Each cycle of our directed evolution strategy (see Fig. 6) began with random mutagenesis to identify those positions that affected either the maturation or brightness of the red fluorescent protein. Once several residues were identified, expanded libraries were constructed in which several of these key positions were simultaneously mutated to a number of reasonable substitutions (see Tables 2 and 3). These directed libraries combine the benefits of shuffling of improved mutant genes (1) with an efficient method of overcoming the limited number of substitutions accessible during random mutagenesis by error-prone PCR. Most methods of in vitro recombination (1, 2) rely on random gene fragmentation but our approach, and a similar method for gene shuffling (3), instead use PCR to generate designed fragments that can be reassembled (4) to give the full-length shuffled gene. A related modification of the DNA shuffling procedure used random gene fragmentation but included synthetic oligonucleotides in the reassembly step to randomly mutate residues of interest (5).

Libraries generated by error-prone PCR had a typical error rate of 2–3 mutations per 1,000 bp and approximately 25,000 to 50,000 colonies of E. coli were manually screened for each library. Directed libraries were of variable genetic diversity (approximately 400 to 6,000,000 unique sequences, see Table 3) though diversity inside the b-barrel was never greater than about 4,000 unique sequences. For each directed library approximately 10,000 to 30,000 colonies were screened. E. coli colonies expressing mutant red fluorescent proteins were evaluated on both the magnitude of their red fluorescence under direct excitation at 540 nm and the ratio of emission intensities at 540 nm over 470-nm excitation. Whereas the former constraint selected for very bright or fast maturing mutants, the latter constraint selected for mutants with decreased 470-nm excitation or red-shifted excitation spectra. Our screening strategy was likely biased toward fast maturing mutants with high folding efficiency, so we have not yet screened specifically for maximum fluorescence quantum yield and bleach resistance and we do not know whether monomers will be inherently inferior in these two properties.

Although reasonable progress toward rescuing DsRed-I125R had been made, our focus quickly turned to the fast tetramer T1 (7) when we discovered that introduction of the I125R mutation into this protein resulted in a dimer that matured in only a few days, similar to our best DsRed dimers at that time. We therefore proceeded with a similar, yet less ambitious, directed mutagenesis strategy starting from T1-I125R (Table 2, library D1) and eventually identified dimer1. Dimer1 was somewhat better than wild-type DsRed both in terms of brightness and rate of maturation but had a substantial green peak equivalent to that of T1. Dimer1 was also somewhat blue-shifted with an excitation maximum at 551 nm and an emission maximum at 579 nm. Error-prone PCR on dimer1 (Table 2, library D2) resulted in the discovery of dimer1.02 containing the mutation V71A in the hydrophobic core of the protein and effectively no green component in the excitation spectra. A second round of random mutagenesis identified the mutations K70R, which further decreased the green excitation, S197A, which red-shifted the dimer back to DsRed wavelengths, and T217S, which greatly improved the rate of maturation. Unfortunately, K70R and S197A matured relatively slowly and T217S had a green excitation peak equivalent to DsRed. With dimer1.02 as the template, two more rounds of directed mutagenesis were performed; the first focusing on the three positions identified above (Table 2, library D3) and the second on C117, F118, F124, and V127 (Table 2, library D4). Our best dimer variant contained a total of 17 mutations and has been designated dimer2 (Fig. 1B).

In a simultaneous effort, we undertook a more direct approach to breaking up the AC interface through introduction of dimer-breaking mutations. Dimer1 was the template for the first such library (Table 1, library M1) in which nine different positions were targeted, including two key AC interface residues, H162 and A164, which were forced to be either lysine or arginine. The brightest colonies from this library were difficult to distinguish from the background red fluorescence of the E. coli colonies even after prolonged imaging with a digital camera. Suspect colonies were restreaked on LB/agar and allowed to mature at room temperature for 2 weeks, and a crude protein preparation was analyzed by SDS/PAGE. Imaging of the gel revealed a single faint band consistent with the expected mass of the monomer and thus mRFP0.1 (monomeric red fluorescent protein) was identified. Sequencing of this clone revealed that mRFP0.1 was equivalent to dimer1 with mutations E144A, A145R, H162K, K163M, A164R, H222G, and H224G. Of these mutations, only K163M is inside the b-barrel and is probably responsible for rescuing what little residual red fluorescence there is. The six external mutations all contribute to the disruption of the former AC dimer interface. Random mutagenesis on mRFP0.1 (Table 2, library M2) resulted in the discovery of the much brighter mRFP0.2, which gave an unambiguous red fluorescent and monomeric band by SDS/PAGE, and which contained the single additional mutation Y192C. In the tetramer, and presumably dimer, of DsRed, Y192 is a central component of the extensive solvent bridged hydrogen bond network found in the AC interface (9, 10). In the monomeric protein, former interface interactions are no longer relevant though Y192 may participate in a solvent-exposed hydrogen bond network. Inspection of the crystal structure of DsRed gives no clue as to how these interactions could interfere with chromophore formation. Both mRFP0.1 and mRFP0.2 displayed at least 3-fold more green fluorescence than red fluorescence, but as expected for the monomer there was no FRET between the green and red components.

With the suspicion that mutations that were beneficial to the dimer could also benefit the monomer, we subjected a template mixture including mRFP0.2, dimer1.56, HF2Ga, and HG2Gb to a combination of PCR-based template shuffling and directed mutagenesis (Table 2, library M3). The top clone identified in this library, mRFP0.3, was relatively bright and had a greatly diminished green fluorescent component. In addition, mRFP0.3 was approximately 10 nm red-shifted from DsRed and was derived primarily from dimer1.56 though at least one mutation, the A71V reversion, must have resulted from a crossover event. The goal of the next directed library (Table 2, library M4) was to investigate the effect of mutations at K83, which have previously been shown to cause a red shift in DsRed (11). The top two clones, designated mRFP0.4a and mRFP0.4b, contained the K83I or L mutation, respectively, were 25 nm red-shifted relative to DsRed, and were very similar in terms of maturation rate and brightness. Unlike all the previous generations of the monomer, colonies of E. coli transformed with mRFP0.4a were red fluorescent within 12 h after transformation when excited with 540 nm light and viewed through a red filter. A template mixture of mRFP0.4a and mRFP0.4b was subjected to random mutagenesis (Table 2, library M5) and the resulting library was thoroughly screened. The five fastest maturing clones from this library were derived from mRFP0.4a and contained individual mutations L174P, V175A (two clones), F177C and F177S. The F177S clone or mRFP0.5a, appeared to mature slightly faster and had the smallest green peak in the absorbance spectra. One colony isolated from this library was exceptionally bright when grown on LB/agar but expressed very poorly when grown in liquid culture. This clone, designated mRFP0.5b, was derived from mRFP0.4b and contained two new mutations: L150M inside the barrel and V156A outside. Inspection of the crystal structure of DsRed suggests that the L150M could be filling an interior cavity left by the K83L mutation.

The next library (Table 2 library M6) was intended to optimize the region around residues V175 and F177 in both mRFP0.5a and the increasingly divergent mRFP0.5b. The top clone in this library, designated mRFP0.6, was derived from mRFP0.5b, though of three other top clones, one was derived from mRFP0.5b, one was from mRFP0.5a, and one appeared to have resulted from multiple crossovers between the two templates. The final library (Table 2, library M7) targeted residues in the vicinity of L150 because this was the one remaining critical mutation that was derived from random mutagenesis and had not been reoptimized. Top clones had combinations of mutations at all targeted positions though the clone with the single mutation R153E was found to express slightly better in culture. This clone was further modified through deletion of the unnecessary V1a insertion and replacement of the cysteine at position 222 with a serine. The final clone, designated mRFP1 (Fig. 1C), contains a total of 33 mutations relative to DsRed.

There is considerable variation in the published values for several of the critical spectral properties of DsRed. Our current estimate of the extinction coefficient of DsRed, 57,000 M^–1•cm^–1, is significantly less than our previous estimate of 75,000 M^–1•cm^–1 (11), a discrepancy that we attribute to variations in temperature and concentration during the critical maturation period. The fraction of aged protein that is red, green, or even nonabsorbing may be exquisitely sensitive to the conditions of the maturation (12). Published values from other research groups have included 22,500 (13), 52,000 (7), and 72,500 M^–1•cm^–1 (14).

With the exception of one particularly low estimate of 0.23 (13), literature values for the fluorescence quantum yield of DsRed have been relatively consistent. Published values include 0.68 (14), 0.68 (7), and 0.7 (11). It is not clear why our current estimate of 0.79 is slightly higher than the apparent consensus value of 0.7. This value for the quantum yield was obtained consistently with different protein preparations.

A photolabile species with molar extinction coefficient e (in M^–1•cm^–1) and photobleaching quantum yield Q, exposed to illumination of intensity I (in einsteins cm^–2•s^–1) at the relevant wavelength, should bleach with an exponential rate constant k = (1,000 cm³/liter)(ln 10)I = eQ (15). The photobleaching experiment only determines the product eQ; a value for e must be known or assumed to calculate Q. We observed double exponential time courses for photobleaching of DsRed and all variants, even mRFP1 where interactions between subunits have been eliminated. Such double exponentials imply that at least two forms of protein participate in the photochemical reaction mechanism, but we can estimate e only for the initial state. Therefore in Table 4, we report k/[(1,000 cm³/liter)(ln 10)I]eQ for both the fast and slow components, then calculate the corresponding Qs assuming that e remains at its initial value. Thus our current extinction coefficient for DsRed gives Q = 1.9 ´ 10^–6 and 1.8 ´ 10^–7 for the fast and slow components, respectively, of the photobleach. These values straddle our previous estimate of 7.7 ´ 10^–7 (11), which was based on a single exponential fit and an extinction coefficient of 75,000 M^–1•cm^–1. This same extinction coefficient with our current photobleach data gives Q = 1.4 ´ 10^–6 and 1.3 ´ 10^–7, in excellent agreement with values of 1.4 ´ 10^–6 and 1.2 ´ 10^–7 reported in the only previous study (16) in which double exponential behavior was acknowledged. Similarly, our current value for the quantum yield of photobleach of EGFP (8.3 ´ 10^–6), which closely approximates a single exponential, is in excellent agreement with a previous study (8 ± 2 ´ 10^–6) (17). Other studies that have not accounted for the double exponential nature of DsRed photobleach have determined values of 9.5 ´ 10^–6 (18) and 1.5 ´ 10^–4 (19).

1. Minshull, J. & Stemmer, W. P. C. (1999) Curr. Opin. Chem. Biol. 3, 284–290.

2. Giver, L. & Arnold, F. H. (1998) Curr. Opin. Chem. Biol. 2, 335–338.

3. Gibbs, M. D., Nevalainen, K. M. H.& Bergquist, P. L. (2001) Gene 271, 13–20.

4. Ho, S. N., Hunt, H. D., Horton, R. M., Pullen, J. K. & Pease, L. R. (1989) Gene 77, 51–59.

5. Liu, D. R., Magliery, T. J., Pastrnak, M. & Schultz, P. G. (1997) Proc. Natl. Acad. Sci. USA 94, 10092–10097.

6. Baird, G. S. (2001) Ph.D. thesis (Univ. of California, San Diego).

7. Bevis, B. J. & Glick, B. S. (2002) Nat. Biotechnol. 20, 83–87.

8. Terskikh, A. V., Fradkov, A. F., Zaraisky, A. G., Kajava, A. V. & Angres, B. (2002) J. Biol. Chem. 277, 7633–7636.

9. Yarbrough, D., Wachter, R. M., Kallio, K., Matz, M. V. & Remington, S. J. (2001) Proc. Natl. Acad. Sci. USA 98, 462–467.

10. Wall, M. A., Socolich, M. & Ranganathan, R. (2000) Nat. Struct. Biol. 7, 1133–1138.

11. Baird, G. S., Zacharias, D. A. & Tsien, R. Y. (2000) Proc. Natl. Acad. Sci. USA 97, 11984–11989.

12. Mizuno, H., Sawano, A., Eli, P., Hama, H. & Miyawaki, A. (2001) Biochemistry 40, 2502–2510.

13. Matz, M. V., Fradkov, A. F., Labas, Y. A., Savitsky, A. P., Zaraisky, A. G., Markelov, M. L. & Lukyanov, S. A. (1999) Nat. Biotechnol. 17, 969–973.

14. Patterson, G., Day, R. N. & Piston, D. (2001) J. Cell Sci. 114, 837–838.

15. Adams, S. R., Kao, J. P. Y., Grynkiewicz, G., Minta, A. & Tsien, R. Y. (1988) J. Am. Chem. Soc. 110, 3212–3220.

16. Lounis, B., Deich, J., Rosell, F. I., Boxer, S. G. & Moerner, W. E. (2001) J. Phys. Chem. 105, 5048–5054.

17. Peterman, E. J. G., Brasselet, S. & Moerner, W. E. (1999) J. Phys. Chem. A 103, 10553–10560.

18. Heikal, A. A., Hess, S. T., Baird, G. S., Tsien, R. Y. & Webb, W. W. (2000) Proc. Natl. Acad. Sci. U S A 97, 11996–12001.

19. Harms, G. S., Cognet, L., Lommerse, P. H., Blab, G. A. & Schmidt, T. (2001) Biophys. J. 80, 2396–2408.