Y. & Urzhumtsev, A. (2007). Acta Cryst. D63, 1194-1197.]) which allows a routine application of the method described in this work.

The method considers that the scattering function of a given atom or ion are known, in particular for charged atoms (e.g. Marques et al., 2019[Marques, M. A., Purdy, M. D. & Yeager, M. (2019). Curr. Opin. Struct. Biol. 58, 214-223.]). Nevertheless, one may dream that, in the eventual case of a large amount of data, it could be possible to start from the scattering functions of neutral atoms and then refine the parameters of these functions against the given map recovering such modified scattering function experimentally.

Even when the decomposition (8[link]) is illustrated by examples in cryoEM, it may be applied to oscillating spherically symmetric functions in other research domains, allowing in particular the calculation of electrostatic potential maps in small and macromolecular charge-density studies (Ghermani et al., 1993[Ghermani, N., Lecomte, C. & Bouhmaida, N. (1993). Z. Naturforsch. 48, 91-98.]; Muzet et al., 2003[Muzet, N., Guillot, B., Jelsch, C., Howard, E. & Lecomte, C. (2003). Proc. Natl Acad. Sci. USA, 100, 8742-8747.]). It may be applied to situations when the atomic density cannot be approximated by a sum of Gaussians, for example, when working with the Coulomb potential. In this case, we can calculate first its image with a relatively small B0 value, which makes the result `Gauss-decomposable'. Then the presented method allows us to obtain images at any resolution and for any BB0.

The suggested method currently ignores one more source of map errors typical for crystallography: incompleteness of datasets (Urzhumtseva et al., 2013[Urzhumtseva, L., Klaholz, B. & Urzhumtsev, A. (2013). Acta Cryst. D69, 1921-1934.]). In fact, the effect of even a small amount of missing data, distributed randomly, is stronger than expected (Urzhumtsev et al., 2014[Urzhumtsev, A. G., Afonine, P. V., Lunin, V. Y., Terwilliger, T. C. & Adams, P. D. (2014). Acta Cryst. D70, 2593-2606.]). This is the reason why an alternative approach to excluding a test set of data (Brünger, 1992[Brünger, A. T. (1992). Nature, 355, 472-475.]) has been suggested (Pražnikar & Turk, 2014[Pražnikar, J. & Turk, D. (2014). Acta Cryst. D70, 3124-3134.]). Otherwise, the effect of missed and weighted reflections can be modeled by respective modification of atomic scattering functions when calculating atomic images.

The programs for the suggested decomposition of 3D oscillating functions (Urzhumtseva et al., 2022[Urzhumtseva, L., Lunin, V. Y. & Urzhumtsev, A. (2022). arXiv:2206.08935.]) are of a general interest and their scripts, in Python3 and in Fortran77, can be obtained upon request from the authors or from the site https://ibmc.cnrs.fr/en/laboratoire/arn-en/presentation/structures-software-and-websites/.

Acknowledgements

We thank L. Urzhumtseva for her help with programming and computing, and B. Klaholz and A. Ben Shem for stimulating discussions of the results and the manuscript.

Funding information

AU acknowledges Instruct-ERIC and the French Infrastructure for Integrated Structural Biology FRISBI (grant no. ANR-10-INBS-05).

References

First citationAfonine, P. V., Grosse-Kunstleve, R. W., Adams, P. D., Lunin, V. Y. & Urzhumtsev, A. (2007). Acta Cryst. D63, 1194–1197.  Web of Science CrossRef CAS IUCr Journals Google Scholar
First citationAfonine, P. V., Klaholz, B. P., Moriarty, N. W., Poon, B. K., Sobolev, O. V., Terwilliger, T. C., Adams, P. D. & Urzhumtsev, A. (2018). Acta Cryst. D74, 814–840.  Web of Science CrossRef IUCr Journals Google Scholar
First citationAfonine, P. V., Poon, B. K., Read, R. J., Sobolev, O. V., Terwilliger, T. C., Urzhumtsev, A. & Adams, P. D. (2018). Acta Cryst. D74, 531–544.  Web of Science CrossRef IUCr Journals Google Scholar
First citationAgarwal, R. C. (1978). Acta Cryst. A34, 791–809.  CrossRef CAS IUCr Journals Web of Science Google Scholar
First citationBaek, M., DiMaio, F., Anishchenko, I., Dauparas, J., Ovchinnikov, S., Lee, G. R., Wang, J., Cong, Q., Kinch, L. N., Schaeffer, R. D., Millán, C., Park, H., Adams, C., Glassman, C. R., DeGiovanni, A., Pereira, J. H., Rodrigues, A. V., van Dijk, A. A., Ebrecht, A. C., Opperman, D. J., Sagmeister, T., Buhlheller, C., Pavkov-Keller, T., Rathinaswamy, M. K., Dalwadi, U., Yip, C. K., Burke, J. E., Garcia, K. C., Grishin, N. V., Adams, P. D., Read, R. J. & Baker, D. (2021). Science, 373, 871–876.  Web of Science CrossRef CAS PubMed Google Scholar
First citationBrown, P. J., Fox, A. G., Maslen, E. N., O'Keefe, M. A. & Willis, B. T. M. (2006). International Tables for Crystallography, Vol. C, pp. 554–595. Chester: International Union of Crystallography.  Google Scholar
First citationBrünger, A. T. (1992). Nature, 355, 472–475.  PubMed Web of Science Google Scholar
First citationBurley, S. K. (2021). Nucleic Acids Res. 49, D437–D451.  Web of Science CrossRef CAS PubMed Google Scholar
First citationCardone, G., Heymann, J. B. & Steven, A. C. (2013). J. Struct. Biol. 184, 226–236.  Web of Science CrossRef PubMed Google Scholar
First citationChapman, M. S. (1995). Acta Cryst. A51, 69–80.  CrossRef CAS Web of Science IUCr Journals Google Scholar
First citationChapman, M. S., Trzynka, A. & Chapman, B. K. (2013). J. Struct. Biol. 182, 10–21.  Web of Science CrossRef CAS PubMed Google Scholar
First citationDeLano, W. L. (2002). PyMol http://www.pymol.orgGoogle Scholar
First citationDiamond, R. (1971). Acta Cryst. A27, 436–452.  CrossRef CAS IUCr Journals Web of Science Google Scholar
First citationDiMaio, F., Song, Y., Li, X., Brunner, M. J., Xu, C., Conticello, V., Egelman, E., Marlovits, T., Cheng, Y. & Baker, D. (2015). Nat. Methods, 12, 361–365.  Web of Science CrossRef CAS PubMed Google Scholar
First citationDoyle, P. A. & Turner, P. S. (1968). Acta Cryst. A24, 390–397.  CrossRef IUCr Journals Web of Science Google Scholar
First citationFox, A. G., O'Keefe, M. A. & Tabbernor, M. A. (1989). Acta Cryst. A45, 786–793.  CrossRef CAS Web of Science IUCr Journals Google Scholar
First citationGhermani, N., Lecomte, C. & Bouhmaida, N. (1993). Z. Naturforsch. 48, 91–98.  CrossRef CAS Google Scholar
First citationGrosse-Kunstleve, R. W., Sauter, N. K. & Adams, P. D. (2004). IUCr Comput. Comm. Newsl. 3, 22–31.  Google Scholar
First citationHansen, N. K. & Coppens, P. (1978). Acta Cryst. A34, 909–921.  CrossRef CAS IUCr Journals Web of Science Google Scholar
First citationHelliwell, J. (2022). IUCr Newsl. 30(1).  Google Scholar
First citationJakobi, A. J., Wilmanns, M. & Sachse, C. (2017). eLife, 6, e27131.  Web of Science CrossRef PubMed Google Scholar
First citationJelsch, C., Pichon-Pesme, V., Lecomte, C. & Aubry, A. (1998). Acta Cryst. D54, 1306–1318.  Web of Science CrossRef CAS IUCr Journals Google Scholar
First citationJumper, J., Evans, R., Pritzel, A., Green, T., Figurnov, M., Ronneberger, O., Tunyasuvunakool, K., Bates, R., Žídek, A., Potapenko, A., Bridgland, A., Meyer, C., Kohl, S. A. A., Ballard, A. J., Cowie, A., Romera-Paredes, B., Nikolov, S., Jain, R., Adler, J., Back, T., Petersen, S., Reiman, D., Clancy, E., Zielinski, M., Steinegger, M., Pacholska, M., Berghammer, T., Bodenstein, S., Silver, D., Vinyals, O., Senior, A. W., Kavukcuoglu, K., Kohli, P. & Hassabis, D. (2021). Nature, 596, 583–589.  Web of Science CrossRef CAS PubMed Google Scholar
First citationKucukelbir, A., Sigworth, F. J. & Tagare, H. D. (2014). Nat. Methods, 11, 63–65.  Web of Science CrossRef CAS PubMed Google Scholar
First citationKühlbrandt, W. (2014). Science, 343, 1443–1444.  Web of Science PubMed Google Scholar
First citationLoeffelholz, O. von, Papai, G., Danev, R., Myasnikov, A. G., Natchiar, S. K., Hazemann, I., Ménétret, J. & Klaholz, B. P. (2018). J. Struct. Biol. 202, 191–199.  Web of Science PubMed Google Scholar
First citationLunin, V. Yu. & Urzhumtsev, A. G. (1984). Acta Cryst. A40, 269–277.  CrossRef CAS Web of Science IUCr Journals Google Scholar
First citationMarques, M. A., Purdy, M. D. & Yeager, M. (2019). Curr. Opin. Struct. Biol. 58, 214–223.  Web of Science CrossRef CAS PubMed Google Scholar
First citationMerritt, E. A. (2012). Acta Cryst. D68, 468–477.  Web of Science CrossRef CAS IUCr Journals Google Scholar
First citationMooij, W. T. M., Hartshorn, M. J., Tickle, I. J., Sharff, A. J., Verdonk, M. L. & Jhoti, H. (2006). ChemMedChem, 1, 827–838.  Web of Science CrossRef PubMed CAS Google Scholar
First citationMurshudov, G. N. (2016). Methods Enzymol. 579, 277–305.  Web of Science CrossRef CAS PubMed Google Scholar
First citationMuzet, N., Guillot, B., Jelsch, C., Howard, E. & Lecomte, C. (2003). Proc. Natl Acad. Sci. USA, 100, 8742–8747.  Web of Science CrossRef PubMed CAS Google Scholar
First citationPalmer, C. M. & Aylett, C. H. S. (2022). Acta Cryst. D78, 136–143.  Web of Science CrossRef IUCr Journals Google Scholar
First citationPeng, L.-M. (1999). Micron, 30, 625–648.  Web of Science CrossRef CAS Google Scholar
First citationPražnikar, J. & Turk, D. (2014). Acta Cryst. D70, 3124–3134.  Web of Science CrossRef IUCr Journals Google Scholar
First citationRamírez-Aportela, E., Mota, J., Conesa, P., Carazo, J. M. & Sorzano, C. O. S. (2019). IUCrJ, 6, 1054–1063.  Web of Science CrossRef PubMed IUCr Journals Google Scholar
First citationRoversi, P. & Tronrud, D. E. (2021). Acta Cryst. D77, 1497–1515.  Web of Science CrossRef IUCr Journals Google Scholar
First citationRupp, B. (2010). Biomolecular Crystallography. Principles, Practice, and Applications to Structural Biology. NY: Garland Science.  Google Scholar
First citationSorzano, C. O. S., Vargas, J., Otón, J., Abrishami, V., de la Rosa-Trevín, J. M., del Riego, S., Fernández-Alderete, A., Martínez-Rey, C., Marabini, R. & Carazo, J. M. (2015). AIMS Biophys. 2, 8–20.  CAS Google Scholar
First citationUrzhumtsev, A. G., Afonine, P. V., Lunin, V. Y., Terwilliger, T. C. & Adams, P. D. (2014). Acta Cryst. D70, 2593–2606.  Web of Science CrossRef IUCr Journals Google Scholar
First citationUrzhumtsev, A. G. & Lunin, V. Y. (2019). Crystallogr. Rev. 25, 164–262.  Web of Science CrossRef Google Scholar
First citationUrzhumtsev, A. G., Urzhumtseva, L. M. & Lunin, V. Y. (2022). bioRxiv, 10.1101/2022.07.17.500345.  Google Scholar
First citationUrzhumtseva, L., Klaholz, B. & Urzhumtsev, A. (2013). Acta Cryst. D69, 1921–1934.  Web of Science CrossRef IUCr Journals Google Scholar
First citationUrzhumtseva, L., Lunin, V. Y. & Urzhumtsev, A. (2022). arXiv:2206.08935.  Google Scholar
First citationVilas, J. L., Gómez-Blanco, J., Conesa, P., Melero, R., Miguel de la Rosa-Trevín, J., Otón, J., Cuenca, J., Marabini, R., Carazo, J. M., Vargas, J. & Sorzano, C. O. S. (2018). Structure, 26, 337–344.  Web of Science CrossRef CAS PubMed Google Scholar
First citationWaasmaier, D. & Kirfel, A. (1995). Acta Cryst. A51, 416–431.  CrossRef CAS Web of Science IUCr Journals Google Scholar
First citationYamashita, K., Palmer, C. M., Burnley, T. & Murshudov, G. N. (2021). Acta Cryst. D77, 1282–1291.  Web of Science CrossRef IUCr Journals Google Scholar
First citationYonekura, K., Kato, K., Ogasawara, M., Tomita, M. & Toyoshima, C. (2015). Proc. Natl Acad. Sci. USA, 112, 3368–3373.  Web of Science CrossRef CAS PubMed Google Scholar

This is an open-access article distributed under the terms of the Creative Commons Attribution (CC-BY) Licence, which permits unrestricted use, distribution, and reproduction in any medium, provided the original authors and source are cited.

IUCrJ
Volume 9| Part 6| November 2022| Pages 728-734
ISSN: 2052-2525