Diversity of rhodopsin cyclases in zoospore-forming fungi

While searching for RGC genes in available fungal genomes (10), we have identified these photoreceptors in the majority of early diverging fungal phyla (Fig. 1), supporting the idea that RGC photoreceptors were already present in the common ancestor of fungi (11). Phylogenetically, the RGC genes group along three branches, separating the modulating, far-red light-absorbing NeoR sequences from the two RGC(a) and RGC(b) branches, with conventional RGCs responsible for light-triggered enzymatic catalysis (Fig. 2A). We assigned the NeoR and RGC(a) clade to the two subunits of heterodimeric RGCs, while RGC(b) shows homodimeric RGCs. Homodimeric RGC(b) sequences are present in various phyla, such as Blastocladiomycota, including the previously characterized homodimeric BeRGC and CaRGC (3–5), Sanchytridiomycota (12), and Monoblepharidomycota (Fig. 2A). In Chytridiomycota, which has the greatest species diversity of early diverging fungi, they are found in the orders Synchytriales, Cladochytriales, Rhizophydiales, and Chytridiales (Figs. 1 and 2A). In contrast, the heterodimeric NeoR/RGC(a) photoreceptors are restricted to Chytridiales and Cladochytrium replicatum^*. C. replicatum and species of Chytridiaceae, an early branching family in the Chytridiales, comprise NeoR, RGC(a), and RGC(b) sequences, indicating that these organisms utilize homo- and heterodimeric RGCs (Fig. 2B).

By reviewing high-covering genome data, we found that NeoR- and RGC(a)-coding genes cluster closely together but are encoded by complementary DNA strands (Fig. 2C) so that the respective NeoR and RGC(a) sequences are transcribed in opposite directions. Similar gene clusters are known to exist in other fungi, where they allow bidirectional gene transcription from a single transcription start region (14). This conserved gene arrangement and the monophyletic branching of NeoR and RGC(a) when rooting the RGC phylogenetic tree along the Blastocladiomycota branch (SI Appendix, Fig. S1) supports the hypothesis of their evolution from a common ancestor and a gene duplication and reverse reintegration for the heterodimeric RGCs. Both the species tree and our phylogenetic RGC reconstructions share the same topology (SI Appendix, Fig. S1 A and B), showing that heterodimeric RGCs (NeoR and RGC(a)) evolved from common precursors along the homodimeric RGC(b) branch and are still closely related to homodimeric RGC(b) sequences in the Chytridiaceae. Therefore, the NeoR and RGC(a) sequences in C. replicatum that both relate to the respective RGCs of Cladochytrium polystomum (Fig. 2A and SI Appendix, Fig. S1) may be a result of horizontal gene transfer.

In the Rhizophydiales Globomyces pollinis-pini Arg68 v1.0 and Gorgonomyces haynaldii MP57 v1.0, we identified two (GpRGC1 and GpRGC2) and three homodimeric RGC(b) genes (GhRGC1, GhRGC2, and GhRGC3), respectively, with the latter linearly clustered in a single locus (Fig. 2C). Surprisingly, the homodimeric GpRGC2 and GhRGC2 are interspaced by or linked to NeoR-like rhodopsins (GhNeoRh and GpNeoRh) that lack linker and cyclase and are encoded on the reverse complementary DNA strand.

For functional characterization of heterodimeric RGC, we selected C. confervae, C. lagenaria, and C. polystomum (Fig. 2C) as model systems that reflect the large RGC variation. The C. confervae system is complex, with seven RGC sequences arranged in three clusters, whereas the two sequences in C. polystomum represent single heterodimers. We recorded the photocurrents under voltage clamp conditions in Xenopus laevis oocytes expressing different combinations of NeoR and RGC(a) proteins together with a cGMP-gated ion channel of olfactory neurons (A2-CNGC) (6). Co-expression of the respective NeoR and RGC(a) proteins^† showed functional heterodimeric photoreceptors, except for CcNeoR4 from C. confervae, which was nonfunctional in all tested combinations (SI Appendix, Fig. S2A). To evaluate trans-species compatibility, we coexpressed RgNeoR with CpRGC, which resulted in photocurrents, whereas the combination of CpNeoR with RgRGC1 was unresponsive (SI Appendix, Fig. S2B). As homodimers, if any, none of these proteins produced photocurrents with the coexpressed A2-CNGC reporter channel, strengthening the claim that these photoreceptors function as obligate heterodimers.

We expected the proteins of the RGC(b) branch to be functional as homodimers similar to RGCs from Blastocladiomycota fungi (3–5). Indeed, GpRGC1 from G. pollinis-pini Arg68 v1.0 was active as a homodimer, and both the activation and inactivation kinetics were fast compared to all functional heterodimers (SI Appendix, Fig. S2C) but similar to previously reported homodimeric RGC (4). Moreover, we tested two variants GpRGC2 (G. pollinis-pini) and GhRGC2 (G. haynaldii MP57 v1.0) that were expected to have red-shifted absorption (see below) in mammalian cells (ND7/23), which enables a more detailed analysis than in oocytes. Both function as homodimers with half-saturation light intensities at 2.8 and 3.2 mW/mm², respectively (Fig. 3A).

The now-available high number of NeoR-sequences allowed us to compare in greater detail the mechanism underlying the near-infrared sensitivity of NeoR chromophores. Originally, a triad of potential counterions (referred to as ci1 to ci3) of the protonated retinal Schiff base (RSBH⁺) chromophore: E136, D140, and E262 has been suggested to be crucial for far-red absorption in RgNeoR (R. globosum) (Fig. 4A) (6). These counterion positions are indeed conserved in most NeoR sequences, including CcNeoR1, CcNeoR2 (C. confervae), and CpNeoR (C. polystomum) (Fig. 4A and SI Appendix, Fig. S3). Although CcNeoR2 barely expresses in mammalian cells (HEK293-T), recombinant CcNeoR1 and CpNeoR both express well, are bistable, and absorb either in the far red or in the UV range (SI Appendix, Table S1 and Fig. 4B) (6). CpNeoR absorbs maximally at 721 nm and is the first identified rhodopsin with an absorption maximum beyond 700 nm. The narrow spectral shape of CpNeoR₇₂₁ (HBW = 854 cm⁻¹) resembles that of RgNeoR, resulting in an extraordinary extinction coefficient of 167,000 M^-1cm⁻¹, which is more than 3-fold higher than most other rhodopsins (Fig. 5B). To study the origin of the extra redshift in CpNeoR, we looked for amino acid residues within the retinal binding pocket that are unique to this rhodopsin. CpNeoR-M138 and CpNeoR-N192 were interesting candidates because mutating the homolog position in RgNeoR (E141 and Y195) resulted in a red-shifted absorption (6). However, neither the CpNeoR-M138Q nor the CpNeoR-N192Y mutant showed visible absorption, whereas the RgNeoR-E141M (λ_max 698 nm; +8 nm red shift) and Y195N (λ_max 681 nm; −9 nm blue shift) mutants with CpNeoR-like retinal-binding pockets caused some absorption shift, albeit in opposite directions.

Some native NeoRs have either a glutamine (CcNeoR3; CsNeoR1) or alanine (CcNeoR4) at the ci1 position, which raises the question of how the absorption is changed accordingly (Fig. 2C). CcNeoR3 with glutamine at ci1 (CcNeoR3 Q137, Fig. 4A and SI Appendix, Fig. S4A) absorbs maximally at ~410 nm, likely reflecting a deprotonated retinal Schiff-base (RSB). The glutamate mutant spectrum (CcNeoR3-Q137E) still absorbs in the UV close to 400 nm (SI Appendix, Fig. S4A) and does not recover the far-red absorption. The purified CcNeoR4 protein with alanine at ci1 showed no retinylidene absorption, indicating the lack of retinal binding. The function of CcNeoR4 in heterodimeric RGCs remains puzzling because the protein lacks a canonical cyclase-transducer-element (CTE) (16) and widely conserved residues crucial for metal- and guanine-base binding in its cyclase domain, which explain the loss of CcNeoR4 functionality. Another interesting candidate was ClNeoR1 (C. lagenaria) with Ser at position ci2 but still absorbing maximally at 695 nm (Fig. 4 A and B and SI Appendix, Fig. S3) although with broader bandwidth compared to RgNeoR₆₉₀ (HBW = 1,658 cm⁻¹). The closely related ClNeoR2 did not express in HEK293-T cells. Although ci2 (D140) in RgNeoR was previously considered important for near-infrared absorption, ClNeoR1-S140D lost visible absorption. Thus, we introduced the reverse mutation in the well-expressing RgNeoR. Both the RgNeoR-D140S and the nonpolar RgNeoR-D140A remain far-red absorbing and bistable (λ_max = 686 nm and 665 nm; SI Appendix, Table S1 and Fig. S4B), whereas the D140T mutation resulted in a significant blue shift of the maximum absorption (λ_max = 630 nm) (6).

To gain insights into the changes caused by ci2 (D140) mutation in RgNeoR at a molecular level, we conducted atomistic molecular dynamics (MD) simulations of RgNeoR, RgNeoR-D140S, and RgNeoR-D140T. These calculations involved protonated E136, D140, and E262 (model 1), as well as a model with deprotonated E262 (model 2) according to previously theoretical studies (6, 17). For all simulations, the root-mean-square deviations (RMSDs) of the protein backbone reached equilibrium after 200-ns simulation time (SI Appendix, Fig. S8 C and D). Hydrogen bond analysis in model 1, involving the retinal, adjacent water (within 3 Å of the RSBH⁺), and all the ci positions protonated (E136, D140, and E262), showed no or very rare interactions between water and the protonated retinal Schiff base RSBH⁺ in RgNeoR and the RgNeoR-D140S mutant. In contrast, a high occupancy of water molecules near the RSBH⁺ was observed in RgNeoR-D140T (SI Appendix, Fig. S9C). The low water occupancy in RgNeoR and RgNeoR-D140S was attributed to the hydrogen bond between the RSBH⁺ and protonated E262, that together with the close contact between neighboring E136 and E262 effectively protected the pocket around the RSBH⁺ from water penetration (SI Appendix, Fig. S9 E and F). However, in RgNeoR-D140T, the methyl group of T140 perturbed the dynamics of E136 as indicated by the broader dihedral angle distribution compared to the wild type and the D140S mutant (SI Appendix, Fig. S9 G and H). This induced flexibility of E136 caused destabilization of the contact to E262 and, consequently, allowed water to enter the RSBH⁺ pocket. This water intrusion was not observed in model 2, where E262 was deprotonated. In this configuration, the sidechain of E136 remains rigid in all three variants (RgNeoR, RgNeoR-D140S, and RgNeoR-D140T), thereby efficiently shielding the RSBH+ pocket from water influx (SI Appendix, Fig. S9 B, D, F, and H). The presence of water is expected to elevate the dark-state dynamics in RgNeoR-D140T, leading to the spectral broadening observed. Accordingly, our simulations suggest that both E136 and E262 are protonated, at least in the context of the D140T mutant.

Like the prototypical RgNeoR, the far-red absorbing states of CcNeoR1, ClNeoR1, and CpNeoR are highly fluorescent with narrow emission spectra and small Stokes shifts of 341 cm⁻¹ (CpNeoR); 424 cm⁻¹ (CcNeoR1); and 557 cm⁻¹ (ClNeoR1) (Fig. 4C and SI Appendix, Table S1), and we expected that the excited state potential energy surfaces of the retinal share similar features. To explore the infrared limits of the chromophores, we reconstituted CcNeoR1, ClNeoR1, and CpNeoR with the (A2)-3,4 dihydroretinal, which caused extra redshifts (~1,300 to 1,400 cm⁻¹) for all three, touching 800 nm in case of CpNeoR (λ_max = 793 nm; Fig. 4B).

NeoR-like rhodopsin fragments (GhNeoRh, GpNeoRh) identified in Rhizophydiales share 24% sequence identity with the rhodopsin domain of RgNeoR. Surprisingly, these sequences lack many highly conserved residues known to constitute the retinal binding pocket in microbial rhodopsins, including the phylogenetically distant members, such as bacteriorhodopsin and heliorhodopsins (Fig. 5A). Particularly, all three aromatic residues surrounding the retinal, namely W86, W182, and Y185 in bacteriorhodopsin (BR) (aromatic triad in SI Appendix, Fig. S3), are replaced by glutamine (GhNeoRh-Q118; GpNeoRh-Q137), asparagine (GhNeoRh-N214; GpNeoRh-N233) and isoleucine/valine (GhNeoRh-I217; GpNeoRh-V236) (Fig. 5A and SI Appendix, Fig. S5). Considering the polar nature of these residues, it is difficult to envision how such a protein can bind the hydrophobic retinal cofactor. Recombinant GhNeoRh was colorless, but GpNeoRh revealed a spectrum with a maximum at 630 nm (Fig. 5 B and D) and bright fluorescence (FQY = 0.1). However, the rhodopsin is neither photochromic nor bistable and does not bleach in red light or undergo a photocycle. Despite the altered chromophore environment, the narrow spectral shape and the large extinction coefficient of 100,000 M⁻¹ cm⁻¹ are conserved, indicating, again, a rigid fluorescent chromophore with a fixed and water-less active site (Fig. 5B) and small Stokes shift of 36 nm (861 cm⁻¹). This property makes it suitable as a fluorescent marker protein. Upon expression in ND7/23, the protein can be visualized as bright clusters of red fluorescence with good expression, but the low plasma membrane-targeting requires further improvement (Fig. 5C). Reintroduction of the aromatic residues N233W caused hardly any spectral change (λ_max = 633 nm) but did cause a loss of fluorescence (FQF = 0.005, Fig. 5D), whereas I236F was colorless and Q137W showed a blue-shifted spectrum with λ_max at 592 nm (SI Appendix, Fig. S6 and Table S1) and decreased fluorescence (FQY: 0.045). Finally, the reintroduction of a second counterion Q261E blue-shifted the spectrum by 42 nm (SI Appendix, Fig. S6).

Since the light sensitivity of the catalytic activity of heterodimeric RGC-complexes is defined by the RGC(a) subunit, we explored the spectral properties of the respective RCG(a) proteins. For RgRGC1 (R. globosum), which is already described to absorb maximally at 550 nm (6), we recorded the photocycle dynamics that pass through the three photointermediates of K₆₀₅, L1₄₈₀, and L2₄₈₀ before returning to the dark state within 84 ms (Fig. 3C).

RGC(a)s from C. confervae, C. lagenaria, and C. polystomum showed absorption spectra that peak between 500 nm and 600 nm (Fig. 4C), with ClRGC1 and CpRGC being sensitive to yellow light (λ_max = 586 nm) and red-light (λ_max = 600 nm), respectively.

Notably, all RGCs(a) with λ_max ≥ 550 nm lack the widely conserved proline residue near the β-ionone ring known to be involved in color determination in other microbial rhodopsins (SI Appendix, Fig. S3) (18). Since this proline is absent in GpRGC2 and GhRGC2, we hypothesized that they are also spectrally red-shifted homodimers, and we obtained light sensitivity by action spectroscopy. We found that GpRGC2 and GhRGC2 peaked at 565 nm and 571 nm, respectively (Fig. 3B). A single mutation T137A in GhRGC2 results in a 24-nm red-shifted action spectrum peaking at 595 nm, making it the first GC that can be half-maximally activated with light at 640 nm. We note a spectral shoulder in the 500- to 550-nm range in the action spectra of GpRGC2, GhRGC2, and GhRGC2 (TA), which results in spectral broadening and may indicate some chromophore heterogeneity. The significant spectral redshifts relative to the previously described BeRGC [~525 nm (4, 5, 19), and CaRGCs (531 nm; Fig. 3B) provide expectations for red light-activated homodimeric cGMP actuators, facilitating their use as optogenetic tools.

The cyclase domain of all RGCs, including NeoR, RGC(a), or RGC(b), exhibited high conservation, with even phylogenetically distant GCs sharing about 50% sequence identity. Recombinant GC domains of the homodimeric BeRGC and CaRGCs are constitutively active, leading to the conclusion that the rhodopsin forces the enzyme into an inactive conformation in darkness and releases the constraints in the presence of light. In heterodimers, the RGC(a) alone imposed the main constraints because NeoR-activation does not release the catalytic activity. We expressed the GC domains of the heterodimeric RgNeoR and RgRGC1 separately as soluble proteins in Escherichia coli to test their cyclase function as homodimers in solution. The cyclase of RgNeoR was found to be inactive, while the GC module of RgRGC1 was catalytically active (K_M = 5.53 mM; V_max = 0.96 µmol min⁻¹ mg⁻¹) (SI Appendix, Fig. S7), albeit with lower activity than reported for CaRGC [K_M = 5.78 mM; V_max = 6.30 µmol min⁻¹ mg⁻¹ (3)]. This raises the question of what prevents the full-length RgRGC1 to function as a homodimer. Notably, the photocycle of RgRGC1 reveals no M-like state with deprotonated RSB, which has been proposed to be the enzyme-activating state in homodimeric CaRGC (3, 19, 20). However, after fusion of the RgRGC1 cyclase (RGC(a) heterodimer, Fig. 2) to the CaRGC rhodopsin (RGC(b) homodimer, Fig. 2), cGMP was neither produced in the dark-treated nor the green-illuminated samples (see CaCh2 in Fig. 3B), indicating that either the resulted chimera protein does not dimerize or that otherwise no active conformation of the enzyme is achieved. In contrast, a connection of the CaRGC cyclase to RgRGC1 rhodopsin resulted in light-induced cGMP production with variable activities depending on the connection site. Nevertheless, these chimera constructs showed substantial dark activity, which was lowest for RgCh3 with the CaRGC N terminus [~4%; (Fig. 3B)]. The photocycle of the RgCh2 chimera, which is active as a homodimer, remained similar to the wt RgRGC1 (Fig. 3C). Therefore, a specific structure of the cyclase domain prevents light-activated cGMP production in native RgRGC1 and consequently prevents its functioning as a homodimeric photoreceptor.

Given that optogenetic strategies aimed at targeting second messengers, such as cGMP could benefit from red-sensitive homodimeric RGCs, we transferred the chimera approach to the red-absorbing ClRGC1 and CpRGC, and RgNeoR. Indeed, after fusing the linker and cyclase of CaRGC with the rhodopsin of ClRGC1 and CpRGC, we observed some cGMP production, albeit with low performance, whereas the fusion construct with RgNeoR was inactive (SI Appendix, Fig. S7).

High-covering genomes from fungi were received from the JGI Fungal Genome portal MycoCosm (https://mycocosm.jgi.doe.gov/mycocosm/home), and reliable RGC sequences were manually annotated based on genomic and transcriptomic data. Genome assemblies of low-covering genomic data were either received from the NCBI database or raw sequencing data stored by JGI or in the Sequence Read Archive (NCBI) and were trimmed using BBduk and assembled by SPAdes (3.15.5) (24). RGC sequences were manually extracted from the DNA assemblies using known variants from closely related fungi as templates in CLC Main Workbench 8 (Qiagen GmbH, Hilden, Germany). Human codon-optimized DNA synthesized by IDT (Integrated DNA Technologies, Inc.; Coralville, USA) was cloned into pEGFP-C1 or pICZ (Invitrogene) for protein expression in mammalian cells or yeast. For phylogenetic reconstruction, we used 152 RGC protein sequences (including four RGCs from environmental source and excluding partial RGCs from Q. haematococci), aligned with Mafft and trimmed using TrimAI (-gt 0.95) (25), and the maximum likelihood (ML) tree was obtained using the IQ-TREE web server (http://iqtree.cibiv.univie.ac.at/) under the LG+F+R6 substitution model derived from the automatic model finding option (26) with 1,000 ultrafast bootstrap replicates. Alphafold2 modeling of RgNeoR was performed on an automated pipeline implemented by Enrico Schiewer based on the available applications (27).

Expression of various RGCs or rhodopsin fragments in mammalian cells was done as previously described (6). Briefly, RGCs were cloned into the EGFP-C1 vector (Clonetech, Takara Bio, Kusatsu, Shiga Prefecture, Japan) with C-terminal 1D4-tags (TETSQVAPA), and HEK-293T cells were transfected using TurboFect (Thermo Fisher Scientific, Waltham, MA, USA), according to the manufacturer’s protocol. After 48 h post-transfection, cells were harvested, and proteins were solubilized in HBS buffer (50 mM HEPES pH 7.4; 100 mM NaCl) containing 2% Dodecylmaltoside (DDM; Glycon Luckenwalde, Germany)/0.4% Cholesteryl hemisuccinate (CHS, Merck, Darmstadt, Germany) and purified using affinity-beads (Rho1D4, CUBE Biotech GmbH, Monheim, Germany). Site-directed mutagenesis was performed according to the Quick-change protocol from Agilent, using Pfu polymerase (Agilent Technologies, Santa Clara, CA, USA). For expression in P. pastoris, RgRGC1, RgCh2, and the rhodopsin of CpNeoR (aa 1-275), human codon-optimized sequences with C-term StrepII-tag were cloned into the pPICZ plasmid (Thermo Fisher Scientific), and their transformation and selection were performed according to the instruction of the manufacturer. Expression was induced by adding 2.5% methanol to the culture media and carried out for 24 h. Cells were lysed by a high-pressure homogenizer (HTU Digi-French-Press, G. Heinemann, Germany), and the protein was solubilized in HBS with 2% DDM/0.04% CHS from crude membranes and purified by affinity beads (Strep-TactinXT 4Flow, IBA Lifesciences GmbH, Göttingen, Germany).

To create chimeras, the desired parts were amplified by PCR and cloned into the EGFP-C1 vector. The sequences of the chimeras were designed as follows: RgCh1 (RgRGC1: aa 1-364; CaRGC: aa 443–626); RgCh2 (RgRGC1: aa 1-318; CaRGC: aa 397-626), RgCh3 (CaRGC: aa 1-145; RgRGC1: aa 69-318; CaRGC: aa 442-626), ClCh2 (ClRGC1: aa 1-321; CaRGC: aa 397-626), CpCh2 (CpRGC: aa 1-322; CaRGC: aa 397-626), RgNeoRCh2 (RgNeoR: aa 1-278; CaRGC: aa 397-626).

His-tagged GC domains (RgRGC1: aa 365-551; RgNeoR: aa 325-509) were expressed, as described previously (3), with slight modifications. Briefly, proteins were expressed in E. coli ArcticExpress (DE3, Agilent) cells overnight at 18 °C in Luria-Bertani broth with 100 µg/mL ampicillin. After binding to a 5-mL HisTrap Nickel-column (GE Healthcare), the cell extract was then washed with sodium phosphate buffer (20 mM sodium phosphate, 0.5 M NaCl, and 20 mM imidazole, pH 7.4) and subsequently eluted with 500 mM imidazole.

To measure cyclase activity in HEK293-T cells, the cells were harvested 2 d after transfection. The cell pellet was then resuspended in HBS, containing 3 mM MgCl₂, and homogenized with 20 strokes of a Douncer. The crude membrane was washed by centrifuging (5 min, 21,000 × g, 4 °C) and then resuspended in MgCl₂-containing HBS. For each assay, a reaction was induced by adding 1 mM GTP to triplicates of 1.5 mg total protein in 350 µL total volume, either illuminated (CaRGC, RgRGC1: 530 nm, 0.11 mW/mm²; ClRGC1: 590 nm, 0.12 mW/mm²; CpRGC: 615 nm, 0.18 mW/mm²; RgNeoR: 680 nm, and 0.21 mW/mm²) or in the dark. At various time points, the reaction was stopped by transferring a 100 µL suspension to 100 µL 0.1 N HCl. The denatured protein was spun down (5 min; 21,000 × g; 4 °C), and for cGMP-detection, the supernatant was applied on a C18-column (Supelco, Sigma-Aldrich) in a reversed-phase HPLC equilibrated with 100 mM potassium phosphate (pH 5.9), 4 mM tetrabutylammonium iodide, and 10% (v/v) methanol, as previously described (3). To assess the activity of RGCs purified from Pichia pastoris, a total of 0.1 nmol protein was incubated in 100 µL HBS containing 2 mM MnCl₂ and 1 mM GTP for a duration of 5 min, either illuminated with 530 nm or kept in the dark. The reaction was stopped by adding 100 µL 0.1 N HCl.

Spectra of purified RGC in HBS with 0.02% DDM/0.004% were taken with a Shimadzu UV-2000 photospectrometer with UVProbe v2.34 (Shimadzu Corporation, Kyoto, Japan). Fluorescence spectra were recorded on a Horiba FluoroMax 4 spectrometer with FluorEssence™ 2.5.2 (HORIBA Instruments Inc., NJ, USA), and fluorescence quantum yields (FQY) were determined by the slope after plotting the integrated fluorescence intensity against the absorbance at the excitation wavelength (700 nm: CpNeoR; 650 nm: CcNeoR1, ClNeoR1; 600 nm: CsNeoR2, GpNeoRh) in comparison to fluorescence references [700-nm excitation: DY-706, FQY 32.8% (Dyomics GmbH, Jena, Germany); 650-nm excitation: iRFP, FQY 5.9% and Cy 5.5, FQY 20% (Molecular Probes, Eugene, OR, USA); 600-nm excitation: DID, FQY 20% (Thermo Fisher Scientific)].

Microsecond-to-second transient absorption spectra were obtained with the LKS.60 flash photolysis system (Applied Photophysics Ltd, Leatherhead, UK) equipped with a tunable optical parametric oscillator (MagicPrism, Opotek Inc., Carlsbad, CA, USA) pumped by the third harmonic of an Nd:YAG laser (Brilliant B, Quantel, Les Ulis, France). RgRGC1 and RgCh2 purified from P. pastoris were excited with a 10-ns laser pulse adjusted to 530 nm and probed by the light of a short-arc XBO Xenon lamp (150 W, Osram, München, Germany). The transient absorbance changes (10 ns to 1 s) were detected by an Andor iStar ICCD Camera (DH734, Andor Technology, Belfast, Northern Ireland), and the data were averaged over eight cycles. Flash photolysis on GpNeoRh was performed using 600-nm laser flash to excite the protein expressed and purified from HEK-T cells.

Two-electrode voltage clamp (TEVC) experiments were performed at −40 mV using a TURBO TEC-03X amplifier (NPI Electronic GmbH, Tamm, Germany), pCLAMP v. 9.0 software (Molecular Devices, San Jose, CA, USA), an XBO 75 W Xenon lamp (Osram, Munich, Germany), and a UNIBLITZ LS3 shutter (Vincent Associates, Rochester, NY, USA) as reported earlier (6). cRNAs coding for various RGCs were synthesized using the mMACHINE TM T7 Transcription Kit (Invitrogene) from linearized DNA. Then, 5 to 10 ng of cRNA were injected into Xenopus laevis oocytes, together with 5 ng cRNA coding for the cGMP-sensitive cyclic nucleotide-gated (CNG)A2 ion channel from rat olfactory neurons, and oocytes were incubated at 18 °C in Ringer’s solution supplemented with 1 µM all-trans-retinal for 3 to 5 d before the measurement. The protocols for animal maintenance and oocyte harvesting were approved by the Federation of European Laboratory Animal Science Associations (Berlin, Germany).

The DNA sequences of GpRGC2 and GhRGC2 were obtained by gene synthesis (GenScript, Piscataway, NJ). For electrophysiological characterization, the following construct configurations were used: YFP-GhRGC2-T2A-SthK, YFP-GpRGC2-T2A-SthK, and YFP-CaRGC-T2A-SthK, comprising the cyclase as well as a cGMP-gated potassium channel SthK, separated by a T2A side.

ND7/23 cells (Sigma-Aldrich, Munich, Germany) were cultured in DMEM (Biochrom, Berlin, Germany), supplemented with 5% fetal bovine serum (FBS Superior; Biochrom, Berlin, Germany), 100 µg/ml penicillin/streptomycin (Biochrom, Berlin, Germany), and 1 μM all-trans-retinal. Cells were plated on poly-D-lysine-coated coverslips and 24 h later transfected using Fugene® HD reagent (Promega, Madison, WI). Cells were kept at 37 °C and 5% CO2.

Whole-cell patch-clamp recordings were performed 1 to 2 d post-transfection. Patch pipettes with a resistance between 1.5 and 2.5 MΩ were prepared from borosilicate capillaries (GB150P-9P; Science Products GmbH, Hofheim am Taunus, Germany) using the P1000 Micropipette Puller (Sutter Instrument Co., Novato, CA, USA) and fire polished. An agar bridge containing 140 mM NaCl connected the reference electrode to the bath. All recordings were taken at room temperature in the whole cell configuration. Cells were clamped at −20 mV. The membrane resistance for each recording was ≥1 GΩ, and the access resistance was <10 MΩ. The extracellular solution contained: 10 HEPES, 140 NaCl, 2.4 KCl, 2 CaCl2, 4 MgCl₂, and 10 glucose, pH 7.4. The intracellular solution contained: 17.8 HEPES, 135 KGluc, 4.6 MgCl₂, 4 MgATP, 0.3 NaGTP, 1 EGTA, 12 Na₂-phosphocreatine, and 50 phosphocreatine kinase, pH 7.3 (all amounts in mM).

For the light titration experiments, signals were acquired and filtered using an Axon MultiClamp 700B Amplifier and the Axon Digidata 1550 (Molecular Devices, Sunnyvale, CA). Measurements were taken on an Olympus IX73 Inverted Microscope. Data were recorded using Clampex 10.4. For cyclase activation, a 10-ms light pulse at 580 nm was applied by the CoolLED pE-4000 (CoolLED, Andover, UK). Photon count was gradually increased (0.3, 0.7, 1.9, 3.9, 7.3, 15.7, 22.1, and 25.8 mW mm⁻²) to obtain light intensity-dependent activation curves.

Action spectra were taken on an Axiovert 100 microscope (Carl Zeiss, Jena, Germany). Signals were acquired and filtered with an AxoPatch 200B amplifier and a DigiData 1440 A digitizer (Molecular Devices, Sunnyvale, CA). A Polychrome V monochromator (TILL Photonics, Planegg, Germany) provided monochromatic light with a half bandwidth of 7 nm and was controlled via a shutter system (VS25 and VCM-D1, Vincent Associates). A motorized neutral density filter wheel (Newport, Irvine, CA) in the light path provided equal photon count for different wavelengths. Light-induced currents were recorded from 410 nm to 670 nm in 20-nm steps and normalized. Action spectra were fitted using a skew normal distribution.

The initial structure of heterodimeric RgRGC1/RgNeoR was modeled using AlphaFold-Multimer (28) (SI Appendix, Fig. S8A), and the all-trans retinal was incorporated from the crystal structure of Salpingoeca rosetta rhodopsin phosphodiesterase (PDB 7CJ3). Based on this initial model, two mutants of RgNeoR (D140S and D140T) were generated using MODELLER 10.1 (29). Besides model 1 (E262 protonated), a second model (model 2) with deprotonated E262 was generated according to previous mutational, and quantum mechanics/molecular mechanics (QM/MM) studies of RgNeoR (6, 17). Additionally, E136 and if present D140 were protonated in all models. All other titratable residues were configured in their standard protonation state at physiological pH. For the subsequent simulations, only the truncated RgNeoR domain (aa 1-279) (SI Appendix, Fig. S8B) was used to save computational time. All models were embedded into 1-palmitoyl-2-oleoyl- sn-glycero-3-phosphocholine (POPC) lipid bilayers, solvated with TIP3P water, and neutralized by adding 150 mM NaCl NaCl using the CHARMM-GUI (30). Utilizing the retinal force field parameters from previous study, two models were generated with different protonation states of E262 (model 1 with E262 protonated; model 2 deprotonated) according to previous mutational, and quantum mechanics/molecular mechanics (QM/MM) studies of RgNeoR. All other titratable residues were modelled in their standard protonation state at physiological pH.

The simulations were performed with GROMACS 2019.3 and 2021.3 (31) using the CHARMM36m force field (32) extended by the retinal parameters (33). After minimizing the energy and thermally equilibrating the models with gradually releasing constraints (SI Appendix, Table S3), 200-ns long production runs were performed, at constant temperature (303 K) and pressure (1 bar) (34, 35). A short-range cutoff of 1.2 nm was applied for the Coulomb and Lennard-Jones potentials while long-range electrostatic interactions were treated by the particle mesh Ewald method (36). Each of the models was simulated five times using different initial velocities.

Hydrogen bond analysis of retinal, water, and the counterion-triad (E136, D140, and E262) was performed using the hydrogen bond analysis module in MDAnalysis 2.4.3 (37) with a distance cutoff of 3 Å for donor and acceptor and an angle cutoff of 150° for donor–hydrogen–acceptor. Molecular visualization and data plotting were done using PyMOL 2.4.1 and Matplotlib 3.4.3, respectively.

M.B., H.S., and P.H. designed research; M.B., W.B., A.S., M.R., Y.A.B.S., and S.H. performed research; M.B., W.B., A.S., Y.A.B.S., S.H., T.U., H.S., and P.H. analyzed data; and M.B. and P.H. wrote the paper.

The authors declare no competing interest.