Machine learning-guided reconstruction of cytoskeleton network from live-cell AFM images

doi:10.1016/j.isci.2024.110907

. 2024 Sep 10;27(10):110907.

doi: 10.1016/j.isci.2024.110907. eCollection 2024 Oct 18.

Machine learning-guided reconstruction of cytoskeleton network from live-cell AFM images

Hanqiu Ju^{1

2}, Henrik Skibbe³, Masaya Fukui¹, Shige H Yoshimura^{4

5}, Honda Naoki^{1

6

7}

Affiliations

¹ Laboratory of Data-driven Biology, Graduate School of Integrated Sciences for Life, Hiroshima University, Kagamiyama, Higashi-Hiroshima, Hiroshima 739-8526, Japan.
² Mathematical Sciences Research Laboratory, Advanced Technology Research & Development Division, Nikon Corporation, 1-5-20, Nishioi, Shinagawa-ku, Tokyo 140-8601, Japan.
³ Brain Image Analysis Unit, RIKEN Center for Brain Science, Wako-shi, Japan.
⁴ Laboratory of Plasma Membrane and Nuclear Signaling, Graduate School of Biostudies, Kyoto University, Kyoto, Japan.
⁵ Center for Living Systems Information Science (CeLiSIS), Kyoto University, Yoshida-Konoe-Cho, Sakyo-ku, Kyoto 606-8501, Japan.
⁶ Graduate School of Medicine, Nagoya University, Tsurumai-cho, Showa-ku, Nagoya, Aichi 466-8550, Japan.
⁷ Theoretical Biology Research Group, Exploratory Research Center on Life and Living Systems (ExCELLS), National Institutes of Natural Sciences, Okazaki, Aichi, Japan.

PMID: 39391724
PMCID: PMC11465173
DOI: 10.1016/j.isci.2024.110907

Machine learning-guided reconstruction of cytoskeleton network from live-cell AFM images

Hanqiu Ju et al. iScience. 2024.

. 2024 Sep 10;27(10):110907.

doi: 10.1016/j.isci.2024.110907. eCollection 2024 Oct 18.

Authors

Hanqiu Ju^{1

2}, Henrik Skibbe³, Masaya Fukui¹, Shige H Yoshimura^{4

5}, Honda Naoki^{1

6

7}

Affiliations

¹ Laboratory of Data-driven Biology, Graduate School of Integrated Sciences for Life, Hiroshima University, Kagamiyama, Higashi-Hiroshima, Hiroshima 739-8526, Japan.
² Mathematical Sciences Research Laboratory, Advanced Technology Research & Development Division, Nikon Corporation, 1-5-20, Nishioi, Shinagawa-ku, Tokyo 140-8601, Japan.
³ Brain Image Analysis Unit, RIKEN Center for Brain Science, Wako-shi, Japan.
⁴ Laboratory of Plasma Membrane and Nuclear Signaling, Graduate School of Biostudies, Kyoto University, Kyoto, Japan.
⁵ Center for Living Systems Information Science (CeLiSIS), Kyoto University, Yoshida-Konoe-Cho, Sakyo-ku, Kyoto 606-8501, Japan.
⁶ Graduate School of Medicine, Nagoya University, Tsurumai-cho, Showa-ku, Nagoya, Aichi 466-8550, Japan.
⁷ Theoretical Biology Research Group, Exploratory Research Center on Life and Living Systems (ExCELLS), National Institutes of Natural Sciences, Okazaki, Aichi, Japan.

PMID: 39391724
PMCID: PMC11465173
DOI: 10.1016/j.isci.2024.110907

Abstract

How actin filaments (F-actins) are dynamically reorganized in motile cells at the level of individual filaments is an open question. To find the answer, a high-speed atomic force microscopy (HS-AFM) system has been developed to live-imagine intracellular dynamics of the individual F-actins. However, noise and low resolution made it difficult to fully recognize individual F-actins in the HS-AFM images. To tackle this problem, we developed a new machine learning method that quantitatively recognizes individual F-actins. The method estimates F-actin orientation from the image while improving the resolution. We found that F-actins were oriented at ±35° toward the membrane in the lamellipodia, which is consistent with Arp2/3 complex-induced branching. Furthermore, in the cell cortex our results showed non-random orientation at four specific angles, suggesting a new mechanism for F-actin organization demonstrating the potential of our newly developed method to fundamentally improve our understanding of the structural dynamics of F-actin networks.

Keywords: biological sciences; computer science; engineering; physics.

PubMed Disclaimer

Conflict of interest statement

The authors declare no competing interests.

Figures

**Figure 1**
HS-AFM imaging of F-actin dynamics in living cells (A) Image showing the areas of interest for the HS-AFM imaging system for live cells. Red and black dashed-line boxes represent the areas in which the F-actins in the lamellipodia and cortex were imaged, respectively. Scale bar: $3 μ m$ . (B) HS-AFM images of F-actins in the cell surface area of the lamellipodia and (C) cell cortex. Scale bar: $1 μ m$ .

**Figure 2**
Schematic illustration of the cyto-LOVE processing flow (A) Flow of data processing in cyto-LOVE. FFT/iFFT processing removes scanning noise. SDS clears filamentous structures of the F-actins by estimating the presence and orientation of individual F-actins. MCMC extracts the F-actin structure as a connected particle network. (B) The Notch filtering method is used to remove scanning noise. The HS-AFM image was transferred to the frequency domain using fast Fourier transform (FFT). The notch filter kernel was applied to remove the frequency component of the scanning noise. The denoized image is reconstructed using an inverse Fourier transform (iFFT). Scale bar: $1 μ m$ . (C) Demonstration of the cyto-LOVE process using a real AFM image. Scale bar: $1 μ m$ .

**Figure 3**
Angle orientation distribution (AOD) function for describing orientation of F-actin (A) Representation of F-actin orientation when using the AOD function. The image shows an example of two crossed F-actins. The AOD functions are depicted at positions (a–c) in the image. (B) Clarified image derived from the AOD function estimated from an AFM image. This image was visualized using the maximum values of the AOD function with respect to $n$ at each pixel. Scale bar: $1 μ m$ . (C) Magnified images of the red rectangular ROI in image (B). (D) The estimated AOD functions are depicted at positions (a’, b’, c’) in image (C). At point (a’), the AOD function has a single peak that indicates a filament without crossing. At point (b’), the AOD function has double peaks that indicates filaments with crossing. At point (c’), the AOD function has no peak and remained low because point (c’) is outside the filament.

**Figure 4**
Model describing topological structure of F-actin (A) Connected particle model for topological structure of F-actin. Particles are ellipse-shaped, and their semi-minor axis represents the orientation of the F-actin. (B) Particles in model $M$ with low external energy should be placed on the central line of and orient with F-actin. (C) In model $M$ with low internal energy, (1) two connected particles should orient at the same angle and be placed in a short distance; (2) particles can collide nearly orthogonally to represent the crossing of the two F-actins but should not collide parallelly; and (3) no closed loop appears. (D) Example of the total energy $E {X} (M)$ of two models $M_{1}$ (left) and $M_{2}$ (middle) for a single F-actin. In the right panel, $E {X} (M_{1})$ is lower than $E {X} (M_{2})$ because $M_{1}$ more fits the F-actin structure in the observed image than $M_{2}$ .

**Figure 5**
Efficient sampling of the MCMC method to optimize the connected particle model (A) Particle birth; a new particle is randomly added to the current model. (B) Particle death; a particle is randomly selected to be removed from the current model. (C) Moving and rotating of particles; the positions or orientations of the particles are altered. (D) Connection/reconnection between particles; a new connection is created between two adjacent particles or an existing connection is reconnected to another adjacent particle. (E) Track-birth of particles; F-actin is randomly selected, and a new particle is added to the end, similar to polymerization. (F) Track-death of particles; F-actin is randomly selected and a particle at the end is removed, similar to depolymerization.

**Figure 6**
Demonstration of the developed method using an artificial F-actin image (A) Artificial image containing two crossing F-actins. Noise was added to the image, which mimics the actual AFM images. (B) Visualization of the MCMC method. The F-actin network is represented by connected-ellipse-shaped particles, and the green and red particles are generated by the track-birth and birth samplings, respectively. (C) Estimated model of the F-actin network. (D) Magnifications of the red ROI at 50,000 iterations in (B). Note the colliding particles, each of which are assigned to different F-actins. (E) Change in the total energy during the MCMC iterations. Note that the intrinsic energy $E_{int} {X} (M)$ consisting of the F-actin elastic energy $C_{દ} (u)$ and the collision energy $C_{coll} (p)$ has a positive value as the number of particles increases, while the external energy $E_{ext} {X} (p)$ always has a negative value.

**Figure 7**
Different orientations of estimated F-actins in the lamellipodia and cell cortex (A and G) AFM images of the lamellipodia and cell cortex. The image contains a large amount of scanning noise along the scanning direction of the AFM needle. Scale bar: $1 μ m$ . (B and H) AFM images after processing using the FFT/iFFT and SDS algorithm. The scanning noise in (A and G) was removed using FFT/iFFT. The image was visualized using the maxima of the estimated AOD function with respect to angle at each pixel. (C and I) Estimated F-actin structures. Individual F-actins are depicted in different colors. (D and J) Distribution of the estimated F-actin angle structures in the lamellipodia and cell cortex. The individual F-actin angles were calculated using the average angle of the connecting particles. (E and K) Changes in the total energy during the MCMC iterations in the lamellipodia (E) and cell cortex (K). (F and L) Persistence length distribution of lamellipodia (F) and cell cortex (L) network. The persistence length $L_{p}$ is measured to be $11.77 \pm 5.85 μ m$ and $12.51 \pm 6.04 μ m$ respectively.

**Figure 8**
Possible F-actin organization mechanism (A) Two crossing F-actins (yellow lines) with angles of 0° and −35°. Two F-actins will reorient at −62.5° and 27.5° when linked with FLNa (green triangle), which expands their angles by 27.5°, respectively. (B) Two crossing F-actins with angles of 0° and 35°. Two F-actins will reorient at −27.5° and 62.5° when linked with FLNa, which expands their angles by 27.5°, respectively.

See this image and copyright information in PMC

References

1. Blanchoin L., Boujemaa-Paterski R., Sykes C., Plastino J. Actin Dynamics, Architecture, and Mechanics in Cell Motility. Physiol. Rev. 2014;94:235–263. doi: 10.1152/physrev.00018.2013. - DOI - PubMed
1. Nonaka S., Naoki H., Ishii S. A multiphysical model of cell migration integrating reaction–diffusion, membrane and cytoskeleton. Neural Netw. 2011;24:979–989. doi: 10.1016/j.neunet.2011.06.009. - DOI - PubMed
1. Mullins R.D., Heuser J.A., Pollard T.D. The interaction of Arp2/3 complex with actin: Nucleation, high affinity pointed end capping, and formation of branching networks of filaments. Proc. Natl. Acad. Sci. USA. 1998;95:6181–6186. doi: 10.1073/pnas.95.11.6181. - DOI - PMC - PubMed
1. Yamao M., Naoki H., Kunida K., Aoki K., Matsuda M., Ishii S. Distinct predictive performance of Rac1 and Cdc42 in cell migration. Sci. Rep. 2015;5 doi: 10.1038/srep17527. - DOI - PMC - PubMed
1. Imoto D., Saito N., Nakajima A., Honda G., Ishida M., Sugita T., Ishihara S., Katagiri K., Okimura C., Iwadate Y., Sawai S. Comparative mapping of crawling-cell morphodynamics in deep learning-based feature space. PLoS Comput. Biol. 2021;17 doi: 10.1371/journal.pcbi.1009237. - DOI - PMC - PubMed

LinkOut - more resources

Full Text Sources
- Elsevier Science
- PubMed Central
Miscellaneous
- NCI CPTAC Assay Portal

[1] Blanchoin L., Boujemaa-Paterski R., Sykes C., Plastino J. Actin Dynamics, Architecture, and Mechanics in Cell Motility. Physiol. Rev. 2014;94:235–263. doi: 10.1152/physrev.00018.2013. - DOI - PubMed

[2] Blanchoin L., Boujemaa-Paterski R., Sykes C., Plastino J. Actin Dynamics, Architecture, and Mechanics in Cell Motility. Physiol. Rev. 2014;94:235–263. doi: 10.1152/physrev.00018.2013. - DOI - PubMed

[3] Nonaka S., Naoki H., Ishii S. A multiphysical model of cell migration integrating reaction–diffusion, membrane and cytoskeleton. Neural Netw. 2011;24:979–989. doi: 10.1016/j.neunet.2011.06.009. - DOI - PubMed

[4] Nonaka S., Naoki H., Ishii S. A multiphysical model of cell migration integrating reaction–diffusion, membrane and cytoskeleton. Neural Netw. 2011;24:979–989. doi: 10.1016/j.neunet.2011.06.009. - DOI - PubMed

[5] Mullins R.D., Heuser J.A., Pollard T.D. The interaction of Arp2/3 complex with actin: Nucleation, high affinity pointed end capping, and formation of branching networks of filaments. Proc. Natl. Acad. Sci. USA. 1998;95:6181–6186. doi: 10.1073/pnas.95.11.6181. - DOI - PMC - PubMed

[6] Mullins R.D., Heuser J.A., Pollard T.D. The interaction of Arp2/3 complex with actin: Nucleation, high affinity pointed end capping, and formation of branching networks of filaments. Proc. Natl. Acad. Sci. USA. 1998;95:6181–6186. doi: 10.1073/pnas.95.11.6181. - DOI - PMC - PubMed

[7] Yamao M., Naoki H., Kunida K., Aoki K., Matsuda M., Ishii S. Distinct predictive performance of Rac1 and Cdc42 in cell migration. Sci. Rep. 2015;5 doi: 10.1038/srep17527. - DOI - PMC - PubMed

[8] Yamao M., Naoki H., Kunida K., Aoki K., Matsuda M., Ishii S. Distinct predictive performance of Rac1 and Cdc42 in cell migration. Sci. Rep. 2015;5 doi: 10.1038/srep17527. - DOI - PMC - PubMed

[9] Imoto D., Saito N., Nakajima A., Honda G., Ishida M., Sugita T., Ishihara S., Katagiri K., Okimura C., Iwadate Y., Sawai S. Comparative mapping of crawling-cell morphodynamics in deep learning-based feature space. PLoS Comput. Biol. 2021;17 doi: 10.1371/journal.pcbi.1009237. - DOI - PMC - PubMed

[10] Imoto D., Saito N., Nakajima A., Honda G., Ishida M., Sugita T., Ishihara S., Katagiri K., Okimura C., Iwadate Y., Sawai S. Comparative mapping of crawling-cell morphodynamics in deep learning-based feature space. PLoS Comput. Biol. 2021;17 doi: 10.1371/journal.pcbi.1009237. - DOI - PMC - PubMed

Save citation to file

Email citation

Add to Collections

Add to My Bibliography

Your saved search

Create a file for external citation management software

Your RSS Feed

Machine learning-guided reconstruction of cytoskeleton network from live-cell AFM images

Affiliations

Machine learning-guided reconstruction of cytoskeleton network from live-cell AFM images

Authors

Affiliations

Abstract

Conflict of interest statement

Figures

Similar articles

References

LinkOut - more resources

Full Text Sources

Miscellaneous

Abstract

Conflict of interest statement

Figures

Similar articles

References

Related information

LinkOut - more resources

Full Text Sources

Miscellaneous