Abstract
Broad-scale protein–protein interaction mapping is a major challenge given the cost, time, and sensitivity constraints of existing technologies. Here, we present a massively multiplexed yeast two-hybrid method, CrY2H-seq, which uses a Cre recombinase interaction reporter to intracellularly fuse the coding sequences of two interacting proteins and next-generation DNA sequencing to identify these interactions en masse. We applied CrY2H-seq to investigate sparsely annotated Arabidopsis thaliana transcription factors interactions. By performing ten independent screens testing a total of 36 million binary interaction combinations, and uncovering a network of 8,577 interactions among 1,453 transcription factors, we demonstrate CrY2H-seq′s improved screening capacity, efficiency, and sensitivity over those of existing technologies. The deep-coverage network resource we call AtTFIN-1 recapitulates one-third of previously reported interactions derived from diverse methods, expands the number of known plant transcription factor interactions by three-fold, and reveals previously unknown family-specific interaction module associations with plant reproductive development, root architecture, and circadian coordination.
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 print issues and online access
$259.00 per year
only $21.58 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
Accession codes
References
Yu, H. et al. Next-generation sequencing to generate interactome datasets. Nat. Methods 8, 478–480 (2011).
Rolland, T. et al. A proteome-scale map of the human interactome network. Cell 159, 1212–1226 (2014).
Arabidopsis Interactome Mapping Consortium. Evidence for network evolution in an Arabidopsis interactome map. Science 333, 601–607 (2011).
Stark, C. et al. BioGRID: a general repository for interaction datasets. Nucleic Acids Res. 34, D535–D539 (2006).
Szklarczyk, D. et al. STRING v10: protein–protein interaction networks, integrated over the tree of life. Nucleic Acids Res. 43, D447–D452 (2015).
Lee, T. et al. AraNet v2: an improved database of co-functional gene networks for the study of Arabidopsis thaliana and 27 other nonmodel plant species. Nucleic Acids Res. 43, D996–D1002 (2015).
Wang, X. et al. Three-dimensional reconstruction of protein networks provides insight into human genetic disease. Nat. Biotechnol. 30, 159–164 (2012).
Hofree, M., Shen, J.P., Carter, H., Gross, A. & Ideker, T. Network-based stratification of tumor mutations. Nat. Methods 10, 1108–1115 (2013).
Jiang, Z., Dong, X. & Zhang, Z. Network-based comparative analysis of Arabidopsis immune responses to Golovinomyces orontii and Botrytis cinerea infections. Sci. Rep. 6, 19149 (2016).
Venkatesan, K. et al. An empirical framework for binary interactome mapping. Nat. Methods 6, 83–90 (2009).
Weimann, M. et al. A Y2H-seq approach defines the human protein methyltransferase interactome. Nat. Methods 10, 339–342 (2013).
Yachie, N. et al. Pooled-matrix protein interaction screens using Barcode Fusion Genetics. Mol. Syst. Biol. 12, 863 (2016).
Hastie, A.R. & Pruitt, S.C. Yeast two-hybrid interaction partner screening through in vivo Cre-mediated Binary Interaction Tag generation. Nucleic Acids Res. 35, e141 (2007).
Pruneda-Paz, J.L. et al. A genome-scale resource for the functional characterization of Arabidopsis transcription factors. Cell Rep. 8, 622–632 (2014).
Oberdoerffer, P., Otipoby, K.L., Maruyama, M. & Rajewsky, K. Unidirectional Cre-mediated genetic inversion in mice using the mutant loxP pair lox66/lox71. Nucleic Acids Res. 31, e140 (2003).
Stynen, B., Tournu, H., Tavernier, J. & Van Dijck, P. Diversity in genetic in vivo methods for protein–protein interaction studies: from the yeast two-hybrid system to the mammalian split-luciferase system. Microbiol. Mol. Biol. Rev. 76, 331–382 (2012).
Serebriiskii, I.G. & Golemis, E.A. in Two-hybrid Systems—Methods and Protocols. (ed. MacDonald, P.N.) Ch 8 (Humana Press, 2001).
Dreze, M. et al. High-quality binary interactome mapping. Methods Enzymol. 470, 281–315 (2010).
Braun, P. et al. An experimentally derived confidence score for binary protein–protein interactions. Nat. Methods 6, 91–97 (2009).
He, F. et al. Large-scale atlas of microarray data reveals the distinct expression landscape of different tissues in Arabidopsis. Plant J. 86, 472–480 (2016).
Debernardi, J.M. et al. Post-transcriptional control of GRF transcription factors by microRNA miR396 and GIF co-activator affects leaf size and longevity. Plant J. 79, 413–426 (2014).
Cho, W.K. et al. Time-course RNA-Seq analysis reveals transcriptional changes in rice plants triggered by rice stripe virus infection. PLoS One 10, e0136736 (2015).
Guilfoyle, T.J. The PB1 domain in auxin response factor and Aux/IAA proteins: a versatile protein interaction module in the auxin response. Plant Cell 27, 33–43 (2015).
Mukhtar, M.S. et al. Independently evolved virulence effectors converge onto hubs in a plant immune system network. Science 333, 596–601 (2011).
Villarino, G.H. et al. Transcriptomic signature of the SHATTERPROOF2 expression domain reveals the meristematic nature of Arabidopsis gynoecial medial domain. Plant Physiol. 171, 42–61 (2016).
Madmon, O. et al. Expression of MAX2 under SCARECROW promoter enhances the strigolactone/MAX2 dependent response of Arabidopsis roots to low-phosphate conditions. Planta 243, 1419–1427 (2016).
Sun, L., Song, L., Zhang, Y., Zheng, Z. & Liu, D. Arabidopsis PHL2 and PHR1 act redundantly as the key components of the central regulatory system controlling transcriptional responses to phosphate starvation. Plant Physiol. 170, 499–514 (2016).
Gangappa, S.N. & Botto, J.F. The BBX family of plant transcription factors. Trends Plant Sci. 19, 460–470 (2014).
Preuss, S.B. et al. Expression of the Arabidopsis thaliana BBX32 gene in soybean increases grain yield. PLoS One 7, e30717 (2012).
Huang, H. & Bader, J.S. Precision and recall estimates for two-hybrid screens. Bioinformatics 25, 372–378 (2009).
Llorca, C.M. et al. The elucidation of the interactome of 16 Arabidopsis bZIP factors reveals three independent functional networks. PLoS One 10, e0139884 (2015).
Century, K., Reuber, T.L. & Ratcliffe, O.J. Regulating the regulators: the future prospects for transcription-factor-based agricultural biotechnology products. Plant Physiol. 147, 20–29 (2008).
Snider, J., Kittanakom, S., Curak, J. & Stagljar, I. Split-ubiquitin based membrane yeast two-hybrid (MYTH) system: a powerful tool for identifying protein–protein interactions. J. Vis. Exp. 1698, 1698 10.3791/1698 (2010).
Petschnigg, J. et al. The mammalian-membrane two-hybrid assay (MaMTH) for probing membrane-protein interactions in human cells. Nat. Methods 11, 585–592 (2014).
Wilson, T.E., Fahrner, T.J., Johnston, M. & Milbrandt, J. Identification of the DNA binding site for NGFI-B by genetic selection in yeast. Science 252, 1296–1300 (1991).
Lumba, S. et al. A mesoscale abscisic acid hormone interactome reveals a dynamic signaling landscape in Arabidopsis. Dev. Cell 29, 360–372 (2014).
Cao, S., Siriwardana, C.L., Kumimoto, R.W. & Holt, B.F. III. Construction of high quality Gateway™ entry libraries and their application to yeast two-hybrid for the monocot model plant Brachypodium distachyon. BMC Biotechnol. 11, 53 (2011).
Benatuil, L., Perez, J.M., Belk, J. & Hsieh, C.M. An improved yeast transformation method for the generation of very large human antibody libraries. Protein Eng. Des. Sel. 23, 155–159 (2010).
Trigg, S.A. et al. CrY2H-seq interactome screening. Protoc. Exch. http://dx.doi.org/10.1038/protex.2017.058 (2017).
Abremski, K. & Hoess, R. Bacteriophage P1 site-specific recombination. Purification and properties of the Cre recombinase protein. J. Biol. Chem. 259, 1509–1514 (1984).
Ehlert, A. et al. Two-hybrid protein–protein interaction analysis in Arabidopsis protoplasts: establishment of a heterodimerization map of group C and group S bZIP transcription factors. Plant J. 46, 890–900 (2006).
Albert, H., Dale, E.C., Lee, E. & Ow, D.W. Site-specific integration of DNA into wild-type and mutant lox sites placed in the plant genome. Plant J. 7, 649–659 (1995).
Langmead, B. & Salzberg, S.L. Fast gapped-read alignment with Bowtie 2. Nat. Methods 9, 357–359 (2012).
Keilwagen, J., Grosse, I. & Grau, J. Area under precision–recall curves for weighted and unweighted data. PLoS One 9, e92209 (2014).
Csardi, G. & Nepusz, T. The igraph software package for complex network research. InterJournal Complex Syst. 1695, 1–9 (2006).
Zhao, S., Guo, Y., Sheng, Q. & Shyr, Y. Heatmap3: an improved heatmap package with more powerful and convenient features. BMC Bioinformatics 15, 16 (2014).
Shannon, P. et al. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res. 13, 2498–2504 (2003).
Acknowledgements
This material is based upon work supported by US Department of Energy grant DOE-DE SC0007078 (to J.R.E.); National Science Foundation grants IOS-1650227 (to J.R.E.), IOS1456950, and IOS1546873 (to M.G.); and the Graduate Research Fellowship Program under grant number DGE-1650112 (to S.A.W.). J.R.E. is an Investigator of the Howard Hughes Medical Institute. S.A.W. is supported in part by the Mary K. Chapman Foundation. We thank M. Hofree, B. Haas, S. Navlakha, A.R. Carvunis, H. Carter, and T. Ideker for network analysis advice; S. Heinz, J. Chory, J. Law, L. Song, H. Chen, Y. He, M. Hariharan, B. Kellman, J. Reyna, L. Gai, and V. Lundblad lab members for advice and discussion; J. Pruneda-Paz (UCSD, California) for the TF ORF collection; H. Yu (Cornell University, New York) for pDEST-AD and pDEST-DB plasmids; and D. Hill (CCSB DFCI, Massachusetts) for Y8930 and Y8800 yeast strains.
Author information
Authors and Affiliations
Contributions
J.R.E. conceived the project. S.A.W., R.M.G., R.O., M.G., and J.R.E. designed and/or advised research. S.A.W., R.M.G., A.M., J.R.N., A.B., R.C., A.G., and M.G. performed experiments. S.A.W. established bioinformatics pipelines and performed computational analysis with contributions from J.F., R.O., S.C.H., and Z.Z.Z. S.A.W., M.G., and J.R.E. prepared the manuscript.
Ethics declarations
Competing interests
The authors declare no competing financial interests.
Integrated supplementary information
Supplementary Figure 1 Pilot CrY2H-seq experiments to confirm strain and plasmid functionality.
(a) RT-PCR on HIS3 positive diploid cell lysate containing known interaction partners AT3G62420 (bZIP53) and AT5G28770 (bZIP63). Lanes 2-4, RT reaction with Cre primers, Lanes 8-10, RT reaction with Gal4 DBD primers (expression control). Control lane, genomic DNA from HIS3 positive diploid lysate. M, 100 bp DNA ladder. (b) Colony PCR on HIS3 positive diploids containing interacting pair bZIP53 and bZIP63 using primers to amplify Cre-recombined ORFs. Lanes 1,3,5 show lox plasmids in Y8800/CRY8930; lanes 2,4,6 show lox plasmids in Y8800/ Y8930; lane 7 shows two non-recombined lox plasmids (negative PCR control); lane 8 shows Cre-recombined ORF products (positive PCR control). (c) Colony PCR to amplify Cre-recombined ORFs from cells containing non-interacting pairs. Lanes 1,3,5 and 9,11,13 show pADlox-ZTL and pDBlox-bZIP63 in Y8800/CRY8930; lanes 2,4,6 and 10, 12,14 show pADlox-bZIP53 and pDBlox-ZTL in Y8800/CRY8930; lanes 7 and 15 show two non-recombined lox plasmids (negative PCR control); lanes 8 and 16 show Cre-recombined plasmid (positive PCR control).
Supplementary Figure 2 Comparing diploid strains Y8930/Y8800 and CRY8930/Y8800 using a range of interactions.
(a) Plate locations of interactions screened in standard 1x1 yeast two-hybrid. Green boxes indicate expected positive interactions. CRY signifies the strain CRY8930/Y8800. Protein names Z53, Z63, ASK1, ZTL, ASK2, and PP2 correspond to the genes AT3G62420, AT5G28770, AT1G75950, AT5G57360, AT5G42190, and AT3G61060, respectively. (b) Diploid culture concentrations reported as OD600/mL. (c) Pictures of diploids after three days of selection on either 1x SC –Leu/-Trp (diploid selection) or 1x SC –Leu/-Trp/-His + 1mM 3-AT (interaction selection); (−) indicates Y8930/Y8800 diploids and (+) indicates CRY8930/Y8800 diploids.
Supplementary Figure 3 CrY2H-seq bait, prey, and interaction libraries show high coverage and minimal bias.
(a) Size distribution of transcription factor ORFs detected in prey library (‘AD’, blue) and bait library (‘DB’, teal) compared to the size distribution of all transcription factor ORFs in the ORFeome (‘theoretical’, pink). (b) Sizes of bait library ORFs plotted as a function of their log10 fragments detected, Pearson correlation, 0.0428. (c) Sizes of prey library ORFs plotted as a function of their log10 fragments detected, Pearson correlation, −0.131. (d) Size distribution of all potential PCR products from 1.9 million non-redundant interactions that could have been detected (pink) compared to size distribution of all PCR products that were detected in CrY2H-seq screens (teal). (e) CrY2H-seq detected PCR product sizes plotted as a function of their log10NPIF values. Pearson correlation, −0.1434.
Supplementary Figure 4 Determining optimal sequencing coverage and basal fragment threshold.
(a) Only PPIs (protein-protein interactions) with 3 or more fragments in the 20 million read depth dataset showed the expected 4-fold or greater increase in coverage in the 80 million read depth dataset, while PPIs with only one or two fragments in the 20 million read depth dataset were frequently not even detected (orange) at the 80 million read depth. (b) At 80 million read depth, the majority of new PPIs that were not detected at the 20 million read depth had less than 10 fragments, with over 90% of PPIs showing less than 3 fragments.
Supplementary Figure 5 The CrY2H-seq analysis pipeline.
(a) Reads are first mapped to a custom genome composed of Arabidopsis TF ORF sequences, S. cerevisiae genome, Gal4 AD and Gal4 DB domain sequences amplified by AD and DB primers (primer region), and empty plasmid sequence (not pictured). Overall alignments were as follows: 74.8% Arabidopsis, 16% primer region, 4.8% not aligned, 4.2% Yeast, and 0.2% empty plasmid. (b) Paired-end high-quality mapped reads are paired by read IDs and clonal fragments are removed. (c) Fragments are filtered for DNA strandedness to remove reads mapping to only one ORF. Examples of one protein interaction fragment (blue, green, and orange fragment) and a fragment mapping to only one ORF (purple fragment) are shown. Fragments are filtered by fragment size based on a 300bp size range (~220 bp - ~520 bp). (d) Fraction of fragments remaining after each filtering step in the analysis pipeline. Average and standard deviation were calculated from the ten replicate screen datasets. (e) Fragments for unique ORF combinations are totaled only if read1 maps to a different ORF than read2. Paired-end reads mapping to three example protein interaction PCR products are shown. (f) A basal fragment cutoff is applied for reasons described in Supplementary Figure 4 and Online Methods. This removes any interaction product with less than 3 fragments from the data, as shown by the carry-over of interaction products showing more than 2 fragments (peach/lime and blue/green interaction product) and absence of the blue/red interaction product which had only 2 fragments. (g) Datasets from each screen are normalized by calculating the median total number of interaction fragments and calculating a scale factor based on the fold difference between the screen and median total interaction fragments. Unique interaction fragments are then multiplied by the calculated scale factor and rounded down to the nearest integer. An example is shown for screen 8. See Online Methods for more details.
Supplementary Figure 6 Homodimers do not yield AD-DB PCR products from Cre-recombined plasmids.
(a) Test DB-ORFs and AD-ORFs screened in CRY8930 and Y8800, respectively, using standard 1x1 yeast two-hybrid. (b) Yeast two-hybrid positive spots grown on selection media (1xSC –Leu/-Trp/-His + 1mM 3-AT) that correspond to ORF pairs in a. (c) Colony PCR of yeast two-hybrid positive spots using AD and DB primers to detect the presence of Cre-recombined plasmids. (d) Model of an AD and DB primer-PCR amplicon of a homodimer, that contains on average a 1.66 kilobase pair inverted repeat. Blue boxes indicate homodimer pairs tested.
Supplementary Figure 7 Predicted saturation curve for CrY2H-seq transcription factor screening.
Predicted average number of interactions that would be detected after each CrY2H-seq screen from a Michaelis-Menton model based on the average CrY2H-seq detection rate over ten screens. Error bars, standard deviations.
Supplementary Figure 8 Results from 1x1 yeast two-hybrid retest of AtTFIN-1 interactions detected in CrY2H-seq screens.
(a) Representative scoring of retested interactions. (b) Fraction of AtTFIN-1 interactions positive in 1x1 matrix style Y2H retest screen (retest rate) as a function of normalized protein interaction fragments (NPIFs). Novel PPI bin sizes: 59 (1-2.5), 77 (2.5-2.8), 62 (2.8-3), 78 (3-3.2), 64 (3.2-3.4), 84 (3.4-3.7), 82 (3.7-4.1), 74 (> 4.1). Known + novel PPI bin sizes: 65 (1-2.5), 83 (2.5-2.8), 71 (2.8-3), 91 (3-3.2), 71 (3.2-3.4), 99 (3.4-3.7), 92 (3.7-4.1), 84 (> 4.1).
Supplementary Figure 9 Determining wNAPPA z-score threshold.
Fraction of known and novel interactions detected by CrY2H-seq, and random interactions that scored positive in wNAPPA plotted as a function of z-score thresholds. Dashed line indicates threshold for which the maximum number of known interactions and minimum number of random interactions are detected.
Supplementary Figure 10 Overlap with known interactions.
(a) Fraction of BioGRID interactions positive in AtTFIN-1 binned by experimental derivation, with actual number of AtTFIN-1 interactions listed inside bars. (b) Overlap between interaction datasets used for evaluating AtTFIN-1 quality.
Supplementary Figure 11 Comparing CrY2H-seq to the array-based HT-Y2H approach that was used to generate Arabidopsis Interactome-1 (AI-1).
(a) Fraction of unique TF interactions that were detected out of all TF interactions screened in CrY2H-seq and in HT-Y2H. 8,577 TF interactions were detected among the 1,880,232 unique TF interactions that were screened in CrY2H-seq, while 229 TF interactions were detected among the 270,480 unique TF interactions that were screened in HT-Y2H. (b) Coverage of commonly screened TF interactions between AtTFIN-1 and AI-1. (c) Fraction of 114 commonly screened LCI pairs recalled by CrY2H-seq and by HT-Y2H; CrY2H-seq recalled 38 interactions while HT-Y2H recalled 14.
Supplementary Figure 12 AtTFIN-1 is enriched for biologically relevant interactions.
Distribution of Pearson correlation coefficients (PCC) for AtTFIN-1 interacting pairs and random pairs across 6057 mRNA expression arrays22. Inset, percentage of AtTFIN-1 interacting pairs (9/5723) and random pairs (18/42262) with PCC > 0.75. Error bars, standard error of proportion. P value, one-sided Fisher’s exact test.
Supplementary Figure 13 Intra- and interfamily interactions in AtTFIN-1.
Discrete empirical P values of interactions observed more frequently in AtTFIN-1 than expected by random chance. Families are hierarchically clustered by common family interactions. Color key: ND = not detected, NS = not significant, * p<0.05, ** p< 0.01, *** p<0.001. Examples of known intra-family and inter-family dimers are highlighted in green and purple, respectively.
Supplementary Figure 14 CrY2H-seq shows reduced screening costs and time compared to traditional Y2H and BFG-Y2H methods.
(a) Screening cost estimates for two replicate screens of 1000 × 1000 proteins. (b) Screening cost estimates for one screen of 30,000 × 30,000 proteins. Costs for sequencing of Y2H positive clones have been omitted here because we were unable to find a report of the number of reads required per screen size for traditional Y2H or for BFG-Y2H. (c) Screening time estimates. It should be noted that these cost estimates do not take into account preliminary cloning steps as any cloning strategy could be applied to move ORFs into expression plasmids. It should also be noted that this cost comparison is conservative as the cost difference would be larger from traditional Y2H had labor and pipetting costs been accounted for.
Supplementary Figure 15 Determining bait and prey orientation of interacting proteins.
(a) Bait and prey orientations can be derived from fragments mapping to interactions that contain at least 15 base pairs of Lox77 sequence. Blue and green squares outline the lox sequence parts that reveal prey and bait identity, respectively. The black line exemplifies an orientation-revealing fragment. Amplicon, example fragment, and reads are not drawn to scale; read 1 and read 2 are each a total of 100bp. (b) Overview of fragments mapping to protein-protein interactions (PPIs).
Supplementary information
Supplementary Text and Figures
Supplementary Figures 1–15 (PDF 2697 kb)
Supplementary Table 1
Primers used for adding CRE reporter to yeast strain, adding lox sites to plasmids, and detecting Cre-recombined ORFs. (XLSX 26 kb)
Supplementary Table 2
a) CrY2H-seq prey and bait libraries screened. (XLSX 1023 kb)
b) An expanded Arabidopsis transcrition factor network. AtTFIN-1 interactions indicated by replicate and NPIF values. Literature and database interactions are listed with a value of 1, except for AraNet and STRING for which overlap is listed with the value given by the respective database. Gene family names are taken from TableS3 from Pruneda-Paz et al. Protein names are taken from AraPort (https://www.araport.org/) and from UniProt (http://www.uniprot.org/).
c) Bait and prey orientations that could be determined for 4264 AtTFIN-1 interactions.
Supplementary Table 3
a) All TFs detected with empty pADlox plasmid and determined to be self-activating proteins. (XLSX 8371 kb)
b) All PPIs detected with self-activating proteins including linkages with empty pADlox plasmid that were removed from the data (ccdB_gene denotes the pADlox plasmid sequence). 'Rep' refers to the CrY2Hseq replicate screen each interaction was detected in. The number of normalized protein interaction fragments is listed for each CrY2H-seq replicate the interaction was detected.
Supplementary Table 4
AtTFIN-1 pairs retested in 1x1 Y2H. 1x1 retest score, 0 = negative,1=positve, SA = Self-activated (XLSX 68 kb)
Supplementary Table 5
a) AtTFIN-1 pairs tested in wNAPPA. (XLSX 45 kb)
b) wNAPPA normalization pairs.
Supplementary Table 6
Literature and database protein-protein interaction data sources used for assessing AtTFIN-1. (XLSX 8 kb)
Supplementary Software
1) CrY2H-seq analysis pipeline. (ZIP 10 kb)
2) Generating datasets of random pairs.
3) Estimating screening saturation (related to Fig. 3c and Supplementary Fig. 7).
4) Bait and prey orientation analysis pipeline.
Supplementary Protocol
CrY2H-seq interactome screening (PDF 896 kb)
Rights and permissions
About this article
Cite this article
Wanamaker, S., Garza, R., MacWilliams, A. et al. CrY2H-seq: a massively multiplexed assay for deep-coverage interactome mapping. Nat Methods 14, 819–825 (2017). https://doi.org/10.1038/nmeth.4343
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1038/nmeth.4343
This article is cited by
-
A multiplexed bacterial two-hybrid for rapid characterization of protein–protein interactions and iterative protein design
Nature Communications (2023)
-
Comparative transcriptome analysis reveals the function of SlPRE2 in multiple phytohormones biosynthesis, signal transduction and stomatal development in tomato
Plant Cell Reports (2023)
-
The TCP transcription factor HvTB2 heterodimerizes with VRS5 and controls spike architecture in barley
Plant Reproduction (2022)
-
Active suppression of leaflet emergence as a mechanism of simple leaf development
Nature Plants (2021)
-
Removing auto-activators from yeast-two-hybrid assays by conditional negative selection
Scientific Reports (2021)
