Versatile protein tagging in cells with split fluorescent protein

Abstract

In addition to the popular method of fluorescent protein fusion, live cell protein imaging has now seen more and more application of epitope tags. The small size of these tags may reduce functional perturbation and enable signal amplification. To address their background issue, we adapt self-complementing split fluorescent proteins as epitope tags for live cell protein labelling. The two tags, GFP11 and sfCherry11 are derived from the eleventh β-strand of super-folder GFP and sfCherry, respectively. The small size of FP11-tags enables a cost-effective and scalable way to insert them into endogenous genomic loci via CRISPR-mediated homology-directed repair. Tandem arrangement FP11-tags allows proportional enhancement of fluorescence signal in tracking intraflagellar transport particles, or reduction of photobleaching for live microtubule imaging. Finally, we show the utility of tandem GFP11-tag in scaffolding protein oligomerization. These experiments illustrate the versatility of FP11-tag as a labelling tool as well as a multimerization-control tool for both imaging and non-imaging applications.

ARTICLE

Received 11 Dec 2015 | Accepted 15 Feb 2016 | Published 18 Mar 2016

DOI: 10.1038/ncomms11046 OPEN

Versatile protein tagging in cells with split uorescent protein

Daichi Kamiyama1,*, Sayaka Sekine1,*, Benjamin Barsi-Rhyne2, Jeffrey Hu1, Baohui Chen1, Luke A. Gilbert3, Hiroaki Ishikawa4, Manuel D. Leonetti3, Wallace F. Marshall4, Jonathan S. Weissman3,5 & Bo Huang1,4

In addition to the popular method of uorescent protein fusion, live cell protein imaging has now seen more and more application of epitope tags. The small size of these tags may reduce functional perturbation and enable signal amplication. To address their background issue, we adapt self-complementing split uorescent proteins as epitope tags for live cell protein labelling. The two tags, GFP11 and sfCherry11 are derived from the eleventh b-strand of super-folder GFP and sfCherry, respectively. The small size of FP11-tags enables a cost-effective and scalable way to insert them into endogenous genomic loci via CRISPR-mediated homology-directed repair. Tandem arrangement FP11-tags allows proportional enhancement of uorescence signal in tracking intraagellar transport particles, or reduction of photobleaching for live microtubule imaging. Finally, we show the utility of tandem GFP11-tag in scaffolding protein oligomerization. These experiments illustrate the versatility of FP11-tag as a labelling tool as well as a multimerization-control tool for both imaging and non-imaging applications.

1 Department of Pharmaceutical Chemistry, University of California, San Francisco, California 94143, USA. 2 Tetrad Graduate Program, University of California, San Francisco, California 94143, USA. 3 Department of Cellular and Molecular Pharmacology, University of California, San Francisco, California 94143, USA.

4 Department of Biochemistry and Biophysics, University of California, San Francisco, California 94143, USA. 5 Howard Hughes Medical Institute, San Francisco, California 94143, USA. * These authors contributed equally to this work. Correspondence and requests for materials should be addressed to B.H. (email: mailto:[email protected]

Web End [email protected] ).

NATURE COMMUNICATIONS | 7:11046 | DOI: 10.1038/ncomms11046 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 1

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11046

Complementary to direct uorescent protein fusion, more and more examples of live cell protein labelling and imaging have recently emerged using epitope tags: peptides

that do not function by themselves but can be recognized by other intracellularly expressed proteins. An epitope tag fused to the target protein can become an enzyme substrate to be ligated to a small molecule1. Alternatively, it can bind an intracellularly expressed antibody/nanobody to bring in functional protein units, such as a uorescent protein for imaging2,3 or ubiquitin ligase for protein degradation4. This two-component labelling approach has a number of advantages. It is highly modular because the functional units can be easily replaced2. A small peptide epitope tag may induce less perturbation than a much larger uorescent protein. In addition, small epitope tags can be arranged into a multimerization scaffold. For example, the SunTag2 allows a dramatic enhancement of uorescence signal by bringing as many as 24 green uorescent proteins (GFPs) to a single target protein. This arrangement enables it to successfully overcome an intrinsic problem such as the low afnity between the tagged protein and the binding module. Alternatively, the SunTag could have used a higher afnity and a more precise expression stoichiometry between them, so that there is neither incomplete labelling nor background from excessive binding modules.

One approach to relax this afnity and stoichiometry requirement is to have a binding module that becomes uorescent only when it is on the epitope. In this case, as long as the binding module is expressed in excess of the target protein, the equilibrium can be shifted towards complete labelling. For this purpose, we adopted a split super-folder GFP (sfGFP) that was previously engineered for efcient self-complementation without the assistance of other proteinprotein interactions5. This split construct breaks the sequence of sfGFP between the tenth and the eleventh b-strand into two parts: GFP1-10 and GFP11, with

GFP11 being a short, 16 amino acid peptide (Fig. 1a). The GFP1-10 fragment, which contains the three residues that constitute the GFP chromophore, is nonuorescent by itself because chromophore maturation requires the conserved E222 residue located on GFP11 (ref. 6). Upon complementation, the reconstituted GFP becomes uorescent after the chromophore maturation reaction is completed5,7. This split GFP system has been previously used for protein quantication5, visualization of protein subcellular localization810, single-molecule imaging11, cell-cell contact detection12, as well as in vitro protein complex assembly13. Some of these applications810 fused the small GFP11 fragment to the target protein as an epitope tag. To expand this approach, here we establish 1-10/11 split uorescent protein constructs (FP11 tags and the corresponding FP1-10 fragments) as a general method for cellular protein visualization including multicolour imaging, especially for the labelling of endogenous protein via genetic knock-in and signal amplication with tandem tags.

ResultsLive cell protein labelling using GFP11 and sfCherry11. To validate protein labelling using FP11-tag in mammalian cells, we fused GFP11 to the (amino) N terminus of human b-actin (with an 18 a.a. GSS linker) and co-expressed it with GFP1-10 in HeLa cells. We observed strong actin-like uorescence signal, whereas GFP11::b-actin or GFP1-10 alone gave no detectable signal (Fig. 1b and Supplementary Fig. 1). We compared the average uorescence intensity (representing a combination of expression efciency, complementation efciency and uorophore maturation14) of b-actin labelled with either full-length sfGFP or GFP11 and expressed using the same vector (Fig. 1c). The difference between the two labelling

approaches was small, indicating that GFP11-tag can label the target protein with a similar efciency as direct GFP fusion. We then demonstrated that GFP11-tag is compatible with a variety of cellular proteins for live imaging, including cytoskeleton (b-actin and b-tubulin), clathrin-coated pits (clathrin light chain), histone (H2B) and focal adhesion (zyxin) (Fig. 1d). By adding an endoplasmic reticulum (ER) signal peptide to GFP1-10 (GFP1-10ER), we were able to specically label ER luminal protein (calreticulin) or the extracellular domain of plasma membrane protein (b2 adrenergic receptor; Fig. 1d). In addition, we showed that the Y66W mutation for cyan uorescent protein (CFP) and the T203Y mutation for yellow uorescent protein (YFP)15,16 can be successfully introduced into GFP1-10 to label proteins with the corresponding colour (Fig. 1e).

To enable multicolour imaging, we tested splitting mCherry at the same site of the GFP1-10/11 system. However, no uorescent complementation signal was observed when co-expressing mCherry::b-actin and mCherry1-10 (data not shown), which is consistent with previous characterization of split-mCherry constructs17. Instead, we were able to obtain a functional split from super-folder Cherry (sfCherry, derived from mCherry for potentially better folding18), resulting in an 18 amino-acid peptide for sfCherry11 (Fig. 1a). We veried that sfCherry11::b-actin can be uorescently imaged when co-expressing sfCherry1-10 (Fig. 1b). Nevertheless, its overall uorescence signal level is substantially weaker than that of direct sfCherry fusion (Fig. 1c), potentially caused by the less-efcient self-complementation between sfCherry11 and sfCherry1-10. This problem could be solved by further optimization of the split-construct1 or by tandem tagging which we will describe later.

Labelling endogenous proteins with FP11 tag. A rst unique application of FP11 epitope tag is to generate libraries of uorescently labelled endogenous proteins via genetic knock-in by homology-directed DNA repair. Previously, such a library created for Saccharomyces cerevisiae has demonstrated its wide application in the systematic investigation of protein localization, gene expression and cellular behaviour19,20. Because the DNA sequence for GFP11 is only 57 nt (including a 3 a.a. linker), we could use very short homology arms (70 nt on either side, plus the start or stop codon for N- or carboxy (C)-terminal labelling, respectively), so that the entire donor DNA can be a 200 nt single-strand oligo-DNA (ssDNA). Unlike the long donor DNA for full-length GFP, which must be made through multiple cloning steps, GFP11 donor ssDNA can be directly synthesized, thus making it cost-effective for library generation. As a demonstration, we used synthetic ssDNA donors to knock GFP11-tag into four genes, LMNA, HIST2H2BE, CBX1 and PRKACA, which encode lamin A/C, histone H2B (H2B), heterochromatin protein 1 homologue (HP1) and cAMP-dependent protein kinase catalytic subunit a (PKA), respectively. Using the Cas9/sgRNA plasmid transfection method21, we have obtained a knock-in efciency in the range of 0.5B0.8% (negative controls: 0.01B0.09%) (Fig. 2a and

Supplementary Fig. 2). Other established knock-in approaches, such as by direct nucleofection of the Cas9/sgRNA complex22, may yield higher knock-in efciencies. We imaged these cells and validated the specicity by immunouorescence (Fig. 2b). This approach can be readily extended to other mammalian proteins or even animal systems given the recent advancements in CRISPR-mediate genome engineering technologies.

Signal amplication using tandem FP11 tags. With their small sizes, we can arrange FP11-tags into tandem arrays to amplify the uorescence signal (Fig. 3a), thus addressing the issues of poor

2 NATURE COMMUNICATIONS | 7:11046 | DOI: 10.1038/ncomms11046 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11046 ARTICLE

a

Protein +

FP11 FP1-10

Reassembled FP

GFP11

GFP1-10

sfCherry1-10

...PNEK

RDHMVLHEYVNAAGIT

=

sfCherry11

...HNED

YTIVEQYERAEGRHSTGG

b

c

sfGFP sfCherry

GFP11::-actin + GFP1-10

sfCherry11::-actin + sfCherry1-10

Relative intensity

1.0

0.8

0.6

0.4

0.2

Full (6)

Split (6)

Full (18)

Split (18)

d

GFP11::clathrin + GFP1-10 GFP11::H2B + GFP1-10 GFP11::-tubulin + GFP1-10

zyxin::GFP11 + GFP1-10 Calreticulin::GFP11 +

GFP1-10ER GFP11::2AR+ GFP1-10

ER

e

GFP11::-actin + CFP1-10 or YFP1-10

CFP channel YFP channel Merge

Figure 1 | Cellular protein labelling with FP11-tag. (a) Schematic diagram for FP11-Tag, illustrated on the crystal structure of sfCherry, and the split schemes. (b) Images of HeLa cells co-expressing GFP1-10 and GFP11::b-actin or sfCherry 1-10 and sfCherry 11::b-actin. (c) Average uorescence intensity of whole cells expressing b-actin labelled with full-length sfGFP, sfCherry or the corresponding FP11 tags. n 618. Error bars are s.e.m. (d) Fluorescence

images of Drosophila S2 cells expressing GFP11-tagged of b-tubulin and calreticulin, and HeLa cells expressing GFP11-tagged clathrin light chain, histone H2B, zyxin and b2AR. (e) An image of mixed S2 cells expressing either CFP1-10 GFP11::b-actin or YFP1-10 GFP11::b-actin. The weak uorescence of the

CFP1-10-expressing cell in the YFP channel is due to the bleed-through of CFP emission. All the scale bars indicate 5 mm.

NATURE COMMUNICATIONS | 7:11046 | DOI: 10.1038/ncomms11046 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 3

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11046

a Lamin H2B

neg (2)

HP1 PKA

1.0

0.2

0.8

FAM/HEX ratio (%)

0.6

0.4

+sgRNA (5)

neg (5)

+sgRNA (4)

neg (3)

+sgRNA (5)

neg (9)

+sgRNA (16)

b

H2B HP1

Lamin A/C PKA

GFP

Immunofluorescence

Merge

Figure 2 | Labelling endogenous proteins using GFP11-tag. (a) GFP11 knock-in efciencies by co-transfection with Cas9/sgRNA expression plasmids and donor templates, quantied by the combined TaqMan PCR/ddPCR assay (see Supplementary Fig. 2). (b) GFP uorescence and immunouorescence images of knock-in cells. All the error bars are s.e.m. All the scale bars indicate 5 mm. neg, negative.

signal or photobleaching that live cell imaging often suffers from23. To design tandem GFP11x3, GFP11x7 and sfCherry11x4 tags, we purposely used synonymous codons to avoid deleterious recombination during cloning, which can be caused by repetitive nucleic acid sequences. For GFP11x7, we tested either 5 a.a. or 15 a.a. linker length between the repeats. Using a fused mCherry as the reference to normalize expression level differences, we observed that the uorescence signal for GFP11x3::mCherry::b-tubulin and

GFP11x7::mCherry::b-tubulin increased proportionally compared with GFP11::mCherry::b-tubulin without the tandem repeat (Fig. 3b,c). No signicant signal difference was recorded for GFP11x7 using either the 5 a.a. or 15 a.a. linker (Fig. 3c), indicating a negligible interference between reconstituted GFPs in the tandem tag. We also observed a proportional enhancement of overall cellular uorescence comparing sfCherry11x4::b-actin with sfCherry11::b-actin (Fig. 3d,e). Combining sfCherry11x4-tagged b-actin with GFP11-tagged clathrin light chain, we showed two colour imaging using FP11 tags (Fig. 3f). Compared with SunTag2, tandem FP11 tags have the advantage that FP1-10 can be

overexpressed without causing a uorescence background, thus allowing a much simpler expression control.

The uorescence amplication by tandem FP11-tag can greatly benet imaging experiments that are difcult because of low signal, such as single particle tracking of intraagellar transport (IFT) in primary cilia24. We fused GFP11x7 to the C terminus of IFT20, which moves along ciliary microtubules. Compared with the image of IFT20::GFP, IFT20::GFP11x7 generated a much higher signal from IFT particles in the primary cilium of mouse IMCD3 cells (Fig. 4a, Supplementary Fig. 3 and Supplementary Movie 1). Moreover, analysis of the kymograph showed no signicant difference between the two labelling methods for either anterograde or retrograde transport speed (Fig. 4a,b), suggesting that IFT20 labelling using tandem GFP11-tag did not perturb its function.

Another advantage of a brighter label is to reduce photobleaching and phototoxicity by lowering excitation intensity while maintaining the same signal level. To demonstrate this application, we imaged Drosophila S2 cells expressing

4 NATURE COMMUNICATIONS | 7:11046 | DOI: 10.1038/ncomms11046 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11046 ARTICLE

a

Protein

FP11n

+ =

Protein

FP1-10

b c

GFP11x1 GFP11x3 GFP11x7

15 a.a. linker

5 a.a. linker

GFP

Relative, normalized

fluorescence

8

6

mCherry (reference)

4

2

x1 x3 x7 x1 x7

Number of GFP11 repeats

d e

sfCherry11x1 sfCherry11x4

4

3

Relative fluorescence

Number of sfCherry11 repeats

2

1

X1 (18)

X4 (16)

f GFP11::clathrin

+ GFP110

sfCherry11x4::-actin + sfCherry110

Merge

Figure 3 | Amplication of uorescence signal by tandem FP11. (a) Schematic of uorescence signal amplication by tandem FP11 (b) Images of Drosophila S2 cells expressing GFP11x1, x3 or x7::mCherry::b-tubulin together with overexpressed GFP1-10. The three panel sets were acquired and displayed with identical settings. (c) Quantication of GFP to mCherry uorescence intensity ratio with different number of GFP11 repeats and different linker lengths between the repeats. Fifty cells were analysed in each case. Error bars are s.e.m. (d) Images of HeLa cells expressing sfCherry11x1 or x4::b-actin together

with overexpressed sfCherry1-10. (e) Average whole-cell uorescence intensity with sfCherry11x1 or x4::b-actin. n as indicated in the gure. Error bars are s.e.m. (f) Two-colour imaging of GFP11-labelled clathrin light chain and sfCherry11x4 labelled b-actin. All the scale bars indicate 5 mm.

GFP::b-tubulin, GFP11::b-tubulin or GFP11x7::b-tubulin by spinning disk confocal microscopy (Fig. 4c and Supplementary Movie 2). With one-seventh the excitation laser power, the GFP11x7 sample generated a similar level of uorescence signal compared with either GFP or GFP11, but its photobleaching rate was approximately nine times as slow (Fig. 4d). Surprisingly, even the non-repeating GFP11 displayed approximately three times as slow a photobleaching rate as that of direct GFP fusion. A possible explanation for this effect is that the complementation between GFP11 and GFP1-10 can be reversed, allowing photobleached GFP1-10 to be exchanged for an intact one in the cytoplasm, although such exchange may be limited by the maturation process of the GFP chromophore.

Tandem FP11 tags as multimerization scaffolds. Tandem FP11-tag can also be used in non-imaging applications13 as a

scaffold to assemble multiple copies of proteins for synergistic functioning. Previously, we have developed a CRISPR/Cas9-based transcriptional activation system (CRISPRa)25, which activates a specic gene by targeting it with a nuclease-deactivated Cas9 (dCas9) fused to the transcription activator VP64. We have then observed that the activation of natively suppressed gene by CRISPRa often require multiple copies of transcription activation VP64 domains at the target2,2628. Here, we showed that tandem GFP11-tag could be used to achieve this goal. We chose to target the C-X-C chemokine receptor type 4 gene CXCR4, which is poorly expressed in K265 cells. We fused VP64 to GFP1-10, and then coexpressed it with dCas9::GFP11x7 and a CXCR4 targeting sgRNA26 in K265 cells (Fig. 5a). We immunostained these cells for CXCR4 and quantied its expression level by ow cytometry (Fig. 5b). A 452-fold signal increase was detected compared with the controls, whereas no signicant activation can be observed either with a non-targeting sgRNA or with direct

NATURE COMMUNICATIONS | 7:11046 | DOI: 10.1038/ncomms11046 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 5

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11046

a b

IFT20::GFP

IFT20::GFP117

Tip

Base

Tip

Base

1.4

1.2

1.0

0.8

Average IFT speed (m s1)

0.6

0.4

0.2

Anterograde Retrograde

5 s

5 m

c

GFP11x7::-tubulin

+GFP110 (1/7 excitation)

0 GFP (83)

GFP11 x7 (112)

GFP (46)

GFP11 x7 (120)

GFP::-tubulin

GFP11::-tubulin

+GFP110

d

1.0

0 s 399 s

GFP11x7

Normalized fluorescence intensity

0.8

0.6

GFP11

0.4

0.2

GFP

0.0

0 100 200 300 400

Time (s)

Figure 4 | Live cell imaging using tandem FP11 tag. (a) Tracking of IFT particles in mouse IMCD3 cells expressing either IFT20::GFP or IFT20::GFP11x7 GFP1-10. Kymographs show both retrograde and anterograde transport (see Supplementary Fig. 3 and Supplementary Movie 1).

(b) Comparison of IFT20 movement speed for IFT20::GFP and IFT20::GFP11x7. The number of particles that were analysed for each case is indicated in the gure. Error bars are standard deviations. (c) Comparison of the photobleaching rate in imaging b-tubulin labelled with GFP, GFP11 and GFP11x7, showing snapshots of S2 cells expressing the three different constructs at the beginning and the end of a 400 s movie. GFP11x7 samples were imaged with one-seventh the excitation laser power. Scale bars indicate 5 mm. (d) Fluorescence photobleaching time traces. Five cells were averaged in each condition. The error bars are standard deviations.

dCas9::VP64 fusion. Immunouorescence images conrmed CXCR4 expression in these cells (Fig. 5c). In conjugation with SunTag, tandem GFP11 can serve as an orthogonal multimerization tag multiplexed applications. We also observed more robust expression of GFP1-10::VP64, potentially due to its better solubility than scFv::VP64 in SunTag2,26.

DiscussionThe experiments above showcased the versatility of FP11 as an epitope tag for protein labelling, library generation and signal amplication, in both imaging and non-imaging applications. The functionality of the FP11 tags can be further expanded. For example, we have shown that the point mutations converting GFP to CFP and YFP can be successfully introduced to GFP1-10. Similarly, other functional mutants of GFP, such as blue uorescent protein (BFP)15, photoactivatable GFP (PA-GFP)29 and the pH-sensitive pHluorin30, may possibly be adaptable for FP11 labelling, too. Of course, as in any protein labelling method, validating the proper functionality of the labelled protein is always essential, especially when tandem GFP11-tag is used. In our case, GFP11x7 was well tolerated by multiple proteins including IFT20 and Cas9. Therefore, we expect a broad applicability for tandem GFP11-tag in helping live cell imaging to overcome the issues of poor signal, photobleaching and phototoxicity. We note that when combined with genetic

knock-in, the tandem FP11 tag could help the visualization of proteins with low endogenous expression levels. However, knocking in tandem FP11 tags require longer donor ssDNAs that contain the larger insert and potentially longer homology arms. While such long ssDNAs cannot be directly synthesized, they can be produced by reverse transcription methods31 to avoid the involvement of cloning.

Methods

Molecular cloning. The amino-acid sequence of super-folder mCherry (sfCherry) was obtained from the published literature18. sfCherry was split between 208D and 209Y at the middle of the loop between b-strands 10 and 11 (Fig. 1a). For the nucleotide sequence of sfCherry1-10 and sfCherry11, see Supplementary Tables 1 and 2.

The DNAs of GFP11 (ref. 5) and sfCherry11 were directly synthesized. The DNAs of H2B, zyxin, clathrin light chain, b-actin were subcloned from the corresponding sfGFP or mEmerald fusion plasmids (cDNA source: the Michael Davidson Fluorescent Protein Collection at the UCSF Nikon Imaging Center). We performed the following restriction enzyme digestion (amino-acid linker length shown in parentheses for each): histone H2B (10 a.a.): sfGFP sequence between AgeI and BglII (sfGFP-H2B C-10); zyxin (6 a.a.): sfGFP sequence between BamHI and NotI (sfGFP-zyxin-6); clathrin light chain (15 a.a.): mEmerald sequence between NheI and BglII (mEmerald-clathrin-15); b-actin (18 a.a.): mEmerald sequence between AgeI and BglII (mEmerald-actin-C-18). PCR-amplied GFP11 as well as sfCherry11 fragments were then inserted into the digested vectors using In-Fusion assembly (Life Technologies).

For the cloning of GFP11-tagged Drosophila calreticulin, GFP11 was inserted at the nineteenth a.a. position where the signal peptide ends. GFP11::calreticulin was inserted into Drosophila expression pACUH vectors (source: Yuh-Nung Jan) at the EcoRI/XbaI sites. For the cloning of GFP11-tagged human b2AR, the signalling

6 NATURE COMMUNICATIONS | 7:11046 | DOI: 10.1038/ncomms11046 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11046 ARTICLE

a

dCas9::VP64

dCas9::GFP11x7 GFP1-10::VP64

dCas9::VP64

dCas9

dCas9 dCas9+ =

VP64

VP64

+sgCXCR4 +sgCXCR4

Activation

CXCR4 CXCR4

b

c

+sgNegative control +sgCXCR4

Fold change

BFP GFP CXCR4

10

0 20 30 40 50

dCas9::GFP11 x7

+GFP1-10::VP64dCas9::VP64

+sgNegative

+sgCXCR4

+sgNegative

+sgCXCR4

dCas9::GFP11 x7

+GFP1-10::VP64

Figure 5 | Controlled protein multimerization using GFP11-tag. (a) Schematic of CXCR4 gene activation by either dCas9::VP64 or dCas9::GFP11x7 GFP1-

10::VP64. (b) Plot of CXCR4 uorescence signals measured by ow cytometry. Only green uorescence positive cells were analysed, which indicated the expression of both dCas9::GFP11x7 and GFP1-10::VP64. Error bars are s.e.m. (c) Images of K256 cells infected by lentivirus expressing the two dCas9 constructs with either a targeting or a non-targeting sgRNA. The cells were stained with a uorescently labelled anti-CXCR4 antibody as an indicator of CXCR4 expression. All the scale bars indicate 5 mm.

peptide (5HT3R: MRLCIPQVLLALFLSMLTGPGEGS), b2AR and GFP11 were synthesized and cloned into pcDNA3.1 vectors at the BamHI and XhoI sites.

For the expression of GFP1-10, GFP1-10 was synthesized and cloned into pACUH vectors, pcDNA3.1 vectors as well as modied lentiviral pHR-SFFV vector32. For GFP1-10 sequence information, see Supplementary Table 1. For the expression of sfCherry1-10, sfCherry1-10 was synthesized and cloned into pcDNA3.1 vectors.

To generate an ER-localized GFP1-10 (GFP1-10ER), a signal peptide (50-ATG ATGTGGTG CAAAACAGTGATAGTGTTGCTGGCGACAGTCGGCTTTATTA GTGCC-30) and an ER retention sequence (50-AGCGAACACGACGAATTG-30)

were fused to the N-terminal and the C-terminal of GFP1-10, respectively. The SP::GFP1-10::SEHDEL fragment was cloned into pACUH vectors. For mammalian expression, we cloned SP::GFP1-10 into pcDNA3.1 vectors.

For CFP and YFP imaging, we introduced point mutations in GFP1-10. It has been previously reported that the introduced point mutations Y66W and T203Y in GFP alter the GFP spectral properties to CFP and YFP, respectively. We generated the corresponding mutations in GFP1-10 with the Q5 site-Directed Mutagenesis Kit (NEB). The primers were designed using the NEB online primer design software (NEBaseChanger). The used primers were as follows: YFP_forward (50-CTACCTCTCAtatCAAACAGTCCTGAGCAAAGATC-30), YFP_reverse (50-TGATTATCAGGAAGAAGTACC-30), CFP_forward (50-ACGCTTACGTgg GGAGTTCAGTGC-30), and CFP_reverse (50-TGTTACGAGAGTCGGCCA-30). We fused these colour variants to the N terminus of Drosophila b-actin and cloned them into pACUH vectors.

To prepare GFP11 repeat arrays constructs, we rst synthesized GFP11 x1, x3 or

x7 with long (15 a.a.) or short (5 a.a.) linkers (see the sequence information in Supplementary Table 2). Then, these GFP11 fragments were fused to b-tubulin, mouse IFT20 and dCas9. Specically, for the cloning of GFP11 x1,x3 or x7::

mCherry::b-tubulin, Drosophila b-tubulin (cDNA source: R.Vale) was cloned into pACUH vectors. We then inserted mCherry-fused GFP11 x1,x3 or x7 at the EcoRI

site. For the cloning of GFP11x1 or x7::Ift20, we rst fused Ift20 with GFP11 x1 or x7, and then we inserted it into the EcoRI/NotI sites of pEGFP-N1 vectors. For the cloning dCas9:: GFP11x7, we modied our SunTag vector (pHRdSV40-NLS-dCas9-24xGCN4_v4-NLS-P2A-BFP-dWPRE (addgene #60910)) (ref. 2). 24xGCN4 was cut out by BamHI and NotI, and GFP11x7 was inserted into these sites.

For the cloning of sfCherry::b-actin, we synthesized full-length sfCherry and cloned it into the vector containing b-actin fragment (see above) at the BglII/NheI site. To construct sfCherry11x4::b-actin, we also synthesized sfCherry11x4, and replaced the sfCherry fragment to the tandem one at the BglII/NheI site. For the sequence information of sfCherry11x4, see Supplementary Table 2.

For our CRISPRa assay in Fig. 5, we fused an NLS sequence to the C-terminal of GFP1-10::VP64 and cloned into modied pHR vectors. For the GFP1-10::VP64:: NLS sequence information, see Supplementary Table 3.

Cell culture. The cell lines used for imaging were human HeLa (UCSF cell culture facility), HEK293FT (gifted from Bruce Conklin) and K562 cells (UCSF cell culture facility), mouse IMCD 3 cells (UCSF cell culture facility) and Drosophila S2 cells (UCSF cell culture facility). HeLa and HEK293FT cells were grown in a Dulbeccos modied Eagles medium with 10% FCS (UCSF Cell Culture Facility), 100units ml 1 streptomycin, 100 mg ml 1 penicillin and 2 mM glutamine. K562 cells were grown in RPMI-1640 with 25 mM HEPES and 2.0 g l 1 NaHCo3 in 10% fetal bovine serum (FBS), 100 units ml 1 streptomycin, 100 mg ml 1 penicillin and2 mM glutamine. Mouse IMCD3 cells were maintained in a mixture of DMEM and

Hams F12 medium (1:1 vol/vol) with 10% FBS. Drosophila S2 cells were grown in Schneiders Drosophila Medium with 10% FBS (Life Technologies).

Generation of stable cell lines and lentiviral infection. For viral production, HEK293T were plated in T25 asks. Twenty-four hours later, the 3 mg of lentiviral vector pHR-SFFV-GFP1-10, 3 mg of lentiviral packaging plasmid pCMV-dR8.91 and0.3 mg of envelop plasmid pMD2.G were transfected into the cells using FuGENE HD (Promega) following the manufacturers protocol. Twenty-four hours after transfection, the culture medium was replaced. Fifty-six hours after transfection, the lentiviral supernatant was harvested and kept at 80 C for later use.

HEK293FT cells stably expressing GFP1-10 were generated by infecting with lentivirus diluted 1:10 in DMEM medium and incubating for 24 h in the medium. We further plated the HEK293FT cells expressing GFP1-10 at one cell per well in a 96-well plate and isolated monoclonal cell lines that showed modest expression.

To construct stable cell lines and measure CRISPRa activity, K562 cells were lentivirally transduced with constructs that express either dCas9::GFP11x7 and

GFP1-10::VP64 or dCas9::VP64::BFP. The cells were additionally transduced with negative control and CXCR4 targeting sgRNA expression constructs2.

For IFT imaging, the plasmids encoding IFT20::GFP (ref. 33) and IFT20::GFP11x7 were transfected into IMCD3 cells using Lipofectamine 2000 (Life

Technologies) and then selected with Geneticin (400 mg ml 1: Life Technologies). After isolating a stable monoclonal cell line, the cells were infected by the GFP1-10 virus-particle-containing medium.

Transient expression. Mammalian expression plasmids (200 mg of each construct per well) were transfected using Lipofectamine 2000 (Life Technologies) into cultures

NATURE COMMUNICATIONS | 7:11046 | DOI: 10.1038/ncomms11046 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 7

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11046

of HeLa or HEK293FT cells grown on a 48-well plate (Eppendorf). Drosophila expression plasmids (100 mg of each construct per well) were transfected using

Effectene (QIAGEN) into cultures of S2 cells grown on a 12-well plate (Eppendorf). Note that in Fig. 3 and Supplementary Movie 2, GFP1-10 and GFP11 x7::mCherry:: b-tubulin plasmids were transfected into S2 cells at a concentration ratio of 7:1. Forty-eight hours after transfection, the cells were live-imaged or xed with 4% paraformaldehyde for later imaging. To facilitate cilia generation, IMCD3 cells were cultured in serum-free media 1 day before imaging. For two-colour imaging in Fig. 3f, HeLa cells were infected with GFP1-10 lentivirus and, 24 h later, 50 ng of Clathrin light chain::GFP11, 100 ng of sfCherry1-10 and 100 ng of sfCherry11x4::

b-actin were transfected using Lipofectamine 2000 (Life Technologies).

Imaging. The cells were grown in 96-well glass bottom plates with #1.5 high-performance cover glass (In Vitro Scientic) coated with either Poly-L-Lysine (Sigma-Aldrich) or Fibronectine (Roche) and were imaged on an inverted Nikon Ti-E microscope, Yokogawa CSU-22 confocal scanner unit, 60/1.4 NA oil

objective, an Andor EM-CCD camera (iXon DU897) and Micro-Manager software. Images in Fig. 1e were taken using an inverted Nikon Ti-E microscope with wide-eld illumination, 100/1.4 NA oil objective, and a sCMOS camera (Hamamatsu

Flash 4.0). For IFT20 time-lapse imaging in Fig. 4, the cells were grown in DMEM and Hams F12 (1:1 vol:vol) medium without phenol red, and were imaged in a 37 C chamber on a Nikon Ti-E microscope with Laser TIRF system, an Andor DU897 EMCCD camera and an Apo TIRF 100/1.49 NA oil objective. All the

imaging experiments were performed at UCSF Nikon Image Center.

Image analysis. To measure the brightness of full-length FPs and split-FPs in Figs 1c and 3e, we took the images of b-actin-fused full-length FPs or split-FPs. These mCherry, sfCherry or GFP images were taken with 10 0.3 NA or 60 1.4

NA objectives, respectively. The acquired images were rst background-subtracted; each of an average value of a 10-mm diameter region of interest (ROI) outside a cell was applied for the background subtraction. Then, the mean within a 10-mm diameter ROI inside a cell was calculated for each cell. A total 6B18 cells were measured for each condition.

To characterize the tandem GFP11-tag, we rst set the excitation intensity and exposure time using a GFP11x1-linker-mCherry fusion protein (GFP11x1::mCherry)

so that the camera counts for GFP and mCherry were in the same range. We then imaged the GFP11x1, x3 and x7 -labelled b-tubulin samples using these parameters. To quantify the ratio of signal enhancement, we acquired images on a wide-eld microscope using a 10/0.3 NA objective to achieve a larger eld of view. The

acquired images were rst background-subtracted and then binarized as 0 for outside and 1 for inside the cells. The GFP-to-mCherry signal ratio was calculated for each pixel within the mask, and the mean ratio within a 10-mm diameter ROI was computed for each cell. A total 50 cells were measured for each condition.

To trace the uorescent intensities for live imaging of microtubules in Fig. 4, we used Plot z-axis Prole function in ImageJ. For each photobleaching experiment, we calculated the average uorescence intensity in a 3 mm 3 mm square ROI in

one cell over the duration of the experiment, subtracted the background measured from a cell-free region in the same eld of view, and normalized the trajectory to the average value of the rst 10 time points. Figure 4d reported the mean normalized photobleaching trajectories from ve experiments for each construct (GFP, GFP11 and GFP11x7). For the photobleaching, rates were calculated from the slope of a linear t to the initial linear segment of the decay curves.

IFT20 moving velocities were measured by kymographs analysis using ImageJ. In brief, we drew a segmented line along a cilium in an image sequence and then created a kymograph using the KymoResliceWide plugin. The velocity of IFP particles were determined by drawing straight lines along IFT trajectories in the kymograph and then measuring the angle of the lines. A detailed particle tracking protocol was described in a previous publication34.

Immunocytochemistry. Mouse monoclonal antibodies used were anti-GFP ([1:100] Sigma-Aldrich, G6539), acetylated tubulin ([1:1,000] Sigma-Aldrich, T7451) and anti-histone H2B ([1:50] Abcam, ab52484) antibodies. Rabbit polyclonal antibodies used were anti-g-tubulin ([1:1,000] Sigma-Aldrich, T5192), anti-lamin A/C ([1:20] Santa Cruz Biotechnology, H110), anti-cAMP protein kinase catalytic subunit ([1:1,000] Abcam, ab26322) and anti-CBX/HP1 beta antibodies ([1:100] Abcam, ab10478). TRITC or Alexa Fluor 647 conjugated secondary antibodies (Jackson Immuno Research Laboratories, Inc. or Life Technologies, respectively) were used for indirect immunouorescence detection. BG-conjugated dyes, including SNAP-Surface Alexa Fluor 647 and 546 (NEB), were used for staining SNAP-tag expressed in cells.

The cells were xed with 4% formaldehyde (Electron Microscopy Sciences) for 15 min, permeabilized with 0.1% Triton X-100 in phosphate-buffered saline (PBS), stained for primary antibodies and Alexa-Fluor-conjugated secondary antibodies, or BG-conjugated uorophores for SNAP-tag at 4 C overnight.

Design of sgRNAs for CRISPR-mediated knock-in. We searched the human genome for CRISPR target sites (GN20GG). We used the px330 vector provided through AddGene to generate the sgRNA and Cas9 expression construct. For the individual target sequences, see Supplementary Table 4.

Design of TaqMan PCR primers and probes. For the design of TaqMan PCR primers and probes, we used the PrimerQuest software (Integrated DNA Technologies) and the Primer-BLAST (NCBI). For the probes, the PrimerQuest default setting gave us optical probe sequences, whereas the primers were identied by the Primer-BLAST. We designed that one of the primers would anneal inside of the donor sequence but not another. The probes and primers were purchased from Integrated DNA Technologies. For sequence of the primers and probes, see Supplementary Table 6.

Design of homologous recombination templates. To prepare EGFP-LMNA homologous recombination templates (70 or 200 bp homology arms), we synthesized two DNA fragments (LMNA 50 regulatory region one half of a

GFP coding sequence (1350 bp), and another half of a GFP coding sequence (350717 bp) LMNA coding region). Subsequently, they were cloned into the

pcDNA3.1 vector in which a CMV promoter sequence was removed.ssDNA donors (200 nt in total length) were manually designed by adding 70 nt

homology sequences to both side of the GFP11 sequence. ssDNA donors were purchased from Integrated DNA Technologies. For the sequence of oligo-nucleotide donors, see Supplementary Table 5.

CRISPR-mediated knock-in using plasmid transfection. For the knock-in experiments, 200 ng of Cas9 sgRNA vector and 400 ng of an oligonucleotide

donor DNA were transfected to HEK293FT cells stably expressing GFP1-10 per 24-well plate (Eppendorf). Genomic DNA was extracted from cells 3 days after transfection with the DNAzol reagent (Molecular Research Center, Inc). We further enriched GFP-positive cells by uorescence-activated cell sorting.

Measurement of knock-in efciency using TaqMan/ddPCR. The premixtures of TaqMan probes and primers consisted of 5 mM of TaqMan probe, 18 mM of a forward primer and 18 mM of a reverse primer. To optimize PCR annealing temperatures and characterize detection efciencies, we mixed Bio-Rad ddPCR Supermix for Probes (12.5 ml), FAM- and HEX probe and primer premixture(1.25 ml each), and mixture of a genomic DNA (100 ng) and a knock-in allele plasmid (0.5 pg or series of diluted plasmid) in 25 ml total volume. Droplet generation, PCR reaction and droplets read were performed by a QX100 Droplet Generator, a C1000 Thermal Cycler and a QX100 Droplet Reader (Bio-Rad), respectively according to the instructions from the manufacturer. Droplets were analysed by QuantSoft software (Bio-Rad). The optimal annealing temperatures for LMNA, HIST2H2BE, CBX1 and PRKACA were 62, 60, 60, 59 and 58 C, respectively.

To measure the knock-in efciency, we mixed Bio-Rad ddPCR Supermix for Probes (12.5 ml), FAM- and HEX probe and primer premixture (1.25 ml each), and HEK293FT genomic DNA (100 ng) in 25 ml total volume. After droplet generation and PCR reaction, the knock-in frequencies were calculated by taking concentration ratio of a mutant-allele-specic FAM probe and a wild-type-allele-specic HEX probe (TUBA1A genomic locus).

Transcription activation assay. For CXCR4 gene activation, the cells were dissociated in Gibco Cell Dissociation Buffer and then stained in PBS with 10% FBS for 1 h at room temperature using anti-human CXCR4 antibody (2 mg ml 1;

Biolegend, 12G5) conjugated to allophycocyanine. CXCR4 activation was measured using LSR II ow cytometer (BD Biosciences) at 72 or 96 h post infection as written above25.

References

1. Fernandez-Suarez, M. & Ting, A. Y. Fluorescent probes for super-resolution imaging in living cells. Nat. Rev. Mol. Cell Biol. 9, 929943 (2008).

2. Tanenbaum, M. E., Gilbert, L. A., Qi, L. S., Weissman, J. S. & Vale, R. D. A protein-tagging system for signal amplication in gene expression and uorescence imaging. Cell 159, 635646 (2014).

3. Irannejad, R. et al. Conformational biosensors reveal GPCR signalling from endosomes. Nature 495, 534538 (2013).

4. Caussinus, E., Kanca, O. & Affolter, M. Fluorescent fusion protein knockout mediated by anti-GFP nanobody. Nat. Struct. Mol. Biol. 19, 117121 (2012).

5. Cabantous, S., Terwilliger, T. C. & Waldo, G. S. Protein tagging and detection with engineered self-assembling fragments of green uorescent protein. Nat. Biotechnol. 23, 102107 (2005).

6. Barondeau, D. P., Putnam, C. D., Kassmann, C. J., Tainer, J. A. & Getzoff, E. D. Mechanism and energetics of green uorescent protein chromophore synthesis revealed by trapped intermediate structures. Proc. Natl Acad. Sci. USA 100, 1211112116 (2003).

7. Kent, K. P., Childs, W. & Boxer, S. G. Deconstructing green uorescent protein.J. Am. Chem. Soc. 130, 96649665 (2008).8. Van Engelenburg, S. B. & Palmer, A. E. Imaging type-III secretion reveals dynamics and spatial segregation of Salmonella effectors. Nat. Methods 7, 325330 (2010).

8 NATURE COMMUNICATIONS | 7:11046 | DOI: 10.1038/ncomms11046 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11046 ARTICLE

9. Kaddoum, L., Magdeleine, E., Waldo, G. S., Joly, E. & Cabantous, S. One-step split GFP staining for sensitive protein detection and localization in mammalian cells. Biotechniques 49, 727728 730, 732 passim (2010).

10. Hyun, S. I., Maruri-Avidal, L. & Moss, B. Topology of endoplasmic reticulum-associated cellular and viral proteins determined with split-GFP. Trafc 16, 787795 (2015).

11. Pinaud, F. & Dahan, M. Targeting and imaging single biomolecules in living cells by complementation-activated light microscopy with split-uorescent proteins. Proc. Natl Acad. Sci. USA 108, E201E210 (2011).

12. Feinberg, E. H. et al. GFP reconstitution across synaptic partners (GRASP) denes cell contacts and synapses in living nervous systems. Neuron 57, 353363 (2008).

13. Kim, Y. E., Kim, Y. N., Kim, J. A., Kim, H. M. & Jung, Y. Green uorescent protein nanopolygons as monodisperse supramolecular assemblies of functional proteins with dened valency. Nat. Commun. 6, 7134 (2015).

14. Wang, S., Moftt, J. R., Dempsey, G. T., Xie, X. S. & Zhuang, X. Characterization and development of photoactivatable uorescent proteins for single-molecule-based superresolution imaging. Proc. Natl Acad. Sci. USA 111, 84528457 (2014).

15. Heim, R. & Tsien, R. Y. Engineering green uorescent protein for improved brightness, longer wavelengths and uorescence resonance energy transfer. Curr. Biol. 6, 178182 (1996).

16. Ormo, M. et al. Crystal structure of the Aequorea victoria green uorescent protein. Science 273, 13921395 (1996).

17. Fan, J. Y. et al. Split mCherry as a new red bimolecular uorescence complementation system for visualizing protein-protein interactions in living cells. Biochem. Biophys. Res. Commun. 367, 4753 (2008).

18. Nguyen, H. B., Hung, L. W., Yeates, T. O., Terwilliger, T. C. & Waldo, G. S. Split green uorescent protein as a modular binding partner for protein crystallization. Acta. Crystallogr. D Biol. Crystallogr. 69, 25132523 (2013).

19. Huh, W. K. et al. Global analysis of protein localization in budding yeast. Nature 425, 686691 (2003).

20. Ghaemmaghami, S. et al. Global analysis of protein expression in yeast. Nature 425, 737741 (2003).

21. Miyaoka, Y. et al. Isolation of single-base genome-edited human iPS cells without antibiotic selection. Nat. Methods 11, 291293 (2014).

22. Lin, S., Staahl, B. T., Alla, R. K. & Doudna, J. A. Enhanced homology-directed human genome engineering by controlled timing of CRISPR/Cas9 delivery. Elife 3, e04766 (2014).

23. Carlton, P. M. et al. Fast live simultaneous multiwavelength four-dimensional optical microscopy. Proc. Natl Acad. Sci. USA 107, 1601616022 (2010).

24. Ishikawa, H. & Marshall, W. F. Ciliogenesis: building the cells antenna. Nat. Rev. Mol. Cell Biol. 12, 222234 (2011).

25. Gilbert, L. A. et al. CRISPR-mediated modular RNA-guided regulation of transcription in eukaryotes. Cell 154, 442451 (2013).

26. Gilbert, L. A. et al. Genome-scale CRISPR-mediated control of gene repression and activation. Cell 159, 647661 (2014).

27. Zalatan, J. G. et al. Engineering complex synthetic transcriptional programs with CRISPR RNA scaffolds. Cell 160, 339350 (2015).

28. Chavez, A. et al. Highly efcient Cas9-mediated transcriptional programming. Nat. Methods 12, 326328 (2015).

29. Patterson, G. H. & Lippincott-Schwartz, J. A photoactivatable GFP for selective photolabeling of proteins and cells. Science 297, 18731877 (2002).

30. Miesenbock, G., De Angelis, D. A. & Rothman, J. E. Visualizing secretion and synaptic transmission with pH-sensitive green uorescent proteins. Nature 394, 192195 (1998).

31. Chen, K. H., Boettiger, A. N., Moftt, J. R., Wang, S. & Zhuang, X. Spatially resolved, highly multiplexed RNA proling in single cells. Science 348, aaa6090 (2015).

32. Chen, B. et al. Dynamic imaging of genomic loci in living human cells by an optimized CRISPR/Cas system. Cell 155, 14791491 (2013).

33. Ishikawa, H. et al. TTC26/DYF13 is an intraagellar transport protein required for transport of motility-related proteins into agella. Elife 3, e01566 (2014).

34. Ishikawa, H. & Marshall, W. F. Efcient live uorescence imaging of intraagellar transport in mammalian primary cilia. Methods Cell Biol. 127, 189201 (2015).

Acknowledgements

We thank Yuichiro Miyaoka, Amanda Chan and Bruce Conklin for helping with TaqMan/ddPCR assay, and Wei Qiang Ong for the CD86-SNAP-tag vector and the b2AR vector. This work is supported by NIH DP2OD008479 (to B.H.), R21MH101688 (to B.H. and D.K.), R01DA036858 (to L.A.G., M.D.L. and J.S.W.), R01GM097017(to H.I. and W.F.M.), the W.M. Keck Foundation (to J.H., B.C. and B.H.). S.S. receives support from the Japan Society for the Promotion of Science postdoctoral fellowship for research abroad.

Author contributions

D.K., S.S. and B.H. conceived and designed the experiments. D.K. and S.S. performed the protein labelling and tandem GFP11 tag imaging experiments. B.B.-R. and D.K. performed the split sfCherry experiments. J.H., S.S., M.D.L. and D.K. performed the CRISPR knock-in experiments using plasmid transfection. B.C. generated and validated GFP11x7 tagged dCas9. H.I., D.K. and W.F.M. performed the IFT tracking experiment.

J.S.W. and B.H. conceived the CRISPRa experiment. L.A.G. performed the CRISPRa experiment. D.K. and B.H. wrote the manuscript.

Additional information

Supplementary Information accompanies this paper at http://www.nature.com/naturecommunications

Web End =http://www.nature.com/ http://www.nature.com/naturecommunications

Web End =naturecommunications

Competing nancial interests: The authors declare no competing nancial interests.

Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions

Web End =http://npg.nature.com/ http://npg.nature.com/reprintsandpermissions

Web End =reprintsandpermissions /

How to cite this article: Kamiyama, D. et al. Versatile protein tagging in cells with split uorescent protein. Nat. Commun. 7:11046 doi: 10.1038/ncomms11046 (2016).

This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the articles Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/

Web End =http://creativecommons.org/licenses/by/4.0/

NATURE COMMUNICATIONS | 7:11046 | DOI: 10.1038/ncomms11046 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 9