EGFP was purified in its native sequence form, encompassing residues 1-Met-Val-Ser to Leu-Tyr-Lys-238 without any affinity purification tag attached at the N- or C-termini. The spectral characteristics of the pure protein were similar to those already reported for EGFP [14], with λ_ex and λ_em of 488 nm and 511 nm, respectively, ε of 55 mM⁻¹ cm⁻¹, a quantum efficiency of 0.6 and brightness of 33 mM⁻¹ cm⁻¹. Size exclusion chromatography (Methods S1) confirmed EGFP was predominantly monomeric (Figure S1).

The crystal structure of EGFP showing residues Lys3 to Leu231 was determined as described in the Methods and Materials. Crystals grew in the space group P2₁2₁2₁ and contained a single molecule in the asymmetric unit. The structure was determined to a resolution of 1.35 Å and refined to an R and R_free of 12.8% and 16.8%, respectively (Table 1). The final refinement statistics and model geometry fall within the expected range (Table 1).

The crystal structure of EGFP displays the traditional β-barrel structure with the chromophore located in the core of the protein (Figure 1A). Secondary structure assignment using DSSP revealed that 11 strands make up the β-barrel core, corresponding to 47% of protein secondary structure. Only 13% is helical, with the core helix containing the chromophore being a mixture of the 3₁₀ and α conformations. Whilst secondary structure assignment using DSSP identified the 3₁₀ helix in the core of the protein (Pro56 – Leu60) it failed to identify residues Val61 – Leu64 as being α-helical in nature. Structural analysis of the residues in the central helix identify an i to i+3 hydrogen bonding pattern between residues Pro56 to Leu60 corresponding to a 3₁₀ helix, despite proline being considered as a typical helix breaker, whilst an i to i+4 hydrogen bonding pattern is seen from residues Leu60 to Leu64 corresponding to an α-helix (Figure 1A). The remaining 40% of the structure is comprised of coil mostly from loops located at the two ends of the β-barrel structure (Figure 1A). A full list of secondary structure assignments can be found in Table S1.

A B-factor ‘putty’ (or ‘sausage’) representation of EGFP (Figure 1B) highlights increased B-factor values for residues at the N- and C-termini and in several loops connecting ordered secondary structures. This may be an indication of increased mobility in these regions and could be considered as possible target sites for future protein engineering endeavors. Overall the B-factor values observed for the higher resolution EGFP, 4EUL, are slightly lower in comparison to those observed for the previously determined lower resolution structure, 2Y0G, providing increased confidence in the placement of side chains in the structure determined here. This has implications in terms of the certainty of alternate side chain conformations observed for EGFP (vide infra).

Despite the mature chromophore requiring protection from the external environment to maintain fluorescence several structured water molecules are also found within the core of the protein (Figure 2), some of which are critical to the fluorescent properties. Whilst all of these waters superimpose with waters from the lower resolution EGFP structure 2Y0G, two of the waters (W₅ and W₆) are absent in both structures for wt GFP (1GFL) and for S65T GFP (1EMA).

Superpositioning of the structure obtained for EGFP with that of wt GFP (PDB entry 1GFL [7]) and a S65T GFP mutant (PDB entry 1EMA [8]) shows that the overall structures are very similar (Figure 3A); the RMSD over the backbone and all atoms of EGFP and wt GFP is 0.40 Å and 1.03 Å respectively, whilst the RMSD over backbone and all atoms of EGFP and S65T GFP are 0.29 Å and 0.85 Å, respectively. This indicates that the F64L and S65T mutations do not have a significant effect on the overall protein structure but have more subtle effects. Secondary structure analysis revealed the boundaries between the different elements in wt GFP and EGFP (Table S1) are very similar.

Superposition of the present model, 4EUL, with the lower resolution 2Y0G structure with an extended N-terminal His-tag (MAHHHHHHGHHH) sequence, reveals a RMSD of 0.59 Å when all common atoms are considered, including side chains. This indicates that both structures are largely identical. Thus, the presence of the longer than normal His tag sequence element in 2Y0G does not appear to greatly influence the overall structure. Both suffer from disordered termini. Whilst there are a number of identical residues in both versions of the EGFP structure that have been refined with multiple conformers, there are several that are only present in one or the other structure. These will be discussed in more detail below.

Given the high resolution of the EGFP structure, the exact placement of side chains can be defined with high confidence. The F64L mutation confers increased folding efficiency to GFP at 37°C but the structural consequences of this mutation in the context of the S65T mutation have not been fully investigated. The most obvious effect of the F64L mutation comprises the exchange of the bulky and buried phenylalanine side chain for a smaller leucine side chain in the central chromophore containing helical structure (Figure 4A). The substitution causes β-strand 2 to pack tighter with the core. The largest observed variation between wt GFP and EGFP was centred on residue Val29, with deviation of 1.37 Å across all atoms (Figure 4A). Val29 shifts closer to the chromophore upon the additional space being made available by loss of the aromatic Phe64 side chain. Residue Leu18 also shifts positions with the electron density only fully satisfied by modelling two alternate side chain conformations (Figure 4B), both of which differ from the side chain conformation of wt GFP Leu18 (Figure 4A). To satisfy the electron density the two conformations were modelled with an occupancy of 0.7 for conformer A and 0.3 for conformer B (Figure 4). Conformer A and B of Leu18 have observed RMSDs of 0.53 Å and 1.22 Å between wt GFP and EGFP across all atoms respectively. Both conformers exhibit a rotation of the Leu18 isobutyl side chain away from the edge of the β-barrel towards the core of the protein (Figure 4A). Whilst the lower resolution 2Y0G structure also shows rotation of the Leu18 isobutyl side chain away from the edge of the β-barrel it was not modelled by two conformers as seen in the present structure (Figure S5).

There is also a slight shift of residue Trp57 away from the surface of the protein towards Leu64, resulting in an RMSD over the side chain atoms of 0.39 Å (Figure 4A), decreasing its solvent accessible surface area in EGFP (12.8 Ų) with respect to wt GFP (15.2 Ų). Trp57 lies in the proline rich PVPWP pentapeptide sequence found in a variety of different proteins and, along with Val55 has been reported to be essential for function as mutation to other residues render EGFP non-fluorescent [15], [16]. The bulky side chain of Phe27 also moves towards the core of the protein with an RMSD over the side chain atoms of 0.41 Å resulting in a decreased solvent accessible surface area in EGFP (1.7 Ų) with respect to wt GFP (2.6 Ų) (Figure 4A). The repositioning of these residues could potentially influence the folding of EGFP at 37°C through better packing of the hydrophobic residues surrounding the central helix and of the central helix itself, and by reducing surface exposure of hydrophobic residues.

The S65T mutation has proved to be a more general mutation that can be transplanted to other green fluorescent protein variants to alter their spectral properties through removal of the 395 nm and promotion of the ∼490 nm excitation λ_ex [17], [18], increase the rate of oxidation during chromophore maturation and increase the brightness of the fluorescent proteins [17]. Analysis of the local environment around the chromophore can explain the molecular basis for the observed spectral properties. Replacement of Ser65 with Thr results in the hydroxyl group of Thr65 in EGFP occupying a different position from the corresponding hydroxyl group of Ser65 in wt GFP (Figure 3), probably as a result of steric effects due to the additional methyl group of Thr65. The Thr65 side chain in the EGFP structure determined here is in the same orientation as in the previously determined S65T-GFP structures [18] (Figure 3B).

In contrast to other S65T-GFP structures the electron density of the Glu222 carboxylate side chain in EGFP suggests that it occupies two distinct conformations (Figure 5A). The electron density difference map produced after molecular replacement and structural refinement displays a tridentate density for Glu222 (Figure 5A), which was successfully modeled as two conformations of the carboxylate, with substantial difference in the placement of atoms along the side-chain (Figure 5B). The rationale behind the Glu222 double conformer in EGFP is described in Figure S2. The occupancy of conformers A and B were set to 70% and 30%, respectively, in order to satisfy the electron density for this residue. The Glu222 conformer A matches very closely to the position of Glu222 seen in the S65T GFP mutant but differs significantly from that observed for wt GFP (Figure 5B). This highlights the role of residue 65 in defining the positioning of Glu222. However, conformer B represents a new and different side-chain positioning for this residue, leaving the carboxylate group orientated in a similar direction (but not superimposable) with the Glu222 carboxylate group from wt GFP (Figure 5B). Tridentate density for Glu222 has only been observed to our knowledge once before in a GFP-derived variant, the recently determined His-tagged EGFP structure 2Y0G [13]. While conformer A in both these independently determined structures are very similar, conformer B shows obvious differences. With regards to Y20G, the tridentate electron density was refined to show changes largely related to the carboxyl group and not the side-chain as a whole (Figure S3), as observed here for 4EUL (Figure 5B and Figure S3). The hydroxyl group of Thr65 now donates a hydrogen bond to the backbone carbonyl of V61 instead of Glu222 (Figure 3B). However, the hydrogen bonding network related to the carboxylate of Glu222 depends on the conformation sampled, as outlined in Figure 5. Thus, we have provided important independent proof that Glu222 can exist in two alternate conformations, albeit with slight discrepancies concerning the alternate B conformation, and with the higher resolution structure can position the atoms of the alternate conformer of Glu222 with higher confidence.

The reason for promotion of the anionic chromophore over the neutral form is thought to be a result of disruption of hydrogen bonding with charged Glu222 on mutation of Ser65 to Thr [3], [18], [19]. This in turn prevents ionization within the core of the Glu222 carboxylate group so removing an electrostatic clash with the anionic form of the ground state chromophore. In both conformations Glu222 O_δ1 is hydrogen bonded to two conserved water molecules. With respect to Glu222 O_δ2, in one conformation it donates a hydrogen bond to the hydroxyl group of Thr65 (Figure 5C) and in the alternate conformation it donates a hydrogen bond to the hydroxyl group of Ser205 (Figure 5D). In order for Glu222 to donate hydrogen bonds its carboxylate group must be protonated and therefore neutral. This allows charge stabilization on the deprotonated phenol group of the chromophore by hydrogen-bonding interactions from His148, Thr203 and a conserved water molecule coordinated between the backbone carbonyl group of Asn146 and the side chain hydroxyl group of Ser205 (Figure 5C & 5D). The neutral charge on Glu222 also removes any potential electrostatic clashes between the negative charges on Glu222 and the chromophore, thus allowing the chromophore to be deprotonated in the ground state. This explains why EGFP has a single excitation peak corresponding to the deprotonated state.

However, given the heterogeneity in the local environment of the chromophore due to the alternate conformations of Glu222 it would be expected that this may be reflected by spectral heterogeneity. This is not the case as is evident from single exponential fluorescence lifetime decays (Figure 6); the measured fluorescence lifetime was 2.54 ns, similar to that reported previously for EGFP [20]. Therefore, small alterations in Glu222 conformation are unlikely to have profound effects on the electrostatic environment surrounding the chromophore and thus the fluorescence properties. The two side chain conformations observed for Glu222 could be a crystallographic artifact and both may not be populated in solution. This is unlikely given that the residue is buried within the interior of the protein and alternate conformations for this residue have been observed in a lower resolution structure of the His-tagged EGFP protein [13]. Alternatively both conformations may exist but could be in a rapid dynamic equilibrium and transiently exchanging between the two conformations. A third possibility could be that upon folding the Glu222 side chain is trapped in one conformation or the other. Further structural analysis by NMR could potentially identify and measure Glu222 side-chain exchange rates in solution to confirm the two conformations observed in the crystal structure.

In addition to Leu18 and Glu222 that lie close in space to the chromophore (Figure 7), electron density maps indicated other residues occupied multiple conformations including Ser30, Thr43, Leu44, Gln80, Thr97, Lys101, Asp102, Lys113, Asp117, Glu124, Met153, Glu172, Tyr182, Gln184 and Thr186 (Figure S4). Whilst the majority of these residues are surface exposed and potentially have the ability to sample multiple conformations freely, residues Leu18 and Leu44 are buried in the core of the protein in close proximity to Glu222 and the chromophore (Figure 7). Both Leu18 and Leu44 are modeled by single conformations in the lower resolution 2Y0G (Figure S5). The electron density for all three residues was best fitted when one conformer occupancy was set to 70% and the other conformer occupancy set to 30%. The implications of the dual conformers in terms of fluorescence characteristics, including whether one represents a fluorescent and the other a non-fluorescent state, are not currently known.

Comparison of 4EUL with the lower resolution 2Y0G structure identified a number of common double conformers observed between the two structures (S30, T43, K101, D102, D117, E124, E172, Q184, T186 and E222). Additional solvent exposed residues were modeled as double conformers in 2Y0G that were not observed in 4EUL (I47, K79, K107, N164, D190 and D197). Likewise a number of double conformers were modeled in 4EUL that were not observed in 2Y0G (L18, L44, Q80, T97, K113, M153). Leu18, Leu44 and Glu222 aside, given the solvent exposed nature of these residues and the difference in crystallisation conditions (4EUL: 0.1M MES/NaOH, pH 6.5, 200 mM calcium acetate, 20% (w/v) PEG8000. 2Y0G: 50 mM HEPES, pH 8.2, 24% (w/v), 50 mM magnesium chloride, PEG4000, 10 mM β-mercaptoethanol) the difference in double conformers modeled between the two structures are potentially crystallographic artifacts arising due to fixing of particular conformation populations in the crystal during random rotations around the side chain σ-dihedral bonds.

In conclusion, we have determined the structure of the native sequence to 1.35 Å resolution of the widely used and commercially important enhanced Green Fluorescent Protein. While the core structure and fold of EGFP is very similar to the wild-type GFP, the introduction of the F64L and S65T have subtle yet important effects on the properties of EGFP. These include altered core packing arrangements close to the chromophore and altering hydrogen bonding and charged states of residues close to the chromophore. These changes in turn give rise to the important properties of EGFP that make such a useful tool compared to the wt GFP: better fluorescence excitation characteristics and improved folding at 37°C. Comparison of the EGFP structure determined here (4EUL) with that of an extended N-terminal His-tagged version recently determined at slightly lower resolution (2Y0G) shows that the structures are very similar so the artificial affinity tag is having little overall effect on structure. However, the present higher resolution structure with slightly lower overall B-factor values increases confidence of side chain placement during the model refinement, which becomes crucial when alternate conformations are observed. Our work independently confirms the recent observation of an alternate conformation of Glu222 but here the side chain placement of the alternate conformer is different to that in the lower resolution structure. In addition, alternate conformations were observed for two residues close to the chromophore that were absent from the lower resolution structure suggesting that the presence of alternate conformations go beyond that of just Glu222. Thus, the observed multiple conformations for residues close to the chromophore, including the chromophore charge defining residue Glu222 is intriguing and may be having a yet unknown impact on the fluorescent properties of EGFP.

The production and subsequent purification of EGFP was performed as follows. LB Broth (15 ml) supplemented with 100 µg/ml ampicillin was inoculated with a single colony of Escherichia coli BL21 (DE3) Gold containing the plasmid pNOM-XP3-egfp to generate a starter culture and incubated overnight at 37°C. A 1/200 dilution of the starter culture was used to inoculate 1 l LB broth supplemented with 100 µg/ml ampicillin and grown at 37°C until an optical density of A₆₀₀ = 0.4 was achieved. Protein expression was induced by the addition of 1 mM IPTG. The culture was incubated for 24 hrs at 37°C. The 1 l culture was then harvested by centrifugation (3000× g for 20 mins) and the pellet resuspended in 20 ml 50 mM Tris-HCl pH 8.0 (Buffer A) supplemented with 1 mM PMSF and 1 mM EDTA. The cells were lysed by French press using a chilled pressure cell. The lysate was then centrifuged (20000 rpm in Beckman JA20 rotor for 30 mins) to pellet any cell debris and the supernatant was decanted and stored at 4°C. The cell lysate was applied to a Q-Sepharose (GE healthcare) ion exchange column and elution monitored at 280 nm and 488 nm. Pooled fractions were then subjected to ammonium sulphate precipitation to further purify and concentrate the protein sample. An initial ammonium sulphate concentration of 45% (w/v) was used to precipitate unwanted proteins from solution. Further addition of ammonium sulphate to a final concentration of 75% (w/v) was carried out to precipitate EGFP from solution. The precipitate was resuspended in buffer A (5 ml) and the protein solution was then applied to a SP Superdex 200 gel filtration column (GE Healthcare) with elution monitored at 280 nm and 488 nm. The purified protein sample was finally stored in buffer A supplemented with 150 mM NaCl. A detailed description of absorption and fluorescence methods is provided in Methods S1.

Purified EGFP (10 mg/ml in 50 mM Tris-HCl, pH 8.0 and 150 mM NaCl) was screened for crystal formation by the sitting drop vapour diffusion method with incubation at 4°C. Drops were set up with equal volumes of protein and precipitant solutions (0.5 µl each). Crystals of EGFP were obtained from 0.1 M MES/NaOH, pH 6.5, 200 mM calcium acetate and 20% (w/v) PEG 8000. A crystal was transferred to mother liquor supplemented with 13% (w/v) PEG 200 as a cryoprotectant and vitrified. Data were collected on beamline I02 at the Diamond Light Source, Harwell, UK. Usable diffraction was recorded up to a resolution of 1.35 Å. Data were reduced with the XIA2 package [21], space group assignment was done by POINTLESS [22], scaling and merging were completed with SCALA [22] and TRUNCATE [23]. Initial molecular replacement for the EGFP structure was performed using a previously determined GFP structure (PDB entry 2HQZ) as the search model, using MOLREP [24]. The structure for EGFP was adjusted manually using COOT [25] and refinement of the completed molecule was carried out using REFMAC [26]. Protein atoms were refined anisotropically, but residues shown as sphericity outliers by REFMAC were refined isotropically. All non-protein atoms were refined isotropically. The above routines were used as the CCP4 package [23] (www.ccp4.ac.uk). Graphical representations were made with PyMOL Molecular Graphics System, Schrödinger, LLC.