Reducing the Environmental Sensitivity of Yellow Fluorescent Protein: Mechanism and Applications

Yellow mutants of the green fluorescent protein (YFP) are crucial constituents of genetically encoded indicators of signal transduction and fusions to monitor protein-protein interactions. However, previous YFPs show excessive pH sensitivity, chloride interference, poor photostability, or poor expression at 37 degrees C. Protein evolution in Escherichia coli has produced a new YFP named Citrine, in which the mutation Q69M confers a much lower pK(a) (5.7) than for previous YFPs, indifference to chloride, twice the photostability of previous YFPs, and much better expression at 37 degrees C and in organelles. The halide resistance is explained by a 2.2-A x-ray crystal structure of Citrine, showing that the methionine side chain fills what was once a large halide-binding cavity adjacent to the chromophore. Insertion of calmodulin within Citrine or fusion of cyan fluorescent protein, calmodulin, a calmodulin-binding peptide and Citrine has generated improved calcium indicators. These chimeras can be targeted to multiple cellular locations and have permitted the first single-cell imaging of free [Ca(2+)] in the Golgi. Citrine is superior to all previous YFPs except when pH or halide sensitivity is desired and is particularly advantageous within genetically encoded fluorescent indicators of physiological signals.

The resulting π-π stacking and increased local polarizability immediately adjacent to the chromophore are believed to be responsible for the ~20 nm shift to longer excitation and emission wavelengths (3). However, the changes in internal hydrogen bonding and steric packing also made the fluorescence more vulnerable to photobleaching (4,5), decolorization by protonation (6)(7)(8)(9)(10), and quenching by many anions (10- 12), of which chloride is the physiologically most relevant. These sensitivities can be exploited for specialized applications such as measuring fluorescence recovery after photobleaching and sensing pH and halide concentrations, but are deleterious for using YFPs either as simple fusion tags or as acceptors for fluorescence resonance energy transfer (FRET). YFPs are becoming very popular in such roles, particularly as partners for cyan fluorescent protein (CFP) mutants of GFP (2,5,(13)(14)(15). CFPs and YFPs are spectroscopically well enough separated to be easily distinguishable in either excitation or emission spectra, yet the emission wavelengths of CFPs and excitation wavelengths of YFPs overlap well enough to make them good partners for FRET. They have largely dimensions, with two orders of magnitude higher spatial resolution than simple colocalization of the two colors. This approach has been used to monitor interactions of nuclear receptors and coactivators (18), nuclear transport factors (19), protein kinase A and anchoring proteins (20), Gprotein subunits (21), G-protein-coupled receptors (22), and cytokine receptors (23). FRET can also detect intramolecular conformational changes, particularly within genetically encoded fluorescent indicators for a wide variety of intracellular analytes and processes such as Ca 2+ (8,(24)(25)(26), (Ca 2+ ) 4 -CaM (27), Zn 2+ (5), NO (28), cGMP (29,30), protease activation (31,32) and protein kinase A-dependent phosphorylation (33).
Genetically encoded indicators offer the major advantages of versatile and modular construction, applicability to intact transgenic organisms, and precise targetability to specific tissues, organelles, and subcellular microenvironments. These advantages are particularly important for Ca 2+ indicators, which have been the subject of more effort than any of the other indicator classes. Both ratiometric and non-ratiometric indicators of Ca 2+ have been constructed from CFP, GFP, or YFP (2) as fluorophores and calmodulin as calcium binding moiety in several configurations. In cameleons (26), an N-terminal CFP is fused to calmodulin, the calmodulin binding peptide M13 from myosin light chain kinase, and a C-terminal YFP. Binding of Ca 2+ to calmodulin leads to a conformational change that enhances the fluorescence resonance energy transfer (FRET) from the shorter wavelength emitting CFP to the longer wavelength emitting YFP. Subsequent modifications in the YFP acceptor protein led to improved cameleons with decreased sensitivity to cytosolic pH-changes (8). The YFP portion of these improved cameleons (termed EYFP V68L/Q69K) had a pK a of 6.1, rendering it largely insensitive to pH-changes near neutrality. However, due to poor folding at 37 °C, specific targeting was hard to achieve.
In a different approach, calmodulin was directly inserted into the backbone of YFP in place of Tyr145 to generate a medium affinity Ca 2+ indicator termed camgaroo-1 that increased fluorescence intensity approximately 8-fold upon saturation with Ca 2+ (34). A problem with this non-ratiometric indicator was that the fluorescence of the indicator in transfected cells at resting Ca 2+ levels was almost zero, making it difficult to identify transfected cells for experiments.
Also, the protein did not express well at 37°C.
In an effort to overcome these problems, we undertook an expression screen in E. coli and identified an improved mutant of YFP, consisting of GFP with mutations S65G/V68L/Q69M/S72A/T203Y. For brevity we have named this mutation Citrine to reflect its yellow color and acid resistance. Citrine folds well at 37°C, can be targeted to subcellular compartments, and has a pK a of 5.7. Some aspects of Citrine's photophysics, including twophoton spectra, light-driven flickering, excitation state decay kinetics, and translational and rotational diffusion, were recently described (35), but these measurements were wholly in vitro, did not document Citrine's superiority over previous YFPs, and did not explain why the Q69M mutation conferred beneficial properties. Using Citrine, we have now constructed new genetic indicators of cellular Ca 2+ dynamics and assessed their properties with respect to pHinterference, folding, and targeting in mammalian cells. In addition, we have determined the 2.2 Å x-ray structure of Citrine and propose a structural explanation for the various improvements conferred upon Citrine by the Q69M mutation. 6 primer included a BamHI site and ended at the starting Met of the GFP, and the 3' primer included an EcoRI site and ended at the stop codon, theoretically allowing mutagenesis of every base of camgaroo-1 other than Met1. The PCR (38 cycles with annealing at 55°C) was run in four 100 µL batches, each containing 10 µL of 10× PCR buffer with Mg 2+ (Boehringer), 150 µM Mn 2+ , 250 µM of three nucleotides, 50 µM of the remaining nucleotide, and 5 ng of template.
Mutagenic PCR products were combined, purified by agarose gel electrophoresis, digested with BamHI and EcoRI, and repurified by QiaQuick columns (Qiagen). The resulting fragment was ligated into pRSET B , and the crude ligation mix was transformed into E. coli BL21(DE3) Gold (Stratagene) by electroporation. Bacteria plated on LB/agar plates were imaged as described (34), and colonies that became fluorescent after overnight incubation at 37°C were grown in liquid culture and the plasmid DNA obtained by Miniprep (Qiagen). Protein was expressed and purified as previously described (34). Spectroscopy of purified protein was typically performed in 100 mM KCl, 10 mM MOPS, pH 7.25, in a fluorescence spectrometer (Fluorolog-2, Spex Industries). pH-titrations were performed as described (34). All DNA sequencing was performed by the Molecular Pathology Shared Resource, UCSD Cancer Center, which is funded in part by membrane-anchored protein galactosyltransferase (GT), which has been used to target GFP to this organelle (6). Mitochondrial targeting of camgaroo-2 was achieved by replacing ECFP with the camgaroo-2 coding sequence in the pECFP-Mito vector (Clontech), which uses the targeting sequence of subunit VIII of cytochrome C oxidase. In order to evaluate targeted expression of YFP mutants, identical amounts of DNA (20 µg) of EYFP V68L/Q69K-ER, Citrine-ER or Single cell imaging-Single HeLa cells were imaged with a charge-coupled device camera (Photometrics, Tucson, AZ) as described (26) at room temperature 1-5 days after transfection.
The excitation filter for ratiometric imaging was 440DF10 with a 455DCLP dichroic mirror. The emission filters were 480DF30 (CFP) or 535DF25 (Citrine). Experiments were processed digitally using Metafluor software version 2.75 or 4.01 (Universal Imaging, West Chester, PA).
For imaging camgaroo-2, a 480DF30 excitation filter was used in combination with a fluorescein dichroic mirror and emission filter 535DF25.
Crystallization and Data Collection-Citrine in vector pRSET B was expressed in E. Coli JM109(DE3) and the protein purified as previously described (34). Following enterokinase (Invitrogen) catalyzed proteolysis of the 6-his tag, 1 mL of Ni-NTA agarose (Qiagen) was added to bind residual uncleaved protein and 6-his peptides and the solution was gently agitated (4 °C

RESULTS
Our newest and best YFP arose from efforts to improve camgaroo-1 (34), a genetically encoded Ca 2+ indicator consisting of Xenopus calmodulin inserted in place of residue 145 of EYFP-Q69K (2). Camgaroo-1 had the desirable feature of a rather large (~7-fold) increase in fluorescence in response to Ca 2+ binding, but it unfortunately expressed poorly at 37°C and could not be targeted to organelles such as mitochondria (34). We therefore randomly mutated the induced with ionomycin increased the fluorescence about 6-fold (Fig. 1C). We also targeted camgaroo-2 to mitochondria using the pECFP-Mito vector (Clontech), which uses the targeting sequence of subunit VIII of cytochrome C oxidase. Transfected cells showed a pattern typical of mitochondria ( Fig. 1D), indistinguishable from that of the accepted mitochondrial marker rhodamine 123 (data not shown). Camgaroo-2 is functional in mitochondria because a response to histamine was detected and ionomycin produced a significant fluorescence increase, though lower in dynamic range than in the cytosol (Fig. 1E).
The desirable effects of mutation Q69M in camgaroo-2 prompted transfer of this same mutation into EYFP V68L/Q69K not containing any inserted proteins. This improved variant of YFP, i.e. Citrine, has excitation and emission peaks of 516 and 529 nm respectively, a quantum yield of 0.76 and an extinction coefficient of 7.7 × 10 4 ( Table 1). These properties are comparable to those of previous YFPs. One unexpected spectroscopic difference is that Citrine photobleaches at about half the rate as EYFP V68L/Q69K ( Fig. 2A). Based on the illumination intensity of 1.9 W/cm 2 , we estimate the photobleaching quantum yield of Citrine to be about 2.3 × 10 -5 , in surprisingly good agreement with an estimate of 2.6 × 10 -5 obtained at much higher illumination intensities (35). The corresponding value for EYFP V68L/Q69K is 5 × 10 -5 from Fig. 2A and (42). Citrine also has a considerably lower pK a , 5.7, than previous YFPs such as EYFP V68L/Q69K ( Fig. 2B and Table 1) (Table 1), in contrast to comparable permutations of EYFP V68L/Q69K that become fluorescent only at 20°C or less. In summary, the Q69M mutation improves many of the shortcomings of YFP including pH and chloride sensitivity as well as the inability to fold well in organelles or as a circular permutation.
To investigate why the mutation Q69M improves YFP's chloride and pH-resistance, we determined the x-ray structure of Citrine at 2.2 Å resolution (Table II and PDB accession code   1HUY). As expected, the effect of the Q69M mutation on the overall structure of YFP is minor.
The r.m.s. deviation between Citrine and the same protein with Gln at 69 (PDB accession code 1YFP) is 0.32 Å (3). In the immediate vicinity of the chromophore and the adjacent Met69 residue, small positional shifts (on the order of 0.3 Å when compared to 1YFP) resulting from the introduction of the bulky methionine side chain are apparent (Fig. 3A). There is a localized slight outward displacement of the two closest strands of the β-barrel due to steric contact of the side chains of residues Val 150 and Leu 201 with the methionine. Additional residues in the local environment, including the chromophore and its π-stacked partner Tyr 203, have undergone compensatory shifts and thus the majority of the packing interactions and hydrogen bond network are unchanged.
In previous YFPs, the pK a of the chromophore and the halide binding constant are interdependent such that protonation and halide binding facilitate each other. To explain this effect, it has been proposed that in the presence of halide, the anionic form of the chromophore is destabilized through suppressed delocalization of the negative charge (10). Conversely, neutralization of the chromophore would reduce electrostatic repulsion of an adjacent anion. Previous x-ray structural studies on YFP have shown that iodide binds in a large cavity adjacent to the chromophore and in close contact to the heterocyclic carbonyl oxygen of the chromophore (10). In the absence of halide, the binding cavity (55 Å 3 ) is partially occupied by the side chain of Gln69 (Fig. 3B). In order to form the anion binding cavity, the side chain of this residue must undergo a conformational change and swing out of the cavity thereby expanding the cavity size (91 Å 3 ) and positioning the nitrogen of the carboxamide such that it can hydrogen bond to the anion (Fig. 3C). In Citrine, Gln69 has been replaced with a Met that effectively fills the halidebinding cavity such that it is no longer accessible to a sphere with radius 1.2 Å (Fig. 3A). In the x-ray structure of Citrine, the Met is well ordered (B av = 17.5 Å 3 ) and there is no unexplained difference density in the region of the cavity. This suggests that the Met side chain is tightly packed into the cavity and likely unable to undergo a conformational change that would be analogous to that observed for Gln69 between the free and iodide bound forms of EYFP (10).
Even if such a conformational change was permitted, it is unlikely that the thioether side chain of We wondered whether removal of the two cysteines in GFPs could further improve folding in the oxidative environment of the secretory pathway. For this purpose we introduced the mutations C48L and C70V into GFP mutants. These mutations had previously been found to be the least injurious replacements for the cysteines in GFP itself (R. Ranganathan, personal communication). When introduced into CFP or EYFP V68L/Q69K, these mutants retained fluorescence but became extremely temperature-sensitive and developed bright fluorescence only after overnight growth at 4°C or room temperature. Cysteine-less Citrine was brightly fluorescent, folded well at 37°C and had spectroscopic properties similar to Citrine itself, with only a slight decrease in quantum yield and extinction coefficient (Table 1). However, in HeLa cells, ER-targeted cysteine-less Citrine gave less fluorescence intensity and lower expression than ER-targeted Citrine containing cysteines, as verified by Western blot analysis. Therefore, the cysteines were left in Citrine for all subsequent constructs for either cytosolic or targeted expression.

We then set out to construct a series of improved genetic indicators that incorporated
Citrine in place of previous YFPs. Yellow cameleons YC2.3 and YC3.3 are new ratiometric indicators of high and medium calcium affinity based on previous cameleons (8,26), but incorporating Citrine as the FRET acceptor protein. The spectral changes in the emission of purified YC3.3 from 100 µM EGTA to calcium saturation were as expected (Fig. 4A), indicating that substitution of EYFP V68L/Q69K with Citrine did not alter the Ca 2+ -dependent FRET changes. The ratio of 528/476 nm emissions was stable down to approximately pH 6.5, and then decreased with further acidification (Fig. 4B). The pH effects were greatest at saturating Ca 2+ , at which FRET from the relatively pH-insensitive CFP to the still somewhat pH-sensitive Citrine is maximal. Nevertheless YC2.3 and 3.3 are more resistant than any other cameleon to acidic pH.

YC2.3 and YC3.3 were brightly fluorescent when expressed in the cytosol of HeLa cells and
were homogenously distributed in the cytosol with the nucleus excluded, as expected of a 74 kDa protein without targeting sequences. Responses to submaximal doses of histamine were readily detected, and the maximal ratio change obtained in cells was around 2-fold (data not shown), similar to the results from previous cameleons (8).
The lower pK a of Citrine compared to previous YFPs should allow imaging of free calcium transients in more acidic compartments that so far have been inaccessible to cameleons.
For example, the Golgi was reported to have a pH of 6.58 (6), which should still be in the working range of our new cameleons. To test this we targeted YC3.3 to the Golgi by fusing the 81 N-terminal amino acids of human galactosyltransferase type II to YC3.3 and thereby generating GT-YC3.3 (Fig. 5A). Transfection of HeLa cells resulted in bright punctate labeling of Golgi stacks in a juxtanuclear position (Fig. 5B), identical to cells transfected with GT-EYFP or stained for the medial/trans-Golgi marker α-mannosidase II (6). GT-YC3.3 was saturated at resting conditions (Fig. 5C), indicating a high concentration of free Ca 2+ in the Golgi. Histamine (100 μM) caused a very small decrease. The Golgi calcium store could be depleted with several washes of ionomycin/EGTA and was refilled upon readmitting extracellular calcium (Fig. 5C), demonstrating the feasibility of single cell imaging of free calcium concentrations in the Golgi of mammalian cells. It has to be kept in mind that ionomycin does not perform optimally in acidic compartments. Also it should be noted that YC3.3 is near its lower pH-limit under these conditions. Further improvements in pH-resistance are still desirable, especially if one wants to study even more acidic compartments of interest such as secretory vesicles. YC3.3 was similarly well expressed in the ER (data not shown).

Citrine represents a third generation of YFPs or yellow mutants of Green Fluorescent
Protein. The first generation was exemplified by S65G/S72A/T203Y (26) and "10C" (1), S65G/V68L/S72A/T203Y. These proteins proved to be quite sensitive to pH (e.g. pK a 6.9-7.1) (6), halides such as Cl -(11) and partially reversible photobleaching (4). These sensitivities have been useful for particular purposes such as quantifying cytosolic pH (6), [Cl -] (12,44), or FRET efficiency (18), but are considerable nuisances whenever one simply wants to use YFP as a reliable label or FRET acceptor. In a second generation, the mutation Q69K was introduced into 10C to give S65G/V68L/Q69K/S72A/T203Y (8) or "EYFP V68L/Q69K", which lowered the pK a to 6.1 with little effect on the other sensitivities. We speculated that the positively charged lysine might electrostatically hinder chromophore protonation (8). The Q69K mutation also introduced a disadvantage: folding became noticeably more difficult, especially in organelles at 37ºC. In the most recent version, Citrine, replacement of Q69K by Q69M lowered the pK a yet further to 5.7, eliminated the halide sensitivity, doubled the photostability, and improved the folding. The improvement in folding efficiency was particularly apparent in difficult cases such as functional expression at 37ºC in organelles or with an internally inserted calmodulin, i.e. camgaroo-2. The crystal structure provides some reasonable rationalizations for these improved properties, in that the Met side chain nicely plugs what had been a sizable halide-binding cavity next to the chromophore. The poorer folding of Q69K might well be due to the extra length of a Lys side chain making an uncomfortable fit within the cavity, or the electrostatic penalty for burying a positive charge, or both. Thus a good steric fit with a neutral side chain seems far more effective at lowering the chromophore pK a than an awkward fit with a positively charged side chain. The apparent photobleaching of YFPs probably consists of two components, a reversible proton redistribution or tautomerization and a truly irreversible covalent reaction (4,5). Either or both would be hindered by better packing of the hydrophobic core and elimination of a cavity next to the chromophore.
Despite the inferiority of Q69K, it was an essential stepping stone in the evolution of better properties by random mutagenesis and screening, because direct alteration of the Gln codon CAG to the Met codon ATG would require two base changes in a single codon, a very unlikely event. It was fortunate that there was an easy evolutionary path from CAG to the Lys codon AAG and then to ATG. Many other examples of optimal sequences may remain relatively inaccessible to random mutation due to barriers created by the genetic code.
We have demonstrated the application of Citrine in a series of genetically encoded Ca 2+ indicators based on Citrine, all of which were improved in relation to their predecessors.
Camgaroo-2 may constitute an alternative to cameleons in confocal microscopy given that it can be conveniently excited at the 488 nm argon laser line, or in cases in which targeting of cameleons are not successful. For example, we and others have found targeting of cameleons to mitochondria to be difficult (45), whereas camgaroo-2 was easy to send to the mitochondria with the targeting sequence of cytochrome C oxidase subunit VIII. Single cell imaging of mitochondrial calcium offers exciting new prospects for studying its dynamics in this organelle as well as to address aspects of heterogeneity of the mitochondrial population (46). Camgaroos lack a CaM-binding peptide and therefore have lower Ca 2+ affinities than the newest generic design of GFP-based Ca 2+ indicators, "G-CaMP" (47) or "pericams" (45). These indicators are chimeras of the CaM-binding peptide M13, circularly permuted GFP or YFP, and CaM.
However, many of these molecules still do not express well at 37ºC, so annealing mutations corresponding to Q69M might well be worth incorporating.
Our new improved cameleons expressed well at 37°C and were successfully targeted to the ER and Golgi. Cytosolic pH fluctuations are readily transmitted to the ER (48), therefore it was important to be able to express a pH-resistant functional indicator in this organelle, which had not been possible with previous versions of cameleons. Similarly, previous cameleons did not allow imaging free calcium in the Golgi due to the mild acidity of the compartment, which quenched other YFPs. Little is known about calcium regulation in the Golgi. One study using targeted aequorin identified the Golgi as a major calcium store within the cell (49), but aequorin has many disadvantages, such as lack of intrinsic fluorescence and requirement for an exogenous cofactor, that limit its use as a reliable calcium probe. We believe that Citrine should supersede previous YFPs within fusions for multicolor observation of protein trafficking, protein-protein interaction, and intramolecular conformational change, especially within genetically encoded Ca 2+ indicators.    1.3 a Numbers in parentheses refer to the highest resolution shell. b R merge = Σ|I hkl -I av |/ΣI av . c R free was calculated using about 10 % of the reflections which were omitted from the refinement.