Abstract
Transgenic expression of bacterial nitroreductase (NTR) enzymes sensitizes eukaryotic cells to prodrugs such as metronidazole (MTZ), enabling selective cell-ablation paradigms that have expanded studies of cell function and regeneration in vertebrates. However, first-generation NTRs required confoundingly toxic prodrug treatments to achieve effective cell ablation, and some cell types have proven resistant. Here we used rational engineering and cross-species screening to develop an NTR variant, NTR 2.0, which exhibits ~100-fold improvement in MTZ-mediated cell-specific ablation efficacy, eliminating the need for near-toxic prodrug treatment regimens. NTR 2.0 therefore enables sustained cell-loss paradigms and ablation of previously resistant cell types. These properties permit enhanced interrogations of cell function, extended challenges to the regenerative capacities of discrete stem cell niches, and novel modeling of chronic degenerative diseases. Accordingly, we have created a series of bipartite transgenic reporter/effector resources to facilitate dissemination of NTR 2.0 to the research community.
This is a preview of subscription content, access via your institution
Relevant articles
Open Access articles citing this article.
-
Nanoparticle-based targeting of microglia improves the neural regeneration enhancing effects of immunosuppression in the zebrafish retina
Communications Biology Open Access 18 May 2023
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 print issues and online access
$259.00 per year
only $21.58 per issue
Rent or buy this article
Prices vary by article type
from$1.95
to$39.95
Prices may be subject to local taxes which are calculated during checkout






Data availability
Source data are provided with this paper. All additional source data files generated during and/or analyzed during the current study are available from the corresponding authors upon request.
Code availability
The ARQiv software package is available on the GitHub open-source website (https://github.com/mummlab/ARQiv). All other freely available R code functions used are as listed in the Statistics section.
References
Williams, E. M. et al. Nitroreductase gene-directed enzyme prodrug therapy: insights and advances toward clinical utility. Biochem. J. 471, 131–153 (2015).
Roldán, M. D., Pérez-Reinado, E., Castillo, F. & Moreno-Vivián, C. Reduction of polynitroaromatic compounds: the bacterial nitroreductases. FEMS Microbiol. Rev. 32, 474–500 (2008).
Venitt, S. & Crofton-Sleigh, C. The toxicity and mutagenicity of the anti-tumour drug 5-aziridino-2,4-dinitrobenzamide (CB1954) is greatly reduced in a nitroreductase-deficient strain of E. coli. Mutagenesis 2, 375–381 (1987).
Knox, R. J. et al. The nitroreductase enzyme in Walker cells that activates 5-(aziridin-1-yl)-2,4-dinitrobenzamide (CB 1954) to 5-(aziridin-1-yl)-4-hydroxylamino-2-nitrobenzamide is a form of NAD(P)H dehydrogenase (quinone) (EC 1.6.99.2). Biochem. Pharmacol. 37, 4671–4677 (1988).
Lewis, K. Platforms for antibiotic discovery. Nat. Rev. Drug Discov. 12, 371–387 (2013).
Bridgewater, J. A. et al. Expression of the bacterial nitroreductase enzyme in mammalian cells renders them selectively sensitive to killing by the prodrug CB1954. Eur. J. Cancer 31A, 2362–2370 (1995).
Drabek, D., Guy, J., Craig, R. & Grosveld, F. The expression of bacterial nitroreductase in transgenic mice results in specific cell killing by the prodrug CB1954. Gene Ther. 4, 93–100 (1997).
Clark, A. J. et al. Selective cell ablation in transgenic mice expression E. coli nitroreductase. Gene Ther. 4, 101–110 (1997).
Bridgewater, J. A., Knox, R. J., Pitts, J. D., Collins, M. K. & Springer, C. J. The bystander effect of the nitroreductase/CB1954 enzyme/prodrug system is due to a cell-permeable metabolite. Hum. Gene Ther. 8, 709–717 (1997).
Medico, E., Gambarotta, G., Gentile, A., Comoglio, P. M. & Soriano, P. A gene trap vector system for identifying transcriptionally responsive genes. Nat. Biotechnol. 19, 579–582 (2001).
Curado, S. et al. Conditional targeted cell ablation in zebrafish: a new tool for regeneration studies. Dev. Dyn. 236, 1025–1035 (2007).
White, D. T. & Mumm, J. S. The nitroreductase system of inducible targeted ablation facilitates cell-specific regenerative studies in zebrafish. Methods 62, 232–240 (2013).
Ariga, J., Walker, S. L. & Mumm, J. S. Multicolor time-lapse imaging of transgenic zebrafish: visualizing retinal stem cells activated by targeted neuronal cell ablation. J. Vis. Exp. 43, e2093 (2010).
White, D. T. et al. Immunomodulation-accelerated neuronal regeneration following selective rod photoreceptor cell ablation in the zebrafish retina. Proc. Natl Acad. Sci. 114, E3719–E3728 (2017).
Walker, S. L. et al. Automated reporter quantification in vivo: High-throughput screening method for reporter-based assays in zebrafish. PLoS ONE 7, e29916 (2012).
White, D. T. et al. ARQiv-HTS, a versatile whole-organism screening platform enabling in vivo drug discovery at high-throughput rates. Nat. Protoc. 11, 2432–2453 (2016).
Zhang, L. et al. Large-scale phenotypic drug screen identifies neuroprotectants in zebrafish and mouse models of retinitis pigmentosa. eLife 10, e57245 (2021).
Mathias, J. R., Zhang, Z., Saxena, M. T. & Mumm, J. S. Enhanced cell-specific ablation in zebrafish using a triple mutant of Escherichia coli nitroreductase. Zebrafish 11, 85–97 (2014).
Godoy, R., Noble, S., Yoon, K., Anisman, H. & Ekker, M. Chemogenetic ablation of dopaminergic neurons leads to transient locomotor impairments in zebrafish larvae. J. Neurochem. 135, 249–260 (2015).
Petrie, T. A. et al. Macrophages modulate adult zebrafish tail fin regeneration. Development 141, 2581–2591 (2014).
Guise, C. P., Grove, J. I., Hyde, E. I. & Searle, P. F. Direct positive selection for improved nitroreductase variants using SOS triggering of bacteriophage lambda lytic cycle. Gene Ther. 14, 690–698 (2007).
Tabor, K. M. et al. Direct activation of the Mauthner cell by electric field pulses drives ultrarapid escape responses. J. Neurophysiol. 112, 834–844 (2014).
Williams, E. M. et al. Engineering Escherichia coli NfsB to activate a hypoxia-resistant analogue of the PET probe EF5 to enable non-invasive imaging during enzyme prodrug therapy. Biochemistry 58, 3700–3710 (2019).
Cory, A. H., Owen, T. C., Barltrop, J. A. & Cory, J. G. Use of an aqueous soluble tetrazolium/formazan assay for cell growth assays in culture. Cancer Commun. 3, 207–212 (1991).
Mosmann, T. Rapid colorimetric assay for cellular growth and survival: Application to proliferation and cytotoxicity assays. J. Immunol. Methods 65, 55–63 (1983).
Xie, X. et al. Silencer-delimited transgenesis: NRSE/RE1 sequences promote neural-specific transgene expression in a NRSF/REST-dependent manner. BMC Biol. 10, 93 (2012).
Distel, M., Wullimann, M. F. & Köster, R. W. Optimized Gal4 genetics for permanent gene expression mapping in zebrafish. Proc. Natl Acad. Sci. USA 106, 13365–13370 (2009).
Pisharath, H., Rhee, J. M., Swanson, M. A., Leach, S. D. & Parsons, M. J. Targeted ablation of beta cells in the embryonic zebrafish pancreas using E. coli nitroreductase. Mech. Dev. 124, 218–229 (2007).
Davison, J. M. et al. Transactivation from Gal4-VP16 transgenic insertions for tissue-specific cell labeling and ablation in zebrafish. Dev. Biol. 304, 811–824 (2007).
Hyatt, G. A., Schmitt, E. A., Fadool, J. M. & Dowling, J. E. Retinoic acid alters photoreceptor development in vivo. Proc. Natl Acad. Sci. USA 93, 13298–13303 (1996).
Larison, K. D. & Bremiller, R. Early onset of phenotype and cell patterning in the embryonic zebrafish retina. Development 109, 567–576 (1990).
Cheng, X. et al. Effects of metronidazole on proopiomelanocortin a gene expression in zebrafish. Gen. Comp. Endocrinol. 214, 87–94 (2015).
McClain, R., Downing, J. C. & Edgcomb, J. E. Effect of metronidazole on fertility and testicular function in male rats. Fundam. Appl. Toxicol. 12, 386–396 (1989).
Sørensen, C. G., Karlsson, W. K., Amin, F. M. & Lindelof, M. Metronidazole-induced encephalopathy: a systematic review. J. Neurol. 267, 1–13 (2020).
Morris, A. C., Schroeter, E. H., Bilotta, J., Wong, R. O. L. & Fadool, J. M. Cone survival despite rod degeneration in XOPS-mCFP transgenic zebrafish. Invest. Ophthalmol. Vis. Sci. 46, 4762–4771 (2005).
Narayan, D. S., Wood, J. P. M., Chidlow, G. & Casson, R. J. A review of the mechanisms of cone degeneration in retinitis pigmentosa. Acta Ophthalmologica 94, 748–754 (2016).
Delyfer, M.-N. et al. Inherited retinal degenerations: therapeutic prospects. Biol. Cell 96, 261–269 (2004).
Doughty, M. L., De Jager, P. L., Korsmeyer, S. J. & Heintz, N. Neurodegeneration in Lurcher mice occurs via multiple cell death pathways. J. Neurosci. 20, 3687–3694 (2000).
Eroglu, M. & Derry, W. B. Your neighbours matter-non-autonomous control of apoptosis in development and disease. Cell Death Differ. 23, 1110–1118 (2016).
Choi, D. W. & Rothman, S. M. The role of glutamate neurotoxicity in hypoxic-ischemic neuronal death. Annu. Rev. Neurosci. 13, 171–182 (1990).
Lee, J. Y. et al. Central and peripheral secondary cell death processes after transient global ischemia in nonhuman primate cerebellum and heart. in Methods in Molecular Biology vol. 1919 215–225 (Humana Press, New York, NY, 2019).
Bergemann, D. et al. Nifurpirinol: a more potent and reliable substrate compared to metronidazole for nitroreductase-mediated cell ablations. Wound Repair Regen. 26, 238–244 (2018).
Williams, E. M. et al. A cofactor consumption screen identifies promising NfsB family nitroreductases for dinitrotoluene remediation. Biotechnol. Lett. 41, 1155–1162 (2019).
Crofts, T. S. et al. Discovery and characterization of a nitroreductase capable of conferring bacterial resistance to chloramphenicol. Cell Chem. Biol. 26, 559–570.e6 (2019).
Huang, J. et al. A zebrafish model of conditional targeted podocyte ablation and regeneration. Kidney Int. 83, 1193–1200 (2013).
Montgomery, J. E., Parsons, M. J. & Hyde, D. R. A novel model of retinal ablation demonstrates that the extent of rod cell death regulates the origin of the regenerated zebrafish rod photoreceptors. J. Comp. Neurol. 518, 800–814 (2010).
Ranski, A. H., Kramer, A. C., Morgan, G. W., Perez, J. L. & Thummel, R. Characterization of retinal regeneration in adult zebrafish following multiple rounds of phototoxic lesion. PeerJ 6, e5646 (2018).
Copp, J. N. et al. Engineering a multifunctional nitroreductase for improved activation of prodrugs and PET probes for cancer gene therapy. Cell Chem. Biol. Biol. 24, 391–403 (2017).
Prosser, G. A. et al. Creation and screening of a multi-family bacterial oxidoreductase library to discover novel nitroreductases that efficiently activate the bioreductive prodrugs CB1954 and PR-104A. Biochem. Pharmacol. 85, 1091–1103 (2013).
Swe, P. M. et al. Targeted mutagenesis of the Vibrio fischeri flavin reductase FRase I to improve activation of the anticancer prodrug CB1954. Biochem. Pharmacol. 84, 775–783 (2012).
Lee, Y. et al. Parthanatos mediates AIMP2-activated age-dependent dopaminergic neuronal loss. Nat. Neurosci. 16, 1392–1400 (2013).
Kam, T.-I. et al. Poly(ADP-ribose) drives pathologic α-synuclein neurodegeneration in Parkinson’s disease. Science 362, eaat8407 (2018).
Power, M. et al. Cellular mechanisms of hereditary photoreceptor degeneration—focus on cGMP. Prog. Retin. Eye Res. 74, 100772 (2020).
Halpern, M. E. et al. Gal4/UAS transgenic tools and their application to zebrafish. Zebrafish 5, 97–110 (2008).
Subedi, A. et al. Adoption of the Q transcriptional regulatory system for zebrafish transgenesis. Methods 66, 433–440 (2014).
Copp, J. N., Hanson-Manful, P., Ackerley, D. F. & Patrick, W. M. Error-prone PCR and effective generation of gene variant libraries for directed evolution. Methods Mol. Biol. 1179, 3–22 (2014).
Mumm, J. S. et al. In vivo imaging reveals dendritic targeting of laminated afferents by zebrafish retinal ganglion cells. Neuron 52, 609–621 (2006).
Unal Eroglu, A. et al. Multiplexed CRISPR/Cas9 targeting of genes implicated in retinal regeneration and degeneration. Front. Cell Dev. Biol. 6, 88 (2018).
Acknowledgements
This work was supported by the following grants from the National Institutes of Health, R01OD020376 (J.S.M. and D.F.A.), P30EY001765 (J.S.M., Animal Module and Imaging Cores), R01NS095355 (A.D.L. and A.L.C.), and R01HG009518 (H.J.). Additional support was provided by a Royal Society of New Zealand Marsden grant (contract VUW1902; D.F.A.) and additional consumables support was provided by an HRC Explorer grant (contract 19/750 from the Health Research Council of New Zealand; D.F.A. and J.S.M.). This study was also supported by an unrestricted departmental grant to the Wilmer Eye Institute from Research to Prevent Blindness. The authors wish to thank M. Williams, B. Bich, N. Lu, and G. Lee for technical assistance and J. Fadool and M. Halpern for providing antibodies and a QUAS reporter/effector plasmid, respectively.
Author information
Authors and Affiliations
Contributions
D.F.A. and J.S.M. conceived the research program. A.V.S., K.R.H., E.M.W., and D.F.A. designed, cloned, and/or engineered all NTR variants, designed bacterial experiments, planned, acquired, analyzed, and interpreted data. A.V.S., D.F.A., M.E.L.-B., A.D.L., and A.L.C. designed mammalian cell culture experiments and acquired, analyzed, and interpreted data. T.S.M., D.T.W., L.Z., K.E., F.M., S.N., M.T.S., and J.S.M. designed zebrafish experiments and acquired, analyzed, and interpreted data. A.V.S., K.R.H., T.S.M., L.Z., S.W., K.L., D.M.-L., and O.L.C. generated novel transgenic zebrafish resources. D.D. and H.J. provided statistical data analysis expertise. A.V.S., T.S.M., D.F.A., and J.S.M. wrote the manuscript with input from all co-authors.
Corresponding authors
Ethics declarations
Competing interests
J.S.M. has been awarded patents for the creation (US patent no. 7,514,595) and use of zebrafish expressing nitroreductase enzymes for gene (US patent no. 8,071,838) and drug discovery (US patent no. 8,431,768) applications. M.T.S. is the President and Scientific Director at Luminomics, a biotechnology start-up that offers ARQiv-based screening services. M.T.S. owns stock in Luminomics and J.S.M. serves as a consultant at Luminomics. The remaining authors declare no competing interests.
Peer review
Peer review information
Nature Methods thanks Ruilin Zhang and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Rita Strack was the primary editor on this article and managed its editorial process and peer review in collaboration with the rest of the editorial 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 NfsB_Vv-F70A/F108Y-expressing cells post-MTZ treatment.
Merged GFP (green), mCherry (magenta), and brightfield image of the MTZ-treated 60:40 co-culture shown in Fig. 2, n = 3 biologically independent experiments. Zoomed brightfield and mCherry images of the boxed region show the remaining mCherry fluorescence corresponds to small, round, phase-bright material suggestive of dead or dying cells adhering to healthy GFP-expressing cells. Scale bar = 100 microns.
Extended Data Fig. 2 NTR 2.0/MTZ-induced targeted cell ablation in larval zebrafish.
a-d, Transgenic zebrafish larvae co-expressing YFP and NTR 2.0 in rod photoreceptors were treated ±400 μM MTZ for 24 h (5-6 dpf). Retinas were then fixed at 7 dpf, sectioned and labeled with the nuclear stain DAPI (blue cells) and an α-rhodopsin antibody (α-rho, also known as 1D1) specific to rods (red cells, a), or an α-arrestin 3a antibody (α-arr3a, also known as zpr-1) specific to cones (red cells, c). Representative confocal images of YFP and antibody labeling show effective ablation of NTR 2.0-expressing rod photoreceptors (a, n = 10 retinas imaged per condition) and maintenance of neighboring cone photoreceptors (c, n = 15 and 14 retinas imaged for the 0 and 400 μM MTZ conditions, respectively). Manual quantification of cell numbers confirmed MTZ-induced loss of rod cells (b) and maintenance of neighboring cones (d). Violin plots show first quartiles (25th percentile), medians, third quartiles (75th percentile), and the full distribution of the data, with individual data points (number of measurements per condition) overlaid as a dot plot. A two-tailed nested t test (GraphPad, Prism 9) was used to calculate p-values comparing MTZ-treated and control larvae. Abbreviations: ONL, outer nuclear layer; INL, inner nuclear layer; GCL, ganglion cell layer. Scale bar = 50 microns.
Extended Data Fig. 3 NTR 2.0/MTZ-induced rod cell ablation in adult zebrafish.
a-f, Transgenic adult zebrafish larvae co-expressing YFP and NTR 2.0 in rod photoreceptors were treated ±1 mM MTZ for 3 (a,b) or 7 days (c-f). Retinas were then fixed, sectioned and labeled with the nuclear stain DAPI (blue cells) and antibodies specific to rods, 4C12 (a and e) or α-rho (c). Representative confocal images of YFP (yellow cells) and antibody labeling (red cells) show effective MTZ-induced ablation of NTR 2.0/YFP-expressing rod photoreceptors and concomitant loss of rod-specific antibody labeling (a, c, and e, n = 2 retinas imaged per condition). Manual quantification of cell numbers confirmed MTZ-induced loss of rod cells (b, d, and f. Violin plots show first quartiles (25th percentile), medians, third quartiles (75th percentile), and the full distribution of the data, with individual data points (number of measurements per condition) overlaid as a dot plot). A two-tailed nested t test (GraphPad, Prism 9) was used to calculate p-values comparing MTZ-treated and control larvae. Abbreviations: ONL, outer nuclear layer; INL, inner nuclear layer; GCL, ganglion cell layer. Scale bar = 50 microns.
Extended Data Fig. 4 NTR 2.0 does not enhance ablation efficacy with the prodrug nifurpirinol.
a, Transgenic zebrafish larvae co-expressing YFP and NTR 2.0 in neurons (NRSE:KalTA4;UAS:YFP-NTR2.0) were exposed to the indicated concentrations of NFP for 24 h (5- 6 dpf) and YFP levels were quantified by plate reader at 7 dpf (n = 3 biologically independent experiments, dot plots show the number of larvae examined). Box plots show first quartiles (25th percentile), medians, third quartiles (75th percentile), and whiskers = SD, with individual data points (larvae) overlaid as a dot plot. Fully detailed statistical comparisons (absolute effect sizes, 95% confidence intervals, Bonferroni-corrected p′-values derived from two-tailed t tests, and sample sizes) between NFP-treated and control conditions in graphs b and d are provided in Supplementary Table 5.Symbols: #p′ > 0.05, *p′ ≤ 0.05, ****p′ ≤ 0.0001; ○ = outlier data points.
Supplementary information
Supplementary Information
Guide, Supplementary Methods, Primer Table, Transgenic Line Table.
41592_2021_1364_MOESM3_ESM.xls
Supplementary Table 1 Statistical analysis of Fig. 3. Experimental conditions were compared to a signal window established by normalizing all individual sample values to the averages of the non-ablated 0 MTZ control (set at 100%) and Non-Tg control (set at 0%). Standard R code functions were used to calculate absolute effect sizes, 95% confidence intervals, and Bonferroni-corrected P values (P′ values) derived from two-tailed t tests (see Methods for details). The last row (N) shows the number of larvae analyzed per condition. The number of biologically independent experiments were as follows. 24 h MTZ treatments: 0 µM (n = 6), 12.5 μM (n = 2), 25 μM (n = 4), 50 μM (n = 4), 100 μM (n = 6), 200 μM (n = 2), Non-Tg (n = 6). 48 h MTZ treatments: 0 µM (n = 3), 12.5 μM (n = 2), 25 μM (n = 3), 50 μM (n = 3), 100 μM (n = 3), 200 μM (n = 2), Non-Tg (n = 3). 2 h MTZ treatments: 0 mM (n = 8), 0.5 mM (n = 5), 1 mM (n = 7), 10 mM (n = 7), Non-Tg (n = 8). 24 h MTZ treatments: 0 mM (n = 8), 0.5 mM (n = 4), 1 mM (n = 4), 10 mM (n = 4), Non-Tg (n = 8).
41592_2021_1364_MOESM4_ESM.xls
Supplementary Table 2 Statistical analysis of Fig. 4. Experimental conditions were compared to a signal window established by normalizing all individual sample values to the averages of the non-ablated 0 MTZ control (set at 100%) and Non-Tg control (set at 0%). Standard R code functions were used to calculate absolute effect sizes, 95% confidence intervals, and Bonferroni-corrected P values (P′ values) derived from two-tailed t tests (see Methods for details). The last row (N) shows the number of larvae analyzed per condition, n = 4 biologically independent experiments per condition.
41592_2021_1364_MOESM5_ESM.xls
Supplementary Table 3 Statistical analysis of Fig. 5. Experimental conditions were compared to 0 mM MTZ control conditions. Data was subjected to ANOVA followed by Dunnett’s multiple comparisons test, depending on the outcome of Levene’s test. Standard R code functions were used to calculate absolute effect sizes, 95% confidence intervals, and Bonferroni-corrected P values (P′ values) derived from two-tailed t tests (see Methods for details). The last row (N) shows the number of successful matings analyzed per condition over n = 3 independent mating sessions.
41592_2021_1364_MOESM6_ESM.xls
Supplementary Table 4 Statistical analysis of Fig. 6. Experimental conditions were compared to 0 mM MTZ control conditions. Standard R code functions were used to calculate absolute effect sizes, 95% confidence intervals, and Bonferroni-corrected P values (P′ values) derived from two-tailed t tests (see Methods for details). The last row (N) shows the number of larval images quantified per condition, n = 3 biologically independent experiments.
41592_2021_1364_MOESM7_ESM.xls
Supplementary Table 5 Statistical analysis of Ext. Data Fig. 4. Experimental conditions were compared to 0 mM nifurpirinol and non-transgenic control conditions. Standard R code functions were used to calculate absolute effect sizes, 95% confidence intervals, and Bonferroni-corrected P values (P′ values) derived from two-tailed t tests (see Methods for details). The last row (N) shows the number of larval images quantified per condition, n = 3 biologically independent experiments.
Supplementary Data
DNA sequences of NTR variants and plasmids used or created for this study.
Source data
Source Data Fig. 1
Data used to generate graphs in Fig. 1.
Source Data Fig. 2
Data used to generate graphs in Fig. 2.
Source Data Fig. 3
Data used to generate graphs in Fig. 3.
Source Data Fig. 4
Data used to generate graphs in Fig. 4.
Source Data Fig. 5
Data used to generate graphs in Fig. 5.
Source Data Fig. 6
Data used to generate graphs in Fig. 6.
Source Data Extended Data Fig. 2
Data used to generate graphs in Extended Data Fig. 2.
Source Data Extended Data Fig. 3
Data used to generate graphs in Extended Data Fig. 3.
Source Data Extended Data Fig. 4
Data used to generate graphs in Extended Data Fig. 4.
Rights and permissions
About this article
Cite this article
Sharrock, A.V., Mulligan, T.S., Hall, K.R. et al. NTR 2.0: a rationally engineered prodrug-converting enzyme with substantially enhanced efficacy for targeted cell ablation. Nat Methods 19, 205–215 (2022). https://doi.org/10.1038/s41592-021-01364-4
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1038/s41592-021-01364-4