Generation of longer emission wavelength red fluorescent proteins using computationally designed libraries

Abstract

The longer emission wavelengths of red fluorescent proteins (RFPs) make them attractive for whole-animal imaging because cells are more transparent to red light. Although several useful RFPs have been developed using directed evolution, the quest for further red-shifted and improved RFPs continues. Herein, we report a structure-based rational design approach to red-shift the fluorescence emission of RFPs. We applied a combined computational and experimental approach that uses computational protein design as an in silico prescreen to generate focused combinatorial libraries of mCherry mutants. The computational procedure helped us identify residues that could fulfill interactions hypothesized to cause red-shifts without destabilizing the protein fold. These interactions include stabilization of the excited state through H-bonding to the acylimine oxygen atom, destabilization of the ground state by hydrophobic packing around the charged phenolate, and stabilization of the excited state by a π-stacking interaction. Our methodology allowed us to identify three mCherry mutants (mRojoA, mRojoB, and mRouge) that display emission wavelengths > 630 nm, representing red-shifts of 20–26 nm. Moreover, our approach required the experimental screening of a total of ∼5,000 clones, a number several orders of magnitude smaller than those previously used to achieve comparable red-shifts. Additionally, crystal structures of mRojoA and mRouge allowed us to verify fulfillment of the interactions hypothesized to cause red-shifts, supporting their contribution to the observed red-shifts.

Red fluorescent proteins (RFPs) derived from organisms in the class Anthozoa have found widespread application in cell biology. For example, these proteins have been used as markers of gene expression (1), expressed as fusions to track endogenous protein within cells (2), and applied with other fluorescent proteins (FPs) for use in FRET experiments (3). The availability of monomeric versions of these proteins has bolstered their worth as fusion tags and vaulted them into routine experimental use (4–5).

The red fluorescence displayed by these proteins arises from the presence of an acylimine group conjugated with the standard p-hydroxybenzylideneimidazolinone GFP chromophore (6). The additional double bond extends the size of the chromophore conjugated system leading to an increase in emission wavelength. The longer emission wavelength of RFPs makes them attractive for whole-animal imaging because cells are more transparent to red light. For imaging applications, higher emission wavelengths (650–900 nm) are desirable because they tend to minimize background absorption and light scattering by tissue components and are less damaging to cells, enabling longer acquisition times.

Naturally-occurring Anthozoa RFPs, such as zRFP574 (7), eqFP578 (8), DsRed (9), and eqFP611 (10), are obligate oligomers that display emission wavelengths ranging from 574 nm to 611 nm. Significant effort has been made to monomerize and red-shift the emission wavelength of these RFPs using directed evolution. Starting from various wild-type precursors, these procedures have produced several far-red (λ_em > 630 nm) monomeric RFPs such as mPlum (11), mKate2 (12), and mNeptune (13). Each of these useful monomeric RFPs was developed using random mutagenesis (5, 13, 14). Although directed evolution has successfully yielded red-shifted monomeric RFPs, a strictly rational methodology to red-shift Anthozoa class FPs has not yet been described. Aside from the T203Y mutation in Aequorea victoria GFP that red-shifts emission by 20 nm to yield the yellow variant YFP (15), and mutations leading to a trans-to-cis isomerization in the chromophore of eqFP611 (2, 16), the rational prediction of mutations causing red-shifts has proven difficult.

We were interested in developing a structure-based rational design approach to red-shift the fluorescence emission of RFPs. Toward this end, we proposed three structure-based hypotheses supported by results reported in the literature and tested them using the mCherry scaffold. We applied a combined computational and experimental approach that used computational protein design (CPD) as an in silico prescreen to generate focused combinatorial libraries of mCherry mutants. The computational procedure helped us identify residues that could fulfill the interactions hypothesized to cause red-shifts without destabilizing the protein fold. This methodology allowed us to identify mutants displaying bathochromic emission wavelength shifts of up to 26 nm. Moreover, our approach required the experimental screening of libraries several orders of magnitude smaller than those previously used to achieve comparable red-shifts. Additionally, crystal structures of two of our most red-shifted mutants allowed us to verify fulfillment of our structure-based design hypotheses and gain a better understanding of the causes of red-shifts in RFPs.

Results and Discussion

Design Hypotheses.

To red-shift the emission wavelength of a monomeric RFP, three literature-based hypotheses were tested by screening for red-shifting in libraries of computationally designed mCherry variants. RFP chromophores contain a p-hydroxybenzylideneimidazolinone group prepended with an acylimine substituent at the C1 atom. The acylimine group is the result of an oxidation coupled to trans-to-cis isomerization of the peptide bond between the first chromophore-forming residue and the preceding residue (F65-M66 bond in mCherry) (6). The presence of the acylimine electron-withdrawing substituent creates a resonance structure for the deprotonated form of the chromophore in which the partial negative charge is located on the acylimine oxygen atom (Fig. 1A, bottom right). The deprotonated chromophore has been demonstrated to be the species responsible for red fluorescence in RFPs (5) and quantum mechanical studies have indicated that resonance forms of this species with the partial negative charge localized on the imidazolinone and acylimine oxygen atoms make higher contributions to the description of the excited state (17–19). Design hypothesis I (Fig. 1A) involves stabilization of the excited state by H-bonding the acylimine oxygen of the chromophore.

Fig. 1.

Structure-based design hypotheses tested on mCherry scaffold. (A) Design hypothesis I: stabilization of the excited state form on lower right by H-bonding to acylimine oxygen. H-bonds that stabilize each resonance structure are indicated by wide dashed lines. (B) Design hypothesis II: destabilization of ground state by hydrophobic packing around charged phenolate. (C) Design hypothesis III: stabilization of polarization of excited state by π-stacking the chromophore phenolate with a tyrosine. (A–C) Designed interactions are shown in blue.

In all known crystal structures of FPs (20), a strictly conserved Arg residue forms an H-bond to the imidazolinone oxygen atom (R95, Fig. 1A). However, no conserved residue has been identified as H-bonding to the acylimine oxygen. CPD was used to identify potential H-bond donor residues located within direct or water-mediated H-bonding distance to the acylimine oxygen. The goal was to preferentially stabilize the resonance form of the chromophore in which the negative charge is located on the acylimine oxygen. We hypothesized that such an H-bond would stabilize the excited state, thereby decreasing the energy difference between the excited and ground states, resulting in an increased emission wavelength. Literature results suggest that this type of interaction would have the predicted effect. The crystal structures of mPlum and its E16Q variant demonstrate an H-bond between the acylimine oxygen and E16 or Q16, respectively (21). Mutation of E16 to hydrophobic residues leads to hypsochromic emission wavelength shifts (11). Additionally, in the course of this work, the crystal structure of the far-red RFP Neptune demonstrated a water-mediated H-bond between the acylimine oxygen and neighboring residue S28 that leads to red-shifted excitation and emission (13).

Design hypothesis II (Fig. 1B) involves destabilizing the ground state by hydrophobic packing around the charged chromophore phenolate group. We theorized that destabilization of negative charge localization at the oxygen atom of the phenolate would raise the energy of the ground state relative to the excited state, thereby creating a further red-shifted emission spectrum. To test this hypothesis, CPD was used to identify hydrophobic residues that could be accommodated at positions surrounding the phenolate negative charge (residues 143, 146, 161, and 163). Campbell et al. (4) postulated that residues just above the plane of the chromophore of DsRed (residues 161, 163, 175, and 177) influence polarization of the RFP chromophore, thus affecting the emission wavelength. For example, DsRed has a Lys at position 163, which can stabilize the phenolate negative charge through an electrostatic interaction. In mCherry, an uncharged Gln is found at this position, perhaps partially accounting for its red-shifted emission relative to DsRed.

Design hypothesis III (Fig. 1C) involves stabilizing polarization of the excited state with a π-stacking interaction, similar to that seen in the crystal structures of yellow variants of Aequorea victoria GFP containing the T203Y mutation (22–23). In these mutants, π-stacking between Y203 and the chromophore phenolate is thought to stabilize the polarization of the chromophore excited state (15), leading to a 20 nm bathochromic shift in emission wavelength.

Computational Design.

To test these hypotheses, we applied a unique CPD approach to generate focused combinatorial libraries of mCherry variants containing mutations that could potentially fulfill the interactions hypothesized to cause red-shifts (24). The library design procedure takes as input a list of scored sequences, and two sets of constraints: a list of allowed sets of amino acids, and a range of desired library sizes. The algorithm generates a list of the combinatorial libraries that satisfy these constraints, and then ranks the libraries by the degree to which they reflect the energetic preferences present in the list of scored sequences. Thus, CPD was used to perform an in silico prescreen to eliminate sequences incompatible with the protein fold and generate combinatorial libraries amenable to rapid experimental screening. Thirteen positions with side chains that point towards the chromophore were divided into three groups intended to test the three design hypotheses (Table 1). Design hypothesis I (library 1) was tested by varying residues with side chains ≤ 6 Å from the acylimine oxygen (positions 14, 16, and 120). Design hypothesis II (library 2) was tested by varying residues with side chains within 5 Å of the chromophore phenolate (positions 143, 146, 161, and 163). Positions 175 and 177 were also varied, because they have been suggested to influence polarization of the chromophore (4). Design hypothesis III (library 3-1) was tested by introducing a π-stacking Tyr at position 197 and varying neighboring residues (73, 146, 195, and 217) to stabilize the Tyr in the correct orientation. In all the designs, all 20 amino acids were allowed at each of the designed positions.

Table 1.

Computational library design results

Library	Design hypothesis tested	Protein sequences sampled	Library size *	Designed position †	Wild-type residue	Required amino acids	Library amino acids
1	I. Stabilization of the excited state by H-bonding to acylimine oxygen	8 × 103	540	14, 16, 120	F, V, Y	FNS, STV, Y	FINSTY, ADEGHIKLNPQRSTV, FHILNY
2	II. Destabilization of ground state by hydrophobic packing around charged phenolate	6.4 × 107	520	143, 146, 161, 163, 175, 177	W, S, I, Q, A, V	MW, CST, IM, MQ, A, V	ACFGILMPRSTVW, ACGST, IM, KLMQ, A, V
3-1	III. Stabilization of polarization of excited state by π-stacking the chromophore phenolate with Tyr197	1.6 × 105	250	73, 146, 195, 217	V, S, V, A	V, CST, V, ACS	IV, ACGST, AIMTV, ACGST
3-2‡		1.6 × 105	240	73, 146, 195, 217	V, S, V, A	V, CGS, V, AS	AV, ACGST, AITV, ADGNST

^*Defined by user.

^†Numbering based on DsRed.

^‡Red-shifting mutation Thr197 was included and stabilized by designing neighboring residues.

Prior to computational design, these 13 positions were subjected to site-saturation mutagenesis and screening to identify single mutations producing emission red-shifts. We found several point mutations that caused small red-shifts (3–8 nm): F14S, F14N, V16S, V16T, W143M, S146C, S146T, I161M, Q163M, I197T, I197Y, and A217S. I197Y was already known to cause red-shifts in mRFP1.1 (13) and GFP (23), F14S was shown to red-shift DsRed (25), and V16T was critical in red-shifting mGrape2 (13). I197R and I197H led to hypsochromic shifts (to 593 nm and 605 nm, respectively), a result that correlates well with the fact that basic residues near the chromophore phenolate, such as K163 in DsRed and H199 in amFP486 (26), decrease emission wavelength. This effect can be explained by stabilization of the ground state through a favorable electrostatic interaction with the phenolate negative charge.

For each of the designed positions, the red-shifting point mutations identified above and the wild-type residue were required in the final library composition during computational library design. A217C, which was found to enhance brightness, was also required for some libraries, as was S146G (to allow similar degenerate codon composition for libraries 3-1 and 3-2).

For libraries 1 and 2, we specified a size of ∼500 sequences—a compromise between sufficient sequence diversity and ease of screening. Library 3 was divided into two half-size libraries of ∼250 sequences. Library 3-1 only allowed Tyr in its π-stacking conformation (see Materials and Methods) at position 197 (to test design hypothesis III) and 3-2 only allowed the red-shifting Thr at position 197, but in this case its conformation was allowed to vary.

Computational library design results are listed in Table 1. For library 1, 540 sequences predicted to be the most energetically favorable were identified for screening. The final library composition included a large variety of amino acids of different sizes and properties. Each position included multiple H-bond donor residues, allowing comprehensive testing of design hypothesis I. Interestingly, Glu and Gln found in mPlum and its E16Q mutant were also found to be favorable at position 16 even though they were not required during library design. For library 2, the wild-type residue was the most energetically favorable at positions 175 and 177. Hydrophobic residues were predicted at each designed position, allowing design hypothesis II to be adequately tested. In addition to the required wild-type Gln and red-shifting Met at position 163, another hydrophobic residue (Ile) was predicted as well as Lys, the residue found in parent DsRed. The composition of libraries 3-1 and 3-2 was very similar but not identical, as expected, given the similarity of the required amino acids.

Library Screening.

The four mutant libraries were screened for emission wavelength red-shifts using a 96-well plate fluorimetric assay. Screening of library 1 identified seven mutants with red-shifts ≥4 nm (λ_em≥615 nm) (Table S1). All retained the wild-type Tyr at position 120 and contained an H-bond donor (Ser or Thr) at positions 14 or 16. These findings correlate well with design hypothesis I. Neither Glu nor Gln at position 16 caused a red-shift, which is seen with mPlum and its E16Q variant. However, it has been shown that Glu must be combined with F65I to induce a red-shift in mPlum (11). Although the point mutants F14S and F14N showed small red-shifts, library members containing these mutations did not exhibit λ_em≥615 nm.

Screening of library 2 produced 16 mutants with red-shifts ≥9 nm. All contained a hydrophobic residue at position 163 (Leu or Met), which correlates well with design hypothesis II. I161M and Q163M found in library 2 are also observed in mRaspberry and mPlum, which could partially account for their far-red-shifted emission wavelengths. Neither the wild-type Gln nor Lys (found in DsRed) was found at position 163 in our most red-shifted mutants. All had Cys, Thr, or Ala at position 146, with Cys strongly preferred (14 of 16). Given the low propensity of Cys to form H-bonds, it is possible that no or weak H-bonding to the chromophore phenolate would shift electron density away from it and towards the imidazolinone and acylimine groups, resulting in a red-shift. At position 143, nonpolar, polar, and aromatic amino acids were found, indicating no preference. Screening of library 3-1 identified nine mutants with red-shifts ≥9 nm, whereas only one was found for library 3-2. The brightest from these two libraries were the triple mutants A195-Y197-C217 (AYC) and T195-T197-N217 (TTN). Interestingly, only the small amino acids Ala, Val, and Thr were found at position 195 in red-shifted mutants from library 3-1 (Table S1), indicating that a small residue at this position may be required to accommodate the Y197 mutation. The larger amino acids Met and Ile, although included in the library (Table 1), were not found at this position in the most red-shifted mutants.

We next combined the best red-shifting mutations from each library to determine if the red-shifting effects were additive. Starting with either the AYC or TTN triple mutants from libraries 3-1 and 3-2 as templates, two combinatorial libraries were prepared. V16T from library 1 and S146C/T, I161I/M, and Q163L/M from library 2 were introduced into both templates. Wild-type S146 was also included as it is important in maintaining the chromophore in its deprotonated state (27). Thus, a total of 24 additional mutants were screened using the same 96-well plate assay. These second-generation mutants displayed red-shifted emission wavelengths of up to 26 nm, well over the 8–10 nm induced by AYC or TTN alone (Table 2). The results demonstrate that red-shifting mutations from the different libraries can have an additive effect and that this is independent of the template used. The observed additivity could arise from the fact that these mutations cause red-shifts through independent processes. For example, the V16T mutation, which satisfies design hypothesis I, caused a 4 nm red-shift on either the AYC or TTN templates. Moreover, mutants containing additional mutations that satisfy design hypothesis II (I161M, Q163M, and Q163L) displayed even greater red-shifts, adding 9–18 nm to the red-shifts induced by the template. Noticeably, proteins with combinations of mutations from all three libraries had decreased quantum yields, leading to lower brightness.

Table 2.

Spectral properties of various RFPs

Protein	λex (nm)	λem (nm)	λem shift from mCherry (nm)	Extinction coefficient (M1 cm1)	ΦF	Relative brightness * (% mCherry)	pKa	Mutations away from mCherry	Mutations
									Library 1	Library 2			Libraries 3-1 and 3-2
									16	146	161	163	195	197	217
mFruits
mCherry	588	611	-	101,000	0.22	100	4.7	-	V	S	I	Q	V	I	A
mRaspberry	600	625	+14	62,000	0.15	42	-	10†	-	-	M	M	T	-	-
AYC template
3-1.A3	596	621	+10	63,000	0.08	23	-	3	-	-	-	-	A	Y	C
3-1.A6	595	625	+14	61,000	0.06	16	-	4	T	-	-	-	A	Y	C
mRojoA	597	633	+22	48,000	0.02	4	5.4	6‡	T	-	-	L	A	Y	C
mRojoB	598	631	+20	61,000	0.06	16	5.0	5	T	-	-	M	A	Y	C
TTN template
3-2.A3	596	619	+8	56,000	0.13	33	-	3	-	-	-	-	T	T	N
3-2.A9	596	623	+12	39,000	0.11	19	-	4	T	-	-	-	T	T	N
3-2.H11	597	628	+17	28,000	0.03	4	-	5	T	-	-	L	T	T	N
mRouge	600	637	+26	43,000	0.02	4	6.1	7	T	C	M	M	T	T	N
Point mutants
V16T	590	615	+4	73,000	0.20	66	-	1	T	-	-	-	-	-	-
S146C	593	619	+8	50,000	0.06	14	-	1	-	C	-	-	-	-	-
I161M	590	614	+3	89,000	0.16	64	-	1	-	-	M	-	-	-	-
Q163L	590	614	+3	73,000	0.09	30	-	1	-	-	-	L	-	-	-
Q163M	590	615	+4	65,000	0.25	73	-	1	-	-	-	M	-	-	-
V195A	588	610	−1	48,000	0.21	45	-	1	-	-	-	-	A	-	-
V195T	590	613	+2	71,000	0.23	73	-	1	-	-	-	-	T	-	-
I197Y	598	618	+7	79,000	0.03	11	-	1	-	-	-	-	-	Y	-
I197T	587	616	+5	79,000	0.05	18	-	1	-	-	-	-	-	T	-
A217C	588	613	+2	101,500	0.22	100	-	1	-	-	-	-	-	-	C
A217N	568	597	−14	31,000	0.28	39	-	1	-	-	-	-	-	-	N

^*Brightness is calculated as the product of the extinction coefficient and the fluorescence quantum yield.

^†This number does not include the N- and C-terminal sequences coming from GFP and other spacer sequences.

^‡Contains an additional undesired mutation (R125H).

Spectroscopic Characterization.

Absorption, excitation, and emission spectra were measured for the most red-shifted mutants and are reported in Fig. S1. We identified three second-generation mutants with λ_em > 630 nm and named them mRojoA, mRojoB, and mRouge, after the Spanish and French words for red, respectively. Both mRojoA and mRojoB were derived from the AYC triple mutant, whereas mRouge was derived from TTN. These proteins differ from mCherry by 5–7 mutations, display λ_em of 631–637 nm, and have λ_ex near 600 nm, 9–12 nm higher than λ_ex of mCherry (Table 2, SI Text, and Fig. S1). All three mutants achieve a longer λ_em than mRaspberry with fewer mutations away from mCherry. This result indicates that these mutations mostly result in λ_em bathochromic shifts, which is expected because we screened only for red-shifted emission.

Extinction coefficients are reported for each mutant in Table 2. The extinction coefficients determined for mCherry and mRaspberry differ from those reported in the literature (72,000 M^-1 cm^-1 for mCherry and 86,000 M^-1 cm^-1 for mRaspberry) (5, 11). This discrepancy is caused by differences in the experimental conditions during the critical chromophore maturation phase. These conditions include the availability of soluble oxygen, the time allowed for maturation, and variations in temperature. Nonetheless, all the extinction coefficient values reported in Table 2 are for proteins purified in the same 96-well plate under identical maturation conditions. mRojoA, mRojoB, and mRouge exhibit much lower quantum yields (< 0.10) than mCherry even though their extinction coefficients are not drastically smaller. In all cases, our mutants are not as bright as mCherry. This result is not unexpected as we did not screen for improved brightness nor did our computational design target residues towards this goal. Because mRaspberry is also brighter than our mutants, we expect that this is accounted for by the extra mutations in mRaspberry (with respect to mCherry).

To better understand the role of each mutation on the observed spectral properties, 11 point mutants accounting for all the mutations found in mRojoA, mRojoB, and mRouge were prepared and characterized. As shown in Table 2, mutations V16T, S146C, Q163M, I197T, and I197Y are largely responsible for the red-shifts, while S146C, Q163L, I197T, and I197Y primarily account for the decreases in quantum yield. Lower quantum yields could result from decreased rigidity of the chromophore. It is unclear how these mutations affect this property. A217N is the only point mutant that causes a hypsochromic shift. The Q163M and A217N point mutants display higher quantum yields, but A217C is the only single mutant that is as bright as wild-type mCherry.

Causes of Red-Shifts in RFPs.

Crystal structures were solved for our two most red-shifted mCherry variants, mRojoA and mRouge (Table S2, Fig. 2, Figs. S2, S3, S4, S5, and SI Text). The conformational predictions of the CPD software agree well with the crystal structures obtained (Fig. S2), as indicated by the atomic rmsds for the 13 designed residues (0.30 ± 0.02 Å for mRojoA and 0.23 Å for mRouge). Analysis of the crystal structures demonstrates that two of the three design hypotheses were clearly achieved. Hypothesis I (Fig. 1A) was fulfilled by the T16 mutation found in both variants, which provides a water-mediated H-bond with the acylimine group of the chromophore (Fig. 2). This interaction should stabilize the resonance form of the chromophore in which negative charge is localized on the acylimine oxygen. A similar interaction was demonstrated to red-shift the emission wavelength in Neptune (13). Hypothesis III (Fig. 1C) was fulfilled by the I197Y mutation in mRojoA, which causes a clearly observable π-stacking interaction with the chromophore phenolate group (Fig. 2A), similar to that observed in YFP (23). The I197Y mutation has been shown to contribute to red-shifted emission wavelength in a series of recently reported mFruits termed mGrapes (13), presumably through a similar process. Additionally, a recently reported variant of DsRed-Express2 called E2-Crimson (28) contains the S197Y mutation, which along with Q66F is largely responsible for the observed red-shift (λ_em = 646 nm).

Fig. 2.

Crystal structures of designed positions in the most red-shifted mutants, mRojoA (A) and mRouge (B). The chromophore is shown in magenta. Residue C217 in mRojoA (A) and residues C146 and T195 in mRouge (B) exhibit two conformations in the crystal structure. H-bonds are indicated by dashed lines.

It is more difficult to determine whether design hypothesis II (increasing hydrophobicity around chromophore phenolate) was fulfilled. Compared to mCherry, mRojoA and mRouge have more hydrophobic residues at position 163 directly above the chromophore phenolate ring (Leu and Met, respectively), replacing the more polar Gln found at this position in mCherry. M163 is found in the two far-red mFruits, mRaspberry and mPlum, suggesting that it is involved in the red-shifting observed. The measured pKa values for mRojoA, mRojoB, and mRouge are higher than for mCherry (Table 2, SI Text, and Fig. S6), indicating that the hydroxyl hydrogen of the chromophore phenol group is less acidic. Because a more hydrophobic environment would destabilize the ionized form of the chromophore phenol, a higher pKa would be an expected result of satisfying design hypothesis II.

Far-Red Fluorescent Proteins.

The quest for higher emission wavelength monomeric RFPs is ongoing given the enormous potential for applications in whole-body imaging of research model animals. Far-red RFPs (> 630 nm) that are already available include mKate/mKate2 (12, 14), HcRed (29), RFP639 (2), E2-Crimson (28), mPlum (11), Neptune/mNeptune (13), AQ143 (30), and TagRFP657 (31). With λ_ems ranging from 630–657 nm, most of these far-red RFPs have been successfully used in imaging experiments. However, they all have disadvantages, such as pKa’s near physiological pH (mKate/mKate2), oligomeric states (RFP639, HcRed, E2-Crimson, and AQ143), monomer-dimer equilibriums (Neptune/mNeptune), slower maturation (TagRFP657), low brightness (HcRed, AQ143), and incomplete maturation (mPlum). Recently, monomeric bacterial phytochrome-derived FPs that emit in the IR region (> 700 nm) have been developed (32). These IR-FPs are very useful but require an exogenous cofactor for fluorescence. Thus, engineering GFP homologues that emit brightly in the far-red or near-IR region of the spectrum is still highly desirable. Until now, using structure-based rational design to increase emission wavelength has proven difficult. For this reason, most far-red FPs were discovered by screening very large libraries of mutants generated by random mutagenesis in a directed evolution approach.

Here, using CPD as a prescreen, we combined structure-based rational design with experimental screening to increase the emission wavelength of mCherry, a monomeric RFP. The computational prescreen discarded mutations that were incompatible with the protein fold, allowing us to drastically decrease the size of libraries required for experimental screening. We were thus able to achieve a 26 nm red-shift by screening libraries several orders of magnitude smaller than those previously used to achieve comparable red-shifts in other RFP scaffolds. Performing two rounds of combinatorial mutagenesis, we screened a total of ∼5,000 clones to identify several mutants with λ_ems > 630 nm. In contrast, starting from mRFP1.2 (whose λ_em of 612 nm is similar to that of our starting structure), Wang et al. performed 10 rounds of somatic hypermutation random mutagenesis and screened millions of cells to obtain mRaspberry (λ_em = 625 nm, Φ_F = 0.15) (11). Perhaps the best comparison to mRaspberry is our 3-2.A9 mutant, which is red-shifted by the same amount (λ_em = 623 nm) and has a similar Φ_F (0.11) (Table 2). We were able to isolate 3-2.A9 by combining three mutations (found after screening only ∼700 clones from library 3-2) with one mutation (V16T) that we had previously identified through site-saturation mutagenesis. To obtain mPlum from mRaspberry, Wang et al. performed an additional 13 rounds of somatic hypermutation, screening millions of clones (11).

The development of the far-red-shifted RFP mKate was guided by identification of an optimized dimeric mutant of eqFP578 termed TurboRFP. Using four rounds of random mutagenesis and screening > 100,000 clones, Shcherbo et al. isolated an RFP with an λ_em of 635 nm termed Katushka (14). The red-shifting mutations of Katushka were then inserted into a monomeric variant of TurboRFP, yielding mKate. TagRFP657, a further optimized mKate mutant with an λ_em of 657 nm, is currently the most red-shifted monomeric RFP (31). Recently, Strack et al. (28) developed E2-Crimson, a very red-shifted DsRed-derived RFP. E2-Crimson was isolated after screening more than 1,000,000 colonies in three rounds of targeted combinatorial mutagenesis and two rounds of random mutagenesis. Thus, directed evolution approaches, though generally more time-consuming and costly than CPD-based approaches, have yielded several useful RFPs.

Generating targeted, computationally designed combinatorial libraries has a distinct advantage in that it allows one to identify combinations of mutations that would have been difficult to predict rationally or to obtain through random mutagenesis (which often builds on point mutations discovered in individual rounds). Synergistic effects are more easily obtained by a semirational approach involving combinatorial mutagenesis (33). For instance, under other experimental circumstances the A217N mutation found in the TTN triple mutant motif (Fig. S5) described earlier would have been discarded during screening as it causes a hypsochromic shift. However, these three mutations together resulted in an 8 nm bathochromic shift. Another example of synergistic effects is seen in mutant 3-1.A3 (Table 2), which contains the red-shifting but quantum yield-reducing I197Y mutation. The presence of V195A and A217C compensate for the decrease in quantum yield, bringing it up from 0.03 to 0.08. Because neither A217C nor V195A increase the quantum yield by themselves (Table 2), this increase is the result of the interplay between these three residues. Synergistic effects are similarly seen in mutant 3-2.A3, likely due to the presence of the quantum yield-increasing A217N.

Conclusion

Using a structure-based rational approach that combines CPD with experimental screening, we were able to identify three mutants exhibiting emission red-shifts of 20–26 nm: mRojoA, mRojoB, and mRouge. Although these mutants are not as bright or as red-shifted as other useful RFPs, these results suggest that this approach may be applicable to red-shift the emission wavelength of other RFPs. It could also be used to further red-shift the recently engineered phytochrome-based IR-FPs. We expect that other useful properties, such as increased quantum yield and maturation rate, could also be improved using this method. Additionally, red-shifted mutants developed using our design procedures could serve as templates for optimization of other properties such as brightness through random mutagenesis, thereby combining the benefits of both rational design and directed evolution.

Materials and Methods

Computational Design.

Four independent libraries were computationally designed using the procedure described in SI Text. Residues 14, 16, and 120 of mCherry were designed in library 1. Residues 143, 146, 161, 163, 175, and 177 were designed in library 2. Residues 73, 146, 195, and 217 were designed in libraries 3-1 and 3-2. These latter two libraries differ by the presence of a Tyr (library 3-1) or Thr (library 3-2) residue at position 197. All 20 proteinogenic amino acids were allowed at designed positions; residues with side chains pointing towards and within 4 Å of the designed residues were allowed to sample alternative conformations during the design, but their identities were not modified. The crystallographic conformer at each designed position was also allowed. A standard backbone-dependent side chain rotamer library (34) with expansions by one standard deviation about χ₁ and χ₂ was used. Prior to our design procedure, we generated an in silico structure of mCherry mutant I197Y in which residue Y197 is stacked next to the chromophore phenolate group. To generate this structure, which would serve as the input structure for computational design in library 3-1, we sampled different conformations of Y197 using a large backbone-independent conformer library (35). During computational design of library 3-1, the conformation of the π-stacked Y197 residue was not allowed to vary. The energy function used is described in SI Text.

Screening.

Whole Escherichia coli cells expressing mCherry mutants (prepared as described in SI Text) were screened in 96-well plates using a Tecan Safire2 plate reader. Emission spectra (λ_ex = 565 nm) and excitation spectra (λ_em = 650 nm) were measured. Purification and spectroscopic characterization of mutants is described in SI Text.