Rational conversion of chromophore selectivity of cyanobacteriochromes to accept mammalian intrinsic biliverdin

To check whether AnPixJg2 could incorporate BV, His-tagged AnPixJg2 was purified from the BV-producing Escherichia coli. Although AnPixJg2 could bind BV and showed reversible photoconversion between a Pfr form with an absorbance maximum at 698 nm and a Po form with a maximum at 614 nm, the binding efficiency was quite low (3.7 ± 1.9%) compared with AM1_C0023g2 (70%) and AM1_1557g2 (40%) (Fig. 1 A and B, Table 1, and SI Appendix, Fig. S2 A and B) (25, 26). We have also reported that BV-binding efficiency of AM1_1870g3 is as low as that of AnPixJg2 (27). These facts suggest that a sequence comparison based on the AnPixJg2 structure (24) may provide clues to the molecular mechanism of BV incorporation. From this comparison, we found that nine residues were specifically conserved between AM1_C0023g2 and AM1_1557g2 within 6 Å of the chromophore, namely Val273, Gln307, Tyr310, Lys318, Thr325, Ser334, Tyr335, Asp337, and Val353 in AM1_C0023g2. These correspond to residues Ala256, Glu290, His293, Arg301, Phe308, Gly317, His318, Ser320, and Ile336 in AnPixJg2 (SI Appendix, Fig. S1 C and D). Among these, the Ser334/Gly317 position has already been shown to be important for BV binding in AM1_C0023g2, and replacement of Ser334 with Gly improves the stability of the BV-binding holoprotein (25). This suggests that Gly317 in AnPixJg2 is potentially important for BV-binding capability. Therefore, we focused on the other eight residues in AnPixJg2 (details of the Gly/Ser position are given in the section of “Gly/Ser position” in SI Appendix) and performed individual site-directed mutageneses on AnPixJg2: A₂₅₆V, E₂₉₀Q, H₂₉₃Y, R₃₀₁K, F₃₀₈T, H₃₁₈Y, S₃₂₀D, and I₃₃₆V. All E. coli cells expressing these variants exhibited similar colors and were visually indistinguishable to those expressing the wild-type protein (SI Appendix, Fig. S1E).

Thus, we constructed a mutant, in which the eight residues of AnPixJg2 were simultaneously replaced with the corresponding residues conserved between AM1_ C0023g2 and AM1_1557g2. The mutant, AnPixJg2_BV8 (A₂₅₆V, E₂₉₀Q, H₂₉₃Y, R₃₀₁K, F₃₀₈T, H₃₁₈Y, S₃₂₀D, and I₃₃₆V), exhibited significant improvement in BV incorporation with retaining far-red/orange reversible photoconversion. To evaluate the BV incorporation capability of the variant proteins, we established two parameters: the BV-binding efficiency of purified protein and the expression enhancement of holoprotein relative to wild-type protein (calculation details are described as Materials and Methods). The binding efficiency and expression enhancement of AnPixJg2_BV8 were 52.6 ± 9.1% and 52.6 ± 19.5-fold, respectively (Fig. 1B, Table 1, and SI Appendix, Fig. S2 A and B).

Based on the PCB-binding AnPixJg2 structure, the Ala256, Glu290, and Ser320 side chains are unlikely to be directly involved in chromophore binding (SI Appendix, Fig. S1C). Furthermore, we identified another BV-binding CBCR GAF domain, AM1_6305g2, from the cyanobacterium A. marina (SI Appendix, Fig. S3 and Table S1). This domain possesses a His instead of the Lys in AM1_C0023g2 and AM1_1557g2 at the position corresponding to Arg301 in AnPixJg2 (SI Appendix, Fig. S1D). Thus, we constructed a mutant, AnPixJg2_BV4 (H₂₉₃Y, F₃₀₈T, H₃₁₈Y, and I₃₃₆V), in which the aforementioned four residues were reverted to originals. The mutant could bind BV, exhibited far-red/orange reversible photoconversion, and its binding efficiency and expression enhancement were 74.7 ± 13.0% and 75.7 ± 8.6-fold, respectively, which were significantly higher than those of AnPixJg2_BV8 (Fig. 1 B and C, Table 1, and SI Appendix, Fig. S2 A and B). This suggests not only that the other four residues are dispensable for BV binding, but that some were, in fact, inhibitory.

To reduce the number of mutagenized residues further, we constructed four kinds of mutant in which each mutated residue of AnPixJg2_BV4 was reverted back to the original: AnPixJg2_BV3_H293 (F₃₀₈T, H₃₁₈Y, and I₃₃₆V), AnPixJg2_BV3_F308 (H₂₉₃Y, H₃₁₈Y, and I₃₃₆V), AnPixJg2_BV3_H318 (H₂₉₃Y, F₃₀₈T, and I₃₃₆V), and AnPixJg2_BV3_I336 (H₂₉₃Y, F₃₀₈T, and H₃₁₈Y). All mutants exhibited far-red/orange reversible photoconversion (SI Appendix, Fig. S2A). The binding efficiencies of AnPixJg2_BV3_H293, AnPixJg2_BV3_F308, AnPixJg2_BV3_H318, and AnPixJg2_BV3_I336 were 54.2 ± 18.6%, 4.0 ± 2.1%, 41.4 ± 15.7%, and 38.5 ± 13.8%, respectively, while their expression enhancements were as 44.7 ± 4.0, 1.3 ± 0.3, 49.0 ± 3.4, and 22.5 ± 5.9-fold, respectively (Fig. 1B, Table 1, and SI Appendix, Fig. S2B). The binding efficiency and expression enhancement of all AnPixJg2_BV3 variants were significantly lower than those of AnPixJg2_BV4, but AnPixJg2_BV3_F308 exhibited largest reductions in both binding efficiency and expression enhancement, suggesting that it plays the most important role in BV incorporation. Furthermore, AnPixJg2_BV3_I336 exhibited a larger defect in expression enhancement than AnPixJg2_BV3_H293 and AnPixJg2_BV3_H318, suggesting that the I₃₃₆V replacement may play a more important role than the H₂₉₃Y and H₃₁₈Y replacements. This consideration prompted us to construct AnPixJg2_BV2 (F₃₀₈T and I₃₃₆V). AnPixJg2_BV2 bound BV with a 37.8 ± 21.3% binding efficiency and 23.1 ± 7.4-fold expression enhancement and exhibited far-red/orange reversible photoconversion (Fig. 1B, Table 1, and SI Appendix, Fig. S2 A and B). Although its binding efficiency and expression enhancement were slightly lower than those of AnPixJg2_BV3_H293 and AnPixJg2_BV3_H318, AnPixJg2_BV2 maintained a moderately high BV-incorporation capability compared with the wild-type protein. Furthermore, we had already constructed AnPixJg2_F₃₀₈T and AnPixJg2_I₃₃₆V, and had found that their BV-binding efficiencies (10.1 ± 10.1% and 13.0 ± 6.7%, respectively) and expression enhancements (3.5 ± 1.0 and 5.5 ± 0.8-fold, respectively) were significantly lower than those of AnPixJg2_BV2 (Fig. 1B, Table 1, and SI Appendix, Fig. S2 A and B). Collectively, these results indicated that the residues Tyr293, Thr308, Tyr318, and Val336 are essential for BV incorporation, with Thr308 and Val336 are likely to be more critical than Tyr293 and Tyr318. In conclusion, AnPixJg2_BV4 is the variant that exhibits the most efficient and stable BV incorporation. The section “Spectral comparison of AnPixJg2 and its variants” in SI Appendix presents the details of the spectral properties of these variant proteins (Table 1 and SI Appendix, Fig. S2 A, C, and D).

To obtain direct insights into the molecular mechanism of BV incorporation, we determined the crystal structure (1.6 Å resolution) of the Pfr form of AnPixJg2_BV4 (SI Appendix, Fig. S4 A–C and Table S2, see the section of “SI Appendix, Crystal Structure Determination” for details), which exhibited the best BV incorporation among the AnPixJg2 variants. The overall structure of BV-bound AnPixJg2_BV4 was quite similar to that of PCB-bound AnPixJg2 (the root mean square deviation was 0.908 Å for the Cα atoms of residues 233–388) except that the N-terminal α-helix was oriented in the opposite direction, which may be an artifact derived from crystal packing (SI Appendix, Fig. S4 D and E). Notably, electron density maps (2|F_o| − |F_c|) clearly showed that the canonical Cys residue (Cys321) of AnPixJg2_BV4 bound to C3² position in the vinyl group of BV, whereas that of AnPixJg2 bound to C3¹ position in the ethylidene group of PCB (Fig. 2 A and B and SI Appendix, Fig. S4 D and E). Because additional carbon (C3²) was inserted in the covalent bond linkage between the C3 of the A ring and the sulfur atom of Cys321, the BV incorporated into AnPixJg2_BV4 shifted relative to the PCB of the WT structure, in which the shifting was calculated as about 0.75 Å. It is of note that the ring C shifted toward the β-sheets of the chromophore binding pocket compared with the PCB incorporated into AnPixJg2 (Fig. 2C). Nevertheless, the structural arrangements of five residues (Trp289, Asp291, Arg301, His322, and Tyr352) reported to be important for chromophore stability (24) were highly conserved between these two structures (Fig. 2 A and B and SI Appendix, Fig. S4 F and G, shown as yellow green and deep pink, respectively). Although Tyr352 was located at the β-sheets, its interaction partner, the D ring carbonyl group, hardly shifted toward the β-sheets, and so this interaction should be conserved between these two structures. Arg301 participated in chromophore binding in different ways in the two proteins. In AnPixJg2_BV4, it stabilized the ring C propionate of BV by hydrogen bonding via a water molecule, whereas in AnPixJg2 it directly interacted with the same propionate of PCB by hydrogen bonds.

Notably, although the side chains of the four mutation sites (His293, Phe308, His318, and Ile336) were positioned around ring C of the chromophores in both structures, the protein–chromophore interactions of these side chains were totally different between these two structures (Fig. 2 A and B and SI Appendix, Fig. S4 F and G, shown as blue green and deep red, respectively). In the Phe/Thr308 and Ile/Val336 positions, the bulky Phe308 and Ile336 residues of AnPixJg2 held the C ring by hydrophobic interactions (Fig. 2D, Lower). Because these residues were located at the β-sheets around the C ring, these residues should be affected by the chromophore shifting. The fact that AnPixJg2 could not efficiently bind BV means that these bulky residues may cause steric hindrance of the C ring, which is largely shifted toward the β-sheets in comparison with that of PCB (Fig. 2C). This interpretation is supported by the superimposition of BV on the AnPixJg2 structure (Fig. 2D, Middle). Conversely, the compact Thr308 and Val336 residues of AnPixJg2_BV4 avoided steric hindrance with the C ring (Fig. 2D, Upper). It is noteworthy that the ring C propionate of BV was in a vertical direction toward the B–C plane, whereas that of PCB lay parallel with the plane (Fig. 2 A and B). This suggests that not only replacement with compact residues, but also a flexible rearrangement of the ring C propionate, is important for BV incorporation (Fig. 2 A and B). In this context, the other replacements at the His/Tyr293 and His/Tyr318 positions contribute to the rearrangement (Fig. 2 A and B and SI Appendix, Fig. S4 F and G). The His318 of AnPixJg2 interacted with the ring B propionate of PCB by a direct hydrogen bond, while the Tyr318 of AnPixJg2_BV4 directly hydrogen bonded with the ring C propionate of BV. The His293 of AnPixJg2 stabilized the ring C propionate of PCB by hydrogen bonding via a water molecule, while Tyr293 of AnPixJg2_BV4 did not form any hydrogen bonds with the chromophore. These distinctive interaction networks should facilitate “parallel” and “vertical” conformations of the ring C propionate. Namely, replacement of His318 with Tyr resulted in switching the interaction partner from the ring B propionate to the ring C propionate and directly contributed to the vertical conformation, whereas replacement of His293 with Tyr resulted in removal of hydrogen bonds with the ring C propionate to contribute to destabilization of the parallel conformation. Moreover, the hydroxy group of Thr308 indirectly hydrogen bonded with the ring C propionate of BV via a water molecule (Fig. 2A and SI Appendix, Fig. S4F), which also facilitated the vertical conformation. This indicates that replacement of Phe308 with Thr has effects not only on reduction of steric hindrance, but also on rearrangement of the ring C propionate, which is consistent with the fact that F₃₀₈T replacement had the largest contribution on efficient BV incorporation (Fig. 1B and Table 1). In conclusion, these four residues in AnPixJg2_BV4 work cooperatively to solve problems triggered by chromophore shifting, resulting in stable BV incorporation.

To expand our strategy to other CBCR GAF domains within the XRG lineage, we targeted several domains with a PCB chromophore that have atypical spectral properties: AnPixJg4 (Pg–to–Pr rapid dark reversion) (28); AM1_1186g2 (red/blue reversible photoconversion) (29); AM1_1870g3 (red-shifted red/green reversible photoconversion) (27); NpF2164g3 (violet/orange reversible photoconversion) (30); and NpF2164g5 (intense far-red fluorescence without photoconversion) (31). Because all these wild-type proteins exhibit a little or no BV incorporation (Fig. 3 A–C and SI Appendix, Fig. S5 A–J and Table S1), we engineered these domains for BV incorporation based on the AnPixJg2_BV4 sequence (SI Appendix, Fig. S5K).

Introduction of mutations into AnPixJg4, AM1_1870g3, and NpF2164g5 significantly improved BV-binding efficiencies, which were 44%, 82%, and 70%, respectively (Fig. 3 A–C and SI Appendix, Fig. S5 A–C and F–H and Table S1). In contrast, introduction of the same mutations into AM1_1186g2 and NpF2164g3, whose sequences near the chromophore are greatly diversified from those of typical XRG CBCR GAF domains including AnPixJg2, did not improve BV-binding efficiencies (SI Appendix, Fig. S5 D, E, I, and J and Table S1). AnPixJg4_BV4 exhibited Pfr-to-Po photoconversion and rapid Po-to-Pfr dark reversion, in which the half-life of the AnPixJg4_BV4 dark reversion was 2.8 s ± 0.1 at 25 °C. This is about 400-fold faster than that of AnPixJg2_BV4 (1,199 s ± 12.9 at 25 °C) (Fig. 3 A and D and SI Appendix, Fig. S5A). Faster dark reversion enables the regulation of biological activities by monochromatic far-red light illumination. AM1_1870g3_BV4 exhibited far-red/orange reversible photoconversion, whose difference spectrum (Pfr – Po) possesses a positive peak at 718 nm, which is red-shifted by 18 nm compared with that of AnPixJg2_BV4 (Fig. 3 B and E and SI Appendix, Fig. S5B). This red-shifted property is advantageous for optogenetic control within deep tissues in mammals. NpF2164g5_BV4 stably absorbed far-red light without photoconversion and exhibited fluorescence covering the near-infrared (NIR) region with a 4% quantum yield. Excitation and emission spectra of NpF2164g5_BV4 bound to BV peaked at 680 nm and 700 nm, respectively, and were red-shifted compared with those of NpF2164g5 bound to PCB (Fig. 3 C and F and SI Appendix, Fig. S5C). It is of note that the fluorescence excitation spectrum showed a prominent shoulder at 641 nm, which may reflect heterogenous population. Detailed studies such as structure determination of this molecule is needed to understand this unique fluorescence property. In contrast, variant proteins based on AnPixJg2_BV2 and the AnPixJg2_BV3_H293 sequences could not bind BV efficiently (SI Appendix, Fig. S5 A–C). These results indicate that the “BV4” replacement is a robust engineering design for producing BV-binding capability.

We show that the compact NIR fluorescent protein, NpF2164g5_BV4, is applicable to in vivo imaging in living mice. To determine its applicability, we first tested whether the NIR fluorescence of NpF2164g5_BV4 could be observed in a live mammalian cell. COS-7 cells were transfected with NpF2164g5_BV4. After incubation with BV, the cells were imaged using a confocal fluorescence microscope. To quantitatively evaluate the NIR fluorescence of NpF2164g5_BV4, we fused it with green fluorescent protein (GFP) and normalized variations in its expression levels by dividing the NIR fluorescence by the GFP fluorescence. The cells expressing NpF2164g5_BV4 exhibited bright NIR fluorescence compared with those expressing GFP as a negative control (Fig. 4 A and B). Next, we examined whether the NIR fluorescence of NpF2164g5_BV4 could be imaged in vivo in living mice. We transiently transfected mice liver with NpF2164g5_BV4 by hydrodynamic tail vein (HTV) injection. The transfected mice were i.v. injected with BV and then imaged using an in vivo fluorescence imager. NIR fluorescence of NpF2164g5_BV4 was observed from the liver of the transfected mice (Fig. 4 C and D). Additionally, the isolated liver from the transfected mice was confirmed to exhibit bright NIR fluorescence compared with that from the uninjected mice (Fig. 4 E and F). These results indicate that NpF2164g5_BV4 is applicable to in vivo fluorescence imaging of deep tissues of living mice.

We compared NpF2164g5_BV4 with iRFP, a near-infrared fluorescent protein widely used for in vivo fluorescence imaging (6). COS-7 cells transfected with iRFP were incubated with BV before imaging as in the case of NpF2164g5_BV4. iRFP has previously been engineered by extensive mutagenesis of a BV-bound bacteriophytochrome, RpBphP2, such as D₂₀₂H mutation to inhibit its photoconversion and an additional 13 mutations including those around the D ring of BV to greatly increase its fluorescence (6). However, NpF2164g5_BV4 has mutations around the B and C rings for conferring BV-binding property but has not yet been subjected to mutagenesis around the D ring of BV to improve its fluorescence. Nevertheless, the NIR fluorescence intensity of NpF2164g5_BV4 has already reached 20% of that of iRFP (Fig. 4 A and B), suggesting that NpF2164g5_BV4 could be a platform to develop a powerful, compact NIR fluorescent probe for in vivo fluorescence imaging.

For the denaturation assay, both the dark state (15Z-isomer) and photoproduct (15E-isomer) of the native proteins were fivefold diluted into 7 M guanidinium chloride (GdmCl)/1% (vol/vol) HCl and absorption spectra were recorded at room temperature before and after 3 min of illumination with white light.

To monitor photoconversion and dark reversion processes, absorbance at 698 nm against red light (700 nm) was measured for AnPixJg2_BV4 and AnPixJg4_BV4 for 2 min with dark intervals of 5 min at 25 °C. Half-lives were calculated from the dark reversion kinetics.

To calculate BV-binding efficiency, we measured the absorbance spectra of Z-isomers of free BV (Frontier Scientific) at various concentrations and constructed a standard curve. Using this standard curve, we calculated BV-binding efficiencies of various proteins based on the absorbance values of Z-isomers at 700 nm denatured by 7 M GdmCl/1% (vol/vol) HCl.

To estimate the relative expression enhancements of the holoproteins (AnPixJg2 and its mutants) expressed in E. coli C41 pKT270, lysates were illuminated with orange light (620 nm and 600 nm) and the holoproteins purified in the Pfr form under dark conditions using an NGC Chromatography System (Bio-Rad). Enhancements were estimated based on the area of each peak monitored by absorbance at 700 nm during chromatography and normalized to fresh cell weight.

The molar extinction coefficients of the dark state and photoproduct of the native proteins were calculated from each absorbance value at the peak wavelength per the molar concentration of the holoprotein as determined by Bradford assay.

The mice were transfected with GFP-NpF2164g5_BV4 by hydrodynamic tail vein (HTV) injection using TransIT-EE Hydrodynamic Delivery Solution (Mirus) according to the manufacturer’s instructions. As a negative control, we used uninjected mice. Twenty-three hours after the HTV injection, both the HTV-injected and uninjected mice were i.v. injected with BV (50 nmol/g mouse weight), and the abdominal hair of them was removed using a depilatory cream. One hour after the BV injection, the mice were anesthetized with isoflurane (Wako) and imaged with Lumazone in vivo imaging system (Shoshin EM) equipped with an EMCCD camera (Evolve 512; Photometrics) and an emission filter, ET700/75m (Chroma). As an excitation light source, a liquid-cooled high-output LED light source (LAMBDA HPX; Shutter instrument) with a tunable filter changer (LAMBDA VF-5; Shutter instrument) was used. Fluorescence images were obtained in two channels: channel 1 was acquired with 570 ± 7 nm excitation (1.05 W/m² at the specimen) and 700 ± 37.5 nm emission, and channel 2 was acquired with 630 ± 6.5 nm excitation (0.53 W/m² at the specimen) and 700 ± 37.5 nm emission. The exposure time was 100 ms, and the camera parameters for gain and intensification were set to 1 and 100, respectively. After the in vivo imaging, the liver was isolated from the mice using standard surgical procedure and imaged under the same imaging channels and conditions as the in vivo imaging. Additionally, to verify the expression of GFP-NpF2164g5_BV4, its GFP fluorescence in the isolated liver was also determined using a fluorescence stereo zoom microscope (Axio Zoom.V16; Carl Zeiss) equipped with a filter set for GFP (filter set 38 HE; Carl Zeiss).