Matryoshka: Ratiometric biosensors from a nested cassette of green- and orange-emitting fluorescent proteins

Sensitivity, dynamic and detection range as well as exclusion of expression and instrumental artifacts are critical for the quantitation of data obtained with fluorescent protein (FP)-based biosensors in vivo. Current biosensors designs are, in general, unable to simultaneously meet all these criteria. Here, we describe a generalizable platform to create dual-FP biosensors with large dynamic ranges by employing a single FP-cassette, named GO-(Green-Orange) Matryoshka. The cassette nests a stable reference FP (large Stokes shift LSSmOrange) within a reporter FP (circularly permuted green FP). GO-Matryoshka yields green and orange fluorescence upon blue excitation. As proof of concept, we converted existing, single-emission biosensors into a series of ratiometric calcium sensors (MatryoshCaMP6s) and ammonium transport activity sensors (AmTryoshka1;3). We additionally identified the internal acid-base equilibrium as a key determinant of the GCaMP dynamic range. Matryoshka technology promises flexibility in the design of a wide spectrum of ratiometric biosensors and expanded in vivo applications.


Introduction
Fluorescent protein (FP)-based, genetically-encoded biosensors have enabled the minimally invasive monitoring of dynamic biological processes with high spatial and temporal resolution in living cells and multicellular organisms 1,2 . However, designing novel biosensors for free elements and metabolites in a high throughput manner remains a challenge with the tools available to date.
Presently, genetically encoded biosensors fall into two categories: single-FP sensors and dual-FP sensors. Single-FP sensors either exploit the intrinsic sensitivities of the FP itself to certain stimuli, such as pH 3 , or the FP is fused to a recognition element that is sensitive to a specific analyte or process 1 . Alternatively, the FP can be sensitized to conformational changes of an attached recognition element by circular permutation (cpFP) 4 . Dual-FP sensors contain one sensor with a recognition element that conformationally rearranges in response to the target analyte, which then effects the interaction between it and a second, spectrally distinct FP. Most dual-FP biosensors consist of a recognition domain sandwiched between two FPs with properties that allow for FRET (Förster Resonance Energy Transfer; the non-radiative energy transfer between a donor and an acceptor FP) 1,5,6 . Conformational rearrangements in the recognition domain affect the relative intensity of the two FPs and, therefore, the ratiometric readout.
Both single-FP and FRET biosensors are characterized by specific advantages and limitations.
Single-FP biosensors can achieve a large dynamic range, a wide detection range, and a high signal-to-noise ratio, thus creating great sensitivity. Most single-FP biosensors are intensiometric, as they rely on the readout of a single fluorescence intensity (FI). However, detecting changes only in a single emissions range presents a disadvantage, as the signal is prone to artifacts, such as changes in expression level of the sensor or instrumental effects, due to bleaching and motion, and may thus lead to misinterpretation of the data. Therefore, due to the lack of an internal reference, single-FP biosensors do not provide absolute quantitative information. This is especially crucial during the screening of a random single-FP biosensor library, where quantitation is essential to correct for expression artifacts [7][8][9][10] . This drawback is commonly addressed by co-expression or terminal fusion of a spectrally distinct FP, such as mCherry, as a normalization control [7][8][9][10][11][12] . However, co-expression of the reference is not a reliably accurate control, as for example, variation in expression or protein levels can lead to artifacts of the signal readout. Alternatively, ratiometric single-FP biosensors have been developed, exploiting the two distinct absorbance maxima of the protonated and deprotonated FPchromophore. The resulting biosensors display two excitation or emission maxima with opposite intensity changes [13][14][15][16][17] . However, these sensors rely on excited-state proton transfer from the protonated FP-chromophore, requiring an intact proton network and near-ultraviolet excitation 18 .
In biological systems, ultraviolet radiation can lead to photodamage and is thus problematic 19 . In contrast, FRET biosensors intrinsically provide a ratiometric readout and contain a reference, as the energy accepting FP can be independently excited and monitored. However, FRET biosensors face three limitations. Firstly, except for rare cases, the detection range of FRET biosensors is limited to two orders of magnitude of analyte concentration 20 . Secondly, FRET sensors are restricted in their dynamic range due to the size of the FP barrel, which limits the proximity of the chromophores 21 . Thirdly, the coupling of the fluorophores via rotatable and flexible peptide linkers leads to signal loss due to rotational averaging 22 .
The design of biosensors is still largely empirical, often requiring the analysis of a large number of constructs 23 . We therefore aimed to develop a generalizable platform for rapid engineering of ratiometric biosensors that retain the features of single-FP biosensors but contain an internal reference FP. The peptide loop created by circular permutation of FPs (cpFP), tolerates nested insertion of a second FP 4 . While the cpFP serves as the reporter, the second, nested FP serves as the internal control. The two green cpFP variants tested here, an enhanced cpEGFP 24,25 and a superfolder cpsfGFP 26 both tolerated the nested reference FP, a large Stokes shift (LSS) mOrange 27,28 . The LSS and the excitation overlap of LSSmOrange with the green cpFP allowed for single excitation at λexc 440 nm yielding green and orange emissions, thus providing a ratiometric readout between the two FPs. At the same time, the emission spectrum of the green cpFP shows little spectral overlap with the absorption spectrum of LSSmOrange, thus limiting FRET. FRET compromises the ratiometric readout, as demonstrated by an alternative construct generated with CyOFP1 as nested reference 29 .
We hypothesized that insertion of the nested FPs into suitable positions of a recognition element would allow for creation of ratiometric biosensors in a single cloning step. Because this concept is reminiscent of the nested Russian dolls, we named this technology 'Matryoshka' [mä′trē-ō′shkə].
As proof of concept, we used the Matryoshka technology to engineer ratiometric calcium sensors based on GCaMP6s, the slow version of GCaMP6 7 . The resulting MatryoshCaMP6s variants retained the high dynamic range of GCaMP6s without substantial effects on other in vitro properties (steady-state fluorescence spectra, Kd and pKa) relative to the parent sensors. We further demonstrated their suitability for in vivo applications using mammalian HEK cells and stably transformed Arabidopsis seedlings.
The different dynamic ranges of the calcium sensor iterations prompted us to analyze the calcium sensor responses in quantitative detail. Careful analysis of the data and mathematical modeling led us to discover a major and previously undescribed contributor to the dynamic range arising from the internal FP acid-base equilibrium. This factor is independent of the pH and constitutes the single largest contribution to the dynamic range of GCamp6s.
Furthermore, we demonstrated the suitability of the technology for membrane transporter proteins by engineering and deploying Matryoshka-type transport activity sensors based on AmTrac 25 , named AmTryoshka1;3. An additional step of mutant screening was necessary, which identified two individual mutations required for restoration of ammonium transport sensing. In living yeast cells, the best performing AmTryoshka1;3 reported a FI change of 30% in the green emission channel, while the orange channel remained stable.
Our results demonstrate that Matryoshka is a promising technology for engineering ratiometric biosensors that contain a stoichiometric reference FP as internal standard for improved analyte quantification. GO-Matryoshka combines the reporter FP and the reference FP in a single cassette, which can be inserted into a recognition element of interest in a single cloning step, simplifying sensor construction and analysis. The Matryoshka-based biosensors use a single excitation wavelength to provide reporting via the relative emission of two FPs. Additionally, our mathematical model may provide for a more refined understanding of biosensor dynamic range and the identification of the residues that affect the acid-base equilibrium, which may facilitate future sensor design and optimization.

Design of the GO-Matryoshka cassette for ratiometric biosensors
To design a tool suitable for one-step generation of ratiometric biosensors with wide dynamic ranges, an FP with a large Stokes shift, LSSmOrange, was inserted as a stable reference domain into a green cpFP. Two different variants of cpFPs were tested as reporters: cpEGFP, which has been used to construct GCaMP and other single-FP sensors 14,24,25,30 , and cpsfGFP, a variant with improved brightness and folding properties compared to cpEGFP 26 . LSSmOrange was inserted into the center of the GGT-GGS sequence, which connects the N-and C-termini of both the original EGFP 4 and its variant sfGFP 31,32 . We named the resulting dual-FP combinations eGO-Matryoshka (based on cpEGFP) and sfGO-Matryoshka (based on cpsfGFP) (Fig. 1a).
The residues that flank the cpFP at its N-and C-termini, engineered during the process of circular permutation, have been reported to affect the protonation equilibrium of the chromophore and, thus, the fluorescence properties of the cpFP 25 . They connect the sensor domain with the seventh β-strand of the cpFP. The seventh β-strand interacts with the FPchromophore, and can thus affect the dynamic range and FI of the sensor 26,31 . Therefore, the flanking residues were maintained throughout the GO-Matryoshka characterization. The combination used, an N-terminal amino acid pair of leucine/glutamic acid (LE) and a C-terminal amino acid pair of leucine/proline (LP), had shown the best performance for GCaMP6 7 .
In vitro characterization of the purified GO-Matryoshka iterations revealed two emission maxima, one at λem ~510 nm and a second one at λem ~570 nm, upon excitation at λexc 440 nm   solid line) of eGO-Matryoshka (b) and sfGO-Matryoshka (c) at pH 10.5 (FI is maximal at this pH). Note that the intensities in the green emission channel are affected by the terminal residues, as they affect the cpFP FI, chromophore quantum yield and pKa. The combination LS/FN yielded a more acidic pKa than the LE/LP combination (figure supplement 2). The spectral analysis was repeated four times.
Nesting of the FPs may potentially result in reduced maturation efficiencies of the FP chromophores. In order to assess if the chromophore maturation is compromised in the Matryoshka iterations, we analyzed the individual and nested FPs by optical absorbance spectroscopy and intact protein mass spectrometry (MS) ( Supplementary Fig. 3, Supplementary Discussion). The absorbance data indicate reduced ratios of LSSmOrange to cpFP ( Supplementary Fig. 3a, b), suggesting that the LSSmOrange chromophore does not reach complete maturation, as opposed to the cpFP chromophores. However, our MS data analysis indicates that the reduced LSSmOrange maturation efficiency in the GO-Matryoshkas is very similar to LSSmOrange alone and that the cpFPs do not influence the extend of maturation of the LSSmOrange ( Supplementary Fig. 3c-e). Therefore, we infer that the identified incomplete maturation of LSSmOrange in the GO-Matryoshka versions is likely not a consequence of the nesting approach but a property of the LSSmOrange alone, which is not uncommon for red FPs 33 . Additionally, we found the fraction of immature LSSmOrange to be consistent and should therefore not pose a limitation for the Matryoshka concept.
We conclude that insertion of LSSmOrange did not lead to detectable alterations in the steadystate properties, pKa values and maturation efficiencies of the individual FPs compared to the GO-Matryoshka versions. The distinct green and orange emission maxima allow for spectral separation of both bands and for ratiometric readout. Both GO-Matryoshka cassettes tested as suitable tools for ratiometric biosensor design.

Generation of MatryoshCaMP6s calcium sensors
GCaMP6s is among the most sensitive calcium sensors 7 and therefore was chosen to test the Matryoshka technology. GCaMP6s carries cpEGFP between a calcium-binding calmodulin (CaM) domain and a CaM-interacting M13 peptide. Numerous mutations have been found that further increase the sensitivity of the sensors, i.e. residue mutations at the interface between cpEGFP and CaM or a K78H mutation in the cpEGFP portion 7 . Since the K78H (amino acid numbering according to 7 ) mutation has been implicated in increased sensitivity of GCaMP6s relative to GCaMP5G and other GCaMP6 sensors 7,34 , its effect on cpsfGFP-based sensors was also explored. In cpsfGFP, the residue equivalent to H78 is a threonine (T78), thus a T78H variant of cpsfGFP-based sensors was analyzed.
In vitro characterization of the purified GO-Matryoshka-coupled calcium sensors revealed two emission maxima, at λem ~510 nm and λem ~570 nm, when excited at λexc 440 nm (Fig. 2b). Upon λexc 485 nm excitation, a single emission maximum at λem ~510 nm was detected, with minimal cross-excitation of LSSmOrange ( Supplementary Fig. 4). Calcium treatment yielded a large positive response in the green emission channel for all sensors (Fig. 2b). Upon calcium addition, a small FI increase was detected in the orange channel. This increase can be attributed to the increased FI from the cpFP as part of the sensor response and was calculated as a ~10% fluorescence bleed-through from green emission into the orange emission channel. A fluorescence bleed-through factor of 0.1 was taken into account in subsequent ratiometric analyses of MatryoshCaMP6s data.
GCaMP calcium sensors are pH sensitive, and the pKa varies among the sensor variants. As described for GCaMP6 7 , the pKa values at saturating calcium conditions (pKa,sat) were more acidic compared to the calcium-free conditions (pKa,apo). sfGCaMP6s-T78H and sfMatryoshCaMP6s-T78H showed the largest change between pKa,apo and pKa,sat, with differences of ~2.6 pH units, followed by sfGCaMP6s and sfMatryoshCaMP6s, with differences of ~2.2 pH units. GCaMP6s and MatryoshCaMP6s showed the lowest differences, ~1.8 pH units, between pKa,apo and pKa,sat (Table 1, Supplementary Fig. 5).
The dynamic range (ΔR/R0) calculated for the ratiometric MatryoshCaMP6s variants ranged from 7.6 to 12 (Table 1). sfMatryoshCaMP6s-T78H showed the highest value (11.9±0.6), followed by MatryoshCaMP6s (8.5±0.2) and sfMatryoshCaMP6s (7.6±0.3). The dynamic range for the parent sensors (GCaMP6s, sfGCaMP6s and sfGCaMP6s-T78H) was estimated using ΔF/F0 at λexc 440 nm, which revealed values consistent with the ΔR/R0 ( Table 1). The dynamic ranges were also estimated using ΔF/F0 at λexc 485 nm. As expected, calculation of the dynamic range using ΔF/F0 at λexc 485 nm yielded a much larger dynamic range for MatryoshCaMP6s (41.8±0.9) and its parent GCaMP6s (49.7±0.4; Table 1) than using λexc 440 nm. The cpsfGFP-or sfGO-Matryoshka-based sensors also retained useful dynamic ranges, with higher values for sfMatryoshCaMP6s-T78H (16.4±0.8) and sfGCaMP6s-T78H (19.3±0.28) than for sfMatryoshCaMP6s (9.1±0.4), sfGCaMP6s (12.7±0.92). The detection range for free calcium, which we defined as ΔR/R0 at 0.1 and 0.9 ligand occupancy To better understand the dynamic range differences, we developed a mathematical model that describes the biosensor responses as a function of pH. We went beyond a simple, single-site chromophore titration description and used the more complete inter-site coupling model for FPs model, which emerged over recent years 31,35,36 . With experimental pH-dependent absorbance and fluorescence excitation spectra of apo and saturated GCaMP6s and sfGCaMP6s, we were able to parameterize the model and elucidate the factors responsible for the dynamic range (see Appendix). Besides the well-established factors of differential pKa's and quantum yields of the ligand-saturated and apo species, a new factor was identified: the internal acid-base equilibrium.
Our analysis indicated that the internal acid-base equilibrium plays a dominant role in establishing the large dynamic range for GCaMP6s sensors. sfGCaMP6s, in contrast, derives its response mainly from the differential pKa's between the apo and saturated species (see Appendix).
To test whether other FPs are suitable as stable reference FPs, we constructed variants of MatryoshCaMP6s that contained CyOFP1 instead of LSSmOrange 29 . CyOFP1 is a LSS FP, whose excitation spectrum shows larger spectral overlap with GFP excitation but a more redshifted emission compared to LSSmOrange. CyOFP1 was well tolerated in the resulting CyOFP1-containing MatryoshCaMP6s variants ( Supplementary Fig. 6, Supplementary Table 2).
However, due to extensive overlap of the emission spectrum of cpEGFP/cpsfGFP with the absorption spectrum of CyOFP1 the occurrence of FRET has to be taken into account, which compromises the ratiometric readout. This was demonstrated by ~20% reduced calcium affinities and an up to 50% reduced dynamic range when ΔR/R0 (λexc 485 nm) was evaluated as opposed to ΔF/F0 (λexc 485 nm) (Supplementary Table 2

MatryoshCaMP6s reports cytosolic calcium elevations in plants
Plants respond to diverse forms of environmental stimuli with transient rises in cytosolic free calcium levels. Because exposure to salt (NaCl) stress is a thoroughly documented elicitor of calcium transients in plants 37,38 we used this treatment to validate that MatryoshCaMP6s is suitable for monitoring cytosolic calcium levels in intact plants. One-week old seedlings expressing MatryoshCaMP6s were mounted on cover slips with small reservoirs filled with liquid media. Seedling roots were exposed to salt shock by addition of NaCl in liquid media to the reservoir (final concentration ~50 mM) (Fig. 3a). The seedling was co-excited using 440 and 488 nm laser illumination, and fluorescence intensities were monitored in the green (500-540 nm) and orange (570-650 nm) emission ranges (Fig. 3b). The treatment evoked a rapid and pronounced signal intensity change in the green emission channel (i.e., mean pixel intensity values were elevated by ~96%), whereas the orange emission channel remained comparatively stable (i.e., mean pixel intensity values were elevated by ~10%, which is similar to expected values from in vitro experiments for bleed-through from the green into the orange channel, Fig.   3b, c). Effects from background fluorescence can be excluded since the control seedling without the sensor did not show an observable ratio change. In response to salt shock treatment, fluorescence monitoring indicated that cytosolic calcium levels increased rapidly (within 4 seconds of treatment), reached peak intensity in about 39 seconds, and returned to baseline conditions after about 108 seconds of the peak response without removal of the salt stress (Fig. 3, Supplementary Fig. 7, Supplementary Movies 1-4). These data show that MatryoshCaMP6s is a suitable tool for monitoring calcium transients in intact plants.

MatryoshCaMP6s expression and photobleaching in mammalian cells
Although the primary motivation for Matryoshka conversion of GCaMP6s was to monitor calcium signals in plants, we sought to demonstrate the suitability of MatryoshCaMP6s in other eukaryotic systems. To evaluate the suitability of the MatryoshCaMP6s variants for future applications in mammalian systems, we expressed a selection of the sensors (GCaMP6s or MatryoshCaMP6s) in HEK293T cells and induced calcium spikes using methacholine (MeCh) as described 39 . As expected, the green FI at 440 nm laser excitation was lower compared to 488 nm laser excitation (Fig. 4a, b). Quantitative analysis of the rate of cells exhibiting spiking showed that both sensors can detect calcium spikes in individual cells with similar efficiency (Fig. 4c). In the case of the Matryoshka-based calcium sensor, the 440 nm laser excitation allowed for simultaneous recording of LSSmOrange emission, providing us with the advantage for discriminating between transfected cells and non-transfected cells, since GCaMP6s is dim at resting state. Also, the LSSmOrange served as stable control throughout the MeCh treatment ( Fig. 4b).

Generation of AmTryoshka sensors for ammonium transport activity
A quantitative readout is necessary when biosensor expression is driven by endogenous promoters that can influence the FI of the biosensor. For example, we recently generated AmTrac ammonium transporter activity sensors, where transport and promoter activity can affect the FI of AmTrac. AmTrac is composed of a cpEGFP inserted into the Arabidopsis thaliana Ammonium Transporter 1;3 (AtAMT1;3) and reports ammonium transport in vivo by a 40% reduction of FI 25 . However, as AMT expression in plants is nitrogen-dependent 40,41 , promoter activity can affect the FI readout of AmTrac, which limits AmTrac use in planta. To address this drawback, we explored different ratiometric design approaches, including the generation of dualemission deAmTracs 13 and the fusion of mCherry to the N-and C-terminus of AmTrac as a second, red-shifted FP. We also tested LSSmOrange as N-terminal tag of AmTrac ( Supplementary Fig. 8a). Unfortunately, all resulting fusion proteins showed impaired transport activity, as evidenced by the lack of complementation in the growth assay ( Supplementary Fig.   8b). Spectral analysis of yeast cultures expressing the fusion constructs resulted in a loss of green and red or orange FI for the N-terminally labeled constructs. AmTrac with a C-terminally fused mCherry confirmed the presence of both of the fluorophores (Supplementary Fig. 8c). To restore transport activity, a suppressor screen was performed 25 . However, all the suppressing colonies that grew on ammonium showed mutations in the C-terminus of AmTrac that introduced a STOP codon, resulting in the removal of mCherry and an indication that also the C-terminal mCherry was not tolerated.
The Matryoshka approach provides an alternative design for ratiometric sensors with large dynamic range and proves particularly advantageous when terminal fusion of FPs is not tolerated. Here, cpEGFP in AmTrac was replaced with sfGO-Matryoshka or sfGO-Matryosh-T78H, while retaining the flanking residue combinations that had been optimized for generating AmTrac-LS and -GS 25 (Fig. 5a). The resulting sensor series was termed AmTryoshka1;3. In parallel, we tested the properties of cpsfGFP relative to cpEGFP in AmTrac and found that the set of cpsfGFP-based AmTrac versions, named sfAmTrac, yielded improved brightness of the sensors (see Appendix).
However, we did not detect FI changes in response to ammonium addition. Insertion of the reference FP, LSSmOrange, seemed to impair transport activity of AmTryoshka1;3-GS, as confirmed by the lack of growth complementation (Fig. 5b, middle panel, second last row). To restore activity, a suppressor screen similar to previous experiments was performed 25 . Two mutations, F138I and L255I that allowed for growth on media containing low ammonium levels were identified (Fig. 5b, middle panel; mutations were numbered according to AtAMT1;3 sequence). Analysis of the crystal structure of the archaeal homolog AfAMT1 (PDB: 2B2F) indicate that both residues may line the pore (Supplementary Fig. 10). While no functional description of L255 is available, F138 had been proposed to form part of the external gate 42 .
Thus, the LSSmOrange insertion may have negatively affected the gate, an effect compensated for by the F138I mutation. However, we did not detect any changes in the plasma membrane localization in yeast compared to non-mutated sfAmTrac-LS ( Supplementary Fig. 11a).
To exclude environmental effects, such as accumulation of intracellular ammonium which could affect the cytosolic pH, wild-type yeast with endogenous MEP ammonium transport activity was transformed with AmTryoshka1;3-LS-F138I, AmTryoshka1;3-LS-F138I-T78H or the nonresponsive control AmTryoshka1;3-GS. In the presence of the endogenous MEP ammonium transporter, cytosolic ammonium concentrations are expected to increase independently of the sensors. Sensor responses were comparable between the wild-type strain and the Δmep1,2,3 mutant (Fig. 5f), indicating that intracellular ammonium levels did not affect the sensor. Hence, using Matryoshka technology, we engineered ratiometric ammonium transport sensors that report substrate concentration-dependent ammonium transport activity. The sensor with the highest dynamic range was AmTryoshka1;3-LS-F138I-T78H, with an approximate 30% FI decrease in response to ammonium transport.

Discussion
Here we developed a platform for ratiometric biosensor design. We engineered two cassettes, Our mathematical model (see Appendix) indicates that three mechanisms appear to differentially affect the dynamic range of the individual calcium sensors: (i) pKa differential (ii) relative quantum yield and (iii) internal acid-base equilibrium between the apo and saturated species.
While the cpEGFP-based GCaMP6s is modeled to utilize all three mechanisms, cpsfGFP-based sfGCaMP6s seems to derive most of its response from their greater pKa differences between apo and saturated proteins. Since the theoretical dynamic range is the product of the three factors, even small differences can substantially affect sensor responses (see Appendix). Analysis of crystal structure and systematic mutagenesis followed by spectral characterization and model fitting will help to parse the molecular connections to the three principal determinants of the dynamic range and may thus aid in future optimization of such sensors. Single-FP biosensors are often limited in their response accuracy due to pH sensitivity. The newly identified internal equilibrium factor is expected to be largely independent of cytosolic pH over normal physiological ranges. Thus, we predict that biosensors which maximize the quantum yield difference and the internal equilibrium factor can reach large dynamic ranges while also retaining high signal fidelity even in the presence of cellular pH fluctuations.
We demonstrated that MatryoshCaMP6s can be used to monitor biological processes in intact organisms, such as Arabidopsis seedlings. Although a wide variety of genetically encoded calcium indicators are already in use, MatryoshCaMP6s offers several advantages over previously developed tools in this context. A co-localized reference FP is particularly important for long-term acquisition where quantitative data is desired. The LSSmOrange channel can also be used for binary masking to avoid the inherent challenge of generating masks from channels showing massive intensity changes. Because GCaMP sensors are dim at resting cytosolic calcium levels, we expect that MatryoshCaMP6s calcium sensors will be more suitable for verifying cell-type specific expression. Also for this reason, signal from the reference FP enables more confident interpretation of negative results (lack of intensity changes) in the reporting FP.
We verified that the Matryoshka technology is suitable for constructing integral membrane biosensors by engineering ammonium transport activity sensors. Here, we substituted the cpEGFP in AmTrac with the sfGO-Matryoshka cassette to generate AmTryoshka1;3. The initial impairment of transport activity, and thus lack of FI signal change, triggered by the LSSmOrange insertion was overcome by the individual suppressor mutations F138I and L255I, which restored transport function and sensor response. It is worth noting that AtAMT1;3 is extremely sensitive towards modifications 25 . Therefore, it is not surprising that insertion of a second FP affected the transporter function. The best performing AmTryoshka1;3 ammonium transport sensor (LS-F138I-T78H) demonstrated a ~30% ratio change and bright fluorescence in the apo-state.
Additionally, our results provide insights into the role of histidine 78 in cpFP. For both calcium and ammonium transport activity sensors, the T78H mutation in cpsfGFP yielded a larger dynamic range relative to sensors with the T78. Histidine at position 78 was postulated to be involved in excluding solvent from the FP-chromophore 34  Variants with different spectral properties can quickly be generated, demonstrated by exchanging LSSmOrange for CyOFP1 in MatryoshCaMP6s and sfMatryoshCaMP6s and further efforts will be invested to red-shift the excitation properties. Alternatively, two-photon excitation will be investigated which will eliminate the potential phototoxic effects of λexc 440nm of the current Matryoshka sensors. Additionally, the Matryoshka concept will be used to further improve the dynamic range and SNR of sensors for other small molecules such as sugars, amino acids, neurotransmitters and hormones 22,23,44,45 . To determine the extinction coefficients of cpEGFP, cpsfGFP and LSSmOrange, the FPs were diluted 20-fold, denatured in 0.2 M NaOH and the absorbance was monitored.

DNA constructs
The results represent two biological replicates.

Fluorimetric analyses of fluorescent sensors
All in vitro fluorimetric analyses were carried out using a fluorescence plate reader (Infinite, intensity of about 100 µmol/m 2 sec for 7-9 days before imaging. To confirm fluorescent signal, seedlings were preliminarily screened for GFP fluorescence using a Nikon SMZ18.

Plant imaging conditions
Images were collected on a Leica TCS SP8 equipped with resonant scanner and white light laser (WLL). A HC PL APO 20 x / 0.70 N.A. multi-immersion objective was used with glycerol. Scan speed was 8,000 Hz (resonant mode), and line averaging was set to 8 or 16. Samples were coexcited with the 440 nm pulsed laser and the 488 nm laser line of the WLL. Fluorescence images were captured simultaneously in two windows using HyD SMD detectors; gain was set to 90 for each. Signal from cpEGFP was collected from 500-540 nm. Signal from LSSmOrange was collected from 570-650 nm. Transmitted light images were collected with a PMT. Seedlings were prepared for imaging by gentle transfer to cover slips and stabilized with surgical glue (spray PDMS), as described previously 49 . The reservoir for liquid media was made using vacuum grease and filled with half-strength MS media without sucrose. An additional cover slip was placed over the seedling to hold it in place and facilitate movement of treatment to the root.
Salt shock was applied by addition of half-strength MS containing NaCl to the liquid/cover glass interface. Final concentrations of NaCl in the seedling reservoirs were approximately 10-50mM, depending of the volume of media required for sample preparation.
Images of treatments were excluded from data analysis if extreme shifting of the root occurred upon application of NaCl or the application of treatment did not reach the media within the reservoir.
Average intensity z-stack projections were generated using FIJI (http://fiji.sc). The average ratio of cpEGFP / LSSmOrange signal intensities were calculated using FIJI, and ratio values are portrayed in pseudocolor (16-color lookup table). A binary mask was made using the LSSmOrange channel and applied to the ratio values to remove noise from background signal (Supplementary Movie 5). The average pixel ratio value per time point was extracted from the masked dataset, normalized with the initial ratio set to 1, and plotted using OriginPro 2015 (OriginLab).

HEK293T cell assay and imaging conditions
Transfection of HEK293T human kidney epithelial cells (ATCC, cat. no. CRL-11268) was performed as described previously 50  For photobleaching experiments, a time course of z-stacks (30 slices at 37 µm) was recorded for a duration of 3 min. The 488 nm laser excitation was set to full power. Images were recorded using the sequential scan mode at a time interval of 8 sec per z-stack scan. Evaluation of photobleaching was performed using FIJI. The z-stacks were transformed to an average intensity projection and a mask was created to identify transformed cells. The ROI manager tool was used to quantify the average pixel intensity of individual cells and the results were plotted over time and averaged. Random calcium spikes were observed during the bleaching experiments and such cells traces were removed before averaging. Results of a minimum of three transformations were analyzed and graphs were plotted using OriginPro 2015 software.

Yeast transformation and culture
The in vivo measurements employed the yeast strain 31019b [mep1Δ mep2Δ::LEU2 mep3Δ::KanMX2 ura3] 51 , which lacks all endogenous MEP ammonium transporters 30,52 . Briefly, yeast transformation was performed using the lithium acetate protocol 53 . Transformants were plated on solid YNB (minimal yeast medium without amino acids/ammonium sulfate; Difco BD, Franklin Lakes, NJ) supplemented with 3% glucose and 1 mM arginine. Single colonies were selected and inoculated in 5 ml liquid YNB supplemented with 3% glucose and 0.1% proline under agitation (230 rpm) at 30 °C until OD600nm 0.5 -0.9. AmTryoshka1;3-GS, which did not show a response upon ammonium treatment, was subjected to a suppressor screen as previously described 25 . Briefly, liquid cultures of yeast cells expressing AmTryoshka1;3-GS were washed twice with sterile water. The final resuspension volume was 5 mL and 500 µL were streaked on five plates with a diameter of 150 mm (VWR, Radnor, PA, USA) of solid YNB medium (buffered with 50 mM MES/Tris, pH 5.2, supplemented with 3% glucose and 1 mM NH4Cl). The plates were incubated at 30 °C; single colonies were identified after 7 days. Yeast plasmid DNA was isolated and sequenced, revealing the mutations F138I and L255I. AmTryoshka1;3-GS with a mutation was called AmTryoshka1;3-GS-F138I or -L255I, respectively.
For complementation assays, liquid cultures were diluted 10 −1 , 10 −2 , 10 −3 and 10 −4 in water, from which 5 µl of each dilution was spotted onto solid YNB medium (buffered with 50 mM MES/Tris, pH 5.2 supplemented with 3% glucose). Either NH4Cl (2 mM; 500 mM) or 1 mM arginine was added as sole nitrogen source. After 3 days of incubation at 30 °C, cell growth was documented by scanning the plates at 300 dpi in grayscale mode.
For fluorescence measurements, liquid yeast cultures were washed twice in 50 mM MES pH 6.0, and resuspended to OD600nm ~ 0.5 in MES pH 6.0, supplemented with 5% glycerol to delay cell sedimentation 52 . All graphs and were plotted using OriginPro 2015 software.

Yeast imaging conditions
Confocal z-sections of yeast cells expressing the sfAmTrac and AmTryoshka1;3 sensor variants were acquired with a 63 x oil objective (NA 1.40) on a White Light Laser Confocal Microscope Leica TCS SP8 X using the 488 nm laser line and a 440 nm pulsed laser in sequential scan mode.
The HyD detector range was set to 500-560 nm for cpEGFP or cpsfGFP and 570-630 nm for LSSmOrange detection, respectively. Cells from three individual transformations were imaged.
Images were analyzed using FIJI.

Electrophysiological measurements AmTryoshka1;3 in Xenopus oocytes
Two electrode voltage clamping in oocytes was performed essentially as described previously 25 .