GECIs are indispensable tools for real-time monitoring of Ca2+ signals1 and cellular activities2. The SBR (ΔF/F0), defined as the ratio of the absolute fluorescence (F) changes (F – F0, or ΔF) over the basal fluorescence (F0), is a key parameter used to gauge the performance of monocolored GECIs3. Efforts have been devoted to generate GECIs with faster kinetics (for example, jGCaMP8 series4), but progress toward increased maximal fluorescence change has remained relatively lagging since the development of GECO and GCaMP6 series approximately 10 years ago5,6.

With a Ca2+-sensing module installed within one fluorescent protein (FP), single-FP-based indicators use Ca2+-dependent fluorescence changes to report Ca2+ transients. Calmodulin (CaM), together with its target peptide (such as RS20 or M13) is among the most commonly used Ca2+ sensing module. Two strategies have been applied to link CaM-M13 with FP: (1) GCaMP-like design6 to install CaM and M13 to the C terminus and N terminus of an FP; and (2) NCaMP7-like strategy7 to insert CaM-M13 into the middle of an FP8. Modifications within the linkers or interaction interfaces among CaM, M13 and FP were successful strategies to improve the performance of GCaMP variants4,6,9. Nevertheless, further improvements in dynamic range (DR) are restricted by the brightness of enhanced green fluorescent protein (EGFP). While NCaMP7 or mNG-GECO10 was built upon the brightest monomeric green FP, mNeonGreen (mNG)11, they exhibited a relatively small in cellulo DR7. By combining the advantages of both the GCaMP and NCaMP7 series, we set out to develop substantially improved GECIs with fast speed and high DRs building upon mNG.

Engineering of mNG-based calcium indicators

Single-FP-based indicators share some structural similarities at the sensing module insertion sites8,12. Assuming that the design strategies for CaM-based indicators might be transferable in principle among GECIs, we created a series of mNG-based calcium indicator (NEMO) constructs mostly by applying known GECI design strategies toward mNG, and screened their performance in HEK293 cells (Fig. 1a–c and Supplementary Tables 1 and 2).

Fig. 1: Screening and in vitro characterization of NEMO indicators.
figure 1

a, NEMO sensors are generated by introducing amino acid substitutions in NCaMP7. Top panel, diagram showing the design of NEMO variants; table (middle panel) or NCaMP7 structure7 (bottom panel) showing key amino acids substitutions introduced into NCaMP7 to generate NEMO variants. be, Screening of GCaMP and NCaMP7 variants in HEK293 cells. b, Ca2+ imaging-based screening. A typical trace from NCaMP7-expressing cells is shown; a.f.u. arbitrary fluorescence units. To avoid saturation of the camera, after recording the basal fluorescence (F0) with regular exposure time (approximately 500 ms), time-series for variants with high dynamic range were recorded using one-tenth to one-fifth the exposure time. Afterwards, the fluorescence response curves of each cell were scaled up according to the corresponding F0. After recording F0, endoplasmic reticulum Ca2+ store was depleted using 2.5 μM iono and 1 μM TG. The cells were then incubated in an imaging solution containing 300 μM EGTA, to read minimal GECI fluorescence (Fmin). Finally, the cells were exposed to imaging solution containing 100 mM Ca2+ to obtain the maximal response (Fmax) via SOCE. c, Representative traces of GCaMP6m and NCaMP7 (left) or selected NEMO sensors (right). d, Scatter plot of F0 – mean dynamic range (DR, (Fmax – Fmin)/Fmin) of the indicated GECIs. e, In vitro dose–response curves of NEMO sensors. Kd, dissociation constant for Ca2+. Top, typical traces; bottom, statistics (see Supplementary Table 3 for details) (n = 3 independent biological replicates; >17 cells per repeat). Data are shown as mean ± s.e.m. f, Basal brightness of NEMO, NCaMP7 or GCaMP6m sensors viewed with YFP (top) or GFP (bottom) filters. To achieve better estimation of the basal fluorescence of GECIs (FGECI), FGECI of cells expressing mKate-P2A-GECI constructs was normalized against the fluorescence of mKate, an expression marker (FmKate). (GCaMP6m, n = 99 cells; NEMOm, n = 142 cells; NEMOb, n = 119 cells NEMOc, n = 89 cells; NEMOs, n = 91 cells; NEMOf, n = 86 cells; NCaMP7, n = 114 cells). Three independent biological repeats. Data are shown as mean ± s.e.m. g, F0 – dynamic range of individual cells expressing NEMO variants or GCaMP6m examined with a YFP filter set.

Source data

We first evaluated their basal fluorescence (F0) and the ratio between maximal (Fmax) and minimal (Fmin) fluorescence, or DR (Fmax – Fmin)/Fmin). To allow measurements of Fmin, the endoplasmic reticulum (ER) Ca2+ store was depleted by 10-min incubation with 2.5 μM ionophore ionomycin (iono) and 1 μM thapsigargin (TG—an inhibitor of the sarcoplasmic/endoplasmic reticulum Ca2+ ATPase). A high amount of Ca2+ (100 mM) was added to the bath to induce Fmax via store-operated Ca2+ entry (SOCE) (Fig. 1b). Top candidates (Fig. 1c) were identified based on both the F0 and DR values (Fig. 1d). We found that only the NCaMP7-like7 design (Fig. 1a) showed improved dynamics and speed (Supplementary Tables 1 and 2). We identified five best-performing constructs and named them as NEMO, including medium (NEMOm), high contrast (NEMOc), fast (NEMOf), bright (NEMOb) and sensitive (NEMOs) versions (Fig. 1d and Extended Data Figs. 1 and 2a).

Ex vivo characterization of NEMO sensors

The overall in cellulo DR of NEMO sensors is higher compared with that of top-of-the-range GECI proteins tested side-by-side. The DRs of NEMOs or NEMOb (102.3 ± 4.0 or 128.8 ± 3.1, respectively) were at least 4.5-fold higher than that of GCaMP6m or NCaMP7 (Fig. 1d). And the DRs of NEMOm and NEMOc were further increased to 240.7 ± 7.6 and 422.2 ± 15.3, respectively, 9.5- to 25.7-fold higher than those of GCaMP6m or NCaMP7 (Fig. 1d and Supplementary Video 1).

We further assessed the in vitro performance of NEMO variants. Except for NEMOb, four NEMO sensors showed DR values either close to (NEMOs) or larger than 100-fold (Fig. 1e and Extended Data Fig. 2b). We thus did not focus on NEMOb for further characterization. To our surprise, in vitro DR values were smaller than their corresponding in-cell ones, as GCaMP-like design usually show the opposite13. We speculate that macromolecular crowding and reducing condition in cytosolic environment may account for this higher in-cell DR14,15, which warrants follow-on studies in the future.

We next examined the basal fluorescence of NEMO sensors with a P2A-based bicistronic vector to drive the coexpression of mKate (as an expression marker) and GECIs at a near 1:1 ratio. The basal GECI brightness was indicated by the fluorescence ratio of GECI and mKate (Fig. 1f). Normalized basal brightness of all NEMO sensors was much lower than that of NCaMP7, and the brightness of NEMOc or NEMOf was only around 0.25–0.5 of GCaMP6m. This finding indicates that the lower basal fluorescence of NEMO variants might contribute to the observed large DR of NEMO indicators, in particular for NEMOc and NEMOf. However, although the DRs of NEMOm, NEMOs and NEMOb were over fivefold higher than that of GCaMP6m, their basal fluorescence was either similar to, or brighter than, that of GCaMP6m (Fig. 1d,f). Hence, high DRs for these three indicators could be attributed to their maximal brightness being larger than that of GCaMP6m as well. In consonance with this notion, NEMO-expressing cells with comparable basal fluorescence to those expressing GCaMP6m still exhibited larger dynamics (Fig. 1g).

Using NEMOc as an example, we next set out to decipher the mechanisms underlying the high DR of NEMO sensors (Extended Data Fig. 3a). Similar to most GECIs7,16,17, the fluorophore of NEMOc adopted two configurations: an anionic state (peak at 509 nm) and a neutral state (403 nm) (Extended Data Fig. 3b). The Ca2+-induced brightening of NEMOc fluorescence is similarly caused by increasing both the proportion and molecular brightness of anionic form (Extended Data Fig. 3c–e and Supplementary Table 4)7,16,17. The increase in DR was associated mostly with the considerably dimmer anionic fluorophore of NEMOc (0.22 ± 0.01 mM–1 cm–1) in the absence of Ca2+, which was approximately one-sixth that of NCaMP7 and one-fifth that of GCaMP6m. Compared with GCaMP6m, the high DR of NEMOc was also a result of increased brightness of Ca2+-saturated anionic NEMOc (64.26 ± 2.67 mM–1 cm–1), approximately three times that of GCaMP6m. Moreover, the in vitro and in cellulo DR of NEMOs under two-photon excitation remained largely similar to those observed with one-photon excitation (Extended Data Fig. 4a–c and Supplementary Table 5).

To examine whether it is possible to compensate weaker basal NEMO fluorescence with stronger illumination, we measured the photostability of NEMO sensors. NEMO indicators showed better photostability than GCaMP6m or mNG11 (Extended Data Fig. 5a, top left two panels). It endured nearly 40 times (1.52 mW) higher illumination than GCaMP6m (0.04 mW), and showed no apparent photobleaching. The stronger illumination (from 0.04 mW to 1.52 mW) could potentially enhance the basal fluorescence of NEMOm sensor by over 60-fold (Extended Data Fig. 5a, top right panel), broadening the applicability of NEMO sensors in scenarios requiring stronger light illumination, such as monitoring Ca2+ signals in vivo within subcellular compartments with dim NEMOf indicator.

Performance of NEMO sensors in nonexcitable cells

We next examined the ability of NEMO sensors to report signals induced by submaximal activation of muscarinic acetylcholinergic receptors with carbachol (CCh, 10 μM). As to the CCh-induced Ca2+ transients in HEK293 cells, peak signal-to-baseline (SBR, ΔF/F0) values of NEMOb and NEMOs were at least three times higher than those of GCaMP6m, jGCaMP8f (ref. 4) and NCaMP7 (Fig. 2a and Extended Data Fig. 5b,c). The SBRs of NEMOm (101.9 ± 6.6), NEMOc (112.0 ± 9.8) and NEMOf (194.3 ± 7.7) were 13–25 times higher than that of GCaMP6m. Similarly, the performance of NEMO sensors in reporting subsequent Ca2+ oscillations were also superior to those of jGCaMP8f and NCaMP7 (Fig. 2a, Extended Data Fig. 5b,c and Supplementary Video 2).

Fig. 2: Performance of NEMO sensors in nonexcitable mammalian cells.
figure 2

a, Typical Ca2+ oscillations in HEK293 cells induced by CCh (10 μM), as indicated by GCaMP6m, NCaMP7 and NEMO sensors; n = 3 independent biological replicates, with at least 15 cells per repeat. b, Ca2+ release and SOCE responses induced by TG; n = 3 independent biological replicates, with at least 20 cells per repeat. Data are shown as mean ± s.e.m. c, Fructose-elicited response in cells coexpressing BmGr-9, an insect fructose receptor. Left, typical traces; right, statistics (****P = 2.52 × 10–5 unpaired Student’s t-test, two-tailed) (GCaMP6s, n = 45 cells; NEMOs, n = 28 cells examined over three independent biological replicates). Data are shown as mean ± s.e.m. d,e, Measurements of Ca2+ concentration with NEMO sensors. d, Ratiometric measurements with NEMOs. NEMO transients in cells upon excitation at 488 nm or 405 nm, induced by 100 μM CCh. Left, typical NEMOs fluorescence response when excited at 488 nm or 405 nm. Right, representative intensiometric (black) or ratiometric (red, F488ex/F405ex) responses of the same set of cells shown on the left. n = 3 independent biological replicates, with at least 16 cells per repeat. Data are shown as mean ± s.e.m. e, iPEAQ of Ca2+ levels. Shown are fluorescence intensities (red) and [Ca2+] traces (green) of NEMO-expressing cells in response to 10 μM CCh. In-cell calibration to determine the absolute Ca2+ concentration, or [Ca2+], is shown in Extended Data Fig. 7g. n = 3 independent biological replicates, with at least nine cells per repeat.

Source data

We further examined the performance of NEMO sensors in detecting weak Ca2+ signals in HEK293 cells. In response to TG-induced Ca2+ releases and SOCE, the peak SBR values of NEMO sensors were at least five times higher than that of NCaMP7 (Fig. 2b and Extended Data Fig. 5d). When monitoring Ca2+ transients induced by the Bombyx mori gustatory receptor (BmGr-9), whose amplitude is much smaller than that of SOCE18, the responses of NEMO sensors were much stronger than that of GCaMP6s (Fig. 2c).

We next examined whether the larger DR of NEMO sensors could enable more sensitive discrimination of Ca2+ signals with varying amplitudes by comparing their performance with GECIs bearing comparable Ca2+-binding affinities. We first examined the responses of GECIs responses to the stepwise increase in Ca2+ influx induced by a coexpressed optogenetic tool, Opto-CRAC, when subjected to varying photoactivation duration19,20,21. Compared with the GCaMP6m signals, the NEMOm signals were significantly larger (NEMOm, 1,000 ms versus 300 ms, P = 4.3 × 10–7; GCaMP6m, 1,000 ms versus 300 ms, P = 0.0178; paired Student’s t-test, two-tailed), showing a stepwise increase in response to prolonged photostimulation (Extended Data Fig. 6a and Supplementary Video 3). Second, we compared the graded SOCE signals in responses to increased extracellular Ca2+ concentrations. NEMOm or NEMOs could discriminate more external Ca2+ gradients than did GCaMP6m or NCaMP7 (Extended Data Fig. 6b,c), with SNRs significantly higher than their corresponding counterparts (Extended Data Fig. 6d, 100 ms, P = 2 × 10–15; 300 ms, P = 1.03 × 10–22; 1,000 ms, P = 5.06 × 10–18; Extended Data Fig. 6e, 0.1 mM, P = 6.95 × 10–23; 0.3 mM, P = 7.01 × 10–26; 1 mM, P = 3.79 × 10–36; 3 mM, P = 1.56 × 10–36; 10 mM, P = 5.85 × 10–37; Extended Data Fig. 6f, 0.1 mM, P = 1.13 × 10–13; 0.3 mM, P = 5.13 × 10–16; 0.6 mM, P = 8.85 × 10–20; 1 mM, P = 1.44 × 10–21; 3 mM, P = 3.58 × 10–22; unpaired Student’s t-test, two-tailed).

One major drawback of intensiometric Ca2+ sensors is that they could not report Ca2+ concentration directly. We thus asked whether NEMO could serve as ratiometric sensors to measure absolute Ca2+ concentrations. Indeed, the fluorescence of NEMOs excited by 405-nm light (F405) was reduced as a function of increasing Ca2+ concentration (Extended Data Fig. 7a). Consequently, the in vitro DR indicated by the F490/F405 ratio was 3.4-fold higher than that obtained with F490 only (Extended Data Fig. 7b). Similarly, the in-cell ratiometric DRs were significantly larger than intensiometric ones (P = 0.0002, unpaired Student’s t-test, two-tailed) (Fig. 2d and Extended Data Fig. 7c,d).

Violet light illumination approximately doubled the fluorescence of NEMO sensors under low Ca2+ conditions, indicating the existence of a photochromic effect22 (Extended Data Fig. 7e). Indeed, further tests showed that brief 405-nm illumination superimposed on 488-nm light could increase NEMOf fluorescence in an inversely Ca2+ dependent manner, with the peak fluorescence named as F0. After switching off the violet light, NEMOf quickly relaxed back to its basal state, termed as the minimal fluorescence (Fend) (Extended Data Fig. 7f). One such photochromic cycle would allow the calculation of its photochromism contrast, defined as ((F0 – Fend)/F0)hv (ref. 22). We thus used NEMOs to report Ca2+ concentration with the intermittent photochromism-enabled absolute quantification (iPEAQ) method22. Using basal photochromism contrast together with two in vitro Ca2+-titration curves (Extended Data Fig. 7g), we quantified CCh-induced Ca2+ releases in terms of absolute Ca2+ concentration (Fig. 2e).

Assessing NEMO sensors in neurons and in planta

We next examined the responses of NEMO sensors in dissociated rat neurons excited by electrical field stimulation with a GCaMP-compatible imaging set up (Fig. 3). We observed that all NEMO sensors were able to detect Ca2+ signals elicited by a single action potential (AP) (Fig. 3a), with peak SBR approximately twice as high as GCaMP6s or GCaMP6f. Consistent with its in vitro Ca2+ dissociation rate (Koff) value being higher than that of GCaMP6f (Supplementary Table 3), NEMOf was fast enough to discriminate neuronal responses stimulated with a frequency up to 5 Hz (Fig. 3b).

Fig. 3: Electric field stimulation-induced NEMO responses in rat hippocampal neurons.
figure 3

a, Average Ca2+ responses reported by GECIs, 1 Hz stimulation (mean traces, n = 10–11 cells). b,c, Mean NEMOf and GCaMP6f transients induced by 5 Hz (b, left, mean traces; right, statistics; P = 0.86, unpaired Student’s t-test, two-tailed, n = 12 cells) or 10 Hz (c, mean traces, n = 11 cells) stimulation. Data are shown as mean ± s.e.m.; NS, nonsignificant. d, Mean GECI responses elicited by stimulation at varied frequencies. Inset, enlarged views of responses of reference GECI sensors (SBR magnified by nine times). e, Statistics for data shown in a and d. Each GECI measurement set was analyzed from several dendrites of at least ten neurons in three different primary hippocampal neuron cultures. All data in this figure are shown as mean ± s.e.m. (for stimulation at varied frequencies, GCaMP6s, n = 10, 10, 10, 11, 10 cells; GCaMP6f, n = 10, 11, 11, 11, 10, 10 cells; jGCaMP8f, n = 10, 10, 11, 11, 11 cells; NEMOm, n = 10, 12, 12, 13, 12 cells; NEMOs, n = 10, 10, 11, 14, 10 cells; NEMOf, n = 11, 11, 11, 12, 14 cells; NEMOc, n = 11, 11, 11, 13, 12 cells).

Source data

Similar to what we observed in nonexcitable cells, the high DR of NEMO sensors enabled high-resolution detection of Ca2+ signals of various amplitudes. In response to a 5 Hz field stimulation, the peak amplitude of NEMOf response was approximately three times that of GCaMP6f, placing NEMOf among the most sensitive and fast GECIs that include XCaMP23, jGCaMP7 (ref. 9), or jGCaMP8 series4. As the stimulus frequency increased, the difference between peak NEMOf and GCaMP6f responses became more pronounced (5 or 22.7 times higher than that of GCaMP6f; Fig. 3c,d), often with their SNRs significantly larger than their counterparts (20 Hz, P = 1.05×10–4; 60 Hz, P = 2.24 × 10–7, 100 Hz, P = 2.41 × 10–9; 180 Hz, P = 1.24 × 10–6, unpaired Student’s t-test, two-tailed) (Extended Data Fig. 8a). The NEMO responses were also fairly linear with no apparent saturation even up to 180 Hz (Fig. 3e).

Combining two-photon imaging and whole-cell electrophysiological recording, we further tested NEMO sensors in cortical neurons in acutely prepared mouse brain slices (Extended Data Fig. 8b, left panel). Under a whole-cell patch clamp condition, the intracellular environment could be perturbed and the GECI signal was diluted23. Despite this caveat, the responses of all NEMO sensors induced by AP occurring at 50 Hz or higher frequencies were significantly higher than those of GCaMP6, with NEMOf exhibiting the fastest kinetics (NEMOf versus GCaMP6s at 50 Hz, P = 0.00035; at 100 Hz, P = 1.4 × 10–5, unpaired Student’s t-test, two-tailed) (Extended Data Fig. 8b). Of note, most fast GECI sensors developed to date4,23,24,25 display rather limited DRs. By contrast, NEMOf expressed in both nonexcitable and excitable cells responds fast with high DR.

We also tested the usability of NEMO sensors in detecting subcellular Ca2+ signals in the leaves of Arabidopsis thaliana. Our super-resolution imaging results showed that NEMOm fused to the plasmodesmata-localized protein 1 could readily report Ca2+ oscillations near the plasmodesmata (Extended Data Fig. 9e and Supplementary Video 4)—a structure between plant cells with a diameter about 30–60 nm (ref. 26).

In vivo performance of NEMO sensors in rodent brains

We next tested the in vivo performance of NEMO sensors with two-photon laser microscopy (Fig. 4a) by measuring GECI responses as readout of neuronal activity evoked by drifting grating stimulus in the primary visual cortex27. To ensure direct comparison with GCaMP6 sensors, we used 920 nm light, an excitation wavelength optimized for GCaMP but less ideal for NEMO (980 nm) to excite the GECIs. All indicators reported differential changes of visual stimuli (Fig. 4b and Extended Data Fig. 8c,d). NEMOf was the fastest among all NEMO variants, with the half-decay time (409 ± 54 ms) comparable with that of GCaMP6f (482 ± 48 ms) (Extended Data Fig. 8e).

Fig. 4: In vivo performance of NEMO sensors in monitoring neuronal activities in rodent brain.
figure 4

ac, Fluorescence responses in the visual cortex of mice induced by a visual stimulus. a, Diagram showing the experimental setups for two-photon imaging of neurons in response to drift gratings. b, Typical response curves for GCaMP6s, NEMOs and NEMOf. c, Cumulative distribution of peak SBR transients of GECI sensors (for NEMOf versus GCaMP6f and NEMOs versus GCaMP6s, **P = 0.012; ****P = 2.75 × 10–53, Kolmogorov–Smirnov test, two-tailed) (GCaMP6f, n = 101 cells from two mice; GCaMP6s, n = 167 cells from three mice; NEMOs, n = 223 cells from four mice; NEMOm, n = 115 cells from three mice; NEMOf, n = 40 cells from three mice). df, Ratiometric responses of GCaMP6f and NEMOs in neurons of the mouse corpus striatum recorded by fiber photometry. d, Diagram of the experimental setup for fiber photometry recordings. e, Mean ratiometric responses elicited by pinch stimulation at the mouse tail tip. Left, mean traces; right, statistics (****P = 4.57 × 10–11, unpaired Student’s t-test, two-tailed). NEMOs, data for 102 cells from six mice; GCaMP6f, data for 59 cells from six mice. Data are shown as mean ± s.e.m. f, Cumulative distribution of peak responses shown in e (****P = 1.6 × 10–21; Kolmogorov–Smirnov test, two-tailed).

Source data

We then moved on to compare the sensitivity of GECIs in vivo. As to the fraction of responsive cells (Extended Data Fig. 8f), no significant difference between NEMO variants and the corresponding GCaMP6 indicators was detected (NEMOf versus GCaMP6f, P = 0.1356; NEMOs versus GCaMP6s, P = 0.9805, unpaired Student’s t-test, two-tailed). However, the cumulative distribution of peak ΔF/F0 of NEMOm and NEMOs was substantially right-shifted relative to the GCaMP6 signal (Fig. 4c), indicating that NEMOm and NEMOs are more responsive. The median response of NEMOs (ΔF/F0 = 3) was over four and seven times stronger than that of GCaMP6s and GCaMP6f, respectively. The visual-stimuli-induced response reported by NEMOs was much larger than existing values reported by existing sensitive GECIs4,9,25,28. In parallel, the median response of NEMOf (ΔF/F0 = 0.80) was significantly larger than that of GCaMP6f (ΔF/F0 = 0.44) (P = 0.0202, unpaired Student’s t-test, two-tailed) (right panel in Fig. 4b versus Extended Data Fig. 8c, left panel), as well as those reported by the known fastest GECIs4,9,24,25.

In addition, NEMOs showed appreciably better SNR (Extended Data Fig. 8g) and good basal fluorescence in the mouse V1 that was comparable with that of GCaMP6s even under excitation conditions optimized for GCaMP (Extended Data Fig. 9a,b). NEMOf signal obtained with GCaMP set up showed similar SNR to GCaMP6f (Extended Data Fig. 8g). Since the basal NEMOm fluorescence approximately doubled by switching from GCaMP excitation (920 nm) to a NEMO set up (980 nm), NEMOf under optimized illumination (Extended Data Fig. 9c) retained its large SBR and showed higher SNR (Extended Data Fig. 9d). Since GCaMP6s and NCaMP7 were reported to have similar in vivo SNR7, it is likely that optimally excited NEMOs (that is, at 980 nm) may exhibit a better SNR than NCaMP7.

Last, we recorded NEMOs responses in sensory neurons deeply buried in the mouse brain using fiber photometry and settings optimized for ratiometric GCaMP recordings29. The ratios of GECI fluorescence excited by 410 nm (F410) or 470 nm light (F470) were used to indicate Ca2+ responses of the neurons within the corpus striatum elicited by tail-pinching stimulus (Fig. 4d). Even though the near-UV light (410 nm) excitation reduced the DR of NEMOs and the 470 nm light was not optimal for NEMOs excitation (Extended Data Figs. 3a and 7d), the median peak response of NEMOs was approximately three times that of GCaMP6f (Fig. 4e,f).


Here, we reported a GECI toolkit with improved photochemical properties. Unlike current indicators that partially sacrifice the dynamic range for improved sensitivity and/or faster kinetics, NEMO variants are fast acting while still retaining superior dynamic ranges to report Ca2+ signals. We would like to point out that when measuring indicators with large in-cell DRs (100–300), the minimal fluorescence used for calculating DR has to be kept close to the background readings to avoid saturation of detectors. Hence, inaccurate subtraction of the background intensity might introduce calculation artifacts, resulting in overestimation of in-cell DR values. However, even with the most conserved estimation, both in vitro and in-cell DRs of NEMOc/m/f were >100-fold, much larger than those of GCaMP6. NEMO indicators are more versatile than the most popular GCaMP series, allowing simultaneous imaging with cyan fluorescence, exhibiting higher photostability that can endure substantially stronger illumination, and better resisting pH fluctuation. Overall, the NEMO sensors may serve as the tool-of-choice for monitoring Ca2+ dynamics in mammalian cells, tissue or in vivo, as well as in planta.


Plasmid construction

The coding sequence of NCaMP7 (ref. 7) was synthesized by BGI Geneland Scientific. The corresponding mutations or substitutions were all included on primers (Supplementary Table 6). After PCR, fragments of NEMO were reassembled and inserted into a pCDNA3.1(+) expression vector linearized by digestion with restriction enzymes BamHI and EcoRI via the Ready-to-Use Seamless Cloning Kit (B632219, Sangon Biotech). All plasmids were confirmed by sequencing.

Bacterial expression and protein purification

Transetta (DE3) bacteria (Transgene) transformed with pET28a plasmids containing coding sequences of sensors were cultured with 300 mM isopropyl-β-d-thiogalactoside for 12 h at 20 °C. Pelleted bacterial cells were then suspended in 20 ml buffer 1 (20 mM Tris, 300 mM NaCl and 1 mM imidazole, pH 7.2), and subsequently sonicated. Recombinant proteins were then purified by 1 ml Ni Sepharose (17-5318-01, GE Healthcare). Columns were sequentially washed with 20 ml buffer 1 and 10 ml buffer 2 (20 mM Tris, 500 mM NaCl and 10 mM imidazole, pH 7.2). Purified proteins were then eluted in 5 ml buffer 3 (20 mM Tris, 100 mM NaCl and 300 mM imidazole, pH 7.2)9.

In vitro Ca2+ titrations and kinetic measurements

Ca2+ titration: 50 μg ml–1 GECIs in buffer A (100 mM KCl, 30 mM HEPES, pH 7.2) supplemented with either 10 mM EGTA or 10 mM Ca-EGTA were mixed in various ratios9. GCaMP (excitation: 485 ± 5 nm; emission: 510 ± 5 nm) and NEMO (excitation: 490 ± 5 nm; emission: 520 ± 5 nm) fluorescence were measured with a Flexstation 3 microplate reader (Molecular Devices) controlled by SoftMax Pro v.7.x. The Ca2+ titrations curves were fit with Prism v.7 using specific binding with Hill slope function.

For measurements of Koff (ref. 9), 50 μg ml–1 GECIs in buffer A containing either 10 µM free Ca2+ or 10 mM EGTA were rapidly mixed with 1:1 ratio GCaMP and NEMO (excitation 485 nm, emission 520 nm) fluorescence were measured by POLARstar Omega microplate reader (BMG LABTECH). Koff values were calculated in Prism v.7 using single exponential regression.

Measurement of spectra and quantum yields of GECIs

Emission and excitation spectra of purified GECIs in buffer A were recorded with a FS5 spectrophotometer (Edinburgh Instruments) controlled by Fluoracle. Absorption spectra were recorded with UV2600 spectrophotometer (Shimadzu) controlled by UVprobe. Quantum yields (Φ) were determined with FS5 using 1 cm quartz cuvette. The fluorescence spectra of GECIs with different concentrations were recorded to calculate the corresponding total integrated fluorescence intensities (TIF). Linear regression of TIF minus absorbance curves were used to derive the slopes (S) of GECIs. Φ was then calculated as: Φprotein = Φstandard × (Sprotein/Sstandard) (ref. 5). For anionic chromophore 470 nm excitation; reference, fluorescein (fluo) in 0.1 M NaOH (φ = 0.925)30. For neutral chromophore, 405 nm excitation, TOLLES (φ = 0.79) (ref. 31).

Chromophore extinction coefficients

The absorption spectra of 200 μg ml–1 GECIs in buffers (30 mM trisodium citrate, 30 mM borate, with either 10 mM EGTA or 10 mM CaCl2, with or without 1 mM MgCl2) with varying pH were measured with Flexstation 3. Chromophore extinction coefficients (ε) values were then obtained with the following calculations16,17. Briefly, the corresponding absorbance (OD) in protonated (ODN, peak absorption at around 400 nm) and deprotonated (ODA, peak absorption at around 500 nm) states were first obtained from the absorption spectrum curves. Next, the slope S was calculated via linear regression using ODN – ODA values at pH = 7, 7.2, 8 and 9

$$S = \Delta \mathrm{OD}_\mathrm{A}/\Delta \mathrm{OD}_\mathrm{N}$$

The relationship between the concentration of a chromophore (n), its OD and ε is:

$$n = \mathrm{OD}/\varepsilon$$

For alkaline-denatured (D), protonated (N) and deprotonated (A) chromophores, the corresponding equation would be nD = ODD/εD, nA = ODA/εA and nN = ODN/εN. And green GECIs will be totally denatured at pH 12.5, thus n = nD. Also n will not change during H+ titration, hence

$$n_\mathrm{D} = n_\mathrm{A} + n_\mathrm{N}$$


$$\Delta n_\mathrm{A} = - \Delta n_\mathrm{N}$$

Putting equation (2) or its variants into (3) and (4) would make:

$$\mathrm{OD}_\mathrm{D}/\varepsilon _\mathrm{D} = \mathrm{OD}_\mathrm{A}/\varepsilon _\mathrm{A} + \mathrm{OD}_\mathrm{N}/\varepsilon _\mathrm{N}$$
$$\Delta \mathrm{OD}_\mathrm{A}/\varepsilon _\mathrm{A} = - \Delta \mathrm{OD}_{\it{\mathrm{N}}}/\varepsilon _\mathrm{N}$$

εD of denatured GFP-like chromophores (44,000 M–1 cm–1) was taken as εD of green GECIs at pH 12.5 (ref. 1).

And equation (6) can be rearranged as:

$$\varepsilon _\mathrm{A}/\varepsilon _\mathrm{N} = - \Delta \mathrm{OD}_\mathrm{A}/\Delta \mathrm{OD}_\mathrm{N}$$

Combining equation (1) and (7), we will obtain

$$\varepsilon _\mathrm{A}/\varepsilon _\mathrm{N} = - S$$

By solving equations (5) and (8), we can get

$${\it{\upvarepsilon }}_\mathrm{N} = \frac{{\frac{{\mathrm{OD}_\mathrm{A}}}{{ - S}} + \mathrm{OD}_\mathrm{N}}}{{\mathrm{OD}_\mathrm{D}/\varepsilon _\mathrm{D}}}$$
$$\varepsilon _\mathrm{A} = \varepsilon _\mathrm{N} \times \left( { - S} \right)$$


$$\rho _\mathrm{A} = \frac{{n_\mathrm{A}}}{{n_\mathrm{A} + n_\mathrm{N}}} = \frac{{\frac{{\mathrm{OD}_\mathrm{A}}}{{\varepsilon _\mathrm{A}}}}}{{\frac{{\mathrm{OD}_\mathrm{A}}}{{\varepsilon _\mathrm{A}}} + \frac{{\mathrm{OD}_\mathrm{N}}}{{\varepsilon _\mathrm{N}}}}}$$
$$\rho _\mathrm{N} = \frac{{n_\mathrm{N}}}{{n_\mathrm{A} + n_\mathrm{N}}} = \frac{{\frac{{OD_\mathrm{N}}}{{\varepsilon _\mathrm{N}}}}}{{\frac{{\mathrm{OD}_\mathrm{A}}}{{\varepsilon _\mathrm{A}}} + \frac{{\mathrm{OD}_\mathrm{N}}}{{\varepsilon _\mathrm{N}}}}}$$

The pKa values were calculated with the ‘specific-binding-with-Hill-slope’ function using Prism v.7.

Two-photon imaging

This was undertaken with a ZEISS LSM880 NLO microscope equipped with ×40 water-immersion objective (numerical aperture (NA) 1.0), gallium arsenide phosphide photomultiplier tubes and ZEN v.2.1 software.

In vitro excitation spectra: two-photon cross-section (δ) values were determined32 using the same imaging solutions as those in Φ measurements.

The specific calculation steps are as follows23. The time-averaged fluorescence photon flux < F(t) > is given as:

$$\left\langle {F\left( t \right)} \right\rangle \approx \frac{{\phi \eta _2C\delta g_p}}{{2f\tau }} \times \frac{{8n\left\langle {P\left( t \right)} \right\rangle ^2}}{{\pi \lambda }}$$

where δ of GECIs can be calculated as:

$$\delta \approx \frac{{2\left\langle {F\left( t \right)} \right\rangle \pi \lambda f\tau }}{{\phi g_\mathrm{p}8\eta _2Cn\left\langle {P\left( t \right)} \right\rangle ^2}}$$

Φ is the fluorescence collection efficiency of the measurement system; ŋ represents the quantum yield; ŋ2 is obtained by two-photon excitation and ŋ2 derived from one-photon excitation (we set ŋ1 = ŋ2 = 0.925 for fluorescein, assuming that ŋ was independent of excitation wavelengths. For ŋ1 of GECIs, please see Supplementary Table 4). C is the concentration of the sample; δ is the two-photon cross-section; gp is the degree of second-order temporal coherence, which depends on the pulse shape; f is the pulse repetition rate; τ is the temporal pulse width; n is the refractive index of the sample medium (n = 1.3334 for water); < P(t) > is the time-averaged instantaneous incident power of the laser beam (we consider this as a constant) and λ stands for the excitation wavelength.

As \(\frac{{\phi g_\mathrm{p} \times \left\langle {P\left( t \right)} \right\rangle ^2}}{{f\tau }}\) value is a system-specific constant at a certain wavelength setting, we could obtain δGECI (δ of GECIs) by using fluorescein at 750–990 nm as a standard33. According to the above method, we measured the < F(t) > values of each GECI and calculated the δGECI values of them by equation 15:

$$\delta _{\mathrm{GECI}} \approx \frac{{\left\langle {F\left( t \right)} \right\rangle _{\mathrm{GECI}} \times \eta _{\mathrm{2fluo}} \times C_{\mathrm{fluo}}}}{{\left\langle {F\left( t \right)} \right\rangle _{{{{\mathrm{fluo}}}}} \times \eta _{\mathrm{2GECI}} \times C_{\mathrm{GECI}}}} \times \delta _{\mathrm{fluo}}$$

Time-lapse Ca2+ imaging of HEK293 cells was acquired using 910 nm (for GCaMP6m) or 970 nm (for NCaMP7 or NEMOs) excitation, and the resulting emission between 500 nm and 600 nm was collected every 2.5 s.

Cell culture and transfection

HEK293 and HeLa cells (ATCC, catalog nos. CRL-1573 and CL-0101, respectively) were cultured regularly in DMEM supplemented with 10% FBS (and 5% penicillin and streptomycin at 37 °C with 5% CO2 (refs. 34,35). Transfection was performed by electroporation using the Bio-Rad Gene Pulser Xcell system. Transfected cells were seeded on round coverslips and cultured in OPTI-MEM medium containing 7% FBS. All experiments were carried out 24 h after transfection.

Hippocampal tissue from E18 Wistar rats was first digested with 0.1% trypsin at 37 °C for 15 min, washed with mixed complete medium (l-glutamine DMEM-F12 plus 10%FBS) and then pipetted with a Pasteur glass tube (15 mm) to get a cell suspension. Neurons were then plated at 50,000 cells per 18 mm coverslips coated with 0.1% poly-d-lysine and cultured in a 37 °C, 5% CO2 incubator. Half of the medium was replaced with neurobasal medium (containing 2% B27 supplement, 0.5 mM GlutaMAX-100X) without serum 24 h later, and once per week thereafter. Arac (10 µM) was added 7 days after plating to inhibit gliacyte. Neurons were transfected with GCaMP6 or NEMO-encoding plasmids at 7–9 days after plating using a calcium phosphate transfection kit (Takara Bio).

Fluorescence imaging

Time-lapse fluorescence imaging experiments were carried out using a ZEISS observer Z1 imaging system controlled by SlideBook v.6.0.23 (ref. 12). Cells were imaged in Ca2+ imaging buffer (107 mM NaCl, 7.2 mM KCl, 1.2 mM MgCl2, 11.5 mM glucose and 20 mM HEPES-NaOH, pH 7.2) every 1 or 2 s. Filters: NEMO/NCaMP7, 500 ± 12 nm ex/542 ± 13 nm em; GCaMP, 470 ± 11 nmex/512 ± 13 nmem.

Confocal imaging and iPEAQ measurements

Confocal imaging was undertaken with a ZEISS LSM880 microscope equipped with a ×63 oil objective (NA 1.4) and ZEN v.2.1 software. Excitation was set at 488 and 405 nm, with the emission collected at 500–580 nm.

For the iPEAQ biosensing36, an in vitro Ca2+ titration curve was first generated. NEMOs fluorescence in solutions with different free Ca2+ concentrations was imaged with the same Ca2+ concentrations as those for live-cell imaging. Eight photochromic cycles induced by 2.5 s illumination with 405 nm light were superimposed on 488 nm light to obtain parameters needed to calculate photochromism contrast. The photochromism contrast is defined by peak fluorescence (F0) in the presence of 405 nm light together with 488 nm light, and minimal fluorescence (Fend) in the presence of 488 nm light only: photochromism contrast = ((F0 – Fend)/F0)hv, or (ΔF/F0)hv. Afterwards, dose–response curves (shown as (ΔF/F0)hv) and NEMOf fluorescence were plotted and fit with a Hill equation.

In-cell iPEAQ recording of NEMOs were similarly imaged with 488 nm laser. First, five photochromic cycles with 2.5 s illumination at 405 nm were superimposed on 488 nm light to obtain basal (ΔF/F0)hv. Then these values of each single cell were applied to the (ΔF/F0)hv Ca2+ titration curve to obtain basal [Ca2+]. Afterwards, the measured basal NEMOf fluorescence and calculated basal [Ca2+] were applied to the NEMOf fluorescence Ca2+ titration curve to obtain the normalizing factor needed to convert in cellulo NEMOf fluorescence to the corresponding in vitro NEMOf fluorescence, and then to real-time [Ca2+]36.

Light-tunable activation of Ca2+ entry in HeLa cells

HeLa cells cotransfected with Opto-CRAC (mCherry-LOV2404–546-STIM1336–486) and NEMOm or GCaMP6m were plated on glass-bottomed dishes (Cellvin, catalog no. D35-20-0-TOP), imaged using a Nikon Ti2 Inverted microscope with a Yokogawa W1 dual spinning disk scanhead, Micro-Scanner for photostimulation and stage top incubator for live-cell imaging19. To prevent preactivation of Opto-CRAC, we acquired Ca2+ signals (NEMOm or GCaMP6m) under emission light of 525 nm with 1% excitation strength of 488 nm and 10 ms exposure time. To tune the activation of Opto-CRAC, a Nikon ‘A1 Stimulation’ toolbar was applied with 488 nm stimulation (2% strength). Varied exposure times were applied in ‘A1 Stimulation’ to control Opto-CRAC activation and photo-induced Ca2+ influx.

Ca2+ imaging in neurons

Neurons of DIV 17-20 were imaged using a W1 spinning disc confocal microscope (Ti2-E, NIKON) with a 100 × oil-immersion objective (1.45 NA, NIKON) in Tyrode’s solution (129 mM NaCl, 5 mM KCl, 30 mM glucose, 25 mM HEPES-NaOH, pH 7.4, 1 mM MgCl2 and 2 mM CaCl2). Field stimulations were performed in a stimulation chamber (Warner Instruments, RC-49MFSH) with a programmable stimulator (Master-8, AMPI). Samples were excited with 488 nm laser and fluorescence was collected with a Zyla4.2 sCMOS camera (Andor) by NIS-Elements AR 5.10.00 software.

Data analysis for Ca2+ imaging in cells or cultured neurons

The corresponding mean fluorescence of regions of interest (ROIs) were analyzed by Matlab v.2014a (The MathWorks) and plotted with Prism v.7 software (Matlab codes are available upon request).

Ca2+ imaging and electrophysiology in cortical slices

Cortical slices were prepared from the adult mice (>P50, either of sex), 3 weeks after the stereotaxic injection of various adeno-associated virus (AAV) vectors encoding different Ca2+ probers to the primary visual cortex (V1), following a protocol described in our previous studies37,38. Ca2+ probe-expressing pyramidal cells (PCs) in cortical slices were recorded in a two-photon laser scanning microscopy system (model FV1200MPE, Olympus) equipped with a wavelength-tunable Mai-Tai femtosecond infrared laser (DeepSee, Spectra Physics). Whole-cell current-clamp recording on individual PCs was made with the glass micropipettes, filled with the internal solution containing 130 mM K-gluconate, 20 mM KCl, 10 mM HEPES, 4 mM Mg2ATP, 0.3 mM Na2GTP and 10 mM Na2-phosphocreatine (pH 7.25–7.35, adjusted with KOH, 305 ± 5 mOsm), and a microelectrode amplifier (Axon MultiClamp 700B, Molecular Devices). Membrane potentials were low-pass filtered at 10 kHz, digitized at 20 kHz (DigiData 1440A, Molecular Devices) and acquired by the pClampex v.10.3. Simultaneously time-lapse imaging of Ca2+ fluorescence signals in the soma of recorded PCs, evoked by APs at various frequencies or numbers, was acquired by 17–20 Hz at a 256 × 256 pixel resolution using a ×40 water-immersion objective (LUMPlanFL, NA 0.8, Olympus) and the 920 nm laser wavelength. Ca2+ signals for each tested AP trains were averaged by three sweeps. The signals of neuronal AP and Ca2+ fluorescence were synchronized with an analog connection unit (FV10-ANALOG, Olympus).

The acquired time-lapse images were analyzed offline with the OLYMPUS FV10-ASW v.4.2 (Olympus) and our custom MATLAB (MathWorks) scripts. A subtraction of the background fluorescence region outside the PC soma was made to estimate the basal fluorescence intensity F0, and the average F0 for a 0.5-s duration before the onset of AP trains was used in the calculation of ΔF/F0 (ref. 23). The latter quantification process was the same as that used in a previous study9.

In vivo two-photon laser Ca2+ imaging of mice V1 neurons

Animal surgery and virus injection

The use of animals was approved by the Institutional Animal Care and Use Committees of Beijing Normal University, Peking University and University of Science and Technology of China. Mice at postnatal days 15–20 (P15–P20) were injected transcranially with 700 nl AAV-Syn-GCaMP6f, AAV-Syn-GCaMP6s or AAV-Syn-NEMO virus in the V1. After 3 weeks, animals were used to perform in vivo and ex vivo calcium imaging experiments. A small craniotomy (2.5 mm × 2.5 mm) was made on the V1 area (centered 2.5 mm left, and 0.5 mm anterior to lambda suture) in the mouse under anesthesia ketamine/medetomidine (50 mg kg–1, 0.6 mg kg–1; intraperitoneal)39 and then covered with a 3-mm diameter round glass-coverslip. A chamber around the craniotomy, made by dental cement, was filled with the Ringer’s solution containing: 123 mM NaCl, 1.5 mM CaCl2, 5 mM KCl. After a 30-min recovery from the surgery, mice were transported to the two-photon-laser imaging set up.

Visual stimuli

Visual stimuli were generated by a custom-developed software using LabVIEW v.8.5 (National Instruments) and MATALB (Mathworks), and were presented on a liquid crystal display monitor (ThinkVision, Lenovo) for in vivo calcium imaging experiments39. The simulation covered 0° to 80° horizontal visual field and −35° to 40° vertical visual field. Full screen drifting gratings with eight different orientations (spatial frequency, 0.02 Hz per degree; temporal frequency, 2 Hz, 100% contrast) were presented in a pseudorandom order, and each orientation with 4-s duration was assessed six times at intervals of 7–8 s blank gray-screen stimulus (with identical mean luminance).

Two-photon imaging and Ca2+ signal analysis

Time-lapse Ca2+ imaging from the V1 layer 2/3 PCs was conducted with Scanbox v.4.1 system, or a custom-modified Olympus two-photon laser scanning microscopy system (model FV1200MPE). The excitation wavelength was set at 920 nm. Fluorescence signals were acquired at 11–14 Hz. For each acquired fields of view (FOV), ROIs were set manually on visually identified neuronal cell bodies.

Ca2+ signal analysis was performed, using our custom MATLAB scripts9. F is instantaneous fluorescence, the averaged baseline fluorescence intensity of 1-s duration before visual stimulation onset was calculated as F0, and Ca2+ responses were defined as ΔF/F0 = (F – F0)/F0. GECI responses evoked by optimal visual orientation stimulus with P values <0.01 (Studentʼs t-test) were identified as responsive cells. For each responsive cell, the peak ΔF/F0 responses, the peak SNR, and the half-decay time of the maximal ΔF/F0 responses were calculated, respectively, as follows.

For the half-decay time, an exponential function was used to fit ΔF/F0 responses of GCaMP6s, NEMOm or NEMOs that were averaged over six trials with optimal stimulus, while the same function was used for maximal peak ΔF/F0 responses of GCaMP6f or NEMOf from one trial of optimal stimulus.

The peak SNR was calculated as

$$\mathrm{peak}\,\mathrm{SNR} = \frac{{\mathrm{peak}\,\Delta F/F_0}}{{\mathrm{SD}_{\mathrm{baseline}}}}$$

Where SDbaseline is the standard deviations of ΔF/F0 responses before 1-s visual stimulus presentation.

The orientation selectivity index and direction selectivity index were calculated by using mean ΔF/F0 amplitude (averaging the top 25% of ΔF/F0 responses during 4-s stimulus presentation) over six trials evoked by individual eight orientation stimulation. The orientation selectivity index was calculated as:

$$\mathrm{OSI} = \frac{{\sqrt {\left( {\mathop {\sum }\nolimits_\mathrm{i} \left( {R\left( {\theta _\mathrm{i}} \right) \times \sin \left( {2\theta _\mathrm{i}} \right)} \right)} \right)^2 + \left( {\mathop {\sum }\nolimits_\mathrm{i} \left( {R\left( {\theta _\mathrm{i}} \right) \times \cos \left( {2\theta _\mathrm{i}} \right)} \right)} \right)^2} }}{{{\sum} {R\left( {\theta _\mathrm{i}} \right)} }}$$

where θi is orientation of drifting gratings, R(θi) is the mean ΔF/F0 amplitude at θi.

The direction selectivity index was calculated as

$$\mathrm{DSI} = \frac{{R_{\mathrm{pref}} - R_{\mathrm{opp}}}}{{R_{\mathrm{pref}} + R_{\mathrm{opp}}}}$$

Where Ropp is the mean ΔF/F0 amplitude at the opposite angle to the preferred angle.

Tail-pinching stimulus and optical fiber recording

After 3 days of adaptive feeding, six male C57BL/6 mice (7 weeks old, 20–25 g) were divided randomly into GCaMP6f group and NEMOs group (N = 3). AAV-Syn-GCaMP6f or AAV-Syn-NEMOs virus were injected into the striatal region (AP, +0.5 mm; R, 1.8 mm; DV, −2.5 mm), respectively. Two weeks after transfection, the experiment was performed using the Reward R810 dualcolor multichannel optical fiber recording system controlled by ORFS v.2_14397.

Pinch stimulation at the tail tip was given by a 15-mm long tail clip. Excitation light (410/470 nm; 17.5/65 μw) and emission between 500–550 nm were transferred via an optical fiber implanted into the virus injection area of anesthetized mice fixed on a stereotaxic instrument. GECI signal excited with 410 nm was a Ca2+-independent reference to cancel out motion artifacts40. Pinch-induced GECI responses were recorded by with a rate of 60 fps. One stimulation was defined as one event. For GCaMP6f, n = 97; for NEMOs, n = 101.


All mice were housed in a 12 h light/dark cycle. Food and water were provided ad libitum. The temperature of the room was controlled at 20–25 °C, and the humidity was maintained at 45–60%.

Super-resolution imaging of plasmodesmata

Full-length cDNA of plasmodesmata-localized protein 1 (PDLP1) was tagged with NEMOm and cloned into pCAMBIA1390. PDLP1 is a well-established plasmodesmata marker26. To obtain transgenic plants, Agrobacterium GV3101 containing the resulted construct was transformed into wild-type Arabidopsis thaliana using the floral dip method41. Images were collected from the abaxial leaves of 10-day-old seedlings at 2-s intervals using an Airyscan LSM880 confocal microscope.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.