Abstract
Here we used machine learning to engineer genetically encoded fluorescent indicators, protein-based sensors critical for real-time monitoring of biological activity. We used machine learning to predict the outcomes of sensor mutagenesis by analyzing established libraries that link sensor sequences to functions. Using the GCaMP calcium indicator as a scaffold, we developed an ensemble of three regression models trained on experimentally derived GCaMP mutation libraries. The trained ensemble performed an in silico functional screen on 1,423 novel, uncharacterized GCaMP variants. As a result, we identified the ensemble-derived GCaMP (eGCaMP) variants, eGCaMP and eGCaMP+, which achieve both faster kinetics and larger ∆F/F0 responses upon stimulation than previously published fast variants. Furthermore, we identified a combinatorial mutation with extraordinary dynamic range, eGCaMP2+, which outperforms the tested sixth-, seventh- and eighth-generation GCaMPs. These findings demonstrate the value of machine learning as a tool to facilitate the efficient engineering of proteins for desired biophysical characteristics.
This is a preview of subscription content, access via your institution
Access options
Access Nature and 54 other Nature Portfolio journals
Get Nature+, our best-value online-access subscription
$29.99 / 30 days
cancel any time
Subscribe to this journal
Receive 12 digital issues and online access to articles
$99.00 per year
only $8.25 per issue
Buy this article
- Purchase on Springer Link
- Instant access to full article PDF
Prices may be subject to local taxes which are calculated during checkout
Similar content being viewed by others
Data availability
All of the datasets generated within this study are available on figshare at https://doi.org/10.6084/M9.FIGSHARE.23750682.V1 (ref. 45). We included the Chen4 and Dana5 datasets used to run our model and an amino acid property matrix derived from AAindex30 in the Supplementary Data. The GCaMP crystal structure used in this paper is accessible online (https://www.rcsb.org/structure/3sg3), GCaMP3–D380Y (RCSB: 3SG3) and in the Supplementary Data. Source data are provided with this paper.
Code availability
The source code is available for download from GitHub at https://doi.org/10.5281/ZENODO.8179256 (ref. 46) and CodeOcean at https://doi.org/10.24433/CO.0624159.v1 (refs. 47,48). Custom Python scripts are available from figshare45.
References
Baird, G. S., Zacharias, D. A. & Tsien, R. Y. Circular permutation and receptor insertion within green fluorescent proteins. Proc. Natl Acad. Sci. USA 96, 11241–11246 (1999).
Tian, L. et al. Imaging neural activity in worms, flies and mice with improved GCaMP calcium indicators. Nat. Methods 6, 875–881 (2009).
Akerboom, J. et al. Optimization of a GCaMP calcium indicator for neural activity imaging. J. Neurosci. 32, 13819–13840 (2012).
Chen, T.-W. et al. Ultrasensitive fluorescent proteins for imaging neuronal activity. Nature 499, 295–300 (2013).
Dana, H. et al. High-performance calcium sensors for imaging activity in neuronal populations and microcompartments. Nat. Methods 16, 649–657 (2019).
Zhang, Y. et al. Fast and sensitive GCaMP calcium indicators for imaging neural populations. Nature 615, 884–891 (2023).
Patriarchi, T. et al. Ultrafast neuronal imaging of dopamine dynamics with designed genetically encoded sensors. Science 360, eaat4422 (2018).
Sun, F. et al. Next-generation GRAB sensors for monitoring dopaminergic activity in vivo. Nat. Methods 17, 1156–1166 (2020).
Feng, J. et al. A genetically encoded fluorescent sensor for rapid and specific in vivo detection of norepinephrine. Neuron 102, 745–761.e8 (2019).
Dong, A. et al. A fluorescent sensor for spatiotemporally resolved imaging of endocannabinoid dynamics in vivo. Nat. Biotechnol. 40, 787–798 (2022).
Rappleye, M. et al. Optogenetic microwell array screening system: a high-throughput engineering platform for genetically encoded fluorescent indicators. ACS Sens. 8, 4233–4244 (2023).
Saito, Y. et al. Machine-learning-guided library design cycle for directed evolution of enzymes: the effects of training data composition on sequence space exploration. ACS Catal. 11, 14615–14624 (2021).
Romero, P. A. & Arnold, F. H. Exploring protein fitness landscapes by directed evolution. Nat. Rev. Mol. Cell Biol. 10, 866–876 (2009).
Wu, Z., Kan, S. B. J., Lewis, R. D., Wittmann, B. J. & Arnold, F. H. Machine learning-assisted directed protein evolution with combinatorial libraries. Proc. Natl Acad. Sci. USA. 116, 8852–8858 (2019).
Saito, Y. et al. Machine-learning-guided mutagenesis for directed evolution of fluorescent proteins. ACS Synth. Biol. 7, 2014–2022 (2018).
Bedbrook, C. N. et al. Machine learning-guided channelrhodopsin engineering enables minimally invasive optogenetics. Nat. Methods 16, 1176–1184 (2019).
Unger, E. K. et al. Directed evolution of a selective and sensitive serotonin sensor via machine learning. Cell 183, 1986–2002.e26 (2020).
Tian, L., Akerboom, J., Schreiter, E. R. & Looger, L. L. in Progress in Brain Research Vol. 196 (eds Knöpfel, T. & Boyden, E. S.) 79–94 (Elsevier, 2012).
Nakai, J., Ohkura, M. & Imoto, K. A high signal-to-noise Ca2+ probe composed of a single green fluorescent protein. Nat. Biotechnol. 19, 137–141 (2001).
Wardill, T. J. et al. A neuron-based screening platform for optimizing genetically-encoded calcium indicators. PLoS ONE 8, e77728 (2013).
Kawashima, S. et al. AAindex: amino acid index database, progress report 2008. Nucleic Acids Res. 36, D202–D205 (2008).
Dong, X., Yu, Z., Cao, W., Shi, Y. & Ma, Q. A survey on ensemble learning. Front. Comput. Sci. 14, 241–258 (2020).
Zhou, Z.-H. in Machine Learning (ed. Zhou, Z.-H.) 181–210 (Springer, 2021).
Yang, Y. et al. Improved calcium sensor GCaMP-X overcomes the calcium channel perturbations induced by the calmodulin in GCaMP. Nat. Commun. 9, 1504 (2018).
Song, Z., Wang, Y., Zhang, F., Yao, F. & Sun, C. Calcium signaling pathways: key pathways in the regulation of obesity. Int. J. Mol. Sci. 20, 2768 (2019).
Nausch, B., Heppner, T. J. & Nelson, M. T. Nerve-released acetylcholine contracts urinary bladder smooth muscle by inducing action potentials independently of IP3-mediated calcium release. Am. J. Physiol. Regul. Integr. Comp. Physiol. 299, R878–R888 (2010).
Ding, J., Luo, A. F., Hu, L., Wang, D. & Shao, F. Structural basis of the ultrasensitive calcium indicator GCaMP6. Sci. China Life Sci. 57, 269–274 (2014).
Souslova, E. A. et al. Single fluorescent protein-based Ca2+ sensors with increased dynamic range. BMC Biotechnol. 7, 37 (2007).
Akerboom, J. et al. Crystal structures of the GCaMP calcium sensor reveal the mechanism of fluorescence signal change and aid rational design. J. Biol. Chem. 284, 6455–6464 (2009).
Barnett, L. M., Hughes, T. E. & Drobizhev, M. Deciphering the molecular mechanism responsible for GCaMP6m’s Ca2+- dependent change in fluorescence. PLoS ONE 12, e0170934 (2017).
Nasu, Y., Shen, Y., Kramer, L. & Campbell, R. E. Structure- and mechanism-guided design of single fluorescent protein-based biosensors. Nat. Chem. Biol. 17, 509–518 (2021).
Fenno, L. E. et al. Comprehensive dual- and triple-feature intersectional single-vector delivery of diverse functional payloads to cells of behaving mammals. Neuron 107, 836–853.e11 (2020).
Kim, C. K. et al. Molecular and circuit-dynamical identification of top-down neural mechanisms for restraint of reward seeking. Cell 170, 1013–1027.e14 (2017).
Wolpert, D. H. & Macready, W. G. No free lunch theorems for optimization. IEEE Trans. Evol. Comput. 1, 67–82 (1997).
Yang, K. K., Wu, Z. & Arnold, F. H. Machine-learning-guided directed evolution for protein engineering. Nat. Methods 16, 687–694 (2019).
Yao, Z. & Ruzzo, W. L. A regression-based K nearest neighbor algorithm for gene function prediction from heterogeneous data. BMC Bioinf. 7, S11 (2006).
Starr, T. N. & Thornton, J. W. Epistasis in protein evolution. Protein Sci. 25, 1204–1218 (2016).
Stringer, C., Wang, T., Michaelos, M. & Pachitariu, M. Cellpose: a generalist algorithm for cellular segmentation. Nat. Methods 18, 100–106 (2021).
Pachitariu, M. & Stringer, C. Cellpose 2.0: how to train your own model. Nat. Methods 19, 1634–1641 (2022).
Klima, J. C. et al. Incorporation of sensing modalities into de novo designed fluorescence-activating proteins. Nat. Commun. 12, 856 (2021).
Klima, J. C. et al. Bacterial expression and protein purification of mini-fluorescence-activating proteins. Preprint at Protocol Exchange https://doi.org/10.21203/rs.3.pex-1077/v1 (2021).
Catapano, L. A., Arnold, M. W., Perez, F. A. & Macklis, J. D. Specific neurotrophic factors support the survival of cortical projection neurons at distinct stages of development. J. Neurosci. 21, 8863–8872 (2001).
Martin, D. L. Synthesis and release of neuroactive substances by glial cells. Glia 5, 81–94 (1992).
Kim, C. K. et al. Simultaneous fast measurement of circuit dynamics at multiple sites across the mammalian brain. Nat. Methods 13, 325–328 (2016).
Wait, S. J. et al. Machine learning ensemble directed engineering of genetically encoded fluorescent calcium indicators. figshare https://doi.org/10.6084/M9.FIGSHARE.23750682.V1 (2023).
sarahwaity/ProteiML: v0.1.1. Zenodo https://doi.org/10.5281/ZENODO.8179256 (2023).
Wait, S. J. A. B. ProteiML. CodeOcean https://doi.org/10.24433/CO.0624159.v1 (2024).
BerndtLab—overview. GitHub https://github.com/BerndtLab (2024).
Dragicevic, P. in Modern Statistical Methods for HCI (eds Robertson, J. & Kaptein, M.) 291–330 (Springer, 2016).
Acknowledgements
S.J.W. was supported by the National Science Foundation (DGE-2140004) and the Herbold Foundation. M.E. was supported by ‘La Caixa’ foundation and the Rafael del Pino Foundation. J.D.L. was supported by 1F31DA056121-01A1. S.L. and C.K.K. were supported by the Burroughs Wellcome Fund (CASI 1019469) and the Searle Scholars Program (SSP-2022-107). A.B. was supported by the Brain Research Foundation, University of Washington (UW) Royalty Research Fund, UW Innovation Pilot Award, National Institute of General Medical Sciences (R01 GM139850-01), National Institute of Mental Health (RF1MH130391), National Institute of Neurological Disorders and Stroke (U01NS128537), National Institute on Drug Abuse (R21DA051193, P30 DA048736 01 Pilot) and the McKnight Foundation’s Technological Innovations in Neuroscience Award. The research received additional support from the UW Center of Excellence in Neurobiology of Addiction, Pain, and Emotion Center and Institute for Stem Cell and Regenerative Medicine Shared Equipment. We thank H. R. Daniels (UC Davis) for assistance with histology for in vivo data. We would like to thank M. B. Colby for extensive input and guidance throughout this process.
Author information
Authors and Affiliations
Contributions
S.J.W. conceived and designed the experiments, performed the experiments, analyzed the data, contributed materials and analysis tools, and wrote the paper. S.L. conceived and designed the experiments, performed the experiments, analyzed the data and wrote the paper. M.E., M.Ra. and J.D.L. analyzed the data and contributed materials/analysis tools. S.A.C. analyzed the data. A.S. contributed materials/analysis tools. A.A. performed the experiments, analyzed the data and contributed materials/analysis tools. L.T. performed the experiments and analyzed the data. M.Re., F.M.-H. and D.B. contributed materials/analysis tools. A.B. and C.K.K. conceived and designed the experiments, analyzed the data, contributed materials/analysis tools and wrote the paper.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Peer review
Peer review information
Nature Computational Science thanks the anonymous reviewers for their contribution to the peer review of this work. Peer reviewer reports are available. Primary Handling Editor: Jie Pan, in collaboration with the Nature Computational Science team.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Extended data
Extended Data Fig. 1 Excitation and Emission Spectra of eGCaMP Sensors.
A. Purified GCaMP6s protein diluted into buffer containing either 10 mM EGTA or 10 mM CaEGTA. Emission spectra were calculated using a fixed excitation at 450 nm and excitation spectra were calculated using a fixed emission at 520 nm. B. Purified eGCaMP protein diluted into buffer containing either 10 mM EGTA or 10 mM CaEGTA. Emission spectra were calculated using a fixed excitation at 450 nm and excitation spectra were calculated using a fixed emission at 520 nm. C. Purified eGCaMP+ protein diluted into buffer containing either 10 mM EGTA or 10 mM CaEGTA. Emission spectra were calculated using a fixed excitation at 450 nm and excitation spectra were calculated using a fixed emission at 520 nm. D. Purified eGCaMP2+ protein diluted into buffer containing either 10 mM EGTA or 10 mM CaEGTA. Emission spectra were calculated using a fixed excitation at 450 nm and excitation spectra were calculated using a fixed emission at 520 nm.
Extended Data Fig. 2 In Vivo Performance of eGCaMP+ and eGCaMP2+ expressed in mouse mPFC.
A. Experimental timeline. Mice were injected with an AAV-Cre dependent-GCaMP variant in the mPFC and a retroAAV-Syn-Cre was injected in NAc. An optic fiber was implanted above the mPFC to allow for light delivery and fluorescence recording. B. Representative fluorescence images of GCaMP expression in mPFC and NAc (stained with anti-GFP-Alexafluor488). Scale bar, 130 µm. C. Mean Z-scored fluorescence changes in response to a foot shock (n = 4 total shock trials, collected from 2 mice for each GCaMP variant, Line depicts mean, shading depicts SEM). D. Comparison of the mean shock response between the three GCaMP variants. Top: schematic of how the shock response was calculated (see methods). Bottom: Mean change in Z-scored fluorescence response to shock (n = 4 total shock trials, collected from 2 mice for each GCaMP version). P-values were calculated using a One-way ANOVA followed by Tukey’s multiple comparisons in panels (D) and (E): *P < 0.05. All data show mean +/− SEM. E. Comparison of the mean decay to shock between the three GCaMP variants. Top: schematic of how the decay to shock was calculated (see methods). Bottom: Mean change in Z-scored fluorescence decay to shock (n = 4 total shock trials, collected from 2 mice for each GCaMP version). P-values were calculated using a One-way ANOVA followed by Tukey’s multiple comparisons in panels (D) and (E): *P < 0.05. All data show mean +/- SEM.
Supplementary information
Supplementary Information
Supplementary Figs. 1–8 and Tables 1–26.
Supplementary Data 1
Crystal structure of GCaMP3.
Supplementary Data 2
Matrix to encode 554 amino acid properties used for model implementations.
Supplementary Data 3
Mutational library for engineering GCaMP6, linking GCaMP characteristics to single mutations.
Supplementary Data 4
Mutational library for engineering GCaMP7, linking GCaMP characteristics to single mutations.
Supplementary Data 5
PCR primers used for mutating GCaMP variants to generate eGCaMP variants.
Source data
Source Data Fig. 1
Data used for plots in Fig. 1a,b,e,f.
Source Data Fig. 2
Statistical data for mutation prediction.
Source Data Fig. 3
Fluorescence intensity and decay kinetics, raw data.
Source Data Fig. 4
Fluorescence intensity and decay kinetics, raw data.
Source Data Fig. 5
Fluorescence intensity and decay kinetics, raw data.
Source Data Extended Data Fig. 1
Emission and excitation spectra raw data plots.
Source Data Extended Data Fig. 2
Fiber photometry raw data.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Wait, S.J., Expòsit, M., Lin, S. et al. Machine learning-guided engineering of genetically encoded fluorescent calcium indicators. Nat Comput Sci 4, 224–236 (2024). https://doi.org/10.1038/s43588-024-00611-w
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1038/s43588-024-00611-w
This article is cited by
-
Accelerating protein sensor optimization with machine learning
Nature Computational Science (2024)
