The gene encoding full-length AtphyA (Unicode entry PHYA_ARATH, residues 1-1122) was obtained from M.D. Zurbriggen and subcloned onto the pCDF backbone (Novagen) using PCR amplification and Gibson cloning (Gibson et al., 2009). The gene was thus equipped with a C-terminal hexahistidine tag and placed under control of a lactose-inducible T7 promoter. As in a previous study (Golonka et al., 2019), the resultant plasmid pDG388 also harbored a second expression cassette under lactose-inducible T7 control that comprised the genes for Synechocystis sp. heme oxygenase and the biliverdin reductase PcyA, originally subcloned from the pKT270 plasmid (Mukougawa et al., 2006). A plasmid map of pDG388 is supplied as Supplementary Material. The plasmid pDG388 was confirmed by Sanger DNA sequencing (Microsynth) and transformed into E. coli BL21(DE3) LOBSTR cells (Andersen et al., 2013). Transformed bacteria were grown at 37°C and 225 rpm in 8 × 1,000 ml TB (terrific broth) medium supplemented with 100 μg ml⁻¹ streptomycin. Once the optical density at 600 nm reached around 0.6, the incubation temperature was lowered to 18°C, 0.5 mM δ-aminolevulinic acid was added, and expression was induced by 1 mM isopropyl-β-thiogalactopyranoside. Following incubation at 18°C for 2 days, cells were harvested by centrifugation, resuspended in 50 mM Tris/HCl pH 8.0, 20 mM NaCl, 20 mM imidazole, and lysed by ultrasound. After clearing by centrifugation, the suspension was applied to a 1-ml Protino Ni-NTA column (Macherey-Nagel) using an Äkta prime chromatography system. Protein was eluted using a gradient from 2 to 500 mM imidazole over 50 ml. Fractions of 1 ml were collected and analyzed by denaturing polyacrylamide gel electrophoresis. Covalently bound chromophore and total protein were visualized by Zn-induced fluorescence (Berkelman and Lagarias, 1986) and Coomassie staining, respectively. Fractions were pooled based on purity and dialyzed into 50 mM Tris/HCl pH 8.0, 20 mM NaCl, 10% (w/v) glycerol. Samples were concentrated by spin filtration (Amicon Ultra spin column, molecular-weight cutoff 10,000 Da), flash-frozen in liquid nitrogen and stored at −80°C. In the course of optimizing solution conditions, AtphyA was purified at different pH values. To this end, all solutions were buffered with 50 mM sodium carbonate at pH 9.3 or pH 10.6, respectively.

Absorbance spectroscopy was performed on an Agilent 8453 diode-array spectrophotometer at 22°C. To determine the concentration of AtphyA, a molar extinction coefficient of 21,200 M⁻¹ cm⁻¹ at 672 nm was used. Absorbance spectra were recorded for dark-adapted AtphyA and after saturating illumination with LEDs of (650 ± 15) nm and (720 ± 15) nm emission wavelength. The Pr:Pfr ratio obtained after 650-nm illumination was calculated according to (Butler et al., 1964).

Prior to cryo-electron microscopy experiments, the frozen AtphyA samples were thawed and purified by size-exclusion chromatography. Size-exclusion chromatography (SEC) of full-length AtphyA was performed at 12°C using an Äkta system (GE Healthcare) with a flow rate of 0.3 ml min⁻¹. A Superose 6 10/300 GL column (GE Healthcare) was equilibrated in 20 mM Tris/HCl pH 8.0, or 50 mM sodium carbonate buffer pH 9.3, respectively. Prior to loading, AtphyA samples were converted to Pr state by illuminating with 730 nm far-red light.

Cryo-EM grids were prepared under dim green safe light using a Vitrobot (FEI) with the sample chamber at 4°C and 100% humidity. Three microliter protein sample (1–1.5 mg ml⁻¹) was pre-illuminated with 730 nm far-red light and applied to glow-discharged C-flat Holey carbon R 2/2-300 grids. Excess solution was removed by blotting filter paper, and the grids were plunge-frozen in liquid ethane. The samples were imaged using a Titan Krios operated at 300 kV and a magnification of 130,000-fold. The images were recorded on a Quantum K2 camera with pixel size of 1.06 Å and an exposure rate of 70 electrons per Ų for a total 40 frames. The targeted defocus range was varied from −1 to −2.5 μm using the EPU software (ThermoFisher).

A total of 1,221 movies were collected, motion-corrected, and contrast transfer function (CTF)-estimated using cryoSPARC2 (Punjani et al., 2017). After the micrographs were denoised using Topaz denoise, a small number of particles (~450) were manually picked and subjected to two-dimensional classification to generate references for the template picker in cryoSPARC2 (Bepler et al., 2020). About 271,000 particles were initially autopicked and extracted. Junk particles were removed by several rounds of 2D class averaging, and the good particles were used in Topaz deep pick (Bepler et al., 2019). Particles were further cleaned up with two more rounds of 2D class averaging. A set of 136,173 particles was used to reconstruct an initial model using Ab-initio in cryoSPARC2. The model was further refined using homogeneous refinement with 2-fold symmetry. The final model has a resolution of approximately 17 Å based on gold-standard Fourier shell correlation at 0.143 cutoff (Scheres and Chen, 2012).

AtphyA model analysis and crystal-structure docking to the low-resolution model were performed in UCSF Chimera (Pettersen et al., 2004). The crystal structures of the PAS-GAF domains of phyA from Glycine max (PDB code 6TC7; Nagano et al., 2020), the PHY domain of plant phyB from A. thaliana (4OUR; Burgie et al., 2014), the PAS domain of the BphP from Xanthomonas campestris (5AKP; Otero et al., 2016), and the PAS-linked histidine kinase from Thermotoga maritima (3A0R; Yamada et al., 2009) were used for the docking.

To facilitate biophysical analyses, especially structural investigation, we set out to establish an efficient protocol for the heterologous expression in E. coli of AtphyA at full length. To this end, we adapted a pCDF plasmid system previously employed for the production of the AtphyB PCM at good yield and purity (Golonka et al., 2019, 2020). By embedding the AtphyA gene in this system, it was furnished with a C-terminal hexahistidine affinity tag and placed under control of a T7-lacO promoter (Figure 2A). To supply the phycocyanobilin chromophore during heterologous expression, the plasmid additionally bore a T7-lacO-controlled bicistronic operon which encodes the Synechocystis sp. heme oxygenase (HO) and the biliverdin reductase PcyA (Mukougawa et al., 2006). The plasmid was transformed into the E. coli LOBSTR expression strain (Andersen et al., 2013), a derivative of BL21 that carries the DE3 lysogen and genomic modifications that alleviate the contamination by endogenous bacterial proteins during purification by immobilized metal-ion affinity chromatography (IMAC). Protein expression was performed over 2 days at 18°C. To improve chromophore supply during expression, 0.5 mM δ-aminolevulinic acid was added as a precursor in the biosynthesis of porphyrin (Shemin and Russell, 1953), from which in turn biliverdin and other bilins derive. Even after a single purification step by IMAC, full-length AtphyA could be obtained with covalently attached PCB, as detected by zinc-induced bilin fluorescence (Berkelman and Lagarias, 1986), and but few contaminants, as revealed by Coomassie staining (Figure 2B). Per liter of bacterial culture, around 1 mg AtphyA could routinely be prepared.

We used UV/vis absorbance spectroscopy to assess the photochemical integrity of the protein preparation (Figure 2C). In its dark-adapted state, AtphyA adopted a mixed population of the Pr and Pfr states, with Q-band absorbance maxima at 650 and 715 nm, respectively. Illumination with red light (650 nm) achieved a Pfr:Pr ratio of 0.57:0.43. Under far-red light (720 nm), AtphyA was converted almost entirely to its Pr state with little Pfr remaining. In the Pr state, the absorbance ratio of the Q band and the Soret band at 358 nm amounted to 2.4.

Having ascertained intact photochemistry, we next used size-exclusion chromatography for further purification and to assess the solution properties and homogeneity of the AtphyA sample. The chromatographic analysis revealed higher-order AtphyA aggregates that eluted within the void volume of the column (Figure 3A). Based on the molecular weight range the column can resolve, we estimated a size of these aggregates in the mega-Dalton range. Attempts at dissolution of the AtphyA aggregates by dialysis into different buffers failed.

Reasoning that the aggregates formed irreversibly, we set out to alter the pH value at which protein purification was conducted. Given an isoelectric point (pI) for AtphyA of 5.9, we opted for higher pH values than 8.0 in the initial attempt. At pH 9.3, AtphyA could be purified by IMAC with similar yield and purity as before. Absorbance spectroscopy revealed that the Q-band intensity somewhat decreased relative to that of the Soret band (ratio 1.5:1; Figure 3B). A similar attenuation of the Q band at alkaline pH was for instance reported for both bacterial phytochromes and cyanobacteriochromes and was ascribed to partial deprotonation of the bilin chromophore (Hirose et al., 2013; Rumfeldt et al., 2019). That notwithstanding, AtphyA showed intact and reversible Pr↔Pfr photoconversion at this pH value. Using SEC, we went on to assess the oligomeric state of the protein preparation in solution and found it markedly improved (Figure 3C). In contrast to the SEC run at the lower pH which was dominated by high-weight aggregates, the run at pH 9.3 revealed a mixture of AtphyA dimers and monomers with but few larger aggregates. To assess whether the solution properties could be further enhanced, we elevated the pH to 10.6. Following expression and purification, the UV/vis spectroscopic analysis, however, showed an altered Q-band absorption without fine structure, of lower intensity and shifted to shorter wavelengths (Figure 3D). Given that this spectrum resembled that of isolated PCB, we concluded that at this pH chromophore binding is severely perturbed although covalent attachment to the AtphyA protein evidently still took place during expression in E. coli. Illumination with red light did not elicit any spectral changes, and we hence discontinued the experiments at a pH of 10.6.

We next prepared grids for the structural analysis of full-length AtphyA by cryo-EM. To this end, AtphyA was first illuminated with 730 nm to convert it to its Pr state. Samples were then applied to carbon grids and plunge-frozen. When initially AtphyA was prepared at pH 8.0, the electron micrographs showed pervasive protein aggregation, despite prior filtration of the sample (Figure 4A). These observations concurred with the SEC analysis (see Figure 3A) and effectively precluded further structural analysis at this pH. By contrast, grids prepared for AtphyA purified at pH 9.3 yielded dispersed particles (Figure 4A), again consistent with SEC (see Figure 3C), and were used for subsequent analysis. Following Topaz denoising, single particles were picked manually and based on templates. We tested different box sizes to extract the picked particles, with the best 2D class averages obtained for a box size of 448 pixels (Figure 4B). From 136,173 particles, the reference-free initial model of AtphyA was acquired without imposing any symmetry. After refinement with 2-fold symmetry imposed, a final 3D electron density map of AtphyA was obtained with a resolution of approximately 17 Å (Figure 4C). Two-dimensional projections of the map resembled the class averages of the experimental data (Figure 4B).

The map revealed an oblong shape with approximately 2-fold symmetry and dimensions of around 100 Å × 150 Å × 250 Å. The size and the form of the map immediately implied a parallel (i.e., head-to-head) arrangement of the AtphyA homodimer, similar to the shape previously observed in cryo-EM studies on DrBphP (Li et al., 2010) and AtphyB (Burgie et al., 2017). Even prior to more detailed modeling, the lower and upper parts of the density map can be assigned to the PCM and CTM, respectively. Notably, the middle region of the map exhibited a hole of 60 Å × 50 Å in its middle, reminiscent of observations made for DrBphP (Li et al., 2010). Interestingly, the overall strength of the electron density is weaker around the hole, as illustrated by contouring the map at different levels (Figure 4C). Additional, more diffuse density surrounds an ordered core with strong density, in particular near the bottom (assigned to the PCM) and top (assigned to the CTM) of the map (Figure 4C). We tentatively ascribed these density differences to varying degrees of static and dynamic structural heterogeneity in the PCM and CTM. In support of this notion, we note that sensor histidine kinases, and by that token the HKRD in AtphyA, exhibit inherently high levels of structural malleability that underpins their function (Gushchin and Gordeliy, 2018; Buschiazzo and Trajtenberg, 2019; Möglich, 2019; Multamäki et al., 2020).

To obtain detailed insight into the AtphyA architecture, we docked structures of the isolated PCM and individual CTM domains (PAS-A, PAS-B, and HKRD) inside the electron density (Figure 5). The PAS-GAF fragment of G. max phyA (PDB identifier 6TC7; Nagano et al., 2020) was first positioned in the bottom half of the electron density (Figure 5). To locate the PHY domains, the AtphyB PCM structure (4OUR; Burgie et al., 2014) was then superposed on the previously placed G. max PAS-GAF fragment. Each of the two monomers was placed separately and resulted in a good fit of the electron density for the PHY domains. The model thus indicated that contacts between the two PCM protomers are mainly formed by the PAS and GAF domains, whereas the PHY domains are separated. Notably, this arrangement is supported by ample structural data on bacterial phytochromes across which the position of the PHY domains and the distance between them are quite variable (Takala et al., 2020). The map also revealed additional electron density near the N-terminus of the PCM which can be ascribed to the N-terminal extension (Figure 5, circle). Although no high-resolution structural template is available for the NTE, the spatial location of the surplus density lends credence to our modeling of the PCM.

To model the CTM, a PAS-A fragment was derived from the full-length structure of the X. campestris BphP (5AKP; Otero et al., 2016) and placed adjacent to the PHY domain, again resulting in a good fit of the electron density. Next, we jointly positioned the PAS-B and the DHp-RD subdomain of the HKRD using as a model the PAS-linked histidine kinase ThkA from T. maritima (3A0R; Yamada et al., 2009). Whereas the DHp-RD antiparallel four-helix bundle well-fitted the top part of the strong electron density, the PAS-B domains were splayed apart and accounted for the weaker electron density below the top of the model (see Figure 4B, lower contour level). Although the PAS-B domains well-fitted the density, we deem their placement tentative given the available data quality. By contrast, the CA-RD subdomain of the HKRD could not be positioned in the electron density with confidence, arguably reflecting its mobility with respect to the DHp-RD module, as evidenced by widely different DHp:CA orientations across sensor histidine kinase structures (Marina et al., 2005; Diensthuber et al., 2013; Wang et al., 2013; Trajtenberg et al., 2016; Buschiazzo and Trajtenberg, 2019).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.